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) & 0x00ffffff;
 517 }
 518
 519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 520 {
 521         entry->block = cpu_to_le32(value);
 522 }
 523
 524 static inline unsigned dx_get_hash(struct dx_entry *entry)
 525 {
 526         return le32_to_cpu(entry->hash);
 527 }
 528
 529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 530 {
 531         entry->hash = cpu_to_le32(value);
 532 }
 533
 534 static inline unsigned dx_get_count(struct dx_entry *entries)
 535 {
 536         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 537 }
 538
 539 static inline unsigned dx_get_limit(struct dx_entry *entries)
 540 {
 541         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 542 }
 543
 544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 545 {
 546         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 547 }
 548
 549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 550 {
 551         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 552 }
 553
 554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 555 {
 556         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 557                 EXT4_DIR_REC_LEN(2) - infosize;
 558
 559         if (ext4_has_metadata_csum(dir->i_sb))
 560                 entry_space -= sizeof(struct dx_tail);
 561         return entry_space / sizeof(struct dx_entry);
 562 }
 563
 564 static inline unsigned dx_node_limit(struct inode *dir)
 565 {
 566         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 567
 568         if (ext4_has_metadata_csum(dir->i_sb))
 569                 entry_space -= sizeof(struct dx_tail);
 570         return entry_space / sizeof(struct dx_entry);
 571 }
 572
 573 /*
 574  * Debug
 575  */
 576 #ifdef DX_DEBUG
 577 static void dx_show_index(char * label, struct dx_entry *entries)
 578 {
 579         int i, n = dx_get_count (entries);
 580         printk(KERN_DEBUG "%s index", label);
 581         for (i = 0; i < n; i++) {
 582                 printk(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         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 743         if (IS_ERR(frame->bh))
 744                 return (struct dx_frame *) frame->bh;
 745
 746         root = (struct dx_root *) frame->bh->b_data;
 747         if (root->info.hash_version != DX_HASH_TEA &&
 748             root->info.hash_version != DX_HASH_HALF_MD4 &&
 749             root->info.hash_version != DX_HASH_LEGACY) {
 750                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 751                                    root->info.hash_version);
 752                 goto fail;
 753         }
 754         if (fname)
 755                 hinfo = &fname->hinfo;
 756         hinfo->hash_version = root->info.hash_version;
 757         if (hinfo->hash_version <= DX_HASH_TEA)
 758                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 759         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 760         if (fname && fname_name(fname))
 761                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 762         hash = hinfo->hash;
 763
 764         if (root->info.unused_flags & 1) {
 765                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 766                                    root->info.unused_flags);
 767                 goto fail;
 768         }
 769
 770         indirect = root->info.indirect_levels;
 771         if (indirect > 1) {
 772                 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
 773                                    root->info.indirect_levels);
 774                 goto fail;
 775         }
 776
 777         entries = (struct dx_entry *)(((char *)&root->info) +
 778                                       root->info.info_length);
 779
 780         if (dx_get_limit(entries) != dx_root_limit(dir,
 781                                                    root->info.info_length)) {
 782                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 783                                    dx_get_limit(entries),
 784                                    dx_root_limit(dir, root->info.info_length));
 785                 goto fail;
 786         }
 787
 788         dxtrace(printk("Look up %x", hash));
 789         while (1) {
 790                 count = dx_get_count(entries);
 791                 if (!count || count > dx_get_limit(entries)) {
 792                         ext4_warning_inode(dir,
 793                                            "dx entry: count %u beyond limit %u",
 794                                            count, dx_get_limit(entries));
 795                         goto fail;
 796                 }
 797
 798                 p = entries + 1;
 799                 q = entries + count - 1;
 800                 while (p <= q) {
 801                         m = p + (q - p) / 2;
 802                         dxtrace(printk(KERN_CONT "."));
 803                         if (dx_get_hash(m) > hash)
 804                                 q = m - 1;
 805                         else
 806                                 p = m + 1;
 807                 }
 808
 809                 if (0) { // linear search cross check
 810                         unsigned n = count - 1;
 811                         at = entries;
 812                         while (n--)
 813                         {
 814                                 dxtrace(printk(KERN_CONT ","));
 815                                 if (dx_get_hash(++at) > hash)
 816                                 {
 817                                         at--;
 818                                         break;
 819                                 }
 820                         }
 821                         assert (at == p - 1);
 822                 }
 823
 824                 at = p - 1;
 825                 dxtrace(printk(KERN_CONT " %x->%u\n",
 826                                at == entries ? 0 : dx_get_hash(at),
 827                                dx_get_block(at)));
 828                 frame->entries = entries;
 829                 frame->at = at;
 830                 if (!indirect--)
 831                         return frame;
 832                 frame++;
 833                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 834                 if (IS_ERR(frame->bh)) {
 835                         ret_err = (struct dx_frame *) frame->bh;
 836                         frame->bh = NULL;
 837                         goto fail;
 838                 }
 839                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 840
 841                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 842                         ext4_warning_inode(dir,
 843                                 "dx entry: limit %u != node limit %u",
 844                                 dx_get_limit(entries), dx_node_limit(dir));
 845                         goto fail;
 846                 }
 847         }
 848 fail:
 849         while (frame >= frame_in) {
 850                 brelse(frame->bh);
 851                 frame--;
 852         }
 853
 854         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 855                 ext4_warning_inode(dir,
 856                         "Corrupt directory, running e2fsck is recommended");
 857         return ret_err;
 858 }
 859
 860 static void dx_release(struct dx_frame *frames)
 861 {
 862         if (frames[0].bh == NULL)
 863                 return;
 864
 865         if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
 866                 brelse(frames[1].bh);
 867         brelse(frames[0].bh);
 868 }
 869
 870 /*
 871  * This function increments the frame pointer to search the next leaf
 872  * block, and reads in the necessary intervening nodes if the search
 873  * should be necessary.  Whether or not the search is necessary is
 874  * controlled by the hash parameter.  If the hash value is even, then
 875  * the search is only continued if the next block starts with that
 876  * hash value.  This is used if we are searching for a specific file.
 877  *
 878  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 879  *
 880  * This function returns 1 if the caller should continue to search,
 881  * or 0 if it should not.  If there is an error reading one of the
 882  * index blocks, it will a negative error code.
 883  *
 884  * If start_hash is non-null, it will be filled in with the starting
 885  * hash of the next page.
 886  */
 887 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 888                                  struct dx_frame *frame,
 889                                  struct dx_frame *frames,
 890                                  __u32 *start_hash)
 891 {
 892         struct dx_frame *p;
 893         struct buffer_head *bh;
 894         int num_frames = 0;
 895         __u32 bhash;
 896
 897         p = frame;
 898         /*
 899          * Find the next leaf page by incrementing the frame pointer.
 900          * If we run out of entries in the interior node, loop around and
 901          * increment pointer in the parent node.  When we break out of
 902          * this loop, num_frames indicates the number of interior
 903          * nodes need to be read.
 904          */
 905         while (1) {
 906                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 907                         break;
 908                 if (p == frames)
 909                         return 0;
 910                 num_frames++;
 911                 p--;
 912         }
 913
 914         /*
 915          * If the hash is 1, then continue only if the next page has a
 916          * continuation hash of any value.  This is used for readdir
 917          * handling.  Otherwise, check to see if the hash matches the
 918          * desired contiuation hash.  If it doesn't, return since
 919          * there's no point to read in the successive index pages.
 920          */
 921         bhash = dx_get_hash(p->at);
 922         if (start_hash)
 923                 *start_hash = bhash;
 924         if ((hash & 1) == 0) {
 925                 if ((bhash & ~1) != hash)
 926                         return 0;
 927         }
 928         /*
 929          * If the hash is HASH_NB_ALWAYS, we always go to the next
 930          * block so no check is necessary
 931          */
 932         while (num_frames--) {
 933                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 934                 if (IS_ERR(bh))
 935                         return PTR_ERR(bh);
 936                 p++;
 937                 brelse(p->bh);
 938                 p->bh = bh;
 939                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 940         }
 941         return 1;
 942 }
 943
 944
 945 /*
 946  * This function fills a red-black tree with information from a
 947  * directory block.  It returns the number directory entries loaded
 948  * into the tree.  If there is an error it is returned in err.
 949  */
 950 static int htree_dirblock_to_tree(struct file *dir_file,
 951                                   struct inode *dir, ext4_lblk_t block,
 952                                   struct dx_hash_info *hinfo,
 953                                   __u32 start_hash, __u32 start_minor_hash)
 954 {
 955         struct buffer_head *bh;
 956         struct ext4_dir_entry_2 *de, *top;
 957         int err = 0, count = 0;
 958         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 959
 960         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 961                                                         (unsigned long)block));
 962         bh = ext4_read_dirblock(dir, block, DIRENT);
 963         if (IS_ERR(bh))
 964                 return PTR_ERR(bh);
 965
 966         de = (struct ext4_dir_entry_2 *) bh->b_data;
 967         top = (struct ext4_dir_entry_2 *) ((char *) de +
 968                                            dir->i_sb->s_blocksize -
 969                                            EXT4_DIR_REC_LEN(0));
 970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 971         /* Check if the directory is encrypted */
 972         if (ext4_encrypted_inode(dir)) {
 973                 err = fscrypt_get_encryption_info(dir);
 974                 if (err < 0) {
 975                         brelse(bh);
 976                         return err;
 977                 }
 978                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
 979                                                      &fname_crypto_str);
 980                 if (err < 0) {
 981                         brelse(bh);
 982                         return err;
 983                 }
 984         }
 985 #endif
 986         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 987                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 988                                 bh->b_data, bh->b_size,
 989                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 990                                          + ((char *)de - bh->b_data))) {
 991                         /* silently ignore the rest of the block */
 992                         break;
 993                 }
 994                 ext4fs_dirhash(de->name, de->name_len, hinfo);
 995                 if ((hinfo->hash < start_hash) ||
 996                     ((hinfo->hash == start_hash) &&
 997                      (hinfo->minor_hash < start_minor_hash)))
 998                         continue;
 999                 if (de->inode == 0)
1000                         continue;
1001                 if (!ext4_encrypted_inode(dir)) {
1002                         tmp_str.name = de->name;
1003                         tmp_str.len = de->name_len;
1004                         err = ext4_htree_store_dirent(dir_file,
1005                                    hinfo->hash, hinfo->minor_hash, de,
1006                                    &tmp_str);
1007                 } else {
1008                         int save_len = fname_crypto_str.len;
1009                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1010                                                                 de->name_len);
1011
1012                         /* Directory is encrypted */
1013                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014                                         hinfo->minor_hash, &de_name,
1015                                         &fname_crypto_str);
1016                         if (err) {
1017                                 count = err;
1018                                 goto errout;
1019                         }
1020                         err = ext4_htree_store_dirent(dir_file,
1021                                    hinfo->hash, hinfo->minor_hash, de,
1022                                         &fname_crypto_str);
1023                         fname_crypto_str.len = save_len;
1024                 }
1025                 if (err != 0) {
1026                         count = err;
1027                         goto errout;
1028                 }
1029                 count++;
1030         }
1031 errout:
1032         brelse(bh);
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034         fscrypt_fname_free_buffer(&fname_crypto_str);
1035 #endif
1036         return count;
1037 }
1038
1039
1040 /*
1041  * This function fills a red-black tree with information from a
1042  * directory.  We start scanning the directory in hash order, starting
1043  * at start_hash and start_minor_hash.
1044  *
1045  * This function returns the number of entries inserted into the tree,
1046  * or a negative error code.
1047  */
1048 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049                          __u32 start_minor_hash, __u32 *next_hash)
1050 {
1051         struct dx_hash_info hinfo;
1052         struct ext4_dir_entry_2 *de;
1053         struct dx_frame frames[2], *frame;
1054         struct inode *dir;
1055         ext4_lblk_t block;
1056         int count = 0;
1057         int ret, err;
1058         __u32 hashval;
1059         struct fscrypt_str tmp_str;
1060
1061         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062                        start_hash, start_minor_hash));
1063         dir = file_inode(dir_file);
1064         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066                 if (hinfo.hash_version <= DX_HASH_TEA)
1067                         hinfo.hash_version +=
1068                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070                 if (ext4_has_inline_data(dir)) {
1071                         int has_inline_data = 1;
1072                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1073                                                         &hinfo, start_hash,
1074                                                         start_minor_hash,
1075                                                         &has_inline_data);
1076                         if (has_inline_data) {
1077                                 *next_hash = ~0;
1078                                 return count;
1079                         }
1080                 }
1081                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082                                                start_hash, start_minor_hash);
1083                 *next_hash = ~0;
1084                 return count;
1085         }
1086         hinfo.hash = start_hash;
1087         hinfo.minor_hash = 0;
1088         frame = dx_probe(NULL, dir, &hinfo, frames);
1089         if (IS_ERR(frame))
1090                 return PTR_ERR(frame);
1091
1092         /* Add '.' and '..' from the htree header */
1093         if (!start_hash && !start_minor_hash) {
1094                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095                 tmp_str.name = de->name;
1096                 tmp_str.len = de->name_len;
1097                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1098                                               de, &tmp_str);
1099                 if (err != 0)
1100                         goto errout;
1101                 count++;
1102         }
1103         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106                 tmp_str.name = de->name;
1107                 tmp_str.len = de->name_len;
1108                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109                                               de, &tmp_str);
1110                 if (err != 0)
1111                         goto errout;
1112                 count++;
1113         }
1114
1115         while (1) {
1116                 if (fatal_signal_pending(current)) {
1117                         err = -ERESTARTSYS;
1118                         goto errout;
1119                 }
1120                 cond_resched();
1121                 block = dx_get_block(frame->at);
1122                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123                                              start_hash, start_minor_hash);
1124                 if (ret < 0) {
1125                         err = ret;
1126                         goto errout;
1127                 }
1128                 count += ret;
1129                 hashval = ~0;
1130                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131                                             frame, frames, &hashval);
1132                 *next_hash = hashval;
1133                 if (ret < 0) {
1134                         err = ret;
1135                         goto errout;
1136                 }
1137                 /*
1138                  * Stop if:  (a) there are no more entries, or
1139                  * (b) we have inserted at least one entry and the
1140                  * next hash value is not a continuation
1141                  */
1142                 if ((ret == 0) ||
1143                     (count && ((hashval & 1) == 0)))
1144                         break;
1145         }
1146         dx_release(frames);
1147         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148                        "next hash: %x\n", count, *next_hash));
1149         return count;
1150 errout:
1151         dx_release(frames);
1152         return (err);
1153 }
1154
1155 static inline int search_dirblock(struct buffer_head *bh,
1156                                   struct inode *dir,
1157                                   struct ext4_filename *fname,
1158                                   const struct qstr *d_name,
1159                                   unsigned int offset,
1160                                   struct ext4_dir_entry_2 **res_dir)
1161 {
1162         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163                                fname, d_name, offset, res_dir);
1164 }
1165
1166 /*
1167  * Directory block splitting, compacting
1168  */
1169
1170 /*
1171  * Create map of hash values, offsets, and sizes, stored at end of block.
1172  * Returns number of entries mapped.
1173  */
1174 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175                        unsigned blocksize, struct dx_hash_info *hinfo,
1176                        struct dx_map_entry *map_tail)
1177 {
1178         int count = 0;
1179         char *base = (char *) de;
1180         struct dx_hash_info h = *hinfo;
1181
1182         while ((char *) de < base + blocksize) {
1183                 if (de->name_len && de->inode) {
1184                         ext4fs_dirhash(de->name, de->name_len, &h);
1185                         map_tail--;
1186                         map_tail->hash = h.hash;
1187                         map_tail->offs = ((char *) de - base)>>2;
1188                         map_tail->size = le16_to_cpu(de->rec_len);
1189                         count++;
1190                         cond_resched();
1191                 }
1192                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193                 de = ext4_next_entry(de, blocksize);
1194         }
1195         return count;
1196 }
1197
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200 {
1201         struct dx_map_entry *p, *q, *top = map + count - 1;
1202         int more;
1203         /* Combsort until bubble sort doesn't suck */
1204         while (count > 2) {
1205                 count = count*10/13;
1206                 if (count - 9 < 2) /* 9, 10 -> 11 */
1207                         count = 11;
1208                 for (p = top, q = p - count; q >= map; p--, q--)
1209                         if (p->hash < q->hash)
1210                                 swap(*p, *q);
1211         }
1212         /* Garden variety bubble sort */
1213         do {
1214                 more = 0;
1215                 q = top;
1216                 while (q-- > map) {
1217                         if (q[1].hash >= q[0].hash)
1218                                 continue;
1219                         swap(*(q+1), *q);
1220                         more = 1;
1221                 }
1222         } while(more);
1223 }
1224
1225 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226 {
1227         struct dx_entry *entries = frame->entries;
1228         struct dx_entry *old = frame->at, *new = old + 1;
1229         int count = dx_get_count(entries);
1230
1231         assert(count < dx_get_limit(entries));
1232         assert(old < entries + count);
1233         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234         dx_set_hash(new, hash);
1235         dx_set_block(new, block);
1236         dx_set_count(entries, count + 1);
1237 }
1238
1239 /*
1240  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1241  *
1242  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243  * `de != NULL' is guaranteed by caller.
1244  */
1245 static inline int ext4_match(struct ext4_filename *fname,
1246                              struct ext4_dir_entry_2 *de)
1247 {
1248         const void *name = fname_name(fname);
1249         u32 len = fname_len(fname);
1250
1251         if (!de->inode)
1252                 return 0;
1253
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255         if (unlikely(!name)) {
1256                 if (fname->usr_fname->name[0] == '_') {
1257                         int ret;
1258                         if (de->name_len < 16)
1259                                 return 0;
1260                         ret = memcmp(de->name + de->name_len - 16,
1261                                      fname->crypto_buf.name + 8, 16);
1262                         return (ret == 0) ? 1 : 0;
1263                 }
1264                 name = fname->crypto_buf.name;
1265                 len = fname->crypto_buf.len;
1266         }
1267 #endif
1268         if (de->name_len != len)
1269                 return 0;
1270         return (memcmp(de->name, name, len) == 0) ? 1 : 0;
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                     const struct qstr *d_name,
1279                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281         struct ext4_dir_entry_2 * de;
1282         char * dlimit;
1283         int de_len;
1284         int res;
1285
1286         de = (struct ext4_dir_entry_2 *)search_buf;
1287         dlimit = search_buf + buf_size;
1288         while ((char *) de < dlimit) {
1289                 /* this code is executed quadratically often */
1290                 /* do minimal checking `by hand' */
1291                 if ((char *) de + de->name_len <= dlimit) {
1292                         res = ext4_match(fname, de);
1293                         if (res < 0) {
1294                                 res = -1;
1295                                 goto return_result;
1296                         }
1297                         if (res > 0) {
1298                                 /* found a match - just to be sure, do
1299                                  * a full check */
1300                                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1301                                                 bh->b_data,
1302                                                  bh->b_size, offset)) {
1303                                         res = -1;
1304                                         goto return_result;
1305                                 }
1306                                 *res_dir = de;
1307                                 res = 1;
1308                                 goto return_result;
1309                         }
1310
1311                 }
1312                 /* prevent looping on a bad block */
1313                 de_len = ext4_rec_len_from_disk(de->rec_len,
1314                                                 dir->i_sb->s_blocksize);
1315                 if (de_len <= 0) {
1316                         res = -1;
1317                         goto return_result;
1318                 }
1319                 offset += de_len;
1320                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1321         }
1322
1323         res = 0;
1324 return_result:
1325         return res;
1326 }
1327
1328 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329                                struct ext4_dir_entry *de)
1330 {
1331         struct super_block *sb = dir->i_sb;
1332
1333         if (!is_dx(dir))
1334                 return 0;
1335         if (block == 0)
1336                 return 1;
1337         if (de->inode == 0 &&
1338             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339                         sb->s_blocksize)
1340                 return 1;
1341         return 0;
1342 }
1343
1344 /*
1345  *      ext4_find_entry()
1346  *
1347  * finds an entry in the specified directory with the wanted name. It
1348  * returns the cache buffer in which the entry was found, and the entry
1349  * itself (as a parameter - res_dir). It does NOT read the inode of the
1350  * entry - you'll have to do that yourself if you want to.
1351  *
1352  * The returned buffer_head has ->b_count elevated.  The caller is expected
1353  * to brelse() it when appropriate.
1354  */
1355 static struct buffer_head * ext4_find_entry (struct inode *dir,
1356                                         const struct qstr *d_name,
1357                                         struct ext4_dir_entry_2 **res_dir,
1358                                         int *inlined)
1359 {
1360         struct super_block *sb;
1361         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362         struct buffer_head *bh, *ret = NULL;
1363         ext4_lblk_t start, block, b;
1364         const u8 *name = d_name->name;
1365         int ra_max = 0;         /* Number of bh's in the readahead
1366                                    buffer, bh_use[] */
1367         int ra_ptr = 0;         /* Current index into readahead
1368                                    buffer */
1369         int num = 0;
1370         ext4_lblk_t  nblocks;
1371         int i, namelen, retval;
1372         struct ext4_filename fname;
1373
1374         *res_dir = NULL;
1375         sb = dir->i_sb;
1376         namelen = d_name->len;
1377         if (namelen > EXT4_NAME_LEN)
1378                 return NULL;
1379
1380         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1381         if (retval == -ENOENT)
1382                 return NULL;
1383         if (retval)
1384                 return ERR_PTR(retval);
1385
1386         if (ext4_has_inline_data(dir)) {
1387                 int has_inline_data = 1;
1388                 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1389                                              &has_inline_data);
1390                 if (has_inline_data) {
1391                         if (inlined)
1392                                 *inlined = 1;
1393                         goto cleanup_and_exit;
1394                 }
1395         }
1396
1397         if ((namelen <= 2) && (name[0] == '.') &&
1398             (name[1] == '.' || name[1] == '\0')) {
1399                 /*
1400                  * "." or ".." will only be in the first block
1401                  * NFS may look up ".."; "." should be handled by the VFS
1402                  */
1403                 block = start = 0;
1404                 nblocks = 1;
1405                 goto restart;
1406         }
1407         if (is_dx(dir)) {
1408                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1409                 /*
1410                  * On success, or if the error was file not found,
1411                  * return.  Otherwise, fall back to doing a search the
1412                  * old fashioned way.
1413                  */
1414                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1415                         goto cleanup_and_exit;
1416                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1417                                "falling back\n"));
1418         }
1419         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1420         start = EXT4_I(dir)->i_dir_start_lookup;
1421         if (start >= nblocks)
1422                 start = 0;
1423         block = start;
1424 restart:
1425         do {
1426                 /*
1427                  * We deal with the read-ahead logic here.
1428                  */
1429                 if (ra_ptr >= ra_max) {
1430                         /* Refill the readahead buffer */
1431                         ra_ptr = 0;
1432                         b = block;
1433                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1434                                 /*
1435                                  * Terminate if we reach the end of the
1436                                  * directory and must wrap, or if our
1437                                  * search has finished at this block.
1438                                  */
1439                                 if (b >= nblocks || (num && block == start)) {
1440                                         bh_use[ra_max] = NULL;
1441                                         break;
1442                                 }
1443                                 num++;
1444                                 bh = ext4_getblk(NULL, dir, b++, 0);
1445                                 if (IS_ERR(bh)) {
1446                                         if (ra_max == 0) {
1447                                                 ret = bh;
1448                                                 goto cleanup_and_exit;
1449                                         }
1450                                         break;
1451                                 }
1452                                 bh_use[ra_max] = bh;
1453                                 if (bh)
1454                                         ll_rw_block(REQ_OP_READ,
1455                                                     REQ_META | REQ_PRIO,
1456                                                     1, &bh);
1457                         }
1458                 }
1459                 if ((bh = bh_use[ra_ptr++]) == NULL)
1460                         goto next;
1461                 wait_on_buffer(bh);
1462                 if (!buffer_uptodate(bh)) {
1463                         /* read error, skip block & hope for the best */
1464                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1465                                          (unsigned long) block);
1466                         brelse(bh);
1467                         goto next;
1468                 }
1469                 if (!buffer_verified(bh) &&
1470                     !is_dx_internal_node(dir, block,
1471                                          (struct ext4_dir_entry *)bh->b_data) &&
1472                     !ext4_dirent_csum_verify(dir,
1473                                 (struct ext4_dir_entry *)bh->b_data)) {
1474                         EXT4_ERROR_INODE(dir, "checksumming directory "
1475                                          "block %lu", (unsigned long)block);
1476                         brelse(bh);
1477                         goto next;
1478                 }
1479                 set_buffer_verified(bh);
1480                 i = search_dirblock(bh, dir, &fname, d_name,
1481                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1482                 if (i == 1) {
1483                         EXT4_I(dir)->i_dir_start_lookup = block;
1484                         ret = bh;
1485                         goto cleanup_and_exit;
1486                 } else {
1487                         brelse(bh);
1488                         if (i < 0)
1489                                 goto cleanup_and_exit;
1490                 }
1491         next:
1492                 if (++block >= nblocks)
1493                         block = 0;
1494         } while (block != start);
1495
1496         /*
1497          * If the directory has grown while we were searching, then
1498          * search the last part of the directory before giving up.
1499          */
1500         block = nblocks;
1501         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1502         if (block < nblocks) {
1503                 start = 0;
1504                 goto restart;
1505         }
1506
1507 cleanup_and_exit:
1508         /* Clean up the read-ahead blocks */
1509         for (; ra_ptr < ra_max; ra_ptr++)
1510                 brelse(bh_use[ra_ptr]);
1511         ext4_fname_free_filename(&fname);
1512         return ret;
1513 }
1514
1515 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1516                         struct ext4_filename *fname,
1517                         struct ext4_dir_entry_2 **res_dir)
1518 {
1519         struct super_block * sb = dir->i_sb;
1520         struct dx_frame frames[2], *frame;
1521         const struct qstr *d_name = fname->usr_fname;
1522         struct buffer_head *bh;
1523         ext4_lblk_t block;
1524         int retval;
1525
1526 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1527         *res_dir = NULL;
1528 #endif
1529         frame = dx_probe(fname, dir, NULL, frames);
1530         if (IS_ERR(frame))
1531                 return (struct buffer_head *) frame;
1532         do {
1533                 block = dx_get_block(frame->at);
1534                 bh = ext4_read_dirblock(dir, block, DIRENT);
1535                 if (IS_ERR(bh))
1536                         goto errout;
1537
1538                 retval = search_dirblock(bh, dir, fname, d_name,
1539                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1540                                          res_dir);
1541                 if (retval == 1)
1542                         goto success;
1543                 brelse(bh);
1544                 if (retval == -1) {
1545                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1546                         goto errout;
1547                 }
1548
1549                 /* Check to see if we should continue to search */
1550                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1551                                                frames, NULL);
1552                 if (retval < 0) {
1553                         ext4_warning_inode(dir,
1554                                 "error %d reading directory index block",
1555                                 retval);
1556                         bh = ERR_PTR(retval);
1557                         goto errout;
1558                 }
1559         } while (retval == 1);
1560
1561         bh = NULL;
1562 errout:
1563         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1564 success:
1565         dx_release(frames);
1566         return bh;
1567 }
1568
1569 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1570 {
1571         struct inode *inode;
1572         struct ext4_dir_entry_2 *de;
1573         struct buffer_head *bh;
1574
1575         if (ext4_encrypted_inode(dir)) {
1576                 int res = fscrypt_get_encryption_info(dir);
1577
1578                 /*
1579                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1580                  * created while the directory was encrypted and we
1581                  * have access to the key.
1582                  */
1583                 if (fscrypt_has_encryption_key(dir))
1584                         fscrypt_set_encrypted_dentry(dentry);
1585                 fscrypt_set_d_op(dentry);
1586                 if (res && res != -ENOKEY)
1587                         return ERR_PTR(res);
1588         }
1589
1590        if (dentry->d_name.len > EXT4_NAME_LEN)
1591                return ERR_PTR(-ENAMETOOLONG);
1592
1593         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1594         if (IS_ERR(bh))
1595                 return (struct dentry *) bh;
1596         inode = NULL;
1597         if (bh) {
1598                 __u32 ino = le32_to_cpu(de->inode);
1599                 brelse(bh);
1600                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1601                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1602                         return ERR_PTR(-EFSCORRUPTED);
1603                 }
1604                 if (unlikely(ino == dir->i_ino)) {
1605                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1606                                          dentry);
1607                         return ERR_PTR(-EFSCORRUPTED);
1608                 }
1609                 inode = ext4_iget_normal(dir->i_sb, ino);
1610                 if (inode == ERR_PTR(-ESTALE)) {
1611                         EXT4_ERROR_INODE(dir,
1612                                          "deleted inode referenced: %u",
1613                                          ino);
1614                         return ERR_PTR(-EFSCORRUPTED);
1615                 }
1616                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1617                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1618                     !fscrypt_has_permitted_context(dir, inode)) {
1619                         int nokey = ext4_encrypted_inode(inode) &&
1620                                 !fscrypt_has_encryption_key(inode);
1621                         if (nokey) {
1622                                 iput(inode);
1623                                 return ERR_PTR(-ENOKEY);
1624                         }
1625                         ext4_warning(inode->i_sb,
1626                                      "Inconsistent encryption contexts: %lu/%lu",
1627                                      (unsigned long) dir->i_ino,
1628                                      (unsigned long) inode->i_ino);
1629                         iput(inode);
1630                         return ERR_PTR(-EPERM);
1631                 }
1632         }
1633         return d_splice_alias(inode, dentry);
1634 }
1635
1636
1637 struct dentry *ext4_get_parent(struct dentry *child)
1638 {
1639         __u32 ino;
1640         static const struct qstr dotdot = QSTR_INIT("..", 2);
1641         struct ext4_dir_entry_2 * de;
1642         struct buffer_head *bh;
1643
1644         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1645         if (IS_ERR(bh))
1646                 return (struct dentry *) bh;
1647         if (!bh)
1648                 return ERR_PTR(-ENOENT);
1649         ino = le32_to_cpu(de->inode);
1650         brelse(bh);
1651
1652         if (!ext4_valid_inum(child->d_sb, ino)) {
1653                 EXT4_ERROR_INODE(d_inode(child),
1654                                  "bad parent inode number: %u", ino);
1655                 return ERR_PTR(-EFSCORRUPTED);
1656         }
1657
1658         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1659 }
1660
1661 /*
1662  * Move count entries from end of map between two memory locations.
1663  * Returns pointer to last entry moved.
1664  */
1665 static struct ext4_dir_entry_2 *
1666 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1667                 unsigned blocksize)
1668 {
1669         unsigned rec_len = 0;
1670
1671         while (count--) {
1672                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1673                                                 (from + (map->offs<<2));
1674                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1675                 memcpy (to, de, rec_len);
1676                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1677                                 ext4_rec_len_to_disk(rec_len, blocksize);
1678                 de->inode = 0;
1679                 map++;
1680                 to += rec_len;
1681         }
1682         return (struct ext4_dir_entry_2 *) (to - rec_len);
1683 }
1684
1685 /*
1686  * Compact each dir entry in the range to the minimal rec_len.
1687  * Returns pointer to last entry in range.
1688  */
1689 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1690 {
1691         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1692         unsigned rec_len = 0;
1693
1694         prev = to = de;
1695         while ((char*)de < base + blocksize) {
1696                 next = ext4_next_entry(de, blocksize);
1697                 if (de->inode && de->name_len) {
1698                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1699                         if (de > to)
1700                                 memmove(to, de, rec_len);
1701                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1702                         prev = to;
1703                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1704                 }
1705                 de = next;
1706         }
1707         return prev;
1708 }
1709
1710 /*
1711  * Split a full leaf block to make room for a new dir entry.
1712  * Allocate a new block, and move entries so that they are approx. equally full.
1713  * Returns pointer to de in block into which the new entry will be inserted.
1714  */
1715 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1716                         struct buffer_head **bh,struct dx_frame *frame,
1717                         struct dx_hash_info *hinfo)
1718 {
1719         unsigned blocksize = dir->i_sb->s_blocksize;
1720         unsigned count, continued;
1721         struct buffer_head *bh2;
1722         ext4_lblk_t newblock;
1723         u32 hash2;
1724         struct dx_map_entry *map;
1725         char *data1 = (*bh)->b_data, *data2;
1726         unsigned split, move, size;
1727         struct ext4_dir_entry_2 *de = NULL, *de2;
1728         struct ext4_dir_entry_tail *t;
1729         int     csum_size = 0;
1730         int     err = 0, i;
1731
1732         if (ext4_has_metadata_csum(dir->i_sb))
1733                 csum_size = sizeof(struct ext4_dir_entry_tail);
1734
1735         bh2 = ext4_append(handle, dir, &newblock);
1736         if (IS_ERR(bh2)) {
1737                 brelse(*bh);
1738                 *bh = NULL;
1739                 return (struct ext4_dir_entry_2 *) bh2;
1740         }
1741
1742         BUFFER_TRACE(*bh, "get_write_access");
1743         err = ext4_journal_get_write_access(handle, *bh);
1744         if (err)
1745                 goto journal_error;
1746
1747         BUFFER_TRACE(frame->bh, "get_write_access");
1748         err = ext4_journal_get_write_access(handle, frame->bh);
1749         if (err)
1750                 goto journal_error;
1751
1752         data2 = bh2->b_data;
1753
1754         /* create map in the end of data2 block */
1755         map = (struct dx_map_entry *) (data2 + blocksize);
1756         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1757                              blocksize, hinfo, map);
1758         map -= count;
1759         dx_sort_map(map, count);
1760         /* Split the existing block in the middle, size-wise */
1761         size = 0;
1762         move = 0;
1763         for (i = count-1; i >= 0; i--) {
1764                 /* is more than half of this entry in 2nd half of the block? */
1765                 if (size + map[i].size/2 > blocksize/2)
1766                         break;
1767                 size += map[i].size;
1768                 move++;
1769         }
1770         /* map index at which we will split */
1771         split = count - move;
1772         hash2 = map[split].hash;
1773         continued = hash2 == map[split - 1].hash;
1774         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1775                         (unsigned long)dx_get_block(frame->at),
1776                                         hash2, split, count-split));
1777
1778         /* Fancy dance to stay within two buffers */
1779         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1780                               blocksize);
1781         de = dx_pack_dirents(data1, blocksize);
1782         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1783                                            (char *) de,
1784                                            blocksize);
1785         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1786                                             (char *) de2,
1787                                             blocksize);
1788         if (csum_size) {
1789                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1790                 initialize_dirent_tail(t, blocksize);
1791
1792                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1793                 initialize_dirent_tail(t, blocksize);
1794         }
1795
1796         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1797                         blocksize, 1));
1798         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1799                         blocksize, 1));
1800
1801         /* Which block gets the new entry? */
1802         if (hinfo->hash >= hash2) {
1803                 swap(*bh, bh2);
1804                 de = de2;
1805         }
1806         dx_insert_block(frame, hash2 + continued, newblock);
1807         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1808         if (err)
1809                 goto journal_error;
1810         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1811         if (err)
1812                 goto journal_error;
1813         brelse(bh2);
1814         dxtrace(dx_show_index("frame", frame->entries));
1815         return de;
1816
1817 journal_error:
1818         brelse(*bh);
1819         brelse(bh2);
1820         *bh = NULL;
1821         ext4_std_error(dir->i_sb, err);
1822         return ERR_PTR(err);
1823 }
1824
1825 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1826                       struct buffer_head *bh,
1827                       void *buf, int buf_size,
1828                       struct ext4_filename *fname,
1829                       struct ext4_dir_entry_2 **dest_de)
1830 {
1831         struct ext4_dir_entry_2 *de;
1832         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1833         int nlen, rlen;
1834         unsigned int offset = 0;
1835         char *top;
1836         int res;
1837
1838         de = (struct ext4_dir_entry_2 *)buf;
1839         top = buf + buf_size - reclen;
1840         while ((char *) de <= top) {
1841                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1842                                          buf, buf_size, offset)) {
1843                         res = -EFSCORRUPTED;
1844                         goto return_result;
1845                 }
1846                 /* Provide crypto context and crypto buffer to ext4 match */
1847                 res = ext4_match(fname, de);
1848                 if (res < 0)
1849                         goto return_result;
1850                 if (res > 0) {
1851                         res = -EEXIST;
1852                         goto return_result;
1853                 }
1854                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1855                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1856                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1857                         break;
1858                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1859                 offset += rlen;
1860         }
1861
1862         if ((char *) de > top)
1863                 res = -ENOSPC;
1864         else {
1865                 *dest_de = de;
1866                 res = 0;
1867         }
1868 return_result:
1869         return res;
1870 }
1871
1872 int ext4_insert_dentry(struct inode *dir,
1873                        struct inode *inode,
1874                        struct ext4_dir_entry_2 *de,
1875                        int buf_size,
1876                        struct ext4_filename *fname)
1877 {
1878
1879         int nlen, rlen;
1880
1881         nlen = EXT4_DIR_REC_LEN(de->name_len);
1882         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1883         if (de->inode) {
1884                 struct ext4_dir_entry_2 *de1 =
1885                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1886                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1887                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1888                 de = de1;
1889         }
1890         de->file_type = EXT4_FT_UNKNOWN;
1891         de->inode = cpu_to_le32(inode->i_ino);
1892         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1893         de->name_len = fname_len(fname);
1894         memcpy(de->name, fname_name(fname), fname_len(fname));
1895         return 0;
1896 }
1897
1898 /*
1899  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1900  * it points to a directory entry which is guaranteed to be large
1901  * enough for new directory entry.  If de is NULL, then
1902  * add_dirent_to_buf will attempt search the directory block for
1903  * space.  It will return -ENOSPC if no space is available, and -EIO
1904  * and -EEXIST if directory entry already exists.
1905  */
1906 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1907                              struct inode *dir,
1908                              struct inode *inode, struct ext4_dir_entry_2 *de,
1909                              struct buffer_head *bh)
1910 {
1911         unsigned int    blocksize = dir->i_sb->s_blocksize;
1912         int             csum_size = 0;
1913         int             err;
1914
1915         if (ext4_has_metadata_csum(inode->i_sb))
1916                 csum_size = sizeof(struct ext4_dir_entry_tail);
1917
1918         if (!de) {
1919                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1920                                         blocksize - csum_size, fname, &de);
1921                 if (err)
1922                         return err;
1923         }
1924         BUFFER_TRACE(bh, "get_write_access");
1925         err = ext4_journal_get_write_access(handle, bh);
1926         if (err) {
1927                 ext4_std_error(dir->i_sb, err);
1928                 return err;
1929         }
1930
1931         /* By now the buffer is marked for journaling. Due to crypto operations,
1932          * the following function call may fail */
1933         err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1934         if (err < 0)
1935                 return err;
1936
1937         /*
1938          * XXX shouldn't update any times until successful
1939          * completion of syscall, but too many callers depend
1940          * on this.
1941          *
1942          * XXX similarly, too many callers depend on
1943          * ext4_new_inode() setting the times, but error
1944          * recovery deletes the inode, so the worst that can
1945          * happen is that the times are slightly out of date
1946          * and/or different from the directory change time.
1947          */
1948         dir->i_mtime = dir->i_ctime = current_time(dir);
1949         ext4_update_dx_flag(dir);
1950         dir->i_version++;
1951         ext4_mark_inode_dirty(handle, dir);
1952         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1953         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1954         if (err)
1955                 ext4_std_error(dir->i_sb, err);
1956         return 0;
1957 }
1958
1959 /*
1960  * This converts a one block unindexed directory to a 3 block indexed
1961  * directory, and adds the dentry to the indexed directory.
1962  */
1963 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1964                             struct inode *dir,
1965                             struct inode *inode, struct buffer_head *bh)
1966 {
1967         struct buffer_head *bh2;
1968         struct dx_root  *root;
1969         struct dx_frame frames[2], *frame;
1970         struct dx_entry *entries;
1971         struct ext4_dir_entry_2 *de, *de2;
1972         struct ext4_dir_entry_tail *t;
1973         char            *data1, *top;
1974         unsigned        len;
1975         int             retval;
1976         unsigned        blocksize;
1977         ext4_lblk_t  block;
1978         struct fake_dirent *fde;
1979         int csum_size = 0;
1980
1981         if (ext4_has_metadata_csum(inode->i_sb))
1982                 csum_size = sizeof(struct ext4_dir_entry_tail);
1983
1984         blocksize =  dir->i_sb->s_blocksize;
1985         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1986         BUFFER_TRACE(bh, "get_write_access");
1987         retval = ext4_journal_get_write_access(handle, bh);
1988         if (retval) {
1989                 ext4_std_error(dir->i_sb, retval);
1990                 brelse(bh);
1991                 return retval;
1992         }
1993         root = (struct dx_root *) bh->b_data;
1994
1995         /* The 0th block becomes the root, move the dirents out */
1996         fde = &root->dotdot;
1997         de = (struct ext4_dir_entry_2 *)((char *)fde +
1998                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1999         if ((char *) de >= (((char *) root) + blocksize)) {
2000                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2001                 brelse(bh);
2002                 return -EFSCORRUPTED;
2003         }
2004         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2005
2006         /* Allocate new block for the 0th block's dirents */
2007         bh2 = ext4_append(handle, dir, &block);
2008         if (IS_ERR(bh2)) {
2009                 brelse(bh);
2010                 return PTR_ERR(bh2);
2011         }
2012         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2013         data1 = bh2->b_data;
2014
2015         memcpy (data1, de, len);
2016         de = (struct ext4_dir_entry_2 *) data1;
2017         top = data1 + len;
2018         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2019                 de = de2;
2020         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2021                                            (char *) de,
2022                                            blocksize);
2023
2024         if (csum_size) {
2025                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2026                 initialize_dirent_tail(t, blocksize);
2027         }
2028
2029         /* Initialize the root; the dot dirents already exist */
2030         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2031         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2032                                            blocksize);
2033         memset (&root->info, 0, sizeof(root->info));
2034         root->info.info_length = sizeof(root->info);
2035         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2036         entries = root->entries;
2037         dx_set_block(entries, 1);
2038         dx_set_count(entries, 1);
2039         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2040
2041         /* Initialize as for dx_probe */
2042         fname->hinfo.hash_version = root->info.hash_version;
2043         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2044                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2045         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2046         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2047
2048         memset(frames, 0, sizeof(frames));
2049         frame = frames;
2050         frame->entries = entries;
2051         frame->at = entries;
2052         frame->bh = bh;
2053
2054         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2055         if (retval)
2056                 goto out_frames;
2057         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2058         if (retval)
2059                 goto out_frames;
2060
2061         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2062         if (IS_ERR(de)) {
2063                 retval = PTR_ERR(de);
2064                 goto out_frames;
2065         }
2066
2067         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2068 out_frames:
2069         /*
2070          * Even if the block split failed, we have to properly write
2071          * out all the changes we did so far. Otherwise we can end up
2072          * with corrupted filesystem.
2073          */
2074         if (retval)
2075                 ext4_mark_inode_dirty(handle, dir);
2076         dx_release(frames);
2077         brelse(bh2);
2078         return retval;
2079 }
2080
2081 /*
2082  *      ext4_add_entry()
2083  *
2084  * adds a file entry to the specified directory, using the same
2085  * semantics as ext4_find_entry(). It returns NULL if it failed.
2086  *
2087  * NOTE!! The inode part of 'de' is left at 0 - which means you
2088  * may not sleep between calling this and putting something into
2089  * the entry, as someone else might have used it while you slept.
2090  */
2091 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2092                           struct inode *inode)
2093 {
2094         struct inode *dir = d_inode(dentry->d_parent);
2095         struct buffer_head *bh = NULL;
2096         struct ext4_dir_entry_2 *de;
2097         struct ext4_dir_entry_tail *t;
2098         struct super_block *sb;
2099         struct ext4_filename fname;
2100         int     retval;
2101         int     dx_fallback=0;
2102         unsigned blocksize;
2103         ext4_lblk_t block, blocks;
2104         int     csum_size = 0;
2105
2106         if (ext4_has_metadata_csum(inode->i_sb))
2107                 csum_size = sizeof(struct ext4_dir_entry_tail);
2108
2109         sb = dir->i_sb;
2110         blocksize = sb->s_blocksize;
2111         if (!dentry->d_name.len)
2112                 return -EINVAL;
2113
2114         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2115         if (retval)
2116                 return retval;
2117
2118         if (ext4_has_inline_data(dir)) {
2119                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2120                 if (retval < 0)
2121                         goto out;
2122                 if (retval == 1) {
2123                         retval = 0;
2124                         goto out;
2125                 }
2126         }
2127
2128         if (is_dx(dir)) {
2129                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2130                 if (!retval || (retval != ERR_BAD_DX_DIR))
2131                         goto out;
2132                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2133                 dx_fallback++;
2134                 ext4_mark_inode_dirty(handle, dir);
2135         }
2136         blocks = dir->i_size >> sb->s_blocksize_bits;
2137         for (block = 0; block < blocks; block++) {
2138                 bh = ext4_read_dirblock(dir, block, DIRENT);
2139                 if (IS_ERR(bh)) {
2140                         retval = PTR_ERR(bh);
2141                         bh = NULL;
2142                         goto out;
2143                 }
2144                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2145                                            NULL, bh);
2146                 if (retval != -ENOSPC)
2147                         goto out;
2148
2149                 if (blocks == 1 && !dx_fallback &&
2150                     ext4_has_feature_dir_index(sb)) {
2151                         retval = make_indexed_dir(handle, &fname, dir,
2152                                                   inode, bh);
2153                         bh = NULL; /* make_indexed_dir releases bh */
2154                         goto out;
2155                 }
2156                 brelse(bh);
2157         }
2158         bh = ext4_append(handle, dir, &block);
2159         if (IS_ERR(bh)) {
2160                 retval = PTR_ERR(bh);
2161                 bh = NULL;
2162                 goto out;
2163         }
2164         de = (struct ext4_dir_entry_2 *) bh->b_data;
2165         de->inode = 0;
2166         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2167
2168         if (csum_size) {
2169                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2170                 initialize_dirent_tail(t, blocksize);
2171         }
2172
2173         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2174 out:
2175         ext4_fname_free_filename(&fname);
2176         brelse(bh);
2177         if (retval == 0)
2178                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2179         return retval;
2180 }
2181
2182 /*
2183  * Returns 0 for success, or a negative error value
2184  */
2185 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2186                              struct inode *dir, struct inode *inode)
2187 {
2188         struct dx_frame frames[2], *frame;
2189         struct dx_entry *entries, *at;
2190         struct buffer_head *bh;
2191         struct super_block *sb = dir->i_sb;
2192         struct ext4_dir_entry_2 *de;
2193         int err;
2194
2195         frame = dx_probe(fname, dir, NULL, frames);
2196         if (IS_ERR(frame))
2197                 return PTR_ERR(frame);
2198         entries = frame->entries;
2199         at = frame->at;
2200         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2201         if (IS_ERR(bh)) {
2202                 err = PTR_ERR(bh);
2203                 bh = NULL;
2204                 goto cleanup;
2205         }
2206
2207         BUFFER_TRACE(bh, "get_write_access");
2208         err = ext4_journal_get_write_access(handle, bh);
2209         if (err)
2210                 goto journal_error;
2211
2212         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2213         if (err != -ENOSPC)
2214                 goto cleanup;
2215
2216         /* Block full, should compress but for now just split */
2217         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2218                        dx_get_count(entries), dx_get_limit(entries)));
2219         /* Need to split index? */
2220         if (dx_get_count(entries) == dx_get_limit(entries)) {
2221                 ext4_lblk_t newblock;
2222                 unsigned icount = dx_get_count(entries);
2223                 int levels = frame - frames;
2224                 struct dx_entry *entries2;
2225                 struct dx_node *node2;
2226                 struct buffer_head *bh2;
2227
2228                 if (levels && (dx_get_count(frames->entries) ==
2229                                dx_get_limit(frames->entries))) {
2230                         ext4_warning_inode(dir, "Directory index full!");
2231                         err = -ENOSPC;
2232                         goto cleanup;
2233                 }
2234                 bh2 = ext4_append(handle, dir, &newblock);
2235                 if (IS_ERR(bh2)) {
2236                         err = PTR_ERR(bh2);
2237                         goto cleanup;
2238                 }
2239                 node2 = (struct dx_node *)(bh2->b_data);
2240                 entries2 = node2->entries;
2241                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2242                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2243                                                            sb->s_blocksize);
2244                 BUFFER_TRACE(frame->bh, "get_write_access");
2245                 err = ext4_journal_get_write_access(handle, frame->bh);
2246                 if (err)
2247                         goto journal_error;
2248                 if (levels) {
2249                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2250                         unsigned hash2 = dx_get_hash(entries + icount1);
2251                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2252                                        icount1, icount2));
2253
2254                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2255                         err = ext4_journal_get_write_access(handle,
2256                                                              frames[0].bh);
2257                         if (err)
2258                                 goto journal_error;
2259
2260                         memcpy((char *) entries2, (char *) (entries + icount1),
2261                                icount2 * sizeof(struct dx_entry));
2262                         dx_set_count(entries, icount1);
2263                         dx_set_count(entries2, icount2);
2264                         dx_set_limit(entries2, dx_node_limit(dir));
2265
2266                         /* Which index block gets the new entry? */
2267                         if (at - entries >= icount1) {
2268                                 frame->at = at = at - entries - icount1 + entries2;
2269                                 frame->entries = entries = entries2;
2270                                 swap(frame->bh, bh2);
2271                         }
2272                         dx_insert_block(frames + 0, hash2, newblock);
2273                         dxtrace(dx_show_index("node", frames[1].entries));
2274                         dxtrace(dx_show_index("node",
2275                                ((struct dx_node *) bh2->b_data)->entries));
2276                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2277                         if (err)
2278                                 goto journal_error;
2279                         brelse (bh2);
2280                 } else {
2281                         dxtrace(printk(KERN_DEBUG
2282                                        "Creating second level index...\n"));
2283                         memcpy((char *) entries2, (char *) entries,
2284                                icount * sizeof(struct dx_entry));
2285                         dx_set_limit(entries2, dx_node_limit(dir));
2286
2287                         /* Set up root */
2288                         dx_set_count(entries, 1);
2289                         dx_set_block(entries + 0, newblock);
2290                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2291
2292                         /* Add new access path frame */
2293                         frame = frames + 1;
2294                         frame->at = at = at - entries + entries2;
2295                         frame->entries = entries = entries2;
2296                         frame->bh = bh2;
2297                         err = ext4_journal_get_write_access(handle,
2298                                                              frame->bh);
2299                         if (err)
2300                                 goto journal_error;
2301                 }
2302                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2303                 if (err) {
2304                         ext4_std_error(inode->i_sb, err);
2305                         goto cleanup;
2306                 }
2307         }
2308         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2309         if (IS_ERR(de)) {
2310                 err = PTR_ERR(de);
2311                 goto cleanup;
2312         }
2313         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2314         goto cleanup;
2315
2316 journal_error:
2317         ext4_std_error(dir->i_sb, err);
2318 cleanup:
2319         brelse(bh);
2320         dx_release(frames);
2321         return err;
2322 }
2323
2324 /*
2325  * ext4_generic_delete_entry deletes a directory entry by merging it
2326  * with the previous entry
2327  */
2328 int ext4_generic_delete_entry(handle_t *handle,
2329                               struct inode *dir,
2330                               struct ext4_dir_entry_2 *de_del,
2331                               struct buffer_head *bh,
2332                               void *entry_buf,
2333                               int buf_size,
2334                               int csum_size)
2335 {
2336         struct ext4_dir_entry_2 *de, *pde;
2337         unsigned int blocksize = dir->i_sb->s_blocksize;
2338         int i;
2339
2340         i = 0;
2341         pde = NULL;
2342         de = (struct ext4_dir_entry_2 *)entry_buf;
2343         while (i < buf_size - csum_size) {
2344                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2345                                          bh->b_data, bh->b_size, i))
2346                         return -EFSCORRUPTED;
2347                 if (de == de_del)  {
2348                         if (pde)
2349                                 pde->rec_len = ext4_rec_len_to_disk(
2350                                         ext4_rec_len_from_disk(pde->rec_len,
2351                                                                blocksize) +
2352                                         ext4_rec_len_from_disk(de->rec_len,
2353                                                                blocksize),
2354                                         blocksize);
2355                         else
2356                                 de->inode = 0;
2357                         dir->i_version++;
2358                         return 0;
2359                 }
2360                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2361                 pde = de;
2362                 de = ext4_next_entry(de, blocksize);
2363         }
2364         return -ENOENT;
2365 }
2366
2367 static int ext4_delete_entry(handle_t *handle,
2368                              struct inode *dir,
2369                              struct ext4_dir_entry_2 *de_del,
2370                              struct buffer_head *bh)
2371 {
2372         int err, csum_size = 0;
2373
2374         if (ext4_has_inline_data(dir)) {
2375                 int has_inline_data = 1;
2376                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2377                                                &has_inline_data);
2378                 if (has_inline_data)
2379                         return err;
2380         }
2381
2382         if (ext4_has_metadata_csum(dir->i_sb))
2383                 csum_size = sizeof(struct ext4_dir_entry_tail);
2384
2385         BUFFER_TRACE(bh, "get_write_access");
2386         err = ext4_journal_get_write_access(handle, bh);
2387         if (unlikely(err))
2388                 goto out;
2389
2390         err = ext4_generic_delete_entry(handle, dir, de_del,
2391                                         bh, bh->b_data,
2392                                         dir->i_sb->s_blocksize, csum_size);
2393         if (err)
2394                 goto out;
2395
2396         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2397         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2398         if (unlikely(err))
2399                 goto out;
2400
2401         return 0;
2402 out:
2403         if (err != -ENOENT)
2404                 ext4_std_error(dir->i_sb, err);
2405         return err;
2406 }
2407
2408 /*
2409  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2410  * since this indicates that nlinks count was previously 1.
2411  */
2412 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2413 {
2414         inc_nlink(inode);
2415         if (is_dx(inode) && inode->i_nlink > 1) {
2416                 /* limit is 16-bit i_links_count */
2417                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2418                         set_nlink(inode, 1);
2419                         ext4_set_feature_dir_nlink(inode->i_sb);
2420                 }
2421         }
2422 }
2423
2424 /*
2425  * If a directory had nlink == 1, then we should let it be 1. This indicates
2426  * directory has >EXT4_LINK_MAX subdirs.
2427  */
2428 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2429 {
2430         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2431                 drop_nlink(inode);
2432 }
2433
2434
2435 static int ext4_add_nondir(handle_t *handle,
2436                 struct dentry *dentry, struct inode *inode)
2437 {
2438         int err = ext4_add_entry(handle, dentry, inode);
2439         if (!err) {
2440                 ext4_mark_inode_dirty(handle, inode);
2441                 unlock_new_inode(inode);
2442                 d_instantiate(dentry, inode);
2443                 return 0;
2444         }
2445         drop_nlink(inode);
2446         unlock_new_inode(inode);
2447         iput(inode);
2448         return err;
2449 }
2450
2451 /*
2452  * By the time this is called, we already have created
2453  * the directory cache entry for the new file, but it
2454  * is so far negative - it has no inode.
2455  *
2456  * If the create succeeds, we fill in the inode information
2457  * with d_instantiate().
2458  */
2459 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2460                        bool excl)
2461 {
2462         handle_t *handle;
2463         struct inode *inode;
2464         int err, credits, retries = 0;
2465
2466         err = dquot_initialize(dir);
2467         if (err)
2468                 return err;
2469
2470         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2471                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2472 retry:
2473         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2474                                             NULL, EXT4_HT_DIR, credits);
2475         handle = ext4_journal_current_handle();
2476         err = PTR_ERR(inode);
2477         if (!IS_ERR(inode)) {
2478                 inode->i_op = &ext4_file_inode_operations;
2479                 inode->i_fop = &ext4_file_operations;
2480                 ext4_set_aops(inode);
2481                 err = ext4_add_nondir(handle, dentry, inode);
2482                 if (!err && IS_DIRSYNC(dir))
2483                         ext4_handle_sync(handle);
2484         }
2485         if (handle)
2486                 ext4_journal_stop(handle);
2487         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2488                 goto retry;
2489         return err;
2490 }
2491
2492 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2493                       umode_t mode, dev_t rdev)
2494 {
2495         handle_t *handle;
2496         struct inode *inode;
2497         int err, credits, retries = 0;
2498
2499         err = dquot_initialize(dir);
2500         if (err)
2501                 return err;
2502
2503         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2504                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2505 retry:
2506         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2507                                             NULL, EXT4_HT_DIR, credits);
2508         handle = ext4_journal_current_handle();
2509         err = PTR_ERR(inode);
2510         if (!IS_ERR(inode)) {
2511                 init_special_inode(inode, inode->i_mode, rdev);
2512                 inode->i_op = &ext4_special_inode_operations;
2513                 err = ext4_add_nondir(handle, dentry, inode);
2514                 if (!err && IS_DIRSYNC(dir))
2515                         ext4_handle_sync(handle);
2516         }
2517         if (handle)
2518                 ext4_journal_stop(handle);
2519         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2520                 goto retry;
2521         return err;
2522 }
2523
2524 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2525 {
2526         handle_t *handle;
2527         struct inode *inode;
2528         int err, retries = 0;
2529
2530         err = dquot_initialize(dir);
2531         if (err)
2532                 return err;
2533
2534 retry:
2535         inode = ext4_new_inode_start_handle(dir, mode,
2536                                             NULL, 0, NULL,
2537                                             EXT4_HT_DIR,
2538                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2539                           4 + EXT4_XATTR_TRANS_BLOCKS);
2540         handle = ext4_journal_current_handle();
2541         err = PTR_ERR(inode);
2542         if (!IS_ERR(inode)) {
2543                 inode->i_op = &ext4_file_inode_operations;
2544                 inode->i_fop = &ext4_file_operations;
2545                 ext4_set_aops(inode);
2546                 d_tmpfile(dentry, inode);
2547                 err = ext4_orphan_add(handle, inode);
2548                 if (err)
2549                         goto err_unlock_inode;
2550                 mark_inode_dirty(inode);
2551                 unlock_new_inode(inode);
2552         }
2553         if (handle)
2554                 ext4_journal_stop(handle);
2555         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2556                 goto retry;
2557         return err;
2558 err_unlock_inode:
2559         ext4_journal_stop(handle);
2560         unlock_new_inode(inode);
2561         return err;
2562 }
2563
2564 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2565                           struct ext4_dir_entry_2 *de,
2566                           int blocksize, int csum_size,
2567                           unsigned int parent_ino, int dotdot_real_len)
2568 {
2569         de->inode = cpu_to_le32(inode->i_ino);
2570         de->name_len = 1;
2571         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2572                                            blocksize);
2573         strcpy(de->name, ".");
2574         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2575
2576         de = ext4_next_entry(de, blocksize);
2577         de->inode = cpu_to_le32(parent_ino);
2578         de->name_len = 2;
2579         if (!dotdot_real_len)
2580                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2581                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2582                                         blocksize);
2583         else
2584                 de->rec_len = ext4_rec_len_to_disk(
2585                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2586         strcpy(de->name, "..");
2587         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2588
2589         return ext4_next_entry(de, blocksize);
2590 }
2591
2592 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2593                              struct inode *inode)
2594 {
2595         struct buffer_head *dir_block = NULL;
2596         struct ext4_dir_entry_2 *de;
2597         struct ext4_dir_entry_tail *t;
2598         ext4_lblk_t block = 0;
2599         unsigned int blocksize = dir->i_sb->s_blocksize;
2600         int csum_size = 0;
2601         int err;
2602
2603         if (ext4_has_metadata_csum(dir->i_sb))
2604                 csum_size = sizeof(struct ext4_dir_entry_tail);
2605
2606         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2607                 err = ext4_try_create_inline_dir(handle, dir, inode);
2608                 if (err < 0 && err != -ENOSPC)
2609                         goto out;
2610                 if (!err)
2611                         goto out;
2612         }
2613
2614         inode->i_size = 0;
2615         dir_block = ext4_append(handle, inode, &block);
2616         if (IS_ERR(dir_block))
2617                 return PTR_ERR(dir_block);
2618         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2619         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2620         set_nlink(inode, 2);
2621         if (csum_size) {
2622                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2623                 initialize_dirent_tail(t, blocksize);
2624         }
2625
2626         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2627         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2628         if (err)
2629                 goto out;
2630         set_buffer_verified(dir_block);
2631 out:
2632         brelse(dir_block);
2633         return err;
2634 }
2635
2636 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2637 {
2638         handle_t *handle;
2639         struct inode *inode;
2640         int err, credits, retries = 0;
2641
2642         if (EXT4_DIR_LINK_MAX(dir))
2643                 return -EMLINK;
2644
2645         err = dquot_initialize(dir);
2646         if (err)
2647                 return err;
2648
2649         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2650                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2651 retry:
2652         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2653                                             &dentry->d_name,
2654                                             0, NULL, EXT4_HT_DIR, credits);
2655         handle = ext4_journal_current_handle();
2656         err = PTR_ERR(inode);
2657         if (IS_ERR(inode))
2658                 goto out_stop;
2659
2660         inode->i_op = &ext4_dir_inode_operations;
2661         inode->i_fop = &ext4_dir_operations;
2662         err = ext4_init_new_dir(handle, dir, inode);
2663         if (err)
2664                 goto out_clear_inode;
2665         err = ext4_mark_inode_dirty(handle, inode);
2666         if (!err)
2667                 err = ext4_add_entry(handle, dentry, inode);
2668         if (err) {
2669 out_clear_inode:
2670                 clear_nlink(inode);
2671                 unlock_new_inode(inode);
2672                 ext4_mark_inode_dirty(handle, inode);
2673                 iput(inode);
2674                 goto out_stop;
2675         }
2676         ext4_inc_count(handle, dir);
2677         ext4_update_dx_flag(dir);
2678         err = ext4_mark_inode_dirty(handle, dir);
2679         if (err)
2680                 goto out_clear_inode;
2681         unlock_new_inode(inode);
2682         d_instantiate(dentry, inode);
2683         if (IS_DIRSYNC(dir))
2684                 ext4_handle_sync(handle);
2685
2686 out_stop:
2687         if (handle)
2688                 ext4_journal_stop(handle);
2689         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2690                 goto retry;
2691         return err;
2692 }
2693
2694 /*
2695  * routine to check that the specified directory is empty (for rmdir)
2696  */
2697 bool ext4_empty_dir(struct inode *inode)
2698 {
2699         unsigned int offset;
2700         struct buffer_head *bh;
2701         struct ext4_dir_entry_2 *de, *de1;
2702         struct super_block *sb;
2703
2704         if (ext4_has_inline_data(inode)) {
2705                 int has_inline_data = 1;
2706                 int ret;
2707
2708                 ret = empty_inline_dir(inode, &has_inline_data);
2709                 if (has_inline_data)
2710                         return ret;
2711         }
2712
2713         sb = inode->i_sb;
2714         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2715                 EXT4_ERROR_INODE(inode, "invalid size");
2716                 return true;
2717         }
2718         bh = ext4_read_dirblock(inode, 0, EITHER);
2719         if (IS_ERR(bh))
2720                 return true;
2721
2722         de = (struct ext4_dir_entry_2 *) bh->b_data;
2723         de1 = ext4_next_entry(de, sb->s_blocksize);
2724         if (le32_to_cpu(de->inode) != inode->i_ino ||
2725                         le32_to_cpu(de1->inode) == 0 ||
2726                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2727                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2728                 brelse(bh);
2729                 return true;
2730         }
2731         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2732                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2733         de = ext4_next_entry(de1, sb->s_blocksize);
2734         while (offset < inode->i_size) {
2735                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2736                         unsigned int lblock;
2737                         brelse(bh);
2738                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2739                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2740                         if (IS_ERR(bh))
2741                                 return true;
2742                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2743                 }
2744                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2745                                          bh->b_data, bh->b_size, offset)) {
2746                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2747                                                          sb->s_blocksize);
2748                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2749                         continue;
2750                 }
2751                 if (le32_to_cpu(de->inode)) {
2752                         brelse(bh);
2753                         return false;
2754                 }
2755                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2756                 de = ext4_next_entry(de, sb->s_blocksize);
2757         }
2758         brelse(bh);
2759         return true;
2760 }
2761
2762 /*
2763  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2764  * such inodes, starting at the superblock, in case we crash before the
2765  * file is closed/deleted, or in case the inode truncate spans multiple
2766  * transactions and the last transaction is not recovered after a crash.
2767  *
2768  * At filesystem recovery time, we walk this list deleting unlinked
2769  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2770  *
2771  * Orphan list manipulation functions must be called under i_mutex unless
2772  * we are just creating the inode or deleting it.
2773  */
2774 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2775 {
2776         struct super_block *sb = inode->i_sb;
2777         struct ext4_sb_info *sbi = EXT4_SB(sb);
2778         struct ext4_iloc iloc;
2779         int err = 0, rc;
2780         bool dirty = false;
2781
2782         if (!sbi->s_journal || is_bad_inode(inode))
2783                 return 0;
2784
2785         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2786                      !inode_is_locked(inode));
2787         /*
2788          * Exit early if inode already is on orphan list. This is a big speedup
2789          * since we don't have to contend on the global s_orphan_lock.
2790          */
2791         if (!list_empty(&EXT4_I(inode)->i_orphan))
2792                 return 0;
2793
2794         /*
2795          * Orphan handling is only valid for files with data blocks
2796          * being truncated, or files being unlinked. Note that we either
2797          * hold i_mutex, or the inode can not be referenced from outside,
2798          * so i_nlink should not be bumped due to race
2799          */
2800         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2801                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2802
2803         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2804         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2805         if (err)
2806                 goto out;
2807
2808         err = ext4_reserve_inode_write(handle, inode, &iloc);
2809         if (err)
2810                 goto out;
2811
2812         mutex_lock(&sbi->s_orphan_lock);
2813         /*
2814          * Due to previous errors inode may be already a part of on-disk
2815          * orphan list. If so skip on-disk list modification.
2816          */
2817         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2818             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2819                 /* Insert this inode at the head of the on-disk orphan list */
2820                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2821                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2822                 dirty = true;
2823         }
2824         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2825         mutex_unlock(&sbi->s_orphan_lock);
2826
2827         if (dirty) {
2828                 err = ext4_handle_dirty_super(handle, sb);
2829                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2830                 if (!err)
2831                         err = rc;
2832                 if (err) {
2833                         /*
2834                          * We have to remove inode from in-memory list if
2835                          * addition to on disk orphan list failed. Stray orphan
2836                          * list entries can cause panics at unmount time.
2837                          */
2838                         mutex_lock(&sbi->s_orphan_lock);
2839                         list_del_init(&EXT4_I(inode)->i_orphan);
2840                         mutex_unlock(&sbi->s_orphan_lock);
2841                 }
2842         }
2843         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2844         jbd_debug(4, "orphan inode %lu will point to %d\n",
2845                         inode->i_ino, NEXT_ORPHAN(inode));
2846 out:
2847         ext4_std_error(sb, err);
2848         return err;
2849 }
2850
2851 /*
2852  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2853  * of such inodes stored on disk, because it is finally being cleaned up.
2854  */
2855 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2856 {
2857         struct list_head *prev;
2858         struct ext4_inode_info *ei = EXT4_I(inode);
2859         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2860         __u32 ino_next;
2861         struct ext4_iloc iloc;
2862         int err = 0;
2863
2864         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2865                 return 0;
2866
2867         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2868                      !inode_is_locked(inode));
2869         /* Do this quick check before taking global s_orphan_lock. */
2870         if (list_empty(&ei->i_orphan))
2871                 return 0;
2872
2873         if (handle) {
2874                 /* Grab inode buffer early before taking global s_orphan_lock */
2875                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2876         }
2877
2878         mutex_lock(&sbi->s_orphan_lock);
2879         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2880
2881         prev = ei->i_orphan.prev;
2882         list_del_init(&ei->i_orphan);
2883
2884         /* If we're on an error path, we may not have a valid
2885          * transaction handle with which to update the orphan list on
2886          * disk, but we still need to remove the inode from the linked
2887          * list in memory. */
2888         if (!handle || err) {
2889                 mutex_unlock(&sbi->s_orphan_lock);
2890                 goto out_err;
2891         }
2892
2893         ino_next = NEXT_ORPHAN(inode);
2894         if (prev == &sbi->s_orphan) {
2895                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2896                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2897                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2898                 if (err) {
2899                         mutex_unlock(&sbi->s_orphan_lock);
2900                         goto out_brelse;
2901                 }
2902                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2903                 mutex_unlock(&sbi->s_orphan_lock);
2904                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2905         } else {
2906                 struct ext4_iloc iloc2;
2907                 struct inode *i_prev =
2908                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2909
2910                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2911                           i_prev->i_ino, ino_next);
2912                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2913                 if (err) {
2914                         mutex_unlock(&sbi->s_orphan_lock);
2915                         goto out_brelse;
2916                 }
2917                 NEXT_ORPHAN(i_prev) = ino_next;
2918                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2919                 mutex_unlock(&sbi->s_orphan_lock);
2920         }
2921         if (err)
2922                 goto out_brelse;
2923         NEXT_ORPHAN(inode) = 0;
2924         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2925 out_err:
2926         ext4_std_error(inode->i_sb, err);
2927         return err;
2928
2929 out_brelse:
2930         brelse(iloc.bh);
2931         goto out_err;
2932 }
2933
2934 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2935 {
2936         int retval;
2937         struct inode *inode;
2938         struct buffer_head *bh;
2939         struct ext4_dir_entry_2 *de;
2940         handle_t *handle = NULL;
2941
2942         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2943                 return -EIO;
2944
2945         /* Initialize quotas before so that eventual writes go in
2946          * separate transaction */
2947         retval = dquot_initialize(dir);
2948         if (retval)
2949                 return retval;
2950         retval = dquot_initialize(d_inode(dentry));
2951         if (retval)
2952                 return retval;
2953
2954         retval = -ENOENT;
2955         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2956         if (IS_ERR(bh))
2957                 return PTR_ERR(bh);
2958         if (!bh)
2959                 goto end_rmdir;
2960
2961         inode = d_inode(dentry);
2962
2963         retval = -EFSCORRUPTED;
2964         if (le32_to_cpu(de->inode) != inode->i_ino)
2965                 goto end_rmdir;
2966
2967         retval = -ENOTEMPTY;
2968         if (!ext4_empty_dir(inode))
2969                 goto end_rmdir;
2970
2971         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2972                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2973         if (IS_ERR(handle)) {
2974                 retval = PTR_ERR(handle);
2975                 handle = NULL;
2976                 goto end_rmdir;
2977         }
2978
2979         if (IS_DIRSYNC(dir))
2980                 ext4_handle_sync(handle);
2981
2982         retval = ext4_delete_entry(handle, dir, de, bh);
2983         if (retval)
2984                 goto end_rmdir;
2985         if (!EXT4_DIR_LINK_EMPTY(inode))
2986                 ext4_warning_inode(inode,
2987                              "empty directory '%.*s' has too many links (%u)",
2988                              dentry->d_name.len, dentry->d_name.name,
2989                              inode->i_nlink);
2990         inode->i_version++;
2991         clear_nlink(inode);
2992         /* There's no need to set i_disksize: the fact that i_nlink is
2993          * zero will ensure that the right thing happens during any
2994          * recovery. */
2995         inode->i_size = 0;
2996         ext4_orphan_add(handle, inode);
2997         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2998         ext4_mark_inode_dirty(handle, inode);
2999         ext4_dec_count(handle, dir);
3000         ext4_update_dx_flag(dir);
3001         ext4_mark_inode_dirty(handle, dir);
3002
3003 end_rmdir:
3004         brelse(bh);
3005         if (handle)
3006                 ext4_journal_stop(handle);
3007         return retval;
3008 }
3009
3010 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3011 {
3012         int retval;
3013         struct inode *inode;
3014         struct buffer_head *bh;
3015         struct ext4_dir_entry_2 *de;
3016         handle_t *handle = NULL;
3017
3018         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3019                 return -EIO;
3020
3021         trace_ext4_unlink_enter(dir, dentry);
3022         /* Initialize quotas before so that eventual writes go
3023          * in separate transaction */
3024         retval = dquot_initialize(dir);
3025         if (retval)
3026                 return retval;
3027         retval = dquot_initialize(d_inode(dentry));
3028         if (retval)
3029                 return retval;
3030
3031         retval = -ENOENT;
3032         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3033         if (IS_ERR(bh))
3034                 return PTR_ERR(bh);
3035         if (!bh)
3036                 goto end_unlink;
3037
3038         inode = d_inode(dentry);
3039
3040         retval = -EFSCORRUPTED;
3041         if (le32_to_cpu(de->inode) != inode->i_ino)
3042                 goto end_unlink;
3043
3044         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3045                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3046         if (IS_ERR(handle)) {
3047                 retval = PTR_ERR(handle);
3048                 handle = NULL;
3049                 goto end_unlink;
3050         }
3051
3052         if (IS_DIRSYNC(dir))
3053                 ext4_handle_sync(handle);
3054
3055         if (inode->i_nlink == 0) {
3056                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3057                                    dentry->d_name.len, dentry->d_name.name);
3058                 set_nlink(inode, 1);
3059         }
3060         retval = ext4_delete_entry(handle, dir, de, bh);
3061         if (retval)
3062                 goto end_unlink;
3063         dir->i_ctime = dir->i_mtime = current_time(dir);
3064         ext4_update_dx_flag(dir);
3065         ext4_mark_inode_dirty(handle, dir);
3066         drop_nlink(inode);
3067         if (!inode->i_nlink)
3068                 ext4_orphan_add(handle, inode);
3069         inode->i_ctime = current_time(inode);
3070         ext4_mark_inode_dirty(handle, inode);
3071
3072 end_unlink:
3073         brelse(bh);
3074         if (handle)
3075                 ext4_journal_stop(handle);
3076         trace_ext4_unlink_exit(dentry, retval);
3077         return retval;
3078 }
3079
3080 static int ext4_symlink(struct inode *dir,
3081                         struct dentry *dentry, const char *symname)
3082 {
3083         handle_t *handle;
3084         struct inode *inode;
3085         int err, len = strlen(symname);
3086         int credits;
3087         bool encryption_required;
3088         struct fscrypt_str disk_link;
3089         struct fscrypt_symlink_data *sd = NULL;
3090
3091         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3092                 return -EIO;
3093
3094         disk_link.len = len + 1;
3095         disk_link.name = (char *) symname;
3096
3097         encryption_required = (ext4_encrypted_inode(dir) ||
3098                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3099         if (encryption_required) {
3100                 err = fscrypt_get_encryption_info(dir);
3101                 if (err)
3102                         return err;
3103                 if (!fscrypt_has_encryption_key(dir))
3104                         return -ENOKEY;
3105                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3106                                  sizeof(struct fscrypt_symlink_data));
3107                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3108                 if (!sd)
3109                         return -ENOMEM;
3110         }
3111
3112         if (disk_link.len > dir->i_sb->s_blocksize) {
3113                 err = -ENAMETOOLONG;
3114                 goto err_free_sd;
3115         }
3116
3117         err = dquot_initialize(dir);
3118         if (err)
3119                 goto err_free_sd;
3120
3121         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3122                 /*
3123                  * For non-fast symlinks, we just allocate inode and put it on
3124                  * orphan list in the first transaction => we need bitmap,
3125                  * group descriptor, sb, inode block, quota blocks, and
3126                  * possibly selinux xattr blocks.
3127                  */
3128                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3129                           EXT4_XATTR_TRANS_BLOCKS;
3130         } else {
3131                 /*
3132                  * Fast symlink. We have to add entry to directory
3133                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3134                  * allocate new inode (bitmap, group descriptor, inode block,
3135                  * quota blocks, sb is already counted in previous macros).
3136                  */
3137                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3138                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3139         }
3140
3141         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3142                                             &dentry->d_name, 0, NULL,
3143                                             EXT4_HT_DIR, credits);
3144         handle = ext4_journal_current_handle();
3145         if (IS_ERR(inode)) {
3146                 if (handle)
3147                         ext4_journal_stop(handle);
3148                 err = PTR_ERR(inode);
3149                 goto err_free_sd;
3150         }
3151
3152         if (encryption_required) {
3153                 struct qstr istr;
3154                 struct fscrypt_str ostr =
3155                         FSTR_INIT(sd->encrypted_path, disk_link.len);
3156
3157                 istr.name = (const unsigned char *) symname;
3158                 istr.len = len;
3159                 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3160                 if (err)
3161                         goto err_drop_inode;
3162                 sd->len = cpu_to_le16(ostr.len);
3163                 disk_link.name = (char *) sd;
3164                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3165         }
3166
3167         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3168                 if (!encryption_required)
3169                         inode->i_op = &ext4_symlink_inode_operations;
3170                 inode_nohighmem(inode);
3171                 ext4_set_aops(inode);
3172                 /*
3173                  * We cannot call page_symlink() with transaction started
3174                  * because it calls into ext4_write_begin() which can wait
3175                  * for transaction commit if we are running out of space
3176                  * and thus we deadlock. So we have to stop transaction now
3177                  * and restart it when symlink contents is written.
3178                  *
3179                  * To keep fs consistent in case of crash, we have to put inode
3180                  * to orphan list in the mean time.
3181                  */
3182                 drop_nlink(inode);
3183                 err = ext4_orphan_add(handle, inode);
3184                 ext4_journal_stop(handle);
3185                 handle = NULL;
3186                 if (err)
3187                         goto err_drop_inode;
3188                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3189                 if (err)
3190                         goto err_drop_inode;
3191                 /*
3192                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3193                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3194                  */
3195                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3196                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3197                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3198                 if (IS_ERR(handle)) {
3199                         err = PTR_ERR(handle);
3200                         handle = NULL;
3201                         goto err_drop_inode;
3202                 }
3203                 set_nlink(inode, 1);
3204                 err = ext4_orphan_del(handle, inode);
3205                 if (err)
3206                         goto err_drop_inode;
3207         } else {
3208                 /* clear the extent format for fast symlink */
3209                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3210                 if (!encryption_required) {
3211                         inode->i_op = &ext4_fast_symlink_inode_operations;
3212                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3213                 }
3214                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3215                        disk_link.len);
3216                 inode->i_size = disk_link.len - 1;
3217         }
3218         EXT4_I(inode)->i_disksize = inode->i_size;
3219         err = ext4_add_nondir(handle, dentry, inode);
3220         if (!err && IS_DIRSYNC(dir))
3221                 ext4_handle_sync(handle);
3222
3223         if (handle)
3224                 ext4_journal_stop(handle);
3225         kfree(sd);
3226         return err;
3227 err_drop_inode:
3228         if (handle)
3229                 ext4_journal_stop(handle);
3230         clear_nlink(inode);
3231         unlock_new_inode(inode);
3232         iput(inode);
3233 err_free_sd:
3234         kfree(sd);
3235         return err;
3236 }
3237
3238 static int ext4_link(struct dentry *old_dentry,
3239                      struct inode *dir, struct dentry *dentry)
3240 {
3241         handle_t *handle;
3242         struct inode *inode = d_inode(old_dentry);
3243         int err, retries = 0;
3244
3245         if (inode->i_nlink >= EXT4_LINK_MAX)
3246                 return -EMLINK;
3247         if (ext4_encrypted_inode(dir) &&
3248                         !fscrypt_has_permitted_context(dir, inode))
3249                 return -EPERM;
3250
3251        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3252            (!projid_eq(EXT4_I(dir)->i_projid,
3253                        EXT4_I(old_dentry->d_inode)->i_projid)))
3254                 return -EXDEV;
3255
3256         err = dquot_initialize(dir);
3257         if (err)
3258                 return err;
3259
3260 retry:
3261         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3262                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3263                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3264         if (IS_ERR(handle))
3265                 return PTR_ERR(handle);
3266
3267         if (IS_DIRSYNC(dir))
3268                 ext4_handle_sync(handle);
3269
3270         inode->i_ctime = current_time(inode);
3271         ext4_inc_count(handle, inode);
3272         ihold(inode);
3273
3274         err = ext4_add_entry(handle, dentry, inode);
3275         if (!err) {
3276                 ext4_mark_inode_dirty(handle, inode);
3277                 /* this can happen only for tmpfile being
3278                  * linked the first time
3279                  */
3280                 if (inode->i_nlink == 1)
3281                         ext4_orphan_del(handle, inode);
3282                 d_instantiate(dentry, inode);
3283         } else {
3284                 drop_nlink(inode);
3285                 iput(inode);
3286         }
3287         ext4_journal_stop(handle);
3288         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3289                 goto retry;
3290         return err;
3291 }
3292
3293
3294 /*
3295  * Try to find buffer head where contains the parent block.
3296  * It should be the inode block if it is inlined or the 1st block
3297  * if it is a normal dir.
3298  */
3299 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3300                                         struct inode *inode,
3301                                         int *retval,
3302                                         struct ext4_dir_entry_2 **parent_de,
3303                                         int *inlined)
3304 {
3305         struct buffer_head *bh;
3306
3307         if (!ext4_has_inline_data(inode)) {
3308                 bh = ext4_read_dirblock(inode, 0, EITHER);
3309                 if (IS_ERR(bh)) {
3310                         *retval = PTR_ERR(bh);
3311                         return NULL;
3312                 }
3313                 *parent_de = ext4_next_entry(
3314                                         (struct ext4_dir_entry_2 *)bh->b_data,
3315                                         inode->i_sb->s_blocksize);
3316                 return bh;
3317         }
3318
3319         *inlined = 1;
3320         return ext4_get_first_inline_block(inode, parent_de, retval);
3321 }
3322
3323 struct ext4_renament {
3324         struct inode *dir;
3325         struct dentry *dentry;
3326         struct inode *inode;
3327         bool is_dir;
3328         int dir_nlink_delta;
3329
3330         /* entry for "dentry" */
3331         struct buffer_head *bh;
3332         struct ext4_dir_entry_2 *de;
3333         int inlined;
3334
3335         /* entry for ".." in inode if it's a directory */
3336         struct buffer_head *dir_bh;
3337         struct ext4_dir_entry_2 *parent_de;
3338         int dir_inlined;
3339 };
3340
3341 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3342 {
3343         int retval;
3344
3345         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3346                                               &retval, &ent->parent_de,
3347                                               &ent->dir_inlined);
3348         if (!ent->dir_bh)
3349                 return retval;
3350         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3351                 return -EFSCORRUPTED;
3352         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3353         return ext4_journal_get_write_access(handle, ent->dir_bh);
3354 }
3355
3356 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3357                                   unsigned dir_ino)
3358 {
3359         int retval;
3360
3361         ent->parent_de->inode = cpu_to_le32(dir_ino);
3362         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3363         if (!ent->dir_inlined) {
3364                 if (is_dx(ent->inode)) {
3365                         retval = ext4_handle_dirty_dx_node(handle,
3366                                                            ent->inode,
3367                                                            ent->dir_bh);
3368                 } else {
3369                         retval = ext4_handle_dirty_dirent_node(handle,
3370                                                                ent->inode,
3371                                                                ent->dir_bh);
3372                 }
3373         } else {
3374                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3375         }
3376         if (retval) {
3377                 ext4_std_error(ent->dir->i_sb, retval);
3378                 return retval;
3379         }
3380         return 0;
3381 }
3382
3383 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3384                        unsigned ino, unsigned file_type)
3385 {
3386         int retval;
3387
3388         BUFFER_TRACE(ent->bh, "get write access");
3389         retval = ext4_journal_get_write_access(handle, ent->bh);
3390         if (retval)
3391                 return retval;
3392         ent->de->inode = cpu_to_le32(ino);
3393         if (ext4_has_feature_filetype(ent->dir->i_sb))
3394                 ent->de->file_type = file_type;
3395         ent->dir->i_version++;
3396         ent->dir->i_ctime = ent->dir->i_mtime =
3397                 current_time(ent->dir);
3398         ext4_mark_inode_dirty(handle, ent->dir);
3399         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3400         if (!ent->inlined) {
3401                 retval = ext4_handle_dirty_dirent_node(handle,
3402                                                        ent->dir, ent->bh);
3403                 if (unlikely(retval)) {
3404                         ext4_std_error(ent->dir->i_sb, retval);
3405                         return retval;
3406                 }
3407         }
3408         brelse(ent->bh);
3409         ent->bh = NULL;
3410
3411         return 0;
3412 }
3413
3414 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3415                                   const struct qstr *d_name)
3416 {
3417         int retval = -ENOENT;
3418         struct buffer_head *bh;
3419         struct ext4_dir_entry_2 *de;
3420
3421         bh = ext4_find_entry(dir, d_name, &de, NULL);
3422         if (IS_ERR(bh))
3423                 return PTR_ERR(bh);
3424         if (bh) {
3425                 retval = ext4_delete_entry(handle, dir, de, bh);
3426                 brelse(bh);
3427         }
3428         return retval;
3429 }
3430
3431 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3432                                int force_reread)
3433 {
3434         int retval;
3435         /*
3436          * ent->de could have moved from under us during htree split, so make
3437          * sure that we are deleting the right entry.  We might also be pointing
3438          * to a stale entry in the unused part of ent->bh so just checking inum
3439          * and the name isn't enough.
3440          */
3441         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3442             ent->de->name_len != ent->dentry->d_name.len ||
3443             strncmp(ent->de->name, ent->dentry->d_name.name,
3444                     ent->de->name_len) ||
3445             force_reread) {
3446                 retval = ext4_find_delete_entry(handle, ent->dir,
3447                                                 &ent->dentry->d_name);
3448         } else {
3449                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3450                 if (retval == -ENOENT) {
3451                         retval = ext4_find_delete_entry(handle, ent->dir,
3452                                                         &ent->dentry->d_name);
3453                 }
3454         }
3455
3456         if (retval) {
3457                 ext4_warning_inode(ent->dir,
3458                                    "Deleting old file: nlink %d, error=%d",
3459                                    ent->dir->i_nlink, retval);
3460         }
3461 }
3462
3463 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3464 {
3465         if (ent->dir_nlink_delta) {
3466                 if (ent->dir_nlink_delta == -1)
3467                         ext4_dec_count(handle, ent->dir);
3468                 else
3469                         ext4_inc_count(handle, ent->dir);
3470                 ext4_mark_inode_dirty(handle, ent->dir);
3471         }
3472 }
3473
3474 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3475                                               int credits, handle_t **h)
3476 {
3477         struct inode *wh;
3478         handle_t *handle;
3479         int retries = 0;
3480
3481         /*
3482          * for inode block, sb block, group summaries,
3483          * and inode bitmap
3484          */
3485         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3486                     EXT4_XATTR_TRANS_BLOCKS + 4);
3487 retry:
3488         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3489                                          &ent->dentry->d_name, 0, NULL,
3490                                          EXT4_HT_DIR, credits);
3491
3492         handle = ext4_journal_current_handle();
3493         if (IS_ERR(wh)) {
3494                 if (handle)
3495                         ext4_journal_stop(handle);
3496                 if (PTR_ERR(wh) == -ENOSPC &&
3497                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3498                         goto retry;
3499         } else {
3500                 *h = handle;
3501                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3502                 wh->i_op = &ext4_special_inode_operations;
3503         }
3504         return wh;
3505 }
3506
3507 /*
3508  * Anybody can rename anything with this: the permission checks are left to the
3509  * higher-level routines.
3510  *
3511  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3512  * while new_{dentry,inode) refers to the destination dentry/inode
3513  * This comes from rename(const char *oldpath, const char *newpath)
3514  */
3515 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3516                        struct inode *new_dir, struct dentry *new_dentry,
3517                        unsigned int flags)
3518 {
3519         handle_t *handle = NULL;
3520         struct ext4_renament old = {
3521                 .dir = old_dir,
3522                 .dentry = old_dentry,
3523                 .inode = d_inode(old_dentry),
3524         };
3525         struct ext4_renament new = {
3526                 .dir = new_dir,
3527                 .dentry = new_dentry,
3528                 .inode = d_inode(new_dentry),
3529         };
3530         int force_reread;
3531         int retval;
3532         struct inode *whiteout = NULL;
3533         int credits;
3534         u8 old_file_type;
3535
3536         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3537             (!projid_eq(EXT4_I(new_dir)->i_projid,
3538                         EXT4_I(old_dentry->d_inode)->i_projid)))
3539                 return -EXDEV;
3540
3541         if ((ext4_encrypted_inode(old_dir) &&
3542              !fscrypt_has_encryption_key(old_dir)) ||
3543             (ext4_encrypted_inode(new_dir) &&
3544              !fscrypt_has_encryption_key(new_dir)))
3545                 return -ENOKEY;
3546
3547         retval = dquot_initialize(old.dir);
3548         if (retval)
3549                 return retval;
3550         retval = dquot_initialize(new.dir);
3551         if (retval)
3552                 return retval;
3553
3554         /* Initialize quotas before so that eventual writes go
3555          * in separate transaction */
3556         if (new.inode) {
3557                 retval = dquot_initialize(new.inode);
3558                 if (retval)
3559                         return retval;
3560         }
3561
3562         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3563         if (IS_ERR(old.bh))
3564                 return PTR_ERR(old.bh);
3565         /*
3566          *  Check for inode number is _not_ due to possible IO errors.
3567          *  We might rmdir the source, keep it as pwd of some process
3568          *  and merrily kill the link to whatever was created under the
3569          *  same name. Goodbye sticky bit ;-<
3570          */
3571         retval = -ENOENT;
3572         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3573                 goto end_rename;
3574
3575         if ((old.dir != new.dir) &&
3576             ext4_encrypted_inode(new.dir) &&
3577             !fscrypt_has_permitted_context(new.dir, old.inode)) {
3578                 retval = -EPERM;
3579                 goto end_rename;
3580         }
3581
3582         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3583                                  &new.de, &new.inlined);
3584         if (IS_ERR(new.bh)) {
3585                 retval = PTR_ERR(new.bh);
3586                 new.bh = NULL;
3587                 goto end_rename;
3588         }
3589         if (new.bh) {
3590                 if (!new.inode) {
3591                         brelse(new.bh);
3592                         new.bh = NULL;
3593                 }
3594         }
3595         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3596                 ext4_alloc_da_blocks(old.inode);
3597
3598         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3599                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3600         if (!(flags & RENAME_WHITEOUT)) {
3601                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3602                 if (IS_ERR(handle)) {
3603                         retval = PTR_ERR(handle);
3604                         handle = NULL;
3605                         goto end_rename;
3606                 }
3607         } else {
3608                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3609                 if (IS_ERR(whiteout)) {
3610                         retval = PTR_ERR(whiteout);
3611                         whiteout = NULL;
3612                         goto end_rename;
3613                 }
3614         }
3615
3616         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3617                 ext4_handle_sync(handle);
3618
3619         if (S_ISDIR(old.inode->i_mode)) {
3620                 if (new.inode) {
3621                         retval = -ENOTEMPTY;
3622                         if (!ext4_empty_dir(new.inode))
3623                                 goto end_rename;
3624                 } else {
3625                         retval = -EMLINK;
3626                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3627                                 goto end_rename;
3628                 }
3629                 retval = ext4_rename_dir_prepare(handle, &old);
3630                 if (retval)
3631                         goto end_rename;
3632         }
3633         /*
3634          * If we're renaming a file within an inline_data dir and adding or
3635          * setting the new dirent causes a conversion from inline_data to
3636          * extents/blockmap, we need to force the dirent delete code to
3637          * re-read the directory, or else we end up trying to delete a dirent
3638          * from what is now the extent tree root (or a block map).
3639          */
3640         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3641                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3642
3643         old_file_type = old.de->file_type;
3644         if (whiteout) {
3645                 /*
3646                  * Do this before adding a new entry, so the old entry is sure
3647                  * to be still pointing to the valid old entry.
3648                  */
3649                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3650                                      EXT4_FT_CHRDEV);
3651                 if (retval)
3652                         goto end_rename;
3653                 ext4_mark_inode_dirty(handle, whiteout);
3654         }
3655         if (!new.bh) {
3656                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3657                 if (retval)
3658                         goto end_rename;
3659         } else {
3660                 retval = ext4_setent(handle, &new,
3661                                      old.inode->i_ino, old_file_type);
3662                 if (retval)
3663                         goto end_rename;
3664         }
3665         if (force_reread)
3666                 force_reread = !ext4_test_inode_flag(new.dir,
3667                                                      EXT4_INODE_INLINE_DATA);
3668
3669         /*
3670          * Like most other Unix systems, set the ctime for inodes on a
3671          * rename.
3672          */
3673         old.inode->i_ctime = current_time(old.inode);
3674         ext4_mark_inode_dirty(handle, old.inode);
3675
3676         if (!whiteout) {
3677                 /*
3678                  * ok, that's it
3679                  */
3680                 ext4_rename_delete(handle, &old, force_reread);
3681         }
3682
3683         if (new.inode) {
3684                 ext4_dec_count(handle, new.inode);
3685                 new.inode->i_ctime = current_time(new.inode);
3686         }
3687         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3688         ext4_update_dx_flag(old.dir);
3689         if (old.dir_bh) {
3690                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3691                 if (retval)
3692                         goto end_rename;
3693
3694                 ext4_dec_count(handle, old.dir);
3695                 if (new.inode) {
3696                         /* checked ext4_empty_dir above, can't have another
3697                          * parent, ext4_dec_count() won't work for many-linked
3698                          * dirs */
3699                         clear_nlink(new.inode);
3700                 } else {
3701                         ext4_inc_count(handle, new.dir);
3702                         ext4_update_dx_flag(new.dir);
3703                         ext4_mark_inode_dirty(handle, new.dir);
3704                 }
3705         }
3706         ext4_mark_inode_dirty(handle, old.dir);
3707         if (new.inode) {
3708                 ext4_mark_inode_dirty(handle, new.inode);
3709                 if (!new.inode->i_nlink)
3710                         ext4_orphan_add(handle, new.inode);
3711         }
3712         retval = 0;
3713
3714 end_rename:
3715         brelse(old.dir_bh);
3716         brelse(old.bh);
3717         brelse(new.bh);
3718         if (whiteout) {
3719                 if (retval)
3720                         drop_nlink(whiteout);
3721                 unlock_new_inode(whiteout);
3722                 iput(whiteout);
3723         }
3724         if (handle)
3725                 ext4_journal_stop(handle);
3726         return retval;
3727 }
3728
3729 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3730                              struct inode *new_dir, struct dentry *new_dentry)
3731 {
3732         handle_t *handle = NULL;
3733         struct ext4_renament old = {
3734                 .dir = old_dir,
3735                 .dentry = old_dentry,
3736                 .inode = d_inode(old_dentry),
3737         };
3738         struct ext4_renament new = {
3739                 .dir = new_dir,
3740                 .dentry = new_dentry,
3741                 .inode = d_inode(new_dentry),
3742         };
3743         u8 new_file_type;
3744         int retval;
3745         struct timespec ctime;
3746
3747         if ((ext4_encrypted_inode(old_dir) &&
3748              !fscrypt_has_encryption_key(old_dir)) ||
3749             (ext4_encrypted_inode(new_dir) &&
3750              !fscrypt_has_encryption_key(new_dir)))
3751                 return -ENOKEY;
3752
3753         if ((ext4_encrypted_inode(old_dir) ||
3754              ext4_encrypted_inode(new_dir)) &&
3755             (old_dir != new_dir) &&
3756             (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3757              !fscrypt_has_permitted_context(old_dir, new.inode)))
3758                 return -EPERM;
3759
3760         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3761              !projid_eq(EXT4_I(new_dir)->i_projid,
3762                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3763             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3764              !projid_eq(EXT4_I(old_dir)->i_projid,
3765                         EXT4_I(new_dentry->d_inode)->i_projid)))
3766                 return -EXDEV;
3767
3768         retval = dquot_initialize(old.dir);
3769         if (retval)
3770                 return retval;
3771         retval = dquot_initialize(new.dir);
3772         if (retval)
3773                 return retval;
3774
3775         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3776                                  &old.de, &old.inlined);
3777         if (IS_ERR(old.bh))
3778                 return PTR_ERR(old.bh);
3779         /*
3780          *  Check for inode number is _not_ due to possible IO errors.
3781          *  We might rmdir the source, keep it as pwd of some process
3782          *  and merrily kill the link to whatever was created under the
3783          *  same name. Goodbye sticky bit ;-<
3784          */
3785         retval = -ENOENT;
3786         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3787                 goto end_rename;
3788
3789         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3790                                  &new.de, &new.inlined);
3791         if (IS_ERR(new.bh)) {
3792                 retval = PTR_ERR(new.bh);
3793                 new.bh = NULL;
3794                 goto end_rename;
3795         }
3796
3797         /* RENAME_EXCHANGE case: old *and* new must both exist */
3798         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3799                 goto end_rename;
3800
3801         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3802                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3803                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3804         if (IS_ERR(handle)) {
3805                 retval = PTR_ERR(handle);
3806                 handle = NULL;
3807                 goto end_rename;
3808         }
3809
3810         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3811                 ext4_handle_sync(handle);
3812
3813         if (S_ISDIR(old.inode->i_mode)) {
3814                 old.is_dir = true;
3815                 retval = ext4_rename_dir_prepare(handle, &old);
3816                 if (retval)
3817                         goto end_rename;
3818         }
3819         if (S_ISDIR(new.inode->i_mode)) {
3820                 new.is_dir = true;
3821                 retval = ext4_rename_dir_prepare(handle, &new);
3822                 if (retval)
3823                         goto end_rename;
3824         }
3825
3826         /*
3827          * Other than the special case of overwriting a directory, parents'
3828          * nlink only needs to be modified if this is a cross directory rename.
3829          */
3830         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3831                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3832                 new.dir_nlink_delta = -old.dir_nlink_delta;
3833                 retval = -EMLINK;
3834                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3835                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3836                         goto end_rename;
3837         }
3838
3839         new_file_type = new.de->file_type;
3840         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3841         if (retval)
3842                 goto end_rename;
3843
3844         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3845         if (retval)
3846                 goto end_rename;
3847
3848         /*
3849          * Like most other Unix systems, set the ctime for inodes on a
3850          * rename.
3851          */
3852         ctime = current_time(old.inode);
3853         old.inode->i_ctime = ctime;
3854         new.inode->i_ctime = ctime;
3855         ext4_mark_inode_dirty(handle, old.inode);
3856         ext4_mark_inode_dirty(handle, new.inode);
3857
3858         if (old.dir_bh) {
3859                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3860                 if (retval)
3861                         goto end_rename;
3862         }
3863         if (new.dir_bh) {
3864                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3865                 if (retval)
3866                         goto end_rename;
3867         }
3868         ext4_update_dir_count(handle, &old);
3869         ext4_update_dir_count(handle, &new);
3870         retval = 0;
3871
3872 end_rename:
3873         brelse(old.dir_bh);
3874         brelse(new.dir_bh);
3875         brelse(old.bh);
3876         brelse(new.bh);
3877         if (handle)
3878                 ext4_journal_stop(handle);
3879         return retval;
3880 }
3881
3882 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3883                         struct inode *new_dir, struct dentry *new_dentry,
3884                         unsigned int flags)
3885 {
3886         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3887                 return -EIO;
3888
3889         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3890                 return -EINVAL;
3891
3892         if (flags & RENAME_EXCHANGE) {
3893                 return ext4_cross_rename(old_dir, old_dentry,
3894                                          new_dir, new_dentry);
3895         }
3896
3897         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3898 }
3899
3900 /*
3901  * directories can handle most operations...
3902  */
3903 const struct inode_operations ext4_dir_inode_operations = {
3904         .create         = ext4_create,
3905         .lookup         = ext4_lookup,
3906         .link           = ext4_link,
3907         .unlink         = ext4_unlink,
3908         .symlink        = ext4_symlink,
3909         .mkdir          = ext4_mkdir,
3910         .rmdir          = ext4_rmdir,
3911         .mknod          = ext4_mknod,
3912         .tmpfile        = ext4_tmpfile,
3913         .rename         = ext4_rename2,
3914         .setattr        = ext4_setattr,
3915         .listxattr      = ext4_listxattr,
3916         .get_acl        = ext4_get_acl,
3917         .set_acl        = ext4_set_acl,
3918         .fiemap         = ext4_fiemap,
3919 };
3920
3921 const struct inode_operations ext4_special_inode_operations = {
3922         .setattr        = ext4_setattr,
3923         .listxattr      = ext4_listxattr,
3924         .get_acl        = ext4_get_acl,
3925         .set_acl        = ext4_set_acl,
3926 };