fs/ext4/namei.c

   1 /*
   2  *  linux/fs/ext4/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *  Hash Tree Directory indexing (c)
  20  *      Daniel Phillips, 2001
  21  *  Hash Tree Directory indexing porting
  22  *      Christopher Li, 2002
  23  *  Hash Tree Directory indexing cleanup
  24  *      Theodore Ts'o, 2002
  25  */
  26
  27 #include <linux/fs.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/time.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/stat.h>
  32 #include <linux/string.h>
  33 #include <linux/quotaops.h>
  34 #include <linux/buffer_head.h>
  35 #include <linux/bio.h>
  36 #include "ext4.h"
  37 #include "ext4_jbd2.h"
  38
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 #include <trace/events/ext4.h>
  43 /*
  44  * define how far ahead to read directories while searching them.
  45  */
  46 #define NAMEI_RA_CHUNKS  2
  47 #define NAMEI_RA_BLOCKS  4
  48 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  49
  50 static struct buffer_head *ext4_append(handle_t *handle,
  51                                         struct inode *inode,
  52                                         ext4_lblk_t *block)
  53 {
  54         struct buffer_head *bh;
  55         int err;
  56
  57         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  58                      ((inode->i_size >> 10) >=
  59                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  60                 return ERR_PTR(-ENOSPC);
  61
  62         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  63
  64         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  65         if (IS_ERR(bh))
  66                 return bh;
  67         inode->i_size += inode->i_sb->s_blocksize;
  68         EXT4_I(inode)->i_disksize = inode->i_size;
  69         BUFFER_TRACE(bh, "get_write_access");
  70         err = ext4_journal_get_write_access(handle, bh);
  71         if (err) {
  72                 brelse(bh);
  73                 ext4_std_error(inode->i_sb, err);
  74                 return ERR_PTR(err);
  75         }
  76         return bh;
  77 }
  78
  79 static int ext4_dx_csum_verify(struct inode *inode,
  80                                struct ext4_dir_entry *dirent);
  81
  82 typedef enum {
  83         EITHER, INDEX, DIRENT
  84 } dirblock_type_t;
  85
  86 #define ext4_read_dirblock(inode, block, type) \
  87         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
  88
  89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  90                                                 ext4_lblk_t block,
  91                                                 dirblock_type_t type,
  92                                                 const char *func,
  93                                                 unsigned int line)
  94 {
  95         struct buffer_head *bh;
  96         struct ext4_dir_entry *dirent;
  97         int is_dx_block = 0;
  98
  99         bh = ext4_bread(NULL, inode, block, 0);
 100         if (IS_ERR(bh)) {
 101                 __ext4_warning(inode->i_sb, func, line,
 102                                "inode #%lu: lblock %lu: comm %s: "
 103                                "error %ld reading directory block",
 104                                inode->i_ino, (unsigned long)block,
 105                                current->comm, PTR_ERR(bh));
 106
 107                 return bh;
 108         }
 109         if (!bh) {
 110                 ext4_error_inode(inode, func, line, block,
 111                                  "Directory hole found");
 112                 return ERR_PTR(-EFSCORRUPTED);
 113         }
 114         dirent = (struct ext4_dir_entry *) bh->b_data;
 115         /* Determine whether or not we have an index block */
 116         if (is_dx(inode)) {
 117                 if (block == 0)
 118                         is_dx_block = 1;
 119                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 120                                                 inode->i_sb->s_blocksize) ==
 121                          inode->i_sb->s_blocksize)
 122                         is_dx_block = 1;
 123         }
 124         if (!is_dx_block && type == INDEX) {
 125                 ext4_error_inode(inode, func, line, block,
 126                        "directory leaf block found instead of index block");
 127                 return ERR_PTR(-EFSCORRUPTED);
 128         }
 129         if (!ext4_has_metadata_csum(inode->i_sb) ||
 130             buffer_verified(bh))
 131                 return bh;
 132
 133         /*
 134          * An empty leaf block can get mistaken for a index block; for
 135          * this reason, we can only check the index checksum when the
 136          * caller is sure it should be an index block.
 137          */
 138         if (is_dx_block && type == INDEX) {
 139                 if (ext4_dx_csum_verify(inode, dirent))
 140                         set_buffer_verified(bh);
 141                 else {
 142                         ext4_error_inode(inode, func, line, block,
 143                                          "Directory index failed checksum");
 144                         brelse(bh);
 145                         return ERR_PTR(-EFSBADCRC);
 146                 }
 147         }
 148         if (!is_dx_block) {
 149                 if (ext4_dirent_csum_verify(inode, dirent))
 150                         set_buffer_verified(bh);
 151                 else {
 152                         ext4_error_inode(inode, func, line, block,
 153                                          "Directory block failed checksum");
 154                         brelse(bh);
 155                         return ERR_PTR(-EFSBADCRC);
 156                 }
 157         }
 158         return bh;
 159 }
 160
 161 #ifndef assert
 162 #define assert(test) J_ASSERT(test)
 163 #endif
 164
 165 #ifdef DX_DEBUG
 166 #define dxtrace(command) command
 167 #else
 168 #define dxtrace(command)
 169 #endif
 170
 171 struct fake_dirent
 172 {
 173         __le32 inode;
 174         __le16 rec_len;
 175         u8 name_len;
 176         u8 file_type;
 177 };
 178
 179 struct dx_countlimit
 180 {
 181         __le16 limit;
 182         __le16 count;
 183 };
 184
 185 struct dx_entry
 186 {
 187         __le32 hash;
 188         __le32 block;
 189 };
 190
 191 /*
 192  * dx_root_info is laid out so that if it should somehow get overlaid by a
 193  * dirent the two low bits of the hash version will be zero.  Therefore, the
 194  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 195  */
 196
 197 struct dx_root
 198 {
 199         struct fake_dirent dot;
 200         char dot_name[4];
 201         struct fake_dirent dotdot;
 202         char dotdot_name[4];
 203         struct dx_root_info
 204         {
 205                 __le32 reserved_zero;
 206                 u8 hash_version;
 207                 u8 info_length; /* 8 */
 208                 u8 indirect_levels;
 209                 u8 unused_flags;
 210         }
 211         info;
 212         struct dx_entry entries[0];
 213 };
 214
 215 struct dx_node
 216 {
 217         struct fake_dirent fake;
 218         struct dx_entry entries[0];
 219 };
 220
 221
 222 struct dx_frame
 223 {
 224         struct buffer_head *bh;
 225         struct dx_entry *entries;
 226         struct dx_entry *at;
 227 };
 228
 229 struct dx_map_entry
 230 {
 231         u32 hash;
 232         u16 offs;
 233         u16 size;
 234 };
 235
 236 /*
 237  * This goes at the end of each htree block.
 238  */
 239 struct dx_tail {
 240         u32 dt_reserved;
 241         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 242 };
 243
 244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 246 static inline unsigned dx_get_hash(struct dx_entry *entry);
 247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 248 static unsigned dx_get_count(struct dx_entry *entries);
 249 static unsigned dx_get_limit(struct dx_entry *entries);
 250 static void dx_set_count(struct dx_entry *entries, unsigned value);
 251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 253 static unsigned dx_node_limit(struct inode *dir);
 254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 255                                  struct inode *dir,
 256                                  struct dx_hash_info *hinfo,
 257                                  struct dx_frame *frame);
 258 static void dx_release(struct dx_frame *frames);
 259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 260                        unsigned blocksize, struct dx_hash_info *hinfo,
 261                        struct dx_map_entry map[]);
 262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 264                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 266 static void dx_insert_block(struct dx_frame *frame,
 267                                         u32 hash, ext4_lblk_t block);
 268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 269                                  struct dx_frame *frame,
 270                                  struct dx_frame *frames,
 271                                  __u32 *start_hash);
 272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 273                 struct ext4_filename *fname,
 274                 struct ext4_dir_entry_2 **res_dir);
 275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 276                              struct 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  * Test whether a directory entry matches the filename being searched for.
1241  *
1242  * Return: %true if the directory entry matches, otherwise %false.
1243  */
1244 static inline bool ext4_match(const struct ext4_filename *fname,
1245                               const struct ext4_dir_entry_2 *de)
1246 {
1247         struct fscrypt_name f;
1248
1249         if (!de->inode)
1250                 return false;
1251
1252         f.usr_fname = fname->usr_fname;
1253         f.disk_name = fname->disk_name;
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255         f.crypto_buf = fname->crypto_buf;
1256 #endif
1257         return fscrypt_match_name(&f, de->name, de->name_len);
1258 }
1259
1260 /*
1261  * Returns 0 if not found, -1 on failure, and 1 on success
1262  */
1263 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1264                     struct inode *dir, struct ext4_filename *fname,
1265                     const struct qstr *d_name,
1266                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1267 {
1268         struct ext4_dir_entry_2 * de;
1269         char * dlimit;
1270         int de_len;
1271
1272         de = (struct ext4_dir_entry_2 *)search_buf;
1273         dlimit = search_buf + buf_size;
1274         while ((char *) de < dlimit) {
1275                 /* this code is executed quadratically often */
1276                 /* do minimal checking `by hand' */
1277                 if ((char *) de + de->name_len <= dlimit &&
1278                     ext4_match(fname, de)) {
1279                         /* found a match - just to be sure, do
1280                          * a full check */
1281                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1282                                                  bh->b_size, offset))
1283                                 return -1;
1284                         *res_dir = de;
1285                         return 1;
1286                 }
1287                 /* prevent looping on a bad block */
1288                 de_len = ext4_rec_len_from_disk(de->rec_len,
1289                                                 dir->i_sb->s_blocksize);
1290                 if (de_len <= 0)
1291                         return -1;
1292                 offset += de_len;
1293                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1294         }
1295         return 0;
1296 }
1297
1298 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1299                                struct ext4_dir_entry *de)
1300 {
1301         struct super_block *sb = dir->i_sb;
1302
1303         if (!is_dx(dir))
1304                 return 0;
1305         if (block == 0)
1306                 return 1;
1307         if (de->inode == 0 &&
1308             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1309                         sb->s_blocksize)
1310                 return 1;
1311         return 0;
1312 }
1313
1314 /*
1315  *      ext4_find_entry()
1316  *
1317  * finds an entry in the specified directory with the wanted name. It
1318  * returns the cache buffer in which the entry was found, and the entry
1319  * itself (as a parameter - res_dir). It does NOT read the inode of the
1320  * entry - you'll have to do that yourself if you want to.
1321  *
1322  * The returned buffer_head has ->b_count elevated.  The caller is expected
1323  * to brelse() it when appropriate.
1324  */
1325 static struct buffer_head * ext4_find_entry (struct inode *dir,
1326                                         const struct qstr *d_name,
1327                                         struct ext4_dir_entry_2 **res_dir,
1328                                         int *inlined)
1329 {
1330         struct super_block *sb;
1331         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1332         struct buffer_head *bh, *ret = NULL;
1333         ext4_lblk_t start, block, b;
1334         const u8 *name = d_name->name;
1335         int ra_max = 0;         /* Number of bh's in the readahead
1336                                    buffer, bh_use[] */
1337         int ra_ptr = 0;         /* Current index into readahead
1338                                    buffer */
1339         int num = 0;
1340         ext4_lblk_t  nblocks;
1341         int i, namelen, retval;
1342         struct ext4_filename fname;
1343
1344         *res_dir = NULL;
1345         sb = dir->i_sb;
1346         namelen = d_name->len;
1347         if (namelen > EXT4_NAME_LEN)
1348                 return NULL;
1349
1350         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1351         if (retval == -ENOENT)
1352                 return NULL;
1353         if (retval)
1354                 return ERR_PTR(retval);
1355
1356         if (ext4_has_inline_data(dir)) {
1357                 int has_inline_data = 1;
1358                 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1359                                              &has_inline_data);
1360                 if (has_inline_data) {
1361                         if (inlined)
1362                                 *inlined = 1;
1363                         goto cleanup_and_exit;
1364                 }
1365         }
1366
1367         if ((namelen <= 2) && (name[0] == '.') &&
1368             (name[1] == '.' || name[1] == '\0')) {
1369                 /*
1370                  * "." or ".." will only be in the first block
1371                  * NFS may look up ".."; "." should be handled by the VFS
1372                  */
1373                 block = start = 0;
1374                 nblocks = 1;
1375                 goto restart;
1376         }
1377         if (is_dx(dir)) {
1378                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1379                 /*
1380                  * On success, or if the error was file not found,
1381                  * return.  Otherwise, fall back to doing a search the
1382                  * old fashioned way.
1383                  */
1384                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1385                         goto cleanup_and_exit;
1386                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1387                                "falling back\n"));
1388         }
1389         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1390         start = EXT4_I(dir)->i_dir_start_lookup;
1391         if (start >= nblocks)
1392                 start = 0;
1393         block = start;
1394 restart:
1395         do {
1396                 /*
1397                  * We deal with the read-ahead logic here.
1398                  */
1399                 if (ra_ptr >= ra_max) {
1400                         /* Refill the readahead buffer */
1401                         ra_ptr = 0;
1402                         b = block;
1403                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1404                                 /*
1405                                  * Terminate if we reach the end of the
1406                                  * directory and must wrap, or if our
1407                                  * search has finished at this block.
1408                                  */
1409                                 if (b >= nblocks || (num && block == start)) {
1410                                         bh_use[ra_max] = NULL;
1411                                         break;
1412                                 }
1413                                 num++;
1414                                 bh = ext4_getblk(NULL, dir, b++, 0);
1415                                 if (IS_ERR(bh)) {
1416                                         if (ra_max == 0) {
1417                                                 ret = bh;
1418                                                 goto cleanup_and_exit;
1419                                         }
1420                                         break;
1421                                 }
1422                                 bh_use[ra_max] = bh;
1423                                 if (bh)
1424                                         ll_rw_block(REQ_OP_READ,
1425                                                     REQ_META | REQ_PRIO,
1426                                                     1, &bh);
1427                         }
1428                 }
1429                 if ((bh = bh_use[ra_ptr++]) == NULL)
1430                         goto next;
1431                 wait_on_buffer(bh);
1432                 if (!buffer_uptodate(bh)) {
1433                         /* read error, skip block & hope for the best */
1434                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1435                                          (unsigned long) block);
1436                         brelse(bh);
1437                         goto next;
1438                 }
1439                 if (!buffer_verified(bh) &&
1440                     !is_dx_internal_node(dir, block,
1441                                          (struct ext4_dir_entry *)bh->b_data) &&
1442                     !ext4_dirent_csum_verify(dir,
1443                                 (struct ext4_dir_entry *)bh->b_data)) {
1444                         EXT4_ERROR_INODE(dir, "checksumming directory "
1445                                          "block %lu", (unsigned long)block);
1446                         brelse(bh);
1447                         goto next;
1448                 }
1449                 set_buffer_verified(bh);
1450                 i = search_dirblock(bh, dir, &fname, d_name,
1451                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1452                 if (i == 1) {
1453                         EXT4_I(dir)->i_dir_start_lookup = block;
1454                         ret = bh;
1455                         goto cleanup_and_exit;
1456                 } else {
1457                         brelse(bh);
1458                         if (i < 0)
1459                                 goto cleanup_and_exit;
1460                 }
1461         next:
1462                 if (++block >= nblocks)
1463                         block = 0;
1464         } while (block != start);
1465
1466         /*
1467          * If the directory has grown while we were searching, then
1468          * search the last part of the directory before giving up.
1469          */
1470         block = nblocks;
1471         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1472         if (block < nblocks) {
1473                 start = 0;
1474                 goto restart;
1475         }
1476
1477 cleanup_and_exit:
1478         /* Clean up the read-ahead blocks */
1479         for (; ra_ptr < ra_max; ra_ptr++)
1480                 brelse(bh_use[ra_ptr]);
1481         ext4_fname_free_filename(&fname);
1482         return ret;
1483 }
1484
1485 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1486                         struct ext4_filename *fname,
1487                         struct ext4_dir_entry_2 **res_dir)
1488 {
1489         struct super_block * sb = dir->i_sb;
1490         struct dx_frame frames[2], *frame;
1491         const struct qstr *d_name = fname->usr_fname;
1492         struct buffer_head *bh;
1493         ext4_lblk_t block;
1494         int retval;
1495
1496 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1497         *res_dir = NULL;
1498 #endif
1499         frame = dx_probe(fname, dir, NULL, frames);
1500         if (IS_ERR(frame))
1501                 return (struct buffer_head *) frame;
1502         do {
1503                 block = dx_get_block(frame->at);
1504                 bh = ext4_read_dirblock(dir, block, DIRENT);
1505                 if (IS_ERR(bh))
1506                         goto errout;
1507
1508                 retval = search_dirblock(bh, dir, fname, d_name,
1509                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1510                                          res_dir);
1511                 if (retval == 1)
1512                         goto success;
1513                 brelse(bh);
1514                 if (retval == -1) {
1515                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1516                         goto errout;
1517                 }
1518
1519                 /* Check to see if we should continue to search */
1520                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1521                                                frames, NULL);
1522                 if (retval < 0) {
1523                         ext4_warning_inode(dir,
1524                                 "error %d reading directory index block",
1525                                 retval);
1526                         bh = ERR_PTR(retval);
1527                         goto errout;
1528                 }
1529         } while (retval == 1);
1530
1531         bh = NULL;
1532 errout:
1533         dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1534 success:
1535         dx_release(frames);
1536         return bh;
1537 }
1538
1539 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1540 {
1541         struct inode *inode;
1542         struct ext4_dir_entry_2 *de;
1543         struct buffer_head *bh;
1544
1545         if (ext4_encrypted_inode(dir)) {
1546                 int res = fscrypt_get_encryption_info(dir);
1547
1548                 /*
1549                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1550                  * created while the directory was encrypted and we
1551                  * have access to the key.
1552                  */
1553                 if (fscrypt_has_encryption_key(dir))
1554                         fscrypt_set_encrypted_dentry(dentry);
1555                 fscrypt_set_d_op(dentry);
1556                 if (res && res != -ENOKEY)
1557                         return ERR_PTR(res);
1558         }
1559
1560        if (dentry->d_name.len > EXT4_NAME_LEN)
1561                return ERR_PTR(-ENAMETOOLONG);
1562
1563         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1564         if (IS_ERR(bh))
1565                 return (struct dentry *) bh;
1566         inode = NULL;
1567         if (bh) {
1568                 __u32 ino = le32_to_cpu(de->inode);
1569                 brelse(bh);
1570                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1571                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1572                         return ERR_PTR(-EFSCORRUPTED);
1573                 }
1574                 if (unlikely(ino == dir->i_ino)) {
1575                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1576                                          dentry);
1577                         return ERR_PTR(-EFSCORRUPTED);
1578                 }
1579                 inode = ext4_iget_normal(dir->i_sb, ino);
1580                 if (inode == ERR_PTR(-ESTALE)) {
1581                         EXT4_ERROR_INODE(dir,
1582                                          "deleted inode referenced: %u",
1583                                          ino);
1584                         return ERR_PTR(-EFSCORRUPTED);
1585                 }
1586                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1587                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1588                     !fscrypt_has_permitted_context(dir, inode)) {
1589                         ext4_warning(inode->i_sb,
1590                                      "Inconsistent encryption contexts: %lu/%lu",
1591                                      dir->i_ino, inode->i_ino);
1592                         iput(inode);
1593                         return ERR_PTR(-EPERM);
1594                 }
1595         }
1596         return d_splice_alias(inode, dentry);
1597 }
1598
1599
1600 struct dentry *ext4_get_parent(struct dentry *child)
1601 {
1602         __u32 ino;
1603         static const struct qstr dotdot = QSTR_INIT("..", 2);
1604         struct ext4_dir_entry_2 * de;
1605         struct buffer_head *bh;
1606
1607         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1608         if (IS_ERR(bh))
1609                 return (struct dentry *) bh;
1610         if (!bh)
1611                 return ERR_PTR(-ENOENT);
1612         ino = le32_to_cpu(de->inode);
1613         brelse(bh);
1614
1615         if (!ext4_valid_inum(child->d_sb, ino)) {
1616                 EXT4_ERROR_INODE(d_inode(child),
1617                                  "bad parent inode number: %u", ino);
1618                 return ERR_PTR(-EFSCORRUPTED);
1619         }
1620
1621         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1622 }
1623
1624 /*
1625  * Move count entries from end of map between two memory locations.
1626  * Returns pointer to last entry moved.
1627  */
1628 static struct ext4_dir_entry_2 *
1629 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1630                 unsigned blocksize)
1631 {
1632         unsigned rec_len = 0;
1633
1634         while (count--) {
1635                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1636                                                 (from + (map->offs<<2));
1637                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1638                 memcpy (to, de, rec_len);
1639                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1640                                 ext4_rec_len_to_disk(rec_len, blocksize);
1641                 de->inode = 0;
1642                 map++;
1643                 to += rec_len;
1644         }
1645         return (struct ext4_dir_entry_2 *) (to - rec_len);
1646 }
1647
1648 /*
1649  * Compact each dir entry in the range to the minimal rec_len.
1650  * Returns pointer to last entry in range.
1651  */
1652 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1653 {
1654         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1655         unsigned rec_len = 0;
1656
1657         prev = to = de;
1658         while ((char*)de < base + blocksize) {
1659                 next = ext4_next_entry(de, blocksize);
1660                 if (de->inode && de->name_len) {
1661                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1662                         if (de > to)
1663                                 memmove(to, de, rec_len);
1664                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1665                         prev = to;
1666                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1667                 }
1668                 de = next;
1669         }
1670         return prev;
1671 }
1672
1673 /*
1674  * Split a full leaf block to make room for a new dir entry.
1675  * Allocate a new block, and move entries so that they are approx. equally full.
1676  * Returns pointer to de in block into which the new entry will be inserted.
1677  */
1678 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1679                         struct buffer_head **bh,struct dx_frame *frame,
1680                         struct dx_hash_info *hinfo)
1681 {
1682         unsigned blocksize = dir->i_sb->s_blocksize;
1683         unsigned count, continued;
1684         struct buffer_head *bh2;
1685         ext4_lblk_t newblock;
1686         u32 hash2;
1687         struct dx_map_entry *map;
1688         char *data1 = (*bh)->b_data, *data2;
1689         unsigned split, move, size;
1690         struct ext4_dir_entry_2 *de = NULL, *de2;
1691         struct ext4_dir_entry_tail *t;
1692         int     csum_size = 0;
1693         int     err = 0, i;
1694
1695         if (ext4_has_metadata_csum(dir->i_sb))
1696                 csum_size = sizeof(struct ext4_dir_entry_tail);
1697
1698         bh2 = ext4_append(handle, dir, &newblock);
1699         if (IS_ERR(bh2)) {
1700                 brelse(*bh);
1701                 *bh = NULL;
1702                 return (struct ext4_dir_entry_2 *) bh2;
1703         }
1704
1705         BUFFER_TRACE(*bh, "get_write_access");
1706         err = ext4_journal_get_write_access(handle, *bh);
1707         if (err)
1708                 goto journal_error;
1709
1710         BUFFER_TRACE(frame->bh, "get_write_access");
1711         err = ext4_journal_get_write_access(handle, frame->bh);
1712         if (err)
1713                 goto journal_error;
1714
1715         data2 = bh2->b_data;
1716
1717         /* create map in the end of data2 block */
1718         map = (struct dx_map_entry *) (data2 + blocksize);
1719         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1720                              blocksize, hinfo, map);
1721         map -= count;
1722         dx_sort_map(map, count);
1723         /* Split the existing block in the middle, size-wise */
1724         size = 0;
1725         move = 0;
1726         for (i = count-1; i >= 0; i--) {
1727                 /* is more than half of this entry in 2nd half of the block? */
1728                 if (size + map[i].size/2 > blocksize/2)
1729                         break;
1730                 size += map[i].size;
1731                 move++;
1732         }
1733         /* map index at which we will split */
1734         split = count - move;
1735         hash2 = map[split].hash;
1736         continued = hash2 == map[split - 1].hash;
1737         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1738                         (unsigned long)dx_get_block(frame->at),
1739                                         hash2, split, count-split));
1740
1741         /* Fancy dance to stay within two buffers */
1742         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1743                               blocksize);
1744         de = dx_pack_dirents(data1, blocksize);
1745         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1746                                            (char *) de,
1747                                            blocksize);
1748         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1749                                             (char *) de2,
1750                                             blocksize);
1751         if (csum_size) {
1752                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1753                 initialize_dirent_tail(t, blocksize);
1754
1755                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1756                 initialize_dirent_tail(t, blocksize);
1757         }
1758
1759         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1760                         blocksize, 1));
1761         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1762                         blocksize, 1));
1763
1764         /* Which block gets the new entry? */
1765         if (hinfo->hash >= hash2) {
1766                 swap(*bh, bh2);
1767                 de = de2;
1768         }
1769         dx_insert_block(frame, hash2 + continued, newblock);
1770         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1771         if (err)
1772                 goto journal_error;
1773         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1774         if (err)
1775                 goto journal_error;
1776         brelse(bh2);
1777         dxtrace(dx_show_index("frame", frame->entries));
1778         return de;
1779
1780 journal_error:
1781         brelse(*bh);
1782         brelse(bh2);
1783         *bh = NULL;
1784         ext4_std_error(dir->i_sb, err);
1785         return ERR_PTR(err);
1786 }
1787
1788 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1789                       struct buffer_head *bh,
1790                       void *buf, int buf_size,
1791                       struct ext4_filename *fname,
1792                       struct ext4_dir_entry_2 **dest_de)
1793 {
1794         struct ext4_dir_entry_2 *de;
1795         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1796         int nlen, rlen;
1797         unsigned int offset = 0;
1798         char *top;
1799
1800         de = (struct ext4_dir_entry_2 *)buf;
1801         top = buf + buf_size - reclen;
1802         while ((char *) de <= top) {
1803                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1804                                          buf, buf_size, offset))
1805                         return -EFSCORRUPTED;
1806                 if (ext4_match(fname, de))
1807                         return -EEXIST;
1808                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1809                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1810                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1811                         break;
1812                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1813                 offset += rlen;
1814         }
1815         if ((char *) de > top)
1816                 return -ENOSPC;
1817
1818         *dest_de = de;
1819         return 0;
1820 }
1821
1822 void ext4_insert_dentry(struct inode *inode,
1823                         struct ext4_dir_entry_2 *de,
1824                         int buf_size,
1825                         struct ext4_filename *fname)
1826 {
1827
1828         int nlen, rlen;
1829
1830         nlen = EXT4_DIR_REC_LEN(de->name_len);
1831         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1832         if (de->inode) {
1833                 struct ext4_dir_entry_2 *de1 =
1834                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1835                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1836                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1837                 de = de1;
1838         }
1839         de->file_type = EXT4_FT_UNKNOWN;
1840         de->inode = cpu_to_le32(inode->i_ino);
1841         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1842         de->name_len = fname_len(fname);
1843         memcpy(de->name, fname_name(fname), fname_len(fname));
1844 }
1845
1846 /*
1847  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1848  * it points to a directory entry which is guaranteed to be large
1849  * enough for new directory entry.  If de is NULL, then
1850  * add_dirent_to_buf will attempt search the directory block for
1851  * space.  It will return -ENOSPC if no space is available, and -EIO
1852  * and -EEXIST if directory entry already exists.
1853  */
1854 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1855                              struct inode *dir,
1856                              struct inode *inode, struct ext4_dir_entry_2 *de,
1857                              struct buffer_head *bh)
1858 {
1859         unsigned int    blocksize = dir->i_sb->s_blocksize;
1860         int             csum_size = 0;
1861         int             err;
1862
1863         if (ext4_has_metadata_csum(inode->i_sb))
1864                 csum_size = sizeof(struct ext4_dir_entry_tail);
1865
1866         if (!de) {
1867                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1868                                         blocksize - csum_size, fname, &de);
1869                 if (err)
1870                         return err;
1871         }
1872         BUFFER_TRACE(bh, "get_write_access");
1873         err = ext4_journal_get_write_access(handle, bh);
1874         if (err) {
1875                 ext4_std_error(dir->i_sb, err);
1876                 return err;
1877         }
1878
1879         /* By now the buffer is marked for journaling */
1880         ext4_insert_dentry(inode, de, blocksize, fname);
1881
1882         /*
1883          * XXX shouldn't update any times until successful
1884          * completion of syscall, but too many callers depend
1885          * on this.
1886          *
1887          * XXX similarly, too many callers depend on
1888          * ext4_new_inode() setting the times, but error
1889          * recovery deletes the inode, so the worst that can
1890          * happen is that the times are slightly out of date
1891          * and/or different from the directory change time.
1892          */
1893         dir->i_mtime = dir->i_ctime = current_time(dir);
1894         ext4_update_dx_flag(dir);
1895         dir->i_version++;
1896         ext4_mark_inode_dirty(handle, dir);
1897         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1898         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1899         if (err)
1900                 ext4_std_error(dir->i_sb, err);
1901         return 0;
1902 }
1903
1904 /*
1905  * This converts a one block unindexed directory to a 3 block indexed
1906  * directory, and adds the dentry to the indexed directory.
1907  */
1908 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1909                             struct inode *dir,
1910                             struct inode *inode, struct buffer_head *bh)
1911 {
1912         struct buffer_head *bh2;
1913         struct dx_root  *root;
1914         struct dx_frame frames[2], *frame;
1915         struct dx_entry *entries;
1916         struct ext4_dir_entry_2 *de, *de2;
1917         struct ext4_dir_entry_tail *t;
1918         char            *data1, *top;
1919         unsigned        len;
1920         int             retval;
1921         unsigned        blocksize;
1922         ext4_lblk_t  block;
1923         struct fake_dirent *fde;
1924         int csum_size = 0;
1925
1926         if (ext4_has_metadata_csum(inode->i_sb))
1927                 csum_size = sizeof(struct ext4_dir_entry_tail);
1928
1929         blocksize =  dir->i_sb->s_blocksize;
1930         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1931         BUFFER_TRACE(bh, "get_write_access");
1932         retval = ext4_journal_get_write_access(handle, bh);
1933         if (retval) {
1934                 ext4_std_error(dir->i_sb, retval);
1935                 brelse(bh);
1936                 return retval;
1937         }
1938         root = (struct dx_root *) bh->b_data;
1939
1940         /* The 0th block becomes the root, move the dirents out */
1941         fde = &root->dotdot;
1942         de = (struct ext4_dir_entry_2 *)((char *)fde +
1943                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1944         if ((char *) de >= (((char *) root) + blocksize)) {
1945                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1946                 brelse(bh);
1947                 return -EFSCORRUPTED;
1948         }
1949         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1950
1951         /* Allocate new block for the 0th block's dirents */
1952         bh2 = ext4_append(handle, dir, &block);
1953         if (IS_ERR(bh2)) {
1954                 brelse(bh);
1955                 return PTR_ERR(bh2);
1956         }
1957         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1958         data1 = bh2->b_data;
1959
1960         memcpy (data1, de, len);
1961         de = (struct ext4_dir_entry_2 *) data1;
1962         top = data1 + len;
1963         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1964                 de = de2;
1965         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1966                                            (char *) de,
1967                                            blocksize);
1968
1969         if (csum_size) {
1970                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1971                 initialize_dirent_tail(t, blocksize);
1972         }
1973
1974         /* Initialize the root; the dot dirents already exist */
1975         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1976         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1977                                            blocksize);
1978         memset (&root->info, 0, sizeof(root->info));
1979         root->info.info_length = sizeof(root->info);
1980         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1981         entries = root->entries;
1982         dx_set_block(entries, 1);
1983         dx_set_count(entries, 1);
1984         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1985
1986         /* Initialize as for dx_probe */
1987         fname->hinfo.hash_version = root->info.hash_version;
1988         if (fname->hinfo.hash_version <= DX_HASH_TEA)
1989                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1990         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1991         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1992
1993         memset(frames, 0, sizeof(frames));
1994         frame = frames;
1995         frame->entries = entries;
1996         frame->at = entries;
1997         frame->bh = bh;
1998
1999         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2000         if (retval)
2001                 goto out_frames;
2002         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2003         if (retval)
2004                 goto out_frames;
2005
2006         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2007         if (IS_ERR(de)) {
2008                 retval = PTR_ERR(de);
2009                 goto out_frames;
2010         }
2011
2012         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2013 out_frames:
2014         /*
2015          * Even if the block split failed, we have to properly write
2016          * out all the changes we did so far. Otherwise we can end up
2017          * with corrupted filesystem.
2018          */
2019         if (retval)
2020                 ext4_mark_inode_dirty(handle, dir);
2021         dx_release(frames);
2022         brelse(bh2);
2023         return retval;
2024 }
2025
2026 /*
2027  *      ext4_add_entry()
2028  *
2029  * adds a file entry to the specified directory, using the same
2030  * semantics as ext4_find_entry(). It returns NULL if it failed.
2031  *
2032  * NOTE!! The inode part of 'de' is left at 0 - which means you
2033  * may not sleep between calling this and putting something into
2034  * the entry, as someone else might have used it while you slept.
2035  */
2036 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2037                           struct inode *inode)
2038 {
2039         struct inode *dir = d_inode(dentry->d_parent);
2040         struct buffer_head *bh = NULL;
2041         struct ext4_dir_entry_2 *de;
2042         struct ext4_dir_entry_tail *t;
2043         struct super_block *sb;
2044         struct ext4_filename fname;
2045         int     retval;
2046         int     dx_fallback=0;
2047         unsigned blocksize;
2048         ext4_lblk_t block, blocks;
2049         int     csum_size = 0;
2050
2051         if (ext4_has_metadata_csum(inode->i_sb))
2052                 csum_size = sizeof(struct ext4_dir_entry_tail);
2053
2054         sb = dir->i_sb;
2055         blocksize = sb->s_blocksize;
2056         if (!dentry->d_name.len)
2057                 return -EINVAL;
2058
2059         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2060         if (retval)
2061                 return retval;
2062
2063         if (ext4_has_inline_data(dir)) {
2064                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2065                 if (retval < 0)
2066                         goto out;
2067                 if (retval == 1) {
2068                         retval = 0;
2069                         goto out;
2070                 }
2071         }
2072
2073         if (is_dx(dir)) {
2074                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2075                 if (!retval || (retval != ERR_BAD_DX_DIR))
2076                         goto out;
2077                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2078                 dx_fallback++;
2079                 ext4_mark_inode_dirty(handle, dir);
2080         }
2081         blocks = dir->i_size >> sb->s_blocksize_bits;
2082         for (block = 0; block < blocks; block++) {
2083                 bh = ext4_read_dirblock(dir, block, DIRENT);
2084                 if (IS_ERR(bh)) {
2085                         retval = PTR_ERR(bh);
2086                         bh = NULL;
2087                         goto out;
2088                 }
2089                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2090                                            NULL, bh);
2091                 if (retval != -ENOSPC)
2092                         goto out;
2093
2094                 if (blocks == 1 && !dx_fallback &&
2095                     ext4_has_feature_dir_index(sb)) {
2096                         retval = make_indexed_dir(handle, &fname, dir,
2097                                                   inode, bh);
2098                         bh = NULL; /* make_indexed_dir releases bh */
2099                         goto out;
2100                 }
2101                 brelse(bh);
2102         }
2103         bh = ext4_append(handle, dir, &block);
2104         if (IS_ERR(bh)) {
2105                 retval = PTR_ERR(bh);
2106                 bh = NULL;
2107                 goto out;
2108         }
2109         de = (struct ext4_dir_entry_2 *) bh->b_data;
2110         de->inode = 0;
2111         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2112
2113         if (csum_size) {
2114                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2115                 initialize_dirent_tail(t, blocksize);
2116         }
2117
2118         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2119 out:
2120         ext4_fname_free_filename(&fname);
2121         brelse(bh);
2122         if (retval == 0)
2123                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2124         return retval;
2125 }
2126
2127 /*
2128  * Returns 0 for success, or a negative error value
2129  */
2130 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2131                              struct inode *dir, struct inode *inode)
2132 {
2133         struct dx_frame frames[2], *frame;
2134         struct dx_entry *entries, *at;
2135         struct buffer_head *bh;
2136         struct super_block *sb = dir->i_sb;
2137         struct ext4_dir_entry_2 *de;
2138         int err;
2139
2140         frame = dx_probe(fname, dir, NULL, frames);
2141         if (IS_ERR(frame))
2142                 return PTR_ERR(frame);
2143         entries = frame->entries;
2144         at = frame->at;
2145         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2146         if (IS_ERR(bh)) {
2147                 err = PTR_ERR(bh);
2148                 bh = NULL;
2149                 goto cleanup;
2150         }
2151
2152         BUFFER_TRACE(bh, "get_write_access");
2153         err = ext4_journal_get_write_access(handle, bh);
2154         if (err)
2155                 goto journal_error;
2156
2157         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2158         if (err != -ENOSPC)
2159                 goto cleanup;
2160
2161         /* Block full, should compress but for now just split */
2162         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2163                        dx_get_count(entries), dx_get_limit(entries)));
2164         /* Need to split index? */
2165         if (dx_get_count(entries) == dx_get_limit(entries)) {
2166                 ext4_lblk_t newblock;
2167                 unsigned icount = dx_get_count(entries);
2168                 int levels = frame - frames;
2169                 struct dx_entry *entries2;
2170                 struct dx_node *node2;
2171                 struct buffer_head *bh2;
2172
2173                 if (levels && (dx_get_count(frames->entries) ==
2174                                dx_get_limit(frames->entries))) {
2175                         ext4_warning_inode(dir, "Directory index full!");
2176                         err = -ENOSPC;
2177                         goto cleanup;
2178                 }
2179                 bh2 = ext4_append(handle, dir, &newblock);
2180                 if (IS_ERR(bh2)) {
2181                         err = PTR_ERR(bh2);
2182                         goto cleanup;
2183                 }
2184                 node2 = (struct dx_node *)(bh2->b_data);
2185                 entries2 = node2->entries;
2186                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2187                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2188                                                            sb->s_blocksize);
2189                 BUFFER_TRACE(frame->bh, "get_write_access");
2190                 err = ext4_journal_get_write_access(handle, frame->bh);
2191                 if (err)
2192                         goto journal_error;
2193                 if (levels) {
2194                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2195                         unsigned hash2 = dx_get_hash(entries + icount1);
2196                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2197                                        icount1, icount2));
2198
2199                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2200                         err = ext4_journal_get_write_access(handle,
2201                                                              frames[0].bh);
2202                         if (err)
2203                                 goto journal_error;
2204
2205                         memcpy((char *) entries2, (char *) (entries + icount1),
2206                                icount2 * sizeof(struct dx_entry));
2207                         dx_set_count(entries, icount1);
2208                         dx_set_count(entries2, icount2);
2209                         dx_set_limit(entries2, dx_node_limit(dir));
2210
2211                         /* Which index block gets the new entry? */
2212                         if (at - entries >= icount1) {
2213                                 frame->at = at = at - entries - icount1 + entries2;
2214                                 frame->entries = entries = entries2;
2215                                 swap(frame->bh, bh2);
2216                         }
2217                         dx_insert_block(frames + 0, hash2, newblock);
2218                         dxtrace(dx_show_index("node", frames[1].entries));
2219                         dxtrace(dx_show_index("node",
2220                                ((struct dx_node *) bh2->b_data)->entries));
2221                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2222                         if (err)
2223                                 goto journal_error;
2224                         brelse (bh2);
2225                 } else {
2226                         dxtrace(printk(KERN_DEBUG
2227                                        "Creating second level index...\n"));
2228                         memcpy((char *) entries2, (char *) entries,
2229                                icount * sizeof(struct dx_entry));
2230                         dx_set_limit(entries2, dx_node_limit(dir));
2231
2232                         /* Set up root */
2233                         dx_set_count(entries, 1);
2234                         dx_set_block(entries + 0, newblock);
2235                         ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2236
2237                         /* Add new access path frame */
2238                         frame = frames + 1;
2239                         frame->at = at = at - entries + entries2;
2240                         frame->entries = entries = entries2;
2241                         frame->bh = bh2;
2242                         err = ext4_journal_get_write_access(handle,
2243                                                              frame->bh);
2244                         if (err)
2245                                 goto journal_error;
2246                 }
2247                 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2248                 if (err) {
2249                         ext4_std_error(inode->i_sb, err);
2250                         goto cleanup;
2251                 }
2252         }
2253         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2254         if (IS_ERR(de)) {
2255                 err = PTR_ERR(de);
2256                 goto cleanup;
2257         }
2258         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2259         goto cleanup;
2260
2261 journal_error:
2262         ext4_std_error(dir->i_sb, err);
2263 cleanup:
2264         brelse(bh);
2265         dx_release(frames);
2266         return err;
2267 }
2268
2269 /*
2270  * ext4_generic_delete_entry deletes a directory entry by merging it
2271  * with the previous entry
2272  */
2273 int ext4_generic_delete_entry(handle_t *handle,
2274                               struct inode *dir,
2275                               struct ext4_dir_entry_2 *de_del,
2276                               struct buffer_head *bh,
2277                               void *entry_buf,
2278                               int buf_size,
2279                               int csum_size)
2280 {
2281         struct ext4_dir_entry_2 *de, *pde;
2282         unsigned int blocksize = dir->i_sb->s_blocksize;
2283         int i;
2284
2285         i = 0;
2286         pde = NULL;
2287         de = (struct ext4_dir_entry_2 *)entry_buf;
2288         while (i < buf_size - csum_size) {
2289                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2290                                          bh->b_data, bh->b_size, i))
2291                         return -EFSCORRUPTED;
2292                 if (de == de_del)  {
2293                         if (pde)
2294                                 pde->rec_len = ext4_rec_len_to_disk(
2295                                         ext4_rec_len_from_disk(pde->rec_len,
2296                                                                blocksize) +
2297                                         ext4_rec_len_from_disk(de->rec_len,
2298                                                                blocksize),
2299                                         blocksize);
2300                         else
2301                                 de->inode = 0;
2302                         dir->i_version++;
2303                         return 0;
2304                 }
2305                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2306                 pde = de;
2307                 de = ext4_next_entry(de, blocksize);
2308         }
2309         return -ENOENT;
2310 }
2311
2312 static int ext4_delete_entry(handle_t *handle,
2313                              struct inode *dir,
2314                              struct ext4_dir_entry_2 *de_del,
2315                              struct buffer_head *bh)
2316 {
2317         int err, csum_size = 0;
2318
2319         if (ext4_has_inline_data(dir)) {
2320                 int has_inline_data = 1;
2321                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2322                                                &has_inline_data);
2323                 if (has_inline_data)
2324                         return err;
2325         }
2326
2327         if (ext4_has_metadata_csum(dir->i_sb))
2328                 csum_size = sizeof(struct ext4_dir_entry_tail);
2329
2330         BUFFER_TRACE(bh, "get_write_access");
2331         err = ext4_journal_get_write_access(handle, bh);
2332         if (unlikely(err))
2333                 goto out;
2334
2335         err = ext4_generic_delete_entry(handle, dir, de_del,
2336                                         bh, bh->b_data,
2337                                         dir->i_sb->s_blocksize, csum_size);
2338         if (err)
2339                 goto out;
2340
2341         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2342         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2343         if (unlikely(err))
2344                 goto out;
2345
2346         return 0;
2347 out:
2348         if (err != -ENOENT)
2349                 ext4_std_error(dir->i_sb, err);
2350         return err;
2351 }
2352
2353 /*
2354  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2355  * since this indicates that nlinks count was previously 1.
2356  */
2357 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2358 {
2359         inc_nlink(inode);
2360         if (is_dx(inode) && inode->i_nlink > 1) {
2361                 /* limit is 16-bit i_links_count */
2362                 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2363                         set_nlink(inode, 1);
2364                         ext4_set_feature_dir_nlink(inode->i_sb);
2365                 }
2366         }
2367 }
2368
2369 /*
2370  * If a directory had nlink == 1, then we should let it be 1. This indicates
2371  * directory has >EXT4_LINK_MAX subdirs.
2372  */
2373 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2374 {
2375         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2376                 drop_nlink(inode);
2377 }
2378
2379
2380 static int ext4_add_nondir(handle_t *handle,
2381                 struct dentry *dentry, struct inode *inode)
2382 {
2383         int err = ext4_add_entry(handle, dentry, inode);
2384         if (!err) {
2385                 ext4_mark_inode_dirty(handle, inode);
2386                 unlock_new_inode(inode);
2387                 d_instantiate(dentry, inode);
2388                 return 0;
2389         }
2390         drop_nlink(inode);
2391         unlock_new_inode(inode);
2392         iput(inode);
2393         return err;
2394 }
2395
2396 /*
2397  * By the time this is called, we already have created
2398  * the directory cache entry for the new file, but it
2399  * is so far negative - it has no inode.
2400  *
2401  * If the create succeeds, we fill in the inode information
2402  * with d_instantiate().
2403  */
2404 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2405                        bool excl)
2406 {
2407         handle_t *handle;
2408         struct inode *inode;
2409         int err, credits, retries = 0;
2410
2411         err = dquot_initialize(dir);
2412         if (err)
2413                 return err;
2414
2415         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2416                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2417 retry:
2418         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2419                                             NULL, EXT4_HT_DIR, credits);
2420         handle = ext4_journal_current_handle();
2421         err = PTR_ERR(inode);
2422         if (!IS_ERR(inode)) {
2423                 inode->i_op = &ext4_file_inode_operations;
2424                 inode->i_fop = &ext4_file_operations;
2425                 ext4_set_aops(inode);
2426                 err = ext4_add_nondir(handle, dentry, inode);
2427                 if (!err && IS_DIRSYNC(dir))
2428                         ext4_handle_sync(handle);
2429         }
2430         if (handle)
2431                 ext4_journal_stop(handle);
2432         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2433                 goto retry;
2434         return err;
2435 }
2436
2437 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2438                       umode_t mode, dev_t rdev)
2439 {
2440         handle_t *handle;
2441         struct inode *inode;
2442         int err, credits, retries = 0;
2443
2444         err = dquot_initialize(dir);
2445         if (err)
2446                 return err;
2447
2448         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2449                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2450 retry:
2451         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2452                                             NULL, EXT4_HT_DIR, credits);
2453         handle = ext4_journal_current_handle();
2454         err = PTR_ERR(inode);
2455         if (!IS_ERR(inode)) {
2456                 init_special_inode(inode, inode->i_mode, rdev);
2457                 inode->i_op = &ext4_special_inode_operations;
2458                 err = ext4_add_nondir(handle, dentry, inode);
2459                 if (!err && IS_DIRSYNC(dir))
2460                         ext4_handle_sync(handle);
2461         }
2462         if (handle)
2463                 ext4_journal_stop(handle);
2464         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2465                 goto retry;
2466         return err;
2467 }
2468
2469 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2470 {
2471         handle_t *handle;
2472         struct inode *inode;
2473         int err, retries = 0;
2474
2475         err = dquot_initialize(dir);
2476         if (err)
2477                 return err;
2478
2479 retry:
2480         inode = ext4_new_inode_start_handle(dir, mode,
2481                                             NULL, 0, NULL,
2482                                             EXT4_HT_DIR,
2483                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2484                           4 + EXT4_XATTR_TRANS_BLOCKS);
2485         handle = ext4_journal_current_handle();
2486         err = PTR_ERR(inode);
2487         if (!IS_ERR(inode)) {
2488                 inode->i_op = &ext4_file_inode_operations;
2489                 inode->i_fop = &ext4_file_operations;
2490                 ext4_set_aops(inode);
2491                 d_tmpfile(dentry, inode);
2492                 err = ext4_orphan_add(handle, inode);
2493                 if (err)
2494                         goto err_unlock_inode;
2495                 mark_inode_dirty(inode);
2496                 unlock_new_inode(inode);
2497         }
2498         if (handle)
2499                 ext4_journal_stop(handle);
2500         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2501                 goto retry;
2502         return err;
2503 err_unlock_inode:
2504         ext4_journal_stop(handle);
2505         unlock_new_inode(inode);
2506         return err;
2507 }
2508
2509 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2510                           struct ext4_dir_entry_2 *de,
2511                           int blocksize, int csum_size,
2512                           unsigned int parent_ino, int dotdot_real_len)
2513 {
2514         de->inode = cpu_to_le32(inode->i_ino);
2515         de->name_len = 1;
2516         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2517                                            blocksize);
2518         strcpy(de->name, ".");
2519         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2520
2521         de = ext4_next_entry(de, blocksize);
2522         de->inode = cpu_to_le32(parent_ino);
2523         de->name_len = 2;
2524         if (!dotdot_real_len)
2525                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2526                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2527                                         blocksize);
2528         else
2529                 de->rec_len = ext4_rec_len_to_disk(
2530                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2531         strcpy(de->name, "..");
2532         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2533
2534         return ext4_next_entry(de, blocksize);
2535 }
2536
2537 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2538                              struct inode *inode)
2539 {
2540         struct buffer_head *dir_block = NULL;
2541         struct ext4_dir_entry_2 *de;
2542         struct ext4_dir_entry_tail *t;
2543         ext4_lblk_t block = 0;
2544         unsigned int blocksize = dir->i_sb->s_blocksize;
2545         int csum_size = 0;
2546         int err;
2547
2548         if (ext4_has_metadata_csum(dir->i_sb))
2549                 csum_size = sizeof(struct ext4_dir_entry_tail);
2550
2551         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2552                 err = ext4_try_create_inline_dir(handle, dir, inode);
2553                 if (err < 0 && err != -ENOSPC)
2554                         goto out;
2555                 if (!err)
2556                         goto out;
2557         }
2558
2559         inode->i_size = 0;
2560         dir_block = ext4_append(handle, inode, &block);
2561         if (IS_ERR(dir_block))
2562                 return PTR_ERR(dir_block);
2563         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2564         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2565         set_nlink(inode, 2);
2566         if (csum_size) {
2567                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2568                 initialize_dirent_tail(t, blocksize);
2569         }
2570
2571         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2572         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2573         if (err)
2574                 goto out;
2575         set_buffer_verified(dir_block);
2576 out:
2577         brelse(dir_block);
2578         return err;
2579 }
2580
2581 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2582 {
2583         handle_t *handle;
2584         struct inode *inode;
2585         int err, credits, retries = 0;
2586
2587         if (EXT4_DIR_LINK_MAX(dir))
2588                 return -EMLINK;
2589
2590         err = dquot_initialize(dir);
2591         if (err)
2592                 return err;
2593
2594         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2595                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2596 retry:
2597         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2598                                             &dentry->d_name,
2599                                             0, NULL, EXT4_HT_DIR, credits);
2600         handle = ext4_journal_current_handle();
2601         err = PTR_ERR(inode);
2602         if (IS_ERR(inode))
2603                 goto out_stop;
2604
2605         inode->i_op = &ext4_dir_inode_operations;
2606         inode->i_fop = &ext4_dir_operations;
2607         err = ext4_init_new_dir(handle, dir, inode);
2608         if (err)
2609                 goto out_clear_inode;
2610         err = ext4_mark_inode_dirty(handle, inode);
2611         if (!err)
2612                 err = ext4_add_entry(handle, dentry, inode);
2613         if (err) {
2614 out_clear_inode:
2615                 clear_nlink(inode);
2616                 unlock_new_inode(inode);
2617                 ext4_mark_inode_dirty(handle, inode);
2618                 iput(inode);
2619                 goto out_stop;
2620         }
2621         ext4_inc_count(handle, dir);
2622         ext4_update_dx_flag(dir);
2623         err = ext4_mark_inode_dirty(handle, dir);
2624         if (err)
2625                 goto out_clear_inode;
2626         unlock_new_inode(inode);
2627         d_instantiate(dentry, inode);
2628         if (IS_DIRSYNC(dir))
2629                 ext4_handle_sync(handle);
2630
2631 out_stop:
2632         if (handle)
2633                 ext4_journal_stop(handle);
2634         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2635                 goto retry;
2636         return err;
2637 }
2638
2639 /*
2640  * routine to check that the specified directory is empty (for rmdir)
2641  */
2642 bool ext4_empty_dir(struct inode *inode)
2643 {
2644         unsigned int offset;
2645         struct buffer_head *bh;
2646         struct ext4_dir_entry_2 *de, *de1;
2647         struct super_block *sb;
2648
2649         if (ext4_has_inline_data(inode)) {
2650                 int has_inline_data = 1;
2651                 int ret;
2652
2653                 ret = empty_inline_dir(inode, &has_inline_data);
2654                 if (has_inline_data)
2655                         return ret;
2656         }
2657
2658         sb = inode->i_sb;
2659         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2660                 EXT4_ERROR_INODE(inode, "invalid size");
2661                 return true;
2662         }
2663         bh = ext4_read_dirblock(inode, 0, EITHER);
2664         if (IS_ERR(bh))
2665                 return true;
2666
2667         de = (struct ext4_dir_entry_2 *) bh->b_data;
2668         de1 = ext4_next_entry(de, sb->s_blocksize);
2669         if (le32_to_cpu(de->inode) != inode->i_ino ||
2670                         le32_to_cpu(de1->inode) == 0 ||
2671                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2672                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2673                 brelse(bh);
2674                 return true;
2675         }
2676         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2677                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2678         de = ext4_next_entry(de1, sb->s_blocksize);
2679         while (offset < inode->i_size) {
2680                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2681                         unsigned int lblock;
2682                         brelse(bh);
2683                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2684                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2685                         if (IS_ERR(bh))
2686                                 return true;
2687                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2688                 }
2689                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2690                                          bh->b_data, bh->b_size, offset)) {
2691                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2692                                                          sb->s_blocksize);
2693                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2694                         continue;
2695                 }
2696                 if (le32_to_cpu(de->inode)) {
2697                         brelse(bh);
2698                         return false;
2699                 }
2700                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2701                 de = ext4_next_entry(de, sb->s_blocksize);
2702         }
2703         brelse(bh);
2704         return true;
2705 }
2706
2707 /*
2708  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2709  * such inodes, starting at the superblock, in case we crash before the
2710  * file is closed/deleted, or in case the inode truncate spans multiple
2711  * transactions and the last transaction is not recovered after a crash.
2712  *
2713  * At filesystem recovery time, we walk this list deleting unlinked
2714  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2715  *
2716  * Orphan list manipulation functions must be called under i_mutex unless
2717  * we are just creating the inode or deleting it.
2718  */
2719 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2720 {
2721         struct super_block *sb = inode->i_sb;
2722         struct ext4_sb_info *sbi = EXT4_SB(sb);
2723         struct ext4_iloc iloc;
2724         int err = 0, rc;
2725         bool dirty = false;
2726
2727         if (!sbi->s_journal || is_bad_inode(inode))
2728                 return 0;
2729
2730         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2731                      !inode_is_locked(inode));
2732         /*
2733          * Exit early if inode already is on orphan list. This is a big speedup
2734          * since we don't have to contend on the global s_orphan_lock.
2735          */
2736         if (!list_empty(&EXT4_I(inode)->i_orphan))
2737                 return 0;
2738
2739         /*
2740          * Orphan handling is only valid for files with data blocks
2741          * being truncated, or files being unlinked. Note that we either
2742          * hold i_mutex, or the inode can not be referenced from outside,
2743          * so i_nlink should not be bumped due to race
2744          */
2745         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2746                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2747
2748         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2749         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2750         if (err)
2751                 goto out;
2752
2753         err = ext4_reserve_inode_write(handle, inode, &iloc);
2754         if (err)
2755                 goto out;
2756
2757         mutex_lock(&sbi->s_orphan_lock);
2758         /*
2759          * Due to previous errors inode may be already a part of on-disk
2760          * orphan list. If so skip on-disk list modification.
2761          */
2762         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2763             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2764                 /* Insert this inode at the head of the on-disk orphan list */
2765                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2766                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2767                 dirty = true;
2768         }
2769         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2770         mutex_unlock(&sbi->s_orphan_lock);
2771
2772         if (dirty) {
2773                 err = ext4_handle_dirty_super(handle, sb);
2774                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2775                 if (!err)
2776                         err = rc;
2777                 if (err) {
2778                         /*
2779                          * We have to remove inode from in-memory list if
2780                          * addition to on disk orphan list failed. Stray orphan
2781                          * list entries can cause panics at unmount time.
2782                          */
2783                         mutex_lock(&sbi->s_orphan_lock);
2784                         list_del_init(&EXT4_I(inode)->i_orphan);
2785                         mutex_unlock(&sbi->s_orphan_lock);
2786                 }
2787         }
2788         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2789         jbd_debug(4, "orphan inode %lu will point to %d\n",
2790                         inode->i_ino, NEXT_ORPHAN(inode));
2791 out:
2792         ext4_std_error(sb, err);
2793         return err;
2794 }
2795
2796 /*
2797  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2798  * of such inodes stored on disk, because it is finally being cleaned up.
2799  */
2800 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2801 {
2802         struct list_head *prev;
2803         struct ext4_inode_info *ei = EXT4_I(inode);
2804         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2805         __u32 ino_next;
2806         struct ext4_iloc iloc;
2807         int err = 0;
2808
2809         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2810                 return 0;
2811
2812         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2813                      !inode_is_locked(inode));
2814         /* Do this quick check before taking global s_orphan_lock. */
2815         if (list_empty(&ei->i_orphan))
2816                 return 0;
2817
2818         if (handle) {
2819                 /* Grab inode buffer early before taking global s_orphan_lock */
2820                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2821         }
2822
2823         mutex_lock(&sbi->s_orphan_lock);
2824         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2825
2826         prev = ei->i_orphan.prev;
2827         list_del_init(&ei->i_orphan);
2828
2829         /* If we're on an error path, we may not have a valid
2830          * transaction handle with which to update the orphan list on
2831          * disk, but we still need to remove the inode from the linked
2832          * list in memory. */
2833         if (!handle || err) {
2834                 mutex_unlock(&sbi->s_orphan_lock);
2835                 goto out_err;
2836         }
2837
2838         ino_next = NEXT_ORPHAN(inode);
2839         if (prev == &sbi->s_orphan) {
2840                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2841                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2842                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2843                 if (err) {
2844                         mutex_unlock(&sbi->s_orphan_lock);
2845                         goto out_brelse;
2846                 }
2847                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2848                 mutex_unlock(&sbi->s_orphan_lock);
2849                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2850         } else {
2851                 struct ext4_iloc iloc2;
2852                 struct inode *i_prev =
2853                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2854
2855                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2856                           i_prev->i_ino, ino_next);
2857                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2858                 if (err) {
2859                         mutex_unlock(&sbi->s_orphan_lock);
2860                         goto out_brelse;
2861                 }
2862                 NEXT_ORPHAN(i_prev) = ino_next;
2863                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2864                 mutex_unlock(&sbi->s_orphan_lock);
2865         }
2866         if (err)
2867                 goto out_brelse;
2868         NEXT_ORPHAN(inode) = 0;
2869         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2870 out_err:
2871         ext4_std_error(inode->i_sb, err);
2872         return err;
2873
2874 out_brelse:
2875         brelse(iloc.bh);
2876         goto out_err;
2877 }
2878
2879 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2880 {
2881         int retval;
2882         struct inode *inode;
2883         struct buffer_head *bh;
2884         struct ext4_dir_entry_2 *de;
2885         handle_t *handle = NULL;
2886
2887         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2888                 return -EIO;
2889
2890         /* Initialize quotas before so that eventual writes go in
2891          * separate transaction */
2892         retval = dquot_initialize(dir);
2893         if (retval)
2894                 return retval;
2895         retval = dquot_initialize(d_inode(dentry));
2896         if (retval)
2897                 return retval;
2898
2899         retval = -ENOENT;
2900         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2901         if (IS_ERR(bh))
2902                 return PTR_ERR(bh);
2903         if (!bh)
2904                 goto end_rmdir;
2905
2906         inode = d_inode(dentry);
2907
2908         retval = -EFSCORRUPTED;
2909         if (le32_to_cpu(de->inode) != inode->i_ino)
2910                 goto end_rmdir;
2911
2912         retval = -ENOTEMPTY;
2913         if (!ext4_empty_dir(inode))
2914                 goto end_rmdir;
2915
2916         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2917                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2918         if (IS_ERR(handle)) {
2919                 retval = PTR_ERR(handle);
2920                 handle = NULL;
2921                 goto end_rmdir;
2922         }
2923
2924         if (IS_DIRSYNC(dir))
2925                 ext4_handle_sync(handle);
2926
2927         retval = ext4_delete_entry(handle, dir, de, bh);
2928         if (retval)
2929                 goto end_rmdir;
2930         if (!EXT4_DIR_LINK_EMPTY(inode))
2931                 ext4_warning_inode(inode,
2932                              "empty directory '%.*s' has too many links (%u)",
2933                              dentry->d_name.len, dentry->d_name.name,
2934                              inode->i_nlink);
2935         inode->i_version++;
2936         clear_nlink(inode);
2937         /* There's no need to set i_disksize: the fact that i_nlink is
2938          * zero will ensure that the right thing happens during any
2939          * recovery. */
2940         inode->i_size = 0;
2941         ext4_orphan_add(handle, inode);
2942         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2943         ext4_mark_inode_dirty(handle, inode);
2944         ext4_dec_count(handle, dir);
2945         ext4_update_dx_flag(dir);
2946         ext4_mark_inode_dirty(handle, dir);
2947
2948 end_rmdir:
2949         brelse(bh);
2950         if (handle)
2951                 ext4_journal_stop(handle);
2952         return retval;
2953 }
2954
2955 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2956 {
2957         int retval;
2958         struct inode *inode;
2959         struct buffer_head *bh;
2960         struct ext4_dir_entry_2 *de;
2961         handle_t *handle = NULL;
2962
2963         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2964                 return -EIO;
2965
2966         trace_ext4_unlink_enter(dir, dentry);
2967         /* Initialize quotas before so that eventual writes go
2968          * in separate transaction */
2969         retval = dquot_initialize(dir);
2970         if (retval)
2971                 return retval;
2972         retval = dquot_initialize(d_inode(dentry));
2973         if (retval)
2974                 return retval;
2975
2976         retval = -ENOENT;
2977         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2978         if (IS_ERR(bh))
2979                 return PTR_ERR(bh);
2980         if (!bh)
2981                 goto end_unlink;
2982
2983         inode = d_inode(dentry);
2984
2985         retval = -EFSCORRUPTED;
2986         if (le32_to_cpu(de->inode) != inode->i_ino)
2987                 goto end_unlink;
2988
2989         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2990                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2991         if (IS_ERR(handle)) {
2992                 retval = PTR_ERR(handle);
2993                 handle = NULL;
2994                 goto end_unlink;
2995         }
2996
2997         if (IS_DIRSYNC(dir))
2998                 ext4_handle_sync(handle);
2999
3000         if (inode->i_nlink == 0) {
3001                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3002                                    dentry->d_name.len, dentry->d_name.name);
3003                 set_nlink(inode, 1);
3004         }
3005         retval = ext4_delete_entry(handle, dir, de, bh);
3006         if (retval)
3007                 goto end_unlink;
3008         dir->i_ctime = dir->i_mtime = current_time(dir);
3009         ext4_update_dx_flag(dir);
3010         ext4_mark_inode_dirty(handle, dir);
3011         drop_nlink(inode);
3012         if (!inode->i_nlink)
3013                 ext4_orphan_add(handle, inode);
3014         inode->i_ctime = current_time(inode);
3015         ext4_mark_inode_dirty(handle, inode);
3016
3017 end_unlink:
3018         brelse(bh);
3019         if (handle)
3020                 ext4_journal_stop(handle);
3021         trace_ext4_unlink_exit(dentry, retval);
3022         return retval;
3023 }
3024
3025 static int ext4_symlink(struct inode *dir,
3026                         struct dentry *dentry, const char *symname)
3027 {
3028         handle_t *handle;
3029         struct inode *inode;
3030         int err, len = strlen(symname);
3031         int credits;
3032         bool encryption_required;
3033         struct fscrypt_str disk_link;
3034         struct fscrypt_symlink_data *sd = NULL;
3035
3036         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3037                 return -EIO;
3038
3039         disk_link.len = len + 1;
3040         disk_link.name = (char *) symname;
3041
3042         encryption_required = (ext4_encrypted_inode(dir) ||
3043                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3044         if (encryption_required) {
3045                 err = fscrypt_get_encryption_info(dir);
3046                 if (err)
3047                         return err;
3048                 if (!fscrypt_has_encryption_key(dir))
3049                         return -ENOKEY;
3050                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3051                                  sizeof(struct fscrypt_symlink_data));
3052                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3053                 if (!sd)
3054                         return -ENOMEM;
3055         }
3056
3057         if (disk_link.len > dir->i_sb->s_blocksize) {
3058                 err = -ENAMETOOLONG;
3059                 goto err_free_sd;
3060         }
3061
3062         err = dquot_initialize(dir);
3063         if (err)
3064                 goto err_free_sd;
3065
3066         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3067                 /*
3068                  * For non-fast symlinks, we just allocate inode and put it on
3069                  * orphan list in the first transaction => we need bitmap,
3070                  * group descriptor, sb, inode block, quota blocks, and
3071                  * possibly selinux xattr blocks.
3072                  */
3073                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3074                           EXT4_XATTR_TRANS_BLOCKS;
3075         } else {
3076                 /*
3077                  * Fast symlink. We have to add entry to directory
3078                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3079                  * allocate new inode (bitmap, group descriptor, inode block,
3080                  * quota blocks, sb is already counted in previous macros).
3081                  */
3082                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3083                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3084         }
3085
3086         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3087                                             &dentry->d_name, 0, NULL,
3088                                             EXT4_HT_DIR, credits);
3089         handle = ext4_journal_current_handle();
3090         if (IS_ERR(inode)) {
3091                 if (handle)
3092                         ext4_journal_stop(handle);
3093                 err = PTR_ERR(inode);
3094                 goto err_free_sd;
3095         }
3096
3097         if (encryption_required) {
3098                 struct qstr istr;
3099                 struct fscrypt_str ostr =
3100                         FSTR_INIT(sd->encrypted_path, disk_link.len);
3101
3102                 istr.name = (const unsigned char *) symname;
3103                 istr.len = len;
3104                 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3105                 if (err)
3106                         goto err_drop_inode;
3107                 sd->len = cpu_to_le16(ostr.len);
3108                 disk_link.name = (char *) sd;
3109                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3110         }
3111
3112         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3113                 if (!encryption_required)
3114                         inode->i_op = &ext4_symlink_inode_operations;
3115                 inode_nohighmem(inode);
3116                 ext4_set_aops(inode);
3117                 /*
3118                  * We cannot call page_symlink() with transaction started
3119                  * because it calls into ext4_write_begin() which can wait
3120                  * for transaction commit if we are running out of space
3121                  * and thus we deadlock. So we have to stop transaction now
3122                  * and restart it when symlink contents is written.
3123                  *
3124                  * To keep fs consistent in case of crash, we have to put inode
3125                  * to orphan list in the mean time.
3126                  */
3127                 drop_nlink(inode);
3128                 err = ext4_orphan_add(handle, inode);
3129                 ext4_journal_stop(handle);
3130                 handle = NULL;
3131                 if (err)
3132                         goto err_drop_inode;
3133                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3134                 if (err)
3135                         goto err_drop_inode;
3136                 /*
3137                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3138                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3139                  */
3140                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3141                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3142                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3143                 if (IS_ERR(handle)) {
3144                         err = PTR_ERR(handle);
3145                         handle = NULL;
3146                         goto err_drop_inode;
3147                 }
3148                 set_nlink(inode, 1);
3149                 err = ext4_orphan_del(handle, inode);
3150                 if (err)
3151                         goto err_drop_inode;
3152         } else {
3153                 /* clear the extent format for fast symlink */
3154                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3155                 if (!encryption_required) {
3156                         inode->i_op = &ext4_fast_symlink_inode_operations;
3157                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3158                 }
3159                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3160                        disk_link.len);
3161                 inode->i_size = disk_link.len - 1;
3162         }
3163         EXT4_I(inode)->i_disksize = inode->i_size;
3164         err = ext4_add_nondir(handle, dentry, inode);
3165         if (!err && IS_DIRSYNC(dir))
3166                 ext4_handle_sync(handle);
3167
3168         if (handle)
3169                 ext4_journal_stop(handle);
3170         kfree(sd);
3171         return err;
3172 err_drop_inode:
3173         if (handle)
3174                 ext4_journal_stop(handle);
3175         clear_nlink(inode);
3176         unlock_new_inode(inode);
3177         iput(inode);
3178 err_free_sd:
3179         kfree(sd);
3180         return err;
3181 }
3182
3183 static int ext4_link(struct dentry *old_dentry,
3184                      struct inode *dir, struct dentry *dentry)
3185 {
3186         handle_t *handle;
3187         struct inode *inode = d_inode(old_dentry);
3188         int err, retries = 0;
3189
3190         if (inode->i_nlink >= EXT4_LINK_MAX)
3191                 return -EMLINK;
3192         if (ext4_encrypted_inode(dir) &&
3193                         !fscrypt_has_permitted_context(dir, inode))
3194                 return -EPERM;
3195
3196        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3197            (!projid_eq(EXT4_I(dir)->i_projid,
3198                        EXT4_I(old_dentry->d_inode)->i_projid)))
3199                 return -EXDEV;
3200
3201         err = dquot_initialize(dir);
3202         if (err)
3203                 return err;
3204
3205 retry:
3206         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3207                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3208                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3209         if (IS_ERR(handle))
3210                 return PTR_ERR(handle);
3211
3212         if (IS_DIRSYNC(dir))
3213                 ext4_handle_sync(handle);
3214
3215         inode->i_ctime = current_time(inode);
3216         ext4_inc_count(handle, inode);
3217         ihold(inode);
3218
3219         err = ext4_add_entry(handle, dentry, inode);
3220         if (!err) {
3221                 ext4_mark_inode_dirty(handle, inode);
3222                 /* this can happen only for tmpfile being
3223                  * linked the first time
3224                  */
3225                 if (inode->i_nlink == 1)
3226                         ext4_orphan_del(handle, inode);
3227                 d_instantiate(dentry, inode);
3228         } else {
3229                 drop_nlink(inode);
3230                 iput(inode);
3231         }
3232         ext4_journal_stop(handle);
3233         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3234                 goto retry;
3235         return err;
3236 }
3237
3238
3239 /*
3240  * Try to find buffer head where contains the parent block.
3241  * It should be the inode block if it is inlined or the 1st block
3242  * if it is a normal dir.
3243  */
3244 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3245                                         struct inode *inode,
3246                                         int *retval,
3247                                         struct ext4_dir_entry_2 **parent_de,
3248                                         int *inlined)
3249 {
3250         struct buffer_head *bh;
3251
3252         if (!ext4_has_inline_data(inode)) {
3253                 bh = ext4_read_dirblock(inode, 0, EITHER);
3254                 if (IS_ERR(bh)) {
3255                         *retval = PTR_ERR(bh);
3256                         return NULL;
3257                 }
3258                 *parent_de = ext4_next_entry(
3259                                         (struct ext4_dir_entry_2 *)bh->b_data,
3260                                         inode->i_sb->s_blocksize);
3261                 return bh;
3262         }
3263
3264         *inlined = 1;
3265         return ext4_get_first_inline_block(inode, parent_de, retval);
3266 }
3267
3268 struct ext4_renament {
3269         struct inode *dir;
3270         struct dentry *dentry;
3271         struct inode *inode;
3272         bool is_dir;
3273         int dir_nlink_delta;
3274
3275         /* entry for "dentry" */
3276         struct buffer_head *bh;
3277         struct ext4_dir_entry_2 *de;
3278         int inlined;
3279
3280         /* entry for ".." in inode if it's a directory */
3281         struct buffer_head *dir_bh;
3282         struct ext4_dir_entry_2 *parent_de;
3283         int dir_inlined;
3284 };
3285
3286 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3287 {
3288         int retval;
3289
3290         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3291                                               &retval, &ent->parent_de,
3292                                               &ent->dir_inlined);
3293         if (!ent->dir_bh)
3294                 return retval;
3295         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3296                 return -EFSCORRUPTED;
3297         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3298         return ext4_journal_get_write_access(handle, ent->dir_bh);
3299 }
3300
3301 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3302                                   unsigned dir_ino)
3303 {
3304         int retval;
3305
3306         ent->parent_de->inode = cpu_to_le32(dir_ino);
3307         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3308         if (!ent->dir_inlined) {
3309                 if (is_dx(ent->inode)) {
3310                         retval = ext4_handle_dirty_dx_node(handle,
3311                                                            ent->inode,
3312                                                            ent->dir_bh);
3313                 } else {
3314                         retval = ext4_handle_dirty_dirent_node(handle,
3315                                                                ent->inode,
3316                                                                ent->dir_bh);
3317                 }
3318         } else {
3319                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3320         }
3321         if (retval) {
3322                 ext4_std_error(ent->dir->i_sb, retval);
3323                 return retval;
3324         }
3325         return 0;
3326 }
3327
3328 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3329                        unsigned ino, unsigned file_type)
3330 {
3331         int retval;
3332
3333         BUFFER_TRACE(ent->bh, "get write access");
3334         retval = ext4_journal_get_write_access(handle, ent->bh);
3335         if (retval)
3336                 return retval;
3337         ent->de->inode = cpu_to_le32(ino);
3338         if (ext4_has_feature_filetype(ent->dir->i_sb))
3339                 ent->de->file_type = file_type;
3340         ent->dir->i_version++;
3341         ent->dir->i_ctime = ent->dir->i_mtime =
3342                 current_time(ent->dir);
3343         ext4_mark_inode_dirty(handle, ent->dir);
3344         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3345         if (!ent->inlined) {
3346                 retval = ext4_handle_dirty_dirent_node(handle,
3347                                                        ent->dir, ent->bh);
3348                 if (unlikely(retval)) {
3349                         ext4_std_error(ent->dir->i_sb, retval);
3350                         return retval;
3351                 }
3352         }
3353         brelse(ent->bh);
3354         ent->bh = NULL;
3355
3356         return 0;
3357 }
3358
3359 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3360                                   const struct qstr *d_name)
3361 {
3362         int retval = -ENOENT;
3363         struct buffer_head *bh;
3364         struct ext4_dir_entry_2 *de;
3365
3366         bh = ext4_find_entry(dir, d_name, &de, NULL);
3367         if (IS_ERR(bh))
3368                 return PTR_ERR(bh);
3369         if (bh) {
3370                 retval = ext4_delete_entry(handle, dir, de, bh);
3371                 brelse(bh);
3372         }
3373         return retval;
3374 }
3375
3376 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3377                                int force_reread)
3378 {
3379         int retval;
3380         /*
3381          * ent->de could have moved from under us during htree split, so make
3382          * sure that we are deleting the right entry.  We might also be pointing
3383          * to a stale entry in the unused part of ent->bh so just checking inum
3384          * and the name isn't enough.
3385          */
3386         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3387             ent->de->name_len != ent->dentry->d_name.len ||
3388             strncmp(ent->de->name, ent->dentry->d_name.name,
3389                     ent->de->name_len) ||
3390             force_reread) {
3391                 retval = ext4_find_delete_entry(handle, ent->dir,
3392                                                 &ent->dentry->d_name);
3393         } else {
3394                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3395                 if (retval == -ENOENT) {
3396                         retval = ext4_find_delete_entry(handle, ent->dir,
3397                                                         &ent->dentry->d_name);
3398                 }
3399         }
3400
3401         if (retval) {
3402                 ext4_warning_inode(ent->dir,
3403                                    "Deleting old file: nlink %d, error=%d",
3404                                    ent->dir->i_nlink, retval);
3405         }
3406 }
3407
3408 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3409 {
3410         if (ent->dir_nlink_delta) {
3411                 if (ent->dir_nlink_delta == -1)
3412                         ext4_dec_count(handle, ent->dir);
3413                 else
3414                         ext4_inc_count(handle, ent->dir);
3415                 ext4_mark_inode_dirty(handle, ent->dir);
3416         }
3417 }
3418
3419 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3420                                               int credits, handle_t **h)
3421 {
3422         struct inode *wh;
3423         handle_t *handle;
3424         int retries = 0;
3425
3426         /*
3427          * for inode block, sb block, group summaries,
3428          * and inode bitmap
3429          */
3430         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3431                     EXT4_XATTR_TRANS_BLOCKS + 4);
3432 retry:
3433         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3434                                          &ent->dentry->d_name, 0, NULL,
3435                                          EXT4_HT_DIR, credits);
3436
3437         handle = ext4_journal_current_handle();
3438         if (IS_ERR(wh)) {
3439                 if (handle)
3440                         ext4_journal_stop(handle);
3441                 if (PTR_ERR(wh) == -ENOSPC &&
3442                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3443                         goto retry;
3444         } else {
3445                 *h = handle;
3446                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3447                 wh->i_op = &ext4_special_inode_operations;
3448         }
3449         return wh;
3450 }
3451
3452 /*
3453  * Anybody can rename anything with this: the permission checks are left to the
3454  * higher-level routines.
3455  *
3456  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3457  * while new_{dentry,inode) refers to the destination dentry/inode
3458  * This comes from rename(const char *oldpath, const char *newpath)
3459  */
3460 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3461                        struct inode *new_dir, struct dentry *new_dentry,
3462                        unsigned int flags)
3463 {
3464         handle_t *handle = NULL;
3465         struct ext4_renament old = {
3466                 .dir = old_dir,
3467                 .dentry = old_dentry,
3468                 .inode = d_inode(old_dentry),
3469         };
3470         struct ext4_renament new = {
3471                 .dir = new_dir,
3472                 .dentry = new_dentry,
3473                 .inode = d_inode(new_dentry),
3474         };
3475         int force_reread;
3476         int retval;
3477         struct inode *whiteout = NULL;
3478         int credits;
3479         u8 old_file_type;
3480
3481         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3482             (!projid_eq(EXT4_I(new_dir)->i_projid,
3483                         EXT4_I(old_dentry->d_inode)->i_projid)))
3484                 return -EXDEV;
3485
3486         if ((ext4_encrypted_inode(old_dir) &&
3487              !fscrypt_has_encryption_key(old_dir)) ||
3488             (ext4_encrypted_inode(new_dir) &&
3489              !fscrypt_has_encryption_key(new_dir)))
3490                 return -ENOKEY;
3491
3492         retval = dquot_initialize(old.dir);
3493         if (retval)
3494                 return retval;
3495         retval = dquot_initialize(new.dir);
3496         if (retval)
3497                 return retval;
3498
3499         /* Initialize quotas before so that eventual writes go
3500          * in separate transaction */
3501         if (new.inode) {
3502                 retval = dquot_initialize(new.inode);
3503                 if (retval)
3504                         return retval;
3505         }
3506
3507         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3508         if (IS_ERR(old.bh))
3509                 return PTR_ERR(old.bh);
3510         /*
3511          *  Check for inode number is _not_ due to possible IO errors.
3512          *  We might rmdir the source, keep it as pwd of some process
3513          *  and merrily kill the link to whatever was created under the
3514          *  same name. Goodbye sticky bit ;-<
3515          */
3516         retval = -ENOENT;
3517         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3518                 goto end_rename;
3519
3520         if ((old.dir != new.dir) &&
3521             ext4_encrypted_inode(new.dir) &&
3522             !fscrypt_has_permitted_context(new.dir, old.inode)) {
3523                 retval = -EPERM;
3524                 goto end_rename;
3525         }
3526
3527         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3528                                  &new.de, &new.inlined);
3529         if (IS_ERR(new.bh)) {
3530                 retval = PTR_ERR(new.bh);
3531                 new.bh = NULL;
3532                 goto end_rename;
3533         }
3534         if (new.bh) {
3535                 if (!new.inode) {
3536                         brelse(new.bh);
3537                         new.bh = NULL;
3538                 }
3539         }
3540         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3541                 ext4_alloc_da_blocks(old.inode);
3542
3543         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3544                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3545         if (!(flags & RENAME_WHITEOUT)) {
3546                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3547                 if (IS_ERR(handle)) {
3548                         retval = PTR_ERR(handle);
3549                         handle = NULL;
3550                         goto end_rename;
3551                 }
3552         } else {
3553                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3554                 if (IS_ERR(whiteout)) {
3555                         retval = PTR_ERR(whiteout);
3556                         whiteout = NULL;
3557                         goto end_rename;
3558                 }
3559         }
3560
3561         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3562                 ext4_handle_sync(handle);
3563
3564         if (S_ISDIR(old.inode->i_mode)) {
3565                 if (new.inode) {
3566                         retval = -ENOTEMPTY;
3567                         if (!ext4_empty_dir(new.inode))
3568                                 goto end_rename;
3569                 } else {
3570                         retval = -EMLINK;
3571                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3572                                 goto end_rename;
3573                 }
3574                 retval = ext4_rename_dir_prepare(handle, &old);
3575                 if (retval)
3576                         goto end_rename;
3577         }
3578         /*
3579          * If we're renaming a file within an inline_data dir and adding or
3580          * setting the new dirent causes a conversion from inline_data to
3581          * extents/blockmap, we need to force the dirent delete code to
3582          * re-read the directory, or else we end up trying to delete a dirent
3583          * from what is now the extent tree root (or a block map).
3584          */
3585         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3586                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3587
3588         old_file_type = old.de->file_type;
3589         if (whiteout) {
3590                 /*
3591                  * Do this before adding a new entry, so the old entry is sure
3592                  * to be still pointing to the valid old entry.
3593                  */
3594                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3595                                      EXT4_FT_CHRDEV);
3596                 if (retval)
3597                         goto end_rename;
3598                 ext4_mark_inode_dirty(handle, whiteout);
3599         }
3600         if (!new.bh) {
3601                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3602                 if (retval)
3603                         goto end_rename;
3604         } else {
3605                 retval = ext4_setent(handle, &new,
3606                                      old.inode->i_ino, old_file_type);
3607                 if (retval)
3608                         goto end_rename;
3609         }
3610         if (force_reread)
3611                 force_reread = !ext4_test_inode_flag(new.dir,
3612                                                      EXT4_INODE_INLINE_DATA);
3613
3614         /*
3615          * Like most other Unix systems, set the ctime for inodes on a
3616          * rename.
3617          */
3618         old.inode->i_ctime = current_time(old.inode);
3619         ext4_mark_inode_dirty(handle, old.inode);
3620
3621         if (!whiteout) {
3622                 /*
3623                  * ok, that's it
3624                  */
3625                 ext4_rename_delete(handle, &old, force_reread);
3626         }
3627
3628         if (new.inode) {
3629                 ext4_dec_count(handle, new.inode);
3630                 new.inode->i_ctime = current_time(new.inode);
3631         }
3632         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3633         ext4_update_dx_flag(old.dir);
3634         if (old.dir_bh) {
3635                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3636                 if (retval)
3637                         goto end_rename;
3638
3639                 ext4_dec_count(handle, old.dir);
3640                 if (new.inode) {
3641                         /* checked ext4_empty_dir above, can't have another
3642                          * parent, ext4_dec_count() won't work for many-linked
3643                          * dirs */
3644                         clear_nlink(new.inode);
3645                 } else {
3646                         ext4_inc_count(handle, new.dir);
3647                         ext4_update_dx_flag(new.dir);
3648                         ext4_mark_inode_dirty(handle, new.dir);
3649                 }
3650         }
3651         ext4_mark_inode_dirty(handle, old.dir);
3652         if (new.inode) {
3653                 ext4_mark_inode_dirty(handle, new.inode);
3654                 if (!new.inode->i_nlink)
3655                         ext4_orphan_add(handle, new.inode);
3656         }
3657         retval = 0;
3658
3659 end_rename:
3660         brelse(old.dir_bh);
3661         brelse(old.bh);
3662         brelse(new.bh);
3663         if (whiteout) {
3664                 if (retval)
3665                         drop_nlink(whiteout);
3666                 unlock_new_inode(whiteout);
3667                 iput(whiteout);
3668         }
3669         if (handle)
3670                 ext4_journal_stop(handle);
3671         return retval;
3672 }
3673
3674 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3675                              struct inode *new_dir, struct dentry *new_dentry)
3676 {
3677         handle_t *handle = NULL;
3678         struct ext4_renament old = {
3679                 .dir = old_dir,
3680                 .dentry = old_dentry,
3681                 .inode = d_inode(old_dentry),
3682         };
3683         struct ext4_renament new = {
3684                 .dir = new_dir,
3685                 .dentry = new_dentry,
3686                 .inode = d_inode(new_dentry),
3687         };
3688         u8 new_file_type;
3689         int retval;
3690         struct timespec ctime;
3691
3692         if ((ext4_encrypted_inode(old_dir) &&
3693              !fscrypt_has_encryption_key(old_dir)) ||
3694             (ext4_encrypted_inode(new_dir) &&
3695              !fscrypt_has_encryption_key(new_dir)))
3696                 return -ENOKEY;
3697
3698         if ((ext4_encrypted_inode(old_dir) ||
3699              ext4_encrypted_inode(new_dir)) &&
3700             (old_dir != new_dir) &&
3701             (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3702              !fscrypt_has_permitted_context(old_dir, new.inode)))
3703                 return -EPERM;
3704
3705         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3706              !projid_eq(EXT4_I(new_dir)->i_projid,
3707                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3708             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3709              !projid_eq(EXT4_I(old_dir)->i_projid,
3710                         EXT4_I(new_dentry->d_inode)->i_projid)))
3711                 return -EXDEV;
3712
3713         retval = dquot_initialize(old.dir);
3714         if (retval)
3715                 return retval;
3716         retval = dquot_initialize(new.dir);
3717         if (retval)
3718                 return retval;
3719
3720         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3721                                  &old.de, &old.inlined);
3722         if (IS_ERR(old.bh))
3723                 return PTR_ERR(old.bh);
3724         /*
3725          *  Check for inode number is _not_ due to possible IO errors.
3726          *  We might rmdir the source, keep it as pwd of some process
3727          *  and merrily kill the link to whatever was created under the
3728          *  same name. Goodbye sticky bit ;-<
3729          */
3730         retval = -ENOENT;
3731         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3732                 goto end_rename;
3733
3734         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3735                                  &new.de, &new.inlined);
3736         if (IS_ERR(new.bh)) {
3737                 retval = PTR_ERR(new.bh);
3738                 new.bh = NULL;
3739                 goto end_rename;
3740         }
3741
3742         /* RENAME_EXCHANGE case: old *and* new must both exist */
3743         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3744                 goto end_rename;
3745
3746         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3747                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3748                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3749         if (IS_ERR(handle)) {
3750                 retval = PTR_ERR(handle);
3751                 handle = NULL;
3752                 goto end_rename;
3753         }
3754
3755         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3756                 ext4_handle_sync(handle);
3757
3758         if (S_ISDIR(old.inode->i_mode)) {
3759                 old.is_dir = true;
3760                 retval = ext4_rename_dir_prepare(handle, &old);
3761                 if (retval)
3762                         goto end_rename;
3763         }
3764         if (S_ISDIR(new.inode->i_mode)) {
3765                 new.is_dir = true;
3766                 retval = ext4_rename_dir_prepare(handle, &new);
3767                 if (retval)
3768                         goto end_rename;
3769         }
3770
3771         /*
3772          * Other than the special case of overwriting a directory, parents'
3773          * nlink only needs to be modified if this is a cross directory rename.
3774          */
3775         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3776                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3777                 new.dir_nlink_delta = -old.dir_nlink_delta;
3778                 retval = -EMLINK;
3779                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3780                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3781                         goto end_rename;
3782         }
3783
3784         new_file_type = new.de->file_type;
3785         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3786         if (retval)
3787                 goto end_rename;
3788
3789         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3790         if (retval)
3791                 goto end_rename;
3792
3793         /*
3794          * Like most other Unix systems, set the ctime for inodes on a
3795          * rename.
3796          */
3797         ctime = current_time(old.inode);
3798         old.inode->i_ctime = ctime;
3799         new.inode->i_ctime = ctime;
3800         ext4_mark_inode_dirty(handle, old.inode);
3801         ext4_mark_inode_dirty(handle, new.inode);
3802
3803         if (old.dir_bh) {
3804                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3805                 if (retval)
3806                         goto end_rename;
3807         }
3808         if (new.dir_bh) {
3809                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3810                 if (retval)
3811                         goto end_rename;
3812         }
3813         ext4_update_dir_count(handle, &old);
3814         ext4_update_dir_count(handle, &new);
3815         retval = 0;
3816
3817 end_rename:
3818         brelse(old.dir_bh);
3819         brelse(new.dir_bh);
3820         brelse(old.bh);
3821         brelse(new.bh);
3822         if (handle)
3823                 ext4_journal_stop(handle);
3824         return retval;
3825 }
3826
3827 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3828                         struct inode *new_dir, struct dentry *new_dentry,
3829                         unsigned int flags)
3830 {
3831         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3832                 return -EIO;
3833
3834         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3835                 return -EINVAL;
3836
3837         if (flags & RENAME_EXCHANGE) {
3838                 return ext4_cross_rename(old_dir, old_dentry,
3839                                          new_dir, new_dentry);
3840         }
3841
3842         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3843 }
3844
3845 /*
3846  * directories can handle most operations...
3847  */
3848 const struct inode_operations ext4_dir_inode_operations = {
3849         .create         = ext4_create,
3850         .lookup         = ext4_lookup,
3851         .link           = ext4_link,
3852         .unlink         = ext4_unlink,
3853         .symlink        = ext4_symlink,
3854         .mkdir          = ext4_mkdir,
3855         .rmdir          = ext4_rmdir,
3856         .mknod          = ext4_mknod,
3857         .tmpfile        = ext4_tmpfile,
3858         .rename         = ext4_rename2,
3859         .setattr        = ext4_setattr,
3860         .getattr        = ext4_getattr,
3861         .listxattr      = ext4_listxattr,
3862         .get_acl        = ext4_get_acl,
3863         .set_acl        = ext4_set_acl,
3864         .fiemap         = ext4_fiemap,
3865 };
3866
3867 const struct inode_operations ext4_special_inode_operations = {
3868         .setattr        = ext4_setattr,
3869         .getattr        = ext4_getattr,
3870         .listxattr      = ext4_listxattr,
3871         .get_acl        = ext4_get_acl,
3872         .set_acl        = ext4_set_acl,
3873 };