fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include "ext4.h"
  38 #include "ext4_jbd2.h"
  39
  40 #include "xattr.h"
  41 #include "acl.h"
  42
  43 #include <trace/events/ext4.h>
  44 /*
  45  * define how far ahead to read directories while searching them.
  46  */
  47 #define NAMEI_RA_CHUNKS  2
  48 #define NAMEI_RA_BLOCKS  4
  49 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50
  51 static struct buffer_head *ext4_append(handle_t *handle,
  52                                         struct inode *inode,
  53                                         ext4_lblk_t *block)
  54 {
  55         struct buffer_head *bh;
  56         int err;
  57
  58         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  59                      ((inode->i_size >> 10) >=
  60                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  61                 return ERR_PTR(-ENOSPC);
  62
  63         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  64
  65         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  66         if (IS_ERR(bh))
  67                 return bh;
  68         inode->i_size += inode->i_sb->s_blocksize;
  69         EXT4_I(inode)->i_disksize = inode->i_size;
  70         BUFFER_TRACE(bh, "get_write_access");
  71         err = ext4_journal_get_write_access(handle, bh);
  72         if (err) {
  73                 brelse(bh);
  74                 ext4_std_error(inode->i_sb, err);
  75                 return ERR_PTR(err);
  76         }
  77         return bh;
  78 }
  79
  80 static int ext4_dx_csum_verify(struct inode *inode,
  81                                struct ext4_dir_entry *dirent);
  82
  83 typedef enum {
  84         EITHER, INDEX, DIRENT
  85 } dirblock_type_t;
  86
  87 #define ext4_read_dirblock(inode, block, type) \
  88         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
  89
  90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  91                                                 ext4_lblk_t block,
  92                                                 dirblock_type_t type,
  93                                                 const char *func,
  94                                                 unsigned int line)
  95 {
  96         struct buffer_head *bh;
  97         struct ext4_dir_entry *dirent;
  98         int is_dx_block = 0;
  99
 100         bh = ext4_bread(NULL, inode, block, 0);
 101         if (IS_ERR(bh)) {
 102                 __ext4_warning(inode->i_sb, func, line,
 103                                "inode #%lu: lblock %lu: comm %s: "
 104                                "error %ld reading directory block",
 105                                inode->i_ino, (unsigned long)block,
 106                                current->comm, PTR_ERR(bh));
 107
 108                 return bh;
 109         }
 110         if (!bh) {
 111                 ext4_error_inode(inode, func, line, block,
 112                                  "Directory hole found");
 113                 return ERR_PTR(-EFSCORRUPTED);
 114         }
 115         dirent = (struct ext4_dir_entry *) bh->b_data;
 116         /* Determine whether or not we have an index block */
 117         if (is_dx(inode)) {
 118                 if (block == 0)
 119                         is_dx_block = 1;
 120                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 121                                                 inode->i_sb->s_blocksize) ==
 122                          inode->i_sb->s_blocksize)
 123                         is_dx_block = 1;
 124         }
 125         if (!is_dx_block && type == INDEX) {
 126                 ext4_error_inode(inode, func, line, block,
 127                        "directory leaf block found instead of index block");
 128                 return ERR_PTR(-EFSCORRUPTED);
 129         }
 130         if (!ext4_has_metadata_csum(inode->i_sb) ||
 131             buffer_verified(bh))
 132                 return bh;
 133
 134         /*
 135          * An empty leaf block can get mistaken for a index block; for
 136          * this reason, we can only check the index checksum when the
 137          * caller is sure it should be an index block.
 138          */
 139         if (is_dx_block && type == INDEX) {
 140                 if (ext4_dx_csum_verify(inode, dirent))
 141                         set_buffer_verified(bh);
 142                 else {
 143                         ext4_error_inode(inode, func, line, block,
 144                                          "Directory index failed checksum");
 145                         brelse(bh);
 146                         return ERR_PTR(-EFSBADCRC);
 147                 }
 148         }
 149         if (!is_dx_block) {
 150                 if (ext4_dirent_csum_verify(inode, dirent))
 151                         set_buffer_verified(bh);
 152                 else {
 153                         ext4_error_inode(inode, func, line, block,
 154                                          "Directory block failed checksum");
 155                         brelse(bh);
 156                         return ERR_PTR(-EFSBADCRC);
 157                 }
 158         }
 159         return bh;
 160 }
 161
 162 #ifndef assert
 163 #define assert(test) J_ASSERT(test)
 164 #endif
 165
 166 #ifdef DX_DEBUG
 167 #define dxtrace(command) command
 168 #else
 169 #define dxtrace(command)
 170 #endif
 171
 172 struct fake_dirent
 173 {
 174         __le32 inode;
 175         __le16 rec_len;
 176         u8 name_len;
 177         u8 file_type;
 178 };
 179
 180 struct dx_countlimit
 181 {
 182         __le16 limit;
 183         __le16 count;
 184 };
 185
 186 struct dx_entry
 187 {
 188         __le32 hash;
 189         __le32 block;
 190 };
 191
 192 /*
 193  * dx_root_info is laid out so that if it should somehow get overlaid by a
 194  * dirent the two low bits of the hash version will be zero.  Therefore, the
 195  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 196  */
 197
 198 struct dx_root
 199 {
 200         struct fake_dirent dot;
 201         char dot_name[4];
 202         struct fake_dirent dotdot;
 203         char dotdot_name[4];
 204         struct dx_root_info
 205         {
 206                 __le32 reserved_zero;
 207                 u8 hash_version;
 208                 u8 info_length; /* 8 */
 209                 u8 indirect_levels;
 210                 u8 unused_flags;
 211         }
 212         info;
 213         struct dx_entry entries[0];
 214 };
 215
 216 struct dx_node
 217 {
 218         struct fake_dirent fake;
 219         struct dx_entry entries[0];
 220 };
 221
 222
 223 struct dx_frame
 224 {
 225         struct buffer_head *bh;
 226         struct dx_entry *entries;
 227         struct dx_entry *at;
 228 };
 229
 230 struct dx_map_entry
 231 {
 232         u32 hash;
 233         u16 offs;
 234         u16 size;
 235 };
 236
 237 /*
 238  * This goes at the end of each htree block.
 239  */
 240 struct dx_tail {
 241         u32 dt_reserved;
 242         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 243 };
 244
 245 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 246 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 247 static inline unsigned dx_get_hash(struct dx_entry *entry);
 248 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 249 static unsigned dx_get_count(struct dx_entry *entries);
 250 static unsigned dx_get_limit(struct dx_entry *entries);
 251 static void dx_set_count(struct dx_entry *entries, unsigned value);
 252 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 253 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 254 static unsigned dx_node_limit(struct inode *dir);
 255 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 256                                  struct inode *dir,
 257                                  struct dx_hash_info *hinfo,
 258                                  struct dx_frame *frame);
 259 static void dx_release(struct dx_frame *frames);
 260 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 261                        unsigned blocksize, struct dx_hash_info *hinfo,
 262                        struct dx_map_entry map[]);
 263 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 264 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 265                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 266 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 267 static void dx_insert_block(struct dx_frame *frame,
 268                                         u32 hash, ext4_lblk_t block);
 269 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 270                                  struct dx_frame *frame,
 271                                  struct dx_frame *frames,
 272                                  __u32 *start_hash);
 273 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 274                 struct ext4_filename *fname,
 275                 struct ext4_dir_entry_2 **res_dir);
 276 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 277                              struct inode *dir, struct inode *inode);
 278
 279 /* checksumming functions */
 280 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 281                             unsigned int blocksize)
 282 {
 283         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 284         t->det_rec_len = ext4_rec_len_to_disk(
 285                         sizeof(struct ext4_dir_entry_tail), blocksize);
 286         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 287 }
 288
 289 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 290 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 291                                                    struct ext4_dir_entry *de)
 292 {
 293         struct ext4_dir_entry_tail *t;
 294
 295 #ifdef PARANOID
 296         struct ext4_dir_entry *d, *top;
 297
 298         d = de;
 299         top = (struct ext4_dir_entry *)(((void *)de) +
 300                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 301                 sizeof(struct ext4_dir_entry_tail)));
 302         while (d < top && d->rec_len)
 303                 d = (struct ext4_dir_entry *)(((void *)d) +
 304                     le16_to_cpu(d->rec_len));
 305
 306         if (d != top)
 307                 return NULL;
 308
 309         t = (struct ext4_dir_entry_tail *)d;
 310 #else
 311         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 312 #endif
 313
 314         if (t->det_reserved_zero1 ||
 315             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 316             t->det_reserved_zero2 ||
 317             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 318                 return NULL;
 319
 320         return t;
 321 }
 322
 323 static __le32 ext4_dirent_csum(struct inode *inode,
 324                                struct ext4_dir_entry *dirent, int size)
 325 {
 326         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 327         struct ext4_inode_info *ei = EXT4_I(inode);
 328         __u32 csum;
 329
 330         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 331         return cpu_to_le32(csum);
 332 }
 333
 334 #define warn_no_space_for_csum(inode)                                   \
 335         __warn_no_space_for_csum((inode), __func__, __LINE__)
 336
 337 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 338                                      unsigned int line)
 339 {
 340         __ext4_warning_inode(inode, func, line,
 341                 "No space for directory leaf checksum. Please run e2fsck -D.");
 342 }
 343
 344 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 345 {
 346         struct ext4_dir_entry_tail *t;
 347
 348         if (!ext4_has_metadata_csum(inode->i_sb))
 349                 return 1;
 350
 351         t = get_dirent_tail(inode, dirent);
 352         if (!t) {
 353                 warn_no_space_for_csum(inode);
 354                 return 0;
 355         }
 356
 357         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 358                                                 (void *)t - (void *)dirent))
 359                 return 0;
 360
 361         return 1;
 362 }
 363
 364 static void ext4_dirent_csum_set(struct inode *inode,
 365                                  struct ext4_dir_entry *dirent)
 366 {
 367         struct ext4_dir_entry_tail *t;
 368
 369         if (!ext4_has_metadata_csum(inode->i_sb))
 370                 return;
 371
 372         t = get_dirent_tail(inode, dirent);
 373         if (!t) {
 374                 warn_no_space_for_csum(inode);
 375                 return;
 376         }
 377
 378         t->det_checksum = ext4_dirent_csum(inode, dirent,
 379                                            (void *)t - (void *)dirent);
 380 }
 381
 382 int ext4_handle_dirty_dirent_node(handle_t *handle,
 383                                   struct inode *inode,
 384                                   struct buffer_head *bh)
 385 {
 386         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 387         return ext4_handle_dirty_metadata(handle, inode, bh);
 388 }
 389
 390 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 391                                                struct ext4_dir_entry *dirent,
 392                                                int *offset)
 393 {
 394         struct ext4_dir_entry *dp;
 395         struct dx_root_info *root;
 396         int count_offset;
 397
 398         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 399                 count_offset = 8;
 400         else if (le16_to_cpu(dirent->rec_len) == 12) {
 401                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 402                 if (le16_to_cpu(dp->rec_len) !=
 403                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 404                         return NULL;
 405                 root = (struct dx_root_info *)(((void *)dp + 12));
 406                 if (root->reserved_zero ||
 407                     root->info_length != sizeof(struct dx_root_info))
 408                         return NULL;
 409                 count_offset = 32;
 410         } else
 411                 return NULL;
 412
 413         if (offset)
 414                 *offset = count_offset;
 415         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 416 }
 417
 418 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 419                            int count_offset, int count, struct dx_tail *t)
 420 {
 421         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 422         struct ext4_inode_info *ei = EXT4_I(inode);
 423         __u32 csum;
 424         int size;
 425         __u32 dummy_csum = 0;
 426         int offset = offsetof(struct dx_tail, dt_checksum);
 427
 428         size = count_offset + (count * sizeof(struct dx_entry));
 429         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 430         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 431         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 432
 433         return cpu_to_le32(csum);
 434 }
 435
 436 static int ext4_dx_csum_verify(struct inode *inode,
 437                                struct ext4_dir_entry *dirent)
 438 {
 439         struct dx_countlimit *c;
 440         struct dx_tail *t;
 441         int count_offset, limit, count;
 442
 443         if (!ext4_has_metadata_csum(inode->i_sb))
 444                 return 1;
 445
 446         c = get_dx_countlimit(inode, dirent, &count_offset);
 447         if (!c) {
 448                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 449                 return 0;
 450         }
 451         limit = le16_to_cpu(c->limit);
 452         count = le16_to_cpu(c->count);
 453         if (count_offset + (limit * sizeof(struct dx_entry)) >
 454             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 455                 warn_no_space_for_csum(inode);
 456                 return 0;
 457         }
 458         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 459
 460         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 461                                             count, t))
 462                 return 0;
 463         return 1;
 464 }
 465
 466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 467 {
 468         struct dx_countlimit *c;
 469         struct dx_tail *t;
 470         int count_offset, limit, count;
 471
 472         if (!ext4_has_metadata_csum(inode->i_sb))
 473                 return;
 474
 475         c = get_dx_countlimit(inode, dirent, &count_offset);
 476         if (!c) {
 477                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 478                 return;
 479         }
 480         limit = le16_to_cpu(c->limit);
 481         count = le16_to_cpu(c->count);
 482         if (count_offset + (limit * sizeof(struct dx_entry)) >
 483             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 484                 warn_no_space_for_csum(inode);
 485                 return;
 486         }
 487         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 488
 489         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 490 }
 491
 492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 493                                             struct inode *inode,
 494                                             struct buffer_head *bh)
 495 {
 496         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 497         return ext4_handle_dirty_metadata(handle, inode, bh);
 498 }
 499
 500 /*
 501  * p is at least 6 bytes before the end of page
 502  */
 503 static inline struct ext4_dir_entry_2 *
 504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 505 {
 506         return (struct ext4_dir_entry_2 *)((char *)p +
 507                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 508 }
 509
 510 /*
 511  * Future: use high four bits of block for coalesce-on-delete flags
 512  * Mask them off for now.
 513  */
 514
 515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 516 {
 517         return le32_to_cpu(entry->block) & 0x0fffffff;
 518 }
 519
 520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 521 {
 522         entry->block = cpu_to_le32(value);
 523 }
 524
 525 static inline unsigned dx_get_hash(struct dx_entry *entry)
 526 {
 527         return le32_to_cpu(entry->hash);
 528 }
 529
 530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 531 {
 532         entry->hash = cpu_to_le32(value);
 533 }
 534
 535 static inline unsigned dx_get_count(struct dx_entry *entries)
 536 {
 537         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 538 }
 539
 540 static inline unsigned dx_get_limit(struct dx_entry *entries)
 541 {
 542         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 543 }
 544
 545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 546 {
 547         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 548 }
 549
 550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 551 {
 552         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 553 }
 554
 555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 556 {
 557         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 558                 EXT4_DIR_REC_LEN(2) - infosize;
 559
 560         if (ext4_has_metadata_csum(dir->i_sb))
 561                 entry_space -= sizeof(struct dx_tail);
 562         return entry_space / sizeof(struct dx_entry);
 563 }
 564
 565 static inline unsigned dx_node_limit(struct inode *dir)
 566 {
 567         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 568
 569         if (ext4_has_metadata_csum(dir->i_sb))
 570                 entry_space -= sizeof(struct dx_tail);
 571         return entry_space / sizeof(struct dx_entry);
 572 }
 573
 574 /*
 575  * Debug
 576  */
 577 #ifdef DX_DEBUG
 578 static void dx_show_index(char * label, struct dx_entry *entries)
 579 {
 580         int i, n = dx_get_count (entries);
 581         printk(KERN_DEBUG "%s index", label);
 582         for (i = 0; i < n; i++) {
 583                 printk(KERN_CONT " %x->%lu",
 584                        i ? dx_get_hash(entries + i) : 0,
 585                        (unsigned long)dx_get_block(entries + i));
 586         }
 587         printk(KERN_CONT "\n");
 588 }
 589
 590 struct stats
 591 {
 592         unsigned names;
 593         unsigned space;
 594         unsigned bcount;
 595 };
 596
 597 static struct stats dx_show_leaf(struct inode *dir,
 598                                 struct dx_hash_info *hinfo,
 599                                 struct ext4_dir_entry_2 *de,
 600                                 int size, int show_names)
 601 {
 602         unsigned names = 0, space = 0;
 603         char *base = (char *) de;
 604         struct dx_hash_info h = *hinfo;
 605
 606         printk("names: ");
 607         while ((char *) de < base + size)
 608         {
 609                 if (de->inode)
 610                 {
 611                         if (show_names)
 612                         {
 613 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 614                                 int len;
 615                                 char *name;
 616                                 struct fscrypt_str fname_crypto_str =
 617                                         FSTR_INIT(NULL, 0);
 618                                 int res = 0;
 619
 620                                 name  = de->name;
 621                                 len = de->name_len;
 622                                 if (ext4_encrypted_inode(dir))
 623                                         res = fscrypt_get_encryption_info(dir);
 624                                 if (res) {
 625                                         printk(KERN_WARNING "Error setting up"
 626                                                " fname crypto: %d\n", res);
 627                                 }
 628                                 if (!fscrypt_has_encryption_key(dir)) {
 629                                         /* Directory is not encrypted */
 630                                         ext4fs_dirhash(de->name,
 631                                                 de->name_len, &h);
 632                                         printk("%*.s:(U)%x.%u ", len,
 633                                                name, h.hash,
 634                                                (unsigned) ((char *) de
 635                                                            - base));
 636                                 } else {
 637                                         struct fscrypt_str de_name =
 638                                                 FSTR_INIT(name, len);
 639
 640                                         /* Directory is encrypted */
 641                                         res = fscrypt_fname_alloc_buffer(
 642                                                 dir, len,
 643                                                 &fname_crypto_str);
 644                                         if (res)
 645                                                 printk(KERN_WARNING "Error "
 646                                                         "allocating crypto "
 647                                                         "buffer--skipping "
 648                                                         "crypto\n");
 649                                         res = fscrypt_fname_disk_to_usr(dir,
 650                                                 0, 0, &de_name,
 651                                                 &fname_crypto_str);
 652                                         if (res) {
 653                                                 printk(KERN_WARNING "Error "
 654                                                         "converting filename "
 655                                                         "from disk to usr"
 656                                                         "\n");
 657                                                 name = "??";
 658                                                 len = 2;
 659                                         } else {
 660                                                 name = fname_crypto_str.name;
 661                                                 len = fname_crypto_str.len;
 662                                         }
 663                                         ext4fs_dirhash(de->name, de->name_len,
 664                                                        &h);
 665                                         printk("%*.s:(E)%x.%u ", len, name,
 666                                                h.hash, (unsigned) ((char *) de
 667                                                                    - base));
 668                                         fscrypt_fname_free_buffer(
 669                                                         &fname_crypto_str);
 670                                 }
 671 #else
 672                                 int len = de->name_len;
 673                                 char *name = de->name;
 674                                 ext4fs_dirhash(de->name, de->name_len, &h);
 675                                 printk("%*.s:%x.%u ", len, name, h.hash,
 676                                        (unsigned) ((char *) de - base));
 677 #endif
 678                         }
 679                         space += EXT4_DIR_REC_LEN(de->name_len);
 680                         names++;
 681                 }
 682                 de = ext4_next_entry(de, size);
 683         }
 684         printk(KERN_CONT "(%i)\n", names);
 685         return (struct stats) { names, space, 1 };
 686 }
 687
 688 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 689                              struct dx_entry *entries, int levels)
 690 {
 691         unsigned blocksize = dir->i_sb->s_blocksize;
 692         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 693         unsigned bcount = 0;
 694         struct buffer_head *bh;
 695         printk("%i indexed blocks...\n", count);
 696         for (i = 0; i < count; i++, entries++)
 697         {
 698                 ext4_lblk_t block = dx_get_block(entries);
 699                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 700                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 701                 struct stats stats;
 702                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 703                 bh = ext4_bread(NULL,dir, block, 0);
 704                 if (!bh || IS_ERR(bh))
 705                         continue;
 706                 stats = levels?
 707                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 708                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 709                         bh->b_data, blocksize, 0);
 710                 names += stats.names;
 711                 space += stats.space;
 712                 bcount += stats.bcount;
 713                 brelse(bh);
 714         }
 715         if (bcount)
 716                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 717                        levels ? "" : "   ", names, space/bcount,
 718                        (space/bcount)*100/blocksize);
 719         return (struct stats) { names, space, bcount};
 720 }
 721 #endif /* DX_DEBUG */
 722
 723 /*
 724  * Probe for a directory leaf block to search.
 725  *
 726  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 727  * error in the directory index, and the caller should fall back to
 728  * searching the directory normally.  The callers of dx_probe **MUST**
 729  * check for this error code, and make sure it never gets reflected
 730  * back to userspace.
 731  */
 732 static struct dx_frame *
 733 dx_probe(struct ext4_filename *fname, struct inode *dir,
 734          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 735 {
 736         unsigned count, indirect;
 737         struct dx_entry *at, *entries, *p, *q, *m;
 738         struct dx_root *root;
 739         struct dx_frame *frame = frame_in;
 740         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 741         u32 hash;
 742
 743         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 744         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 745         if (IS_ERR(frame->bh))
 746                 return (struct dx_frame *) frame->bh;
 747
 748         root = (struct dx_root *) frame->bh->b_data;
 749         if (root->info.hash_version != DX_HASH_TEA &&
 750             root->info.hash_version != DX_HASH_HALF_MD4 &&
 751             root->info.hash_version != DX_HASH_LEGACY) {
 752                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 753                                    root->info.hash_version);
 754                 goto fail;
 755         }
 756         if (fname)
 757                 hinfo = &fname->hinfo;
 758         hinfo->hash_version = root->info.hash_version;
 759         if (hinfo->hash_version <= DX_HASH_TEA)
 760                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 761         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 762         if (fname && fname_name(fname))
 763                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 764         hash = hinfo->hash;
 765
 766         if (root->info.unused_flags & 1) {
 767                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 768                                    root->info.unused_flags);
 769                 goto fail;
 770         }
 771
 772         indirect = root->info.indirect_levels;
 773         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 774                 ext4_warning(dir->i_sb,
 775                              "Directory (ino: %lu) htree depth %#06x exceed"
 776                              "supported value", dir->i_ino,
 777                              ext4_dir_htree_level(dir->i_sb));
 778                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 779                         ext4_warning(dir->i_sb, "Enable large directory "
 780                                                 "feature to access it");
 781                 }
 782                 goto fail;
 783         }
 784
 785         entries = (struct dx_entry *)(((char *)&root->info) +
 786                                       root->info.info_length);
 787
 788         if (dx_get_limit(entries) != dx_root_limit(dir,
 789                                                    root->info.info_length)) {
 790                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 791                                    dx_get_limit(entries),
 792                                    dx_root_limit(dir, root->info.info_length));
 793                 goto fail;
 794         }
 795
 796         dxtrace(printk("Look up %x", hash));
 797         while (1) {
 798                 count = dx_get_count(entries);
 799                 if (!count || count > dx_get_limit(entries)) {
 800                         ext4_warning_inode(dir,
 801                                            "dx entry: count %u beyond limit %u",
 802                                            count, dx_get_limit(entries));
 803                         goto fail;
 804                 }
 805
 806                 p = entries + 1;
 807                 q = entries + count - 1;
 808                 while (p <= q) {
 809                         m = p + (q - p) / 2;
 810                         dxtrace(printk(KERN_CONT "."));
 811                         if (dx_get_hash(m) > hash)
 812                                 q = m - 1;
 813                         else
 814                                 p = m + 1;
 815                 }
 816
 817                 if (0) { // linear search cross check
 818                         unsigned n = count - 1;
 819                         at = entries;
 820                         while (n--)
 821                         {
 822                                 dxtrace(printk(KERN_CONT ","));
 823                                 if (dx_get_hash(++at) > hash)
 824                                 {
 825                                         at--;
 826                                         break;
 827                                 }
 828                         }
 829                         assert (at == p - 1);
 830                 }
 831
 832                 at = p - 1;
 833                 dxtrace(printk(KERN_CONT " %x->%u\n",
 834                                at == entries ? 0 : dx_get_hash(at),
 835                                dx_get_block(at)));
 836                 frame->entries = entries;
 837                 frame->at = at;
 838                 if (!indirect--)
 839                         return frame;
 840                 frame++;
 841                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 842                 if (IS_ERR(frame->bh)) {
 843                         ret_err = (struct dx_frame *) frame->bh;
 844                         frame->bh = NULL;
 845                         goto fail;
 846                 }
 847                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 848
 849                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 850                         ext4_warning_inode(dir,
 851                                 "dx entry: limit %u != node limit %u",
 852                                 dx_get_limit(entries), dx_node_limit(dir));
 853                         goto fail;
 854                 }
 855         }
 856 fail:
 857         while (frame >= frame_in) {
 858                 brelse(frame->bh);
 859                 frame--;
 860         }
 861
 862         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 863                 ext4_warning_inode(dir,
 864                         "Corrupt directory, running e2fsck is recommended");
 865         return ret_err;
 866 }
 867
 868 static void dx_release(struct dx_frame *frames)
 869 {
 870         struct dx_root_info *info;
 871         int i;
 872
 873         if (frames[0].bh == NULL)
 874                 return;
 875
 876         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 877         for (i = 0; i <= info->indirect_levels; i++) {
 878                 if (frames[i].bh == NULL)
 879                         break;
 880                 brelse(frames[i].bh);
 881                 frames[i].bh = NULL;
 882         }
 883 }
 884
 885 /*
 886  * This function increments the frame pointer to search the next leaf
 887  * block, and reads in the necessary intervening nodes if the search
 888  * should be necessary.  Whether or not the search is necessary is
 889  * controlled by the hash parameter.  If the hash value is even, then
 890  * the search is only continued if the next block starts with that
 891  * hash value.  This is used if we are searching for a specific file.
 892  *
 893  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 894  *
 895  * This function returns 1 if the caller should continue to search,
 896  * or 0 if it should not.  If there is an error reading one of the
 897  * index blocks, it will a negative error code.
 898  *
 899  * If start_hash is non-null, it will be filled in with the starting
 900  * hash of the next page.
 901  */
 902 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 903                                  struct dx_frame *frame,
 904                                  struct dx_frame *frames,
 905                                  __u32 *start_hash)
 906 {
 907         struct dx_frame *p;
 908         struct buffer_head *bh;
 909         int num_frames = 0;
 910         __u32 bhash;
 911
 912         p = frame;
 913         /*
 914          * Find the next leaf page by incrementing the frame pointer.
 915          * If we run out of entries in the interior node, loop around and
 916          * increment pointer in the parent node.  When we break out of
 917          * this loop, num_frames indicates the number of interior
 918          * nodes need to be read.
 919          */
 920         while (1) {
 921                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 922                         break;
 923                 if (p == frames)
 924                         return 0;
 925                 num_frames++;
 926                 p--;
 927         }
 928
 929         /*
 930          * If the hash is 1, then continue only if the next page has a
 931          * continuation hash of any value.  This is used for readdir
 932          * handling.  Otherwise, check to see if the hash matches the
 933          * desired contiuation hash.  If it doesn't, return since
 934          * there's no point to read in the successive index pages.
 935          */
 936         bhash = dx_get_hash(p->at);
 937         if (start_hash)
 938                 *start_hash = bhash;
 939         if ((hash & 1) == 0) {
 940                 if ((bhash & ~1) != hash)
 941                         return 0;
 942         }
 943         /*
 944          * If the hash is HASH_NB_ALWAYS, we always go to the next
 945          * block so no check is necessary
 946          */
 947         while (num_frames--) {
 948                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 949                 if (IS_ERR(bh))
 950                         return PTR_ERR(bh);
 951                 p++;
 952                 brelse(p->bh);
 953                 p->bh = bh;
 954                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 955         }
 956         return 1;
 957 }
 958
 959
 960 /*
 961  * This function fills a red-black tree with information from a
 962  * directory block.  It returns the number directory entries loaded
 963  * into the tree.  If there is an error it is returned in err.
 964  */
 965 static int htree_dirblock_to_tree(struct file *dir_file,
 966                                   struct inode *dir, ext4_lblk_t block,
 967                                   struct dx_hash_info *hinfo,
 968                                   __u32 start_hash, __u32 start_minor_hash)
 969 {
 970         struct buffer_head *bh;
 971         struct ext4_dir_entry_2 *de, *top;
 972         int err = 0, count = 0;
 973         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 974
 975         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 976                                                         (unsigned long)block));
 977         bh = ext4_read_dirblock(dir, block, DIRENT);
 978         if (IS_ERR(bh))
 979                 return PTR_ERR(bh);
 980
 981         de = (struct ext4_dir_entry_2 *) bh->b_data;
 982         top = (struct ext4_dir_entry_2 *) ((char *) de +
 983                                            dir->i_sb->s_blocksize -
 984                                            EXT4_DIR_REC_LEN(0));
 985 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 986         /* Check if the directory is encrypted */
 987         if (ext4_encrypted_inode(dir)) {
 988                 err = fscrypt_get_encryption_info(dir);
 989                 if (err < 0) {
 990                         brelse(bh);
 991                         return err;
 992                 }
 993                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
 994                                                      &fname_crypto_str);
 995                 if (err < 0) {
 996                         brelse(bh);
 997                         return err;
 998                 }
 999         }
1000 #endif
1001         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1002                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1003                                 bh->b_data, bh->b_size,
1004                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1005                                          + ((char *)de - bh->b_data))) {
1006                         /* silently ignore the rest of the block */
1007                         break;
1008                 }
1009                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1010                 if ((hinfo->hash < start_hash) ||
1011                     ((hinfo->hash == start_hash) &&
1012                      (hinfo->minor_hash < start_minor_hash)))
1013                         continue;
1014                 if (de->inode == 0)
1015                         continue;
1016                 if (!ext4_encrypted_inode(dir)) {
1017                         tmp_str.name = de->name;
1018                         tmp_str.len = de->name_len;
1019                         err = ext4_htree_store_dirent(dir_file,
1020                                    hinfo->hash, hinfo->minor_hash, de,
1021                                    &tmp_str);
1022                 } else {
1023                         int save_len = fname_crypto_str.len;
1024                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1025                                                                 de->name_len);
1026
1027                         /* Directory is encrypted */
1028                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1029                                         hinfo->minor_hash, &de_name,
1030                                         &fname_crypto_str);
1031                         if (err) {
1032                                 count = err;
1033                                 goto errout;
1034                         }
1035                         err = ext4_htree_store_dirent(dir_file,
1036                                    hinfo->hash, hinfo->minor_hash, de,
1037                                         &fname_crypto_str);
1038                         fname_crypto_str.len = save_len;
1039                 }
1040                 if (err != 0) {
1041                         count = err;
1042                         goto errout;
1043                 }
1044                 count++;
1045         }
1046 errout:
1047         brelse(bh);
1048 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1049         fscrypt_fname_free_buffer(&fname_crypto_str);
1050 #endif
1051         return count;
1052 }
1053
1054
1055 /*
1056  * This function fills a red-black tree with information from a
1057  * directory.  We start scanning the directory in hash order, starting
1058  * at start_hash and start_minor_hash.
1059  *
1060  * This function returns the number of entries inserted into the tree,
1061  * or a negative error code.
1062  */
1063 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1064                          __u32 start_minor_hash, __u32 *next_hash)
1065 {
1066         struct dx_hash_info hinfo;
1067         struct ext4_dir_entry_2 *de;
1068         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1069         struct inode *dir;
1070         ext4_lblk_t block;
1071         int count = 0;
1072         int ret, err;
1073         __u32 hashval;
1074         struct fscrypt_str tmp_str;
1075
1076         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1077                        start_hash, start_minor_hash));
1078         dir = file_inode(dir_file);
1079         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1080                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1081                 if (hinfo.hash_version <= DX_HASH_TEA)
1082                         hinfo.hash_version +=
1083                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1084                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1085                 if (ext4_has_inline_data(dir)) {
1086                         int has_inline_data = 1;
1087                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1088                                                         &hinfo, start_hash,
1089                                                         start_minor_hash,
1090                                                         &has_inline_data);
1091                         if (has_inline_data) {
1092                                 *next_hash = ~0;
1093                                 return count;
1094                         }
1095                 }
1096                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1097                                                start_hash, start_minor_hash);
1098                 *next_hash = ~0;
1099                 return count;
1100         }
1101         hinfo.hash = start_hash;
1102         hinfo.minor_hash = 0;
1103         frame = dx_probe(NULL, dir, &hinfo, frames);
1104         if (IS_ERR(frame))
1105                 return PTR_ERR(frame);
1106
1107         /* Add '.' and '..' from the htree header */
1108         if (!start_hash && !start_minor_hash) {
1109                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1110                 tmp_str.name = de->name;
1111                 tmp_str.len = de->name_len;
1112                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1113                                               de, &tmp_str);
1114                 if (err != 0)
1115                         goto errout;
1116                 count++;
1117         }
1118         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1119                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1120                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1121                 tmp_str.name = de->name;
1122                 tmp_str.len = de->name_len;
1123                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1124                                               de, &tmp_str);
1125                 if (err != 0)
1126                         goto errout;
1127                 count++;
1128         }
1129
1130         while (1) {
1131                 if (fatal_signal_pending(current)) {
1132                         err = -ERESTARTSYS;
1133                         goto errout;
1134                 }
1135                 cond_resched();
1136                 block = dx_get_block(frame->at);
1137                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1138                                              start_hash, start_minor_hash);
1139                 if (ret < 0) {
1140                         err = ret;
1141                         goto errout;
1142                 }
1143                 count += ret;
1144                 hashval = ~0;
1145                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1146                                             frame, frames, &hashval);
1147                 *next_hash = hashval;
1148                 if (ret < 0) {
1149                         err = ret;
1150                         goto errout;
1151                 }
1152                 /*
1153                  * Stop if:  (a) there are no more entries, or
1154                  * (b) we have inserted at least one entry and the
1155                  * next hash value is not a continuation
1156                  */
1157                 if ((ret == 0) ||
1158                     (count && ((hashval & 1) == 0)))
1159                         break;
1160         }
1161         dx_release(frames);
1162         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1163                        "next hash: %x\n", count, *next_hash));
1164         return count;
1165 errout:
1166         dx_release(frames);
1167         return (err);
1168 }
1169
1170 static inline int search_dirblock(struct buffer_head *bh,
1171                                   struct inode *dir,
1172                                   struct ext4_filename *fname,
1173                                   unsigned int offset,
1174                                   struct ext4_dir_entry_2 **res_dir)
1175 {
1176         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1177                                fname, offset, res_dir);
1178 }
1179
1180 /*
1181  * Directory block splitting, compacting
1182  */
1183
1184 /*
1185  * Create map of hash values, offsets, and sizes, stored at end of block.
1186  * Returns number of entries mapped.
1187  */
1188 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1189                        unsigned blocksize, struct dx_hash_info *hinfo,
1190                        struct dx_map_entry *map_tail)
1191 {
1192         int count = 0;
1193         char *base = (char *) de;
1194         struct dx_hash_info h = *hinfo;
1195
1196         while ((char *) de < base + blocksize) {
1197                 if (de->name_len && de->inode) {
1198                         ext4fs_dirhash(de->name, de->name_len, &h);
1199                         map_tail--;
1200                         map_tail->hash = h.hash;
1201                         map_tail->offs = ((char *) de - base)>>2;
1202                         map_tail->size = le16_to_cpu(de->rec_len);
1203                         count++;
1204                         cond_resched();
1205                 }
1206                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1207                 de = ext4_next_entry(de, blocksize);
1208         }
1209         return count;
1210 }
1211
1212 /* Sort map by hash value */
1213 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1214 {
1215         struct dx_map_entry *p, *q, *top = map + count - 1;
1216         int more;
1217         /* Combsort until bubble sort doesn't suck */
1218         while (count > 2) {
1219                 count = count*10/13;
1220                 if (count - 9 < 2) /* 9, 10 -> 11 */
1221                         count = 11;
1222                 for (p = top, q = p - count; q >= map; p--, q--)
1223                         if (p->hash < q->hash)
1224                                 swap(*p, *q);
1225         }
1226         /* Garden variety bubble sort */
1227         do {
1228                 more = 0;
1229                 q = top;
1230                 while (q-- > map) {
1231                         if (q[1].hash >= q[0].hash)
1232                                 continue;
1233                         swap(*(q+1), *q);
1234                         more = 1;
1235                 }
1236         } while(more);
1237 }
1238
1239 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1240 {
1241         struct dx_entry *entries = frame->entries;
1242         struct dx_entry *old = frame->at, *new = old + 1;
1243         int count = dx_get_count(entries);
1244
1245         assert(count < dx_get_limit(entries));
1246         assert(old < entries + count);
1247         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1248         dx_set_hash(new, hash);
1249         dx_set_block(new, block);
1250         dx_set_count(entries, count + 1);
1251 }
1252
1253 /*
1254  * Test whether a directory entry matches the filename being searched for.
1255  *
1256  * Return: %true if the directory entry matches, otherwise %false.
1257  */
1258 static inline bool ext4_match(const struct ext4_filename *fname,
1259                               const struct ext4_dir_entry_2 *de)
1260 {
1261         struct fscrypt_name f;
1262
1263         if (!de->inode)
1264                 return false;
1265
1266         f.usr_fname = fname->usr_fname;
1267         f.disk_name = fname->disk_name;
1268 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1269         f.crypto_buf = fname->crypto_buf;
1270 #endif
1271         return fscrypt_match_name(&f, de->name, de->name_len);
1272 }
1273
1274 /*
1275  * Returns 0 if not found, -1 on failure, and 1 on success
1276  */
1277 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1278                     struct inode *dir, struct ext4_filename *fname,
1279                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281         struct ext4_dir_entry_2 * de;
1282         char * dlimit;
1283         int de_len;
1284
1285         de = (struct ext4_dir_entry_2 *)search_buf;
1286         dlimit = search_buf + buf_size;
1287         while ((char *) de < dlimit) {
1288                 /* this code is executed quadratically often */
1289                 /* do minimal checking `by hand' */
1290                 if ((char *) de + de->name_len <= dlimit &&
1291                     ext4_match(fname, de)) {
1292                         /* found a match - just to be sure, do
1293                          * a full check */
1294                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1295                                                  bh->b_size, offset))
1296                                 return -1;
1297                         *res_dir = de;
1298                         return 1;
1299                 }
1300                 /* prevent looping on a bad block */
1301                 de_len = ext4_rec_len_from_disk(de->rec_len,
1302                                                 dir->i_sb->s_blocksize);
1303                 if (de_len <= 0)
1304                         return -1;
1305                 offset += de_len;
1306                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1307         }
1308         return 0;
1309 }
1310
1311 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1312                                struct ext4_dir_entry *de)
1313 {
1314         struct super_block *sb = dir->i_sb;
1315
1316         if (!is_dx(dir))
1317                 return 0;
1318         if (block == 0)
1319                 return 1;
1320         if (de->inode == 0 &&
1321             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1322                         sb->s_blocksize)
1323                 return 1;
1324         return 0;
1325 }
1326
1327 /*
1328  *      ext4_find_entry()
1329  *
1330  * finds an entry in the specified directory with the wanted name. It
1331  * returns the cache buffer in which the entry was found, and the entry
1332  * itself (as a parameter - res_dir). It does NOT read the inode of the
1333  * entry - you'll have to do that yourself if you want to.
1334  *
1335  * The returned buffer_head has ->b_count elevated.  The caller is expected
1336  * to brelse() it when appropriate.
1337  */
1338 static struct buffer_head * ext4_find_entry (struct inode *dir,
1339                                         const struct qstr *d_name,
1340                                         struct ext4_dir_entry_2 **res_dir,
1341                                         int *inlined)
1342 {
1343         struct super_block *sb;
1344         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1345         struct buffer_head *bh, *ret = NULL;
1346         ext4_lblk_t start, block;
1347         const u8 *name = d_name->name;
1348         size_t ra_max = 0;      /* Number of bh's in the readahead
1349                                    buffer, bh_use[] */
1350         size_t ra_ptr = 0;      /* Current index into readahead
1351                                    buffer */
1352         ext4_lblk_t  nblocks;
1353         int i, namelen, retval;
1354         struct ext4_filename fname;
1355
1356         *res_dir = NULL;
1357         sb = dir->i_sb;
1358         namelen = d_name->len;
1359         if (namelen > EXT4_NAME_LEN)
1360                 return NULL;
1361
1362         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1363         if (retval == -ENOENT)
1364                 return NULL;
1365         if (retval)
1366                 return ERR_PTR(retval);
1367
1368         if (ext4_has_inline_data(dir)) {
1369                 int has_inline_data = 1;
1370                 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1371                                              &has_inline_data);
1372                 if (has_inline_data) {
1373                         if (inlined)
1374                                 *inlined = 1;
1375                         goto cleanup_and_exit;
1376                 }
1377         }
1378
1379         if ((namelen <= 2) && (name[0] == '.') &&
1380             (name[1] == '.' || name[1] == '\0')) {
1381                 /*
1382                  * "." or ".." will only be in the first block
1383                  * NFS may look up ".."; "." should be handled by the VFS
1384                  */
1385                 block = start = 0;
1386                 nblocks = 1;
1387                 goto restart;
1388         }
1389         if (is_dx(dir)) {
1390                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1391                 /*
1392                  * On success, or if the error was file not found,
1393                  * return.  Otherwise, fall back to doing a search the
1394                  * old fashioned way.
1395                  */
1396                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1397                         goto cleanup_and_exit;
1398                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1399                                "falling back\n"));
1400         }
1401         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1402         start = EXT4_I(dir)->i_dir_start_lookup;
1403         if (start >= nblocks)
1404                 start = 0;
1405         block = start;
1406 restart:
1407         do {
1408                 /*
1409                  * We deal with the read-ahead logic here.
1410                  */
1411                 if (ra_ptr >= ra_max) {
1412                         /* Refill the readahead buffer */
1413                         ra_ptr = 0;
1414                         if (block < start)
1415                                 ra_max = start - block;
1416                         else
1417                                 ra_max = nblocks - block;
1418                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1419                         retval = ext4_bread_batch(dir, block, ra_max,
1420                                                   false /* wait */, bh_use);
1421                         if (retval) {
1422                                 ret = ERR_PTR(retval);
1423                                 ra_max = 0;
1424                                 goto cleanup_and_exit;
1425                         }
1426                 }
1427                 if ((bh = bh_use[ra_ptr++]) == NULL)
1428                         goto next;
1429                 wait_on_buffer(bh);
1430                 if (!buffer_uptodate(bh)) {
1431                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1432                                          (unsigned long) block);
1433                         brelse(bh);
1434                         ret = ERR_PTR(-EIO);
1435                         goto cleanup_and_exit;
1436                 }
1437                 if (!buffer_verified(bh) &&
1438                     !is_dx_internal_node(dir, block,
1439                                          (struct ext4_dir_entry *)bh->b_data) &&
1440                     !ext4_dirent_csum_verify(dir,
1441                                 (struct ext4_dir_entry *)bh->b_data)) {
1442                         EXT4_ERROR_INODE(dir, "checksumming directory "
1443                                          "block %lu", (unsigned long)block);
1444                         brelse(bh);
1445                         ret = ERR_PTR(-EFSBADCRC);
1446                         goto cleanup_and_exit;
1447                 }
1448                 set_buffer_verified(bh);
1449                 i = search_dirblock(bh, dir, &fname,
1450                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1451                 if (i == 1) {
1452                         EXT4_I(dir)->i_dir_start_lookup = block;
1453                         ret = bh;
1454                         goto cleanup_and_exit;
1455                 } else {
1456                         brelse(bh);
1457                         if (i < 0)
1458                                 goto cleanup_and_exit;
1459                 }
1460         next:
1461                 if (++block >= nblocks)
1462                         block = 0;
1463         } while (block != start);
1464
1465         /*
1466          * If the directory has grown while we were searching, then
1467          * search the last part of the directory before giving up.
1468          */
1469         block = nblocks;
1470         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1471         if (block < nblocks) {
1472                 start = 0;
1473                 goto restart;
1474         }
1475
1476 cleanup_and_exit:
1477         /* Clean up the read-ahead blocks */
1478         for (; ra_ptr < ra_max; ra_ptr++)
1479                 brelse(bh_use[ra_ptr]);
1480         ext4_fname_free_filename(&fname);
1481         return ret;
1482 }
1483
1484 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1485                         struct ext4_filename *fname,
1486                         struct ext4_dir_entry_2 **res_dir)
1487 {
1488         struct super_block * sb = dir->i_sb;
1489         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1490         struct buffer_head *bh;
1491         ext4_lblk_t block;
1492         int retval;
1493
1494 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1495         *res_dir = NULL;
1496 #endif
1497         frame = dx_probe(fname, dir, NULL, frames);
1498         if (IS_ERR(frame))
1499                 return (struct buffer_head *) frame;
1500         do {
1501                 block = dx_get_block(frame->at);
1502                 bh = ext4_read_dirblock(dir, block, DIRENT);
1503                 if (IS_ERR(bh))
1504                         goto errout;
1505
1506                 retval = search_dirblock(bh, dir, fname,
1507                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1508                                          res_dir);
1509                 if (retval == 1)
1510                         goto success;
1511                 brelse(bh);
1512                 if (retval == -1) {
1513                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1514                         goto errout;
1515                 }
1516
1517                 /* Check to see if we should continue to search */
1518                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1519                                                frames, NULL);
1520                 if (retval < 0) {
1521                         ext4_warning_inode(dir,
1522                                 "error %d reading directory index block",
1523                                 retval);
1524                         bh = ERR_PTR(retval);
1525                         goto errout;
1526                 }
1527         } while (retval == 1);
1528
1529         bh = NULL;
1530 errout:
1531         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1532 success:
1533         dx_release(frames);
1534         return bh;
1535 }
1536
1537 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1538 {
1539         struct inode *inode;
1540         struct ext4_dir_entry_2 *de;
1541         struct buffer_head *bh;
1542
1543         if (ext4_encrypted_inode(dir)) {
1544                 int res = fscrypt_get_encryption_info(dir);
1545
1546                 /*
1547                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1548                  * created while the directory was encrypted and we
1549                  * have access to the key.
1550                  */
1551                 if (fscrypt_has_encryption_key(dir))
1552                         fscrypt_set_encrypted_dentry(dentry);
1553                 fscrypt_set_d_op(dentry);
1554                 if (res && res != -ENOKEY)
1555                         return ERR_PTR(res);
1556         }
1557
1558        if (dentry->d_name.len > EXT4_NAME_LEN)
1559                return ERR_PTR(-ENAMETOOLONG);
1560
1561         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1562         if (IS_ERR(bh))
1563                 return (struct dentry *) bh;
1564         inode = NULL;
1565         if (bh) {
1566                 __u32 ino = le32_to_cpu(de->inode);
1567                 brelse(bh);
1568                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1569                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1570                         return ERR_PTR(-EFSCORRUPTED);
1571                 }
1572                 if (unlikely(ino == dir->i_ino)) {
1573                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1574                                          dentry);
1575                         return ERR_PTR(-EFSCORRUPTED);
1576                 }
1577                 inode = ext4_iget_normal(dir->i_sb, ino);
1578                 if (inode == ERR_PTR(-ESTALE)) {
1579                         EXT4_ERROR_INODE(dir,
1580                                          "deleted inode referenced: %u",
1581                                          ino);
1582                         return ERR_PTR(-EFSCORRUPTED);
1583                 }
1584                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1585                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1586                     !fscrypt_has_permitted_context(dir, inode)) {
1587                         ext4_warning(inode->i_sb,
1588                                      "Inconsistent encryption contexts: %lu/%lu",
1589                                      dir->i_ino, inode->i_ino);
1590                         iput(inode);
1591                         return ERR_PTR(-EPERM);
1592                 }
1593         }
1594         return d_splice_alias(inode, dentry);
1595 }
1596
1597
1598 struct dentry *ext4_get_parent(struct dentry *child)
1599 {
1600         __u32 ino;
1601         static const struct qstr dotdot = QSTR_INIT("..", 2);
1602         struct ext4_dir_entry_2 * de;
1603         struct buffer_head *bh;
1604
1605         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1606         if (IS_ERR(bh))
1607                 return (struct dentry *) bh;
1608         if (!bh)
1609                 return ERR_PTR(-ENOENT);
1610         ino = le32_to_cpu(de->inode);
1611         brelse(bh);
1612
1613         if (!ext4_valid_inum(child->d_sb, ino)) {
1614                 EXT4_ERROR_INODE(d_inode(child),
1615                                  "bad parent inode number: %u", ino);
1616                 return ERR_PTR(-EFSCORRUPTED);
1617         }
1618
1619         return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1620 }
1621
1622 /*
1623  * Move count entries from end of map between two memory locations.
1624  * Returns pointer to last entry moved.
1625  */
1626 static struct ext4_dir_entry_2 *
1627 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1628                 unsigned blocksize)
1629 {
1630         unsigned rec_len = 0;
1631
1632         while (count--) {
1633                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1634                                                 (from + (map->offs<<2));
1635                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1636                 memcpy (to, de, rec_len);
1637                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1638                                 ext4_rec_len_to_disk(rec_len, blocksize);
1639                 de->inode = 0;
1640                 map++;
1641                 to += rec_len;
1642         }
1643         return (struct ext4_dir_entry_2 *) (to - rec_len);
1644 }
1645
1646 /*
1647  * Compact each dir entry in the range to the minimal rec_len.
1648  * Returns pointer to last entry in range.
1649  */
1650 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1651 {
1652         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1653         unsigned rec_len = 0;
1654
1655         prev = to = de;
1656         while ((char*)de < base + blocksize) {
1657                 next = ext4_next_entry(de, blocksize);
1658                 if (de->inode && de->name_len) {
1659                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1660                         if (de > to)
1661                                 memmove(to, de, rec_len);
1662                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1663                         prev = to;
1664                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1665                 }
1666                 de = next;
1667         }
1668         return prev;
1669 }
1670
1671 /*
1672  * Split a full leaf block to make room for a new dir entry.
1673  * Allocate a new block, and move entries so that they are approx. equally full.
1674  * Returns pointer to de in block into which the new entry will be inserted.
1675  */
1676 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1677                         struct buffer_head **bh,struct dx_frame *frame,
1678                         struct dx_hash_info *hinfo)
1679 {
1680         unsigned blocksize = dir->i_sb->s_blocksize;
1681         unsigned count, continued;
1682         struct buffer_head *bh2;
1683         ext4_lblk_t newblock;
1684         u32 hash2;
1685         struct dx_map_entry *map;
1686         char *data1 = (*bh)->b_data, *data2;
1687         unsigned split, move, size;
1688         struct ext4_dir_entry_2 *de = NULL, *de2;
1689         struct ext4_dir_entry_tail *t;
1690         int     csum_size = 0;
1691         int     err = 0, i;
1692
1693         if (ext4_has_metadata_csum(dir->i_sb))
1694                 csum_size = sizeof(struct ext4_dir_entry_tail);
1695
1696         bh2 = ext4_append(handle, dir, &newblock);
1697         if (IS_ERR(bh2)) {
1698                 brelse(*bh);
1699                 *bh = NULL;
1700                 return (struct ext4_dir_entry_2 *) bh2;
1701         }
1702
1703         BUFFER_TRACE(*bh, "get_write_access");
1704         err = ext4_journal_get_write_access(handle, *bh);
1705         if (err)
1706                 goto journal_error;
1707
1708         BUFFER_TRACE(frame->bh, "get_write_access");
1709         err = ext4_journal_get_write_access(handle, frame->bh);
1710         if (err)
1711                 goto journal_error;
1712
1713         data2 = bh2->b_data;
1714
1715         /* create map in the end of data2 block */
1716         map = (struct dx_map_entry *) (data2 + blocksize);
1717         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1718                              blocksize, hinfo, map);
1719         map -= count;
1720         dx_sort_map(map, count);
1721         /* Split the existing block in the middle, size-wise */
1722         size = 0;
1723         move = 0;
1724         for (i = count-1; i >= 0; i--) {
1725                 /* is more than half of this entry in 2nd half of the block? */
1726                 if (size + map[i].size/2 > blocksize/2)
1727                         break;
1728                 size += map[i].size;
1729                 move++;
1730         }
1731         /* map index at which we will split */
1732         split = count - move;
1733         hash2 = map[split].hash;
1734         continued = hash2 == map[split - 1].hash;
1735         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1736                         (unsigned long)dx_get_block(frame->at),
1737                                         hash2, split, count-split));
1738
1739         /* Fancy dance to stay within two buffers */
1740         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1741                               blocksize);
1742         de = dx_pack_dirents(data1, blocksize);
1743         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1744                                            (char *) de,
1745                                            blocksize);
1746         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1747                                             (char *) de2,
1748                                             blocksize);
1749         if (csum_size) {
1750                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1751                 initialize_dirent_tail(t, blocksize);
1752
1753                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1754                 initialize_dirent_tail(t, blocksize);
1755         }
1756
1757         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1758                         blocksize, 1));
1759         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1760                         blocksize, 1));
1761
1762         /* Which block gets the new entry? */
1763         if (hinfo->hash >= hash2) {
1764                 swap(*bh, bh2);
1765                 de = de2;
1766         }
1767         dx_insert_block(frame, hash2 + continued, newblock);
1768         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1769         if (err)
1770                 goto journal_error;
1771         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1772         if (err)
1773                 goto journal_error;
1774         brelse(bh2);
1775         dxtrace(dx_show_index("frame", frame->entries));
1776         return de;
1777
1778 journal_error:
1779         brelse(*bh);
1780         brelse(bh2);
1781         *bh = NULL;
1782         ext4_std_error(dir->i_sb, err);
1783         return ERR_PTR(err);
1784 }
1785
1786 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1787                       struct buffer_head *bh,
1788                       void *buf, int buf_size,
1789                       struct ext4_filename *fname,
1790                       struct ext4_dir_entry_2 **dest_de)
1791 {
1792         struct ext4_dir_entry_2 *de;
1793         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1794         int nlen, rlen;
1795         unsigned int offset = 0;
1796         char *top;
1797
1798         de = (struct ext4_dir_entry_2 *)buf;
1799         top = buf + buf_size - reclen;
1800         while ((char *) de <= top) {
1801                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1802                                          buf, buf_size, offset))
1803                         return -EFSCORRUPTED;
1804                 if (ext4_match(fname, de))
1805                         return -EEXIST;
1806                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1807                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1808                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1809                         break;
1810                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1811                 offset += rlen;
1812         }
1813         if ((char *) de > top)
1814                 return -ENOSPC;
1815
1816         *dest_de = de;
1817         return 0;
1818 }
1819
1820 void ext4_insert_dentry(struct inode *inode,
1821                         struct ext4_dir_entry_2 *de,
1822                         int buf_size,
1823                         struct ext4_filename *fname)
1824 {
1825
1826         int nlen, rlen;
1827
1828         nlen = EXT4_DIR_REC_LEN(de->name_len);
1829         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1830         if (de->inode) {
1831                 struct ext4_dir_entry_2 *de1 =
1832                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1833                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1834                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1835                 de = de1;
1836         }
1837         de->file_type = EXT4_FT_UNKNOWN;
1838         de->inode = cpu_to_le32(inode->i_ino);
1839         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1840         de->name_len = fname_len(fname);
1841         memcpy(de->name, fname_name(fname), fname_len(fname));
1842 }
1843
1844 /*
1845  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1846  * it points to a directory entry which is guaranteed to be large
1847  * enough for new directory entry.  If de is NULL, then
1848  * add_dirent_to_buf will attempt search the directory block for
1849  * space.  It will return -ENOSPC if no space is available, and -EIO
1850  * and -EEXIST if directory entry already exists.
1851  */
1852 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1853                              struct inode *dir,
1854                              struct inode *inode, struct ext4_dir_entry_2 *de,
1855                              struct buffer_head *bh)
1856 {
1857         unsigned int    blocksize = dir->i_sb->s_blocksize;
1858         int             csum_size = 0;
1859         int             err;
1860
1861         if (ext4_has_metadata_csum(inode->i_sb))
1862                 csum_size = sizeof(struct ext4_dir_entry_tail);
1863
1864         if (!de) {
1865                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1866                                         blocksize - csum_size, fname, &de);
1867                 if (err)
1868                         return err;
1869         }
1870         BUFFER_TRACE(bh, "get_write_access");
1871         err = ext4_journal_get_write_access(handle, bh);
1872         if (err) {
1873                 ext4_std_error(dir->i_sb, err);
1874                 return err;
1875         }
1876
1877         /* By now the buffer is marked for journaling */
1878         ext4_insert_dentry(inode, de, blocksize, fname);
1879
1880         /*
1881          * XXX shouldn't update any times until successful
1882          * completion of syscall, but too many callers depend
1883          * on this.
1884          *
1885          * XXX similarly, too many callers depend on
1886          * ext4_new_inode() setting the times, but error
1887          * recovery deletes the inode, so the worst that can
1888          * happen is that the times are slightly out of date
1889          * and/or different from the directory change time.
1890          */
1891         dir->i_mtime = dir->i_ctime = current_time(dir);
1892         ext4_update_dx_flag(dir);
1893         inode_inc_iversion(dir);
1894         ext4_mark_inode_dirty(handle, dir);
1895         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1896         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1897         if (err)
1898                 ext4_std_error(dir->i_sb, err);
1899         return 0;
1900 }
1901
1902 /*
1903  * This converts a one block unindexed directory to a 3 block indexed
1904  * directory, and adds the dentry to the indexed directory.
1905  */
1906 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1907                             struct inode *dir,
1908                             struct inode *inode, struct buffer_head *bh)
1909 {
1910         struct buffer_head *bh2;
1911         struct dx_root  *root;
1912         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1913         struct dx_entry *entries;
1914         struct ext4_dir_entry_2 *de, *de2;
1915         struct ext4_dir_entry_tail *t;
1916         char            *data1, *top;
1917         unsigned        len;
1918         int             retval;
1919         unsigned        blocksize;
1920         ext4_lblk_t  block;
1921         struct fake_dirent *fde;
1922         int csum_size = 0;
1923
1924         if (ext4_has_metadata_csum(inode->i_sb))
1925                 csum_size = sizeof(struct ext4_dir_entry_tail);
1926
1927         blocksize =  dir->i_sb->s_blocksize;
1928         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1929         BUFFER_TRACE(bh, "get_write_access");
1930         retval = ext4_journal_get_write_access(handle, bh);
1931         if (retval) {
1932                 ext4_std_error(dir->i_sb, retval);
1933                 brelse(bh);
1934                 return retval;
1935         }
1936         root = (struct dx_root *) bh->b_data;
1937
1938         /* The 0th block becomes the root, move the dirents out */
1939         fde = &root->dotdot;
1940         de = (struct ext4_dir_entry_2 *)((char *)fde +
1941                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1942         if ((char *) de >= (((char *) root) + blocksize)) {
1943                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1944                 brelse(bh);
1945                 return -EFSCORRUPTED;
1946         }
1947         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1948
1949         /* Allocate new block for the 0th block's dirents */
1950         bh2 = ext4_append(handle, dir, &block);
1951         if (IS_ERR(bh2)) {
1952                 brelse(bh);
1953                 return PTR_ERR(bh2);
1954         }
1955         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1956         data1 = bh2->b_data;
1957
1958         memcpy (data1, de, len);
1959         de = (struct ext4_dir_entry_2 *) data1;
1960         top = data1 + len;
1961         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1962                 de = de2;
1963         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1964                                            (char *) de,
1965                                            blocksize);
1966
1967         if (csum_size) {
1968                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1969                 initialize_dirent_tail(t, blocksize);
1970         }
1971
1972         /* Initialize the root; the dot dirents already exist */
1973         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1974         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1975                                            blocksize);
1976         memset (&root->info, 0, sizeof(root->info));
1977         root->info.info_length = sizeof(root->info);
1978         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1979         entries = root->entries;
1980         dx_set_block(entries, 1);
1981         dx_set_count(entries, 1);
1982         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1983
1984         /* Initialize as for dx_probe */
1985         fname->hinfo.hash_version = root->info.hash_version;
1986         if (fname->hinfo.hash_version <= DX_HASH_TEA)
1987                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1988         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1989         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1990
1991         memset(frames, 0, sizeof(frames));
1992         frame = frames;
1993         frame->entries = entries;
1994         frame->at = entries;
1995         frame->bh = bh;
1996
1997         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1998         if (retval)
1999                 goto out_frames;
2000         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2001         if (retval)
2002                 goto out_frames;
2003
2004         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2005         if (IS_ERR(de)) {
2006                 retval = PTR_ERR(de);
2007                 goto out_frames;
2008         }
2009
2010         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2011 out_frames:
2012         /*
2013          * Even if the block split failed, we have to properly write
2014          * out all the changes we did so far. Otherwise we can end up
2015          * with corrupted filesystem.
2016          */
2017         if (retval)
2018                 ext4_mark_inode_dirty(handle, dir);
2019         dx_release(frames);
2020         brelse(bh2);
2021         return retval;
2022 }
2023
2024 /*
2025  *      ext4_add_entry()
2026  *
2027  * adds a file entry to the specified directory, using the same
2028  * semantics as ext4_find_entry(). It returns NULL if it failed.
2029  *
2030  * NOTE!! The inode part of 'de' is left at 0 - which means you
2031  * may not sleep between calling this and putting something into
2032  * the entry, as someone else might have used it while you slept.
2033  */
2034 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2035                           struct inode *inode)
2036 {
2037         struct inode *dir = d_inode(dentry->d_parent);
2038         struct buffer_head *bh = NULL;
2039         struct ext4_dir_entry_2 *de;
2040         struct ext4_dir_entry_tail *t;
2041         struct super_block *sb;
2042         struct ext4_filename fname;
2043         int     retval;
2044         int     dx_fallback=0;
2045         unsigned blocksize;
2046         ext4_lblk_t block, blocks;
2047         int     csum_size = 0;
2048
2049         if (ext4_has_metadata_csum(inode->i_sb))
2050                 csum_size = sizeof(struct ext4_dir_entry_tail);
2051
2052         sb = dir->i_sb;
2053         blocksize = sb->s_blocksize;
2054         if (!dentry->d_name.len)
2055                 return -EINVAL;
2056
2057         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2058         if (retval)
2059                 return retval;
2060
2061         if (ext4_has_inline_data(dir)) {
2062                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2063                 if (retval < 0)
2064                         goto out;
2065                 if (retval == 1) {
2066                         retval = 0;
2067                         goto out;
2068                 }
2069         }
2070
2071         if (is_dx(dir)) {
2072                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2073                 if (!retval || (retval != ERR_BAD_DX_DIR))
2074                         goto out;
2075                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2076                 dx_fallback++;
2077                 ext4_mark_inode_dirty(handle, dir);
2078         }
2079         blocks = dir->i_size >> sb->s_blocksize_bits;
2080         for (block = 0; block < blocks; block++) {
2081                 bh = ext4_read_dirblock(dir, block, DIRENT);
2082                 if (IS_ERR(bh)) {
2083                         retval = PTR_ERR(bh);
2084                         bh = NULL;
2085                         goto out;
2086                 }
2087                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2088                                            NULL, bh);
2089                 if (retval != -ENOSPC)
2090                         goto out;
2091
2092                 if (blocks == 1 && !dx_fallback &&
2093                     ext4_has_feature_dir_index(sb)) {
2094                         retval = make_indexed_dir(handle, &fname, dir,
2095                                                   inode, bh);
2096                         bh = NULL; /* make_indexed_dir releases bh */
2097                         goto out;
2098                 }
2099                 brelse(bh);
2100         }
2101         bh = ext4_append(handle, dir, &block);
2102         if (IS_ERR(bh)) {
2103                 retval = PTR_ERR(bh);
2104                 bh = NULL;
2105                 goto out;
2106         }
2107         de = (struct ext4_dir_entry_2 *) bh->b_data;
2108         de->inode = 0;
2109         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2110
2111         if (csum_size) {
2112                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2113                 initialize_dirent_tail(t, blocksize);
2114         }
2115
2116         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2117 out:
2118         ext4_fname_free_filename(&fname);
2119         brelse(bh);
2120         if (retval == 0)
2121                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2122         return retval;
2123 }
2124
2125 /*
2126  * Returns 0 for success, or a negative error value
2127  */
2128 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2129                              struct inode *dir, struct inode *inode)
2130 {
2131         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2132         struct dx_entry *entries, *at;
2133         struct buffer_head *bh;
2134         struct super_block *sb = dir->i_sb;
2135         struct ext4_dir_entry_2 *de;
2136         int restart;
2137         int err;
2138
2139 again:
2140         restart = 0;
2141         frame = dx_probe(fname, dir, NULL, frames);
2142         if (IS_ERR(frame))
2143                 return PTR_ERR(frame);
2144         entries = frame->entries;
2145         at = frame->at;
2146         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2147         if (IS_ERR(bh)) {
2148                 err = PTR_ERR(bh);
2149                 bh = NULL;
2150                 goto cleanup;
2151         }
2152
2153         BUFFER_TRACE(bh, "get_write_access");
2154         err = ext4_journal_get_write_access(handle, bh);
2155         if (err)
2156                 goto journal_error;
2157
2158         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2159         if (err != -ENOSPC)
2160                 goto cleanup;
2161
2162         err = 0;
2163         /* Block full, should compress but for now just split */
2164         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2165                        dx_get_count(entries), dx_get_limit(entries)));
2166         /* Need to split index? */
2167         if (dx_get_count(entries) == dx_get_limit(entries)) {
2168                 ext4_lblk_t newblock;
2169                 int levels = frame - frames + 1;
2170                 unsigned int icount;
2171                 int add_level = 1;
2172                 struct dx_entry *entries2;
2173                 struct dx_node *node2;
2174                 struct buffer_head *bh2;
2175
2176                 while (frame > frames) {
2177                         if (dx_get_count((frame - 1)->entries) <
2178                             dx_get_limit((frame - 1)->entries)) {
2179                                 add_level = 0;
2180                                 break;
2181                         }
2182                         frame--; /* split higher index block */
2183                         at = frame->at;
2184                         entries = frame->entries;
2185                         restart = 1;
2186                 }
2187                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2188                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2189                                          "reach max htree level :%d",
2190                                          dir->i_ino, levels);
2191                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2192                                 ext4_warning(sb, "Large directory feature is "
2193                                                  "not enabled on this "
2194                                                  "filesystem");
2195                         }
2196                         err = -ENOSPC;
2197                         goto cleanup;
2198                 }
2199                 icount = dx_get_count(entries);
2200                 bh2 = ext4_append(handle, dir, &newblock);
2201                 if (IS_ERR(bh2)) {
2202                         err = PTR_ERR(bh2);
2203                         goto cleanup;
2204                 }
2205                 node2 = (struct dx_node *)(bh2->b_data);
2206                 entries2 = node2->entries;
2207                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2208                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2209                                                            sb->s_blocksize);
2210                 BUFFER_TRACE(frame->bh, "get_write_access");
2211                 err = ext4_journal_get_write_access(handle, frame->bh);
2212                 if (err)
2213                         goto journal_error;
2214                 if (!add_level) {
2215                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2216                         unsigned hash2 = dx_get_hash(entries + icount1);
2217                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2218                                        icount1, icount2));
2219
2220                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2221                         err = ext4_journal_get_write_access(handle,
2222                                                              (frame - 1)->bh);
2223                         if (err)
2224                                 goto journal_error;
2225
2226                         memcpy((char *) entries2, (char *) (entries + icount1),
2227                                icount2 * sizeof(struct dx_entry));
2228                         dx_set_count(entries, icount1);
2229                         dx_set_count(entries2, icount2);
2230                         dx_set_limit(entries2, dx_node_limit(dir));
2231
2232                         /* Which index block gets the new entry? */
2233                         if (at - entries >= icount1) {
2234                                 frame->at = at = at - entries - icount1 + entries2;
2235                                 frame->entries = entries = entries2;
2236                                 swap(frame->bh, bh2);
2237                         }
2238                         dx_insert_block((frame - 1), hash2, newblock);
2239                         dxtrace(dx_show_index("node", frame->entries));
2240                         dxtrace(dx_show_index("node",
2241                                ((struct dx_node *) bh2->b_data)->entries));
2242                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2243                         if (err)
2244                                 goto journal_error;
2245                         brelse (bh2);
2246                         err = ext4_handle_dirty_dx_node(handle, dir,
2247                                                    (frame - 1)->bh);
2248                         if (err)
2249                                 goto journal_error;
2250                         if (restart) {
2251                                 err = ext4_handle_dirty_dx_node(handle, dir,
2252                                                            frame->bh);
2253                                 goto journal_error;
2254                         }
2255                 } else {
2256                         struct dx_root *dxroot;
2257                         memcpy((char *) entries2, (char *) entries,
2258                                icount * sizeof(struct dx_entry));
2259                         dx_set_limit(entries2, dx_node_limit(dir));
2260
2261                         /* Set up root */
2262                         dx_set_count(entries, 1);
2263                         dx_set_block(entries + 0, newblock);
2264                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2265                         dxroot->info.indirect_levels += 1;
2266                         dxtrace(printk(KERN_DEBUG
2267                                        "Creating %d level index...\n",
2268                                        info->indirect_levels));
2269                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2270                         if (err)
2271                                 goto journal_error;
2272                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2273                         brelse(bh2);
2274                         restart = 1;
2275                         goto journal_error;
2276                 }
2277         }
2278         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2279         if (IS_ERR(de)) {
2280                 err = PTR_ERR(de);
2281                 goto cleanup;
2282         }
2283         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2284         goto cleanup;
2285
2286 journal_error:
2287         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2288 cleanup:
2289         brelse(bh);
2290         dx_release(frames);
2291         /* @restart is true means htree-path has been changed, we need to
2292          * repeat dx_probe() to find out valid htree-path
2293          */
2294         if (restart && err == 0)
2295                 goto again;
2296         return err;
2297 }
2298
2299 /*
2300  * ext4_generic_delete_entry deletes a directory entry by merging it
2301  * with the previous entry
2302  */
2303 int ext4_generic_delete_entry(handle_t *handle,
2304                               struct inode *dir,
2305                               struct ext4_dir_entry_2 *de_del,
2306                               struct buffer_head *bh,
2307                               void *entry_buf,
2308                               int buf_size,
2309                               int csum_size)
2310 {
2311         struct ext4_dir_entry_2 *de, *pde;
2312         unsigned int blocksize = dir->i_sb->s_blocksize;
2313         int i;
2314
2315         i = 0;
2316         pde = NULL;
2317         de = (struct ext4_dir_entry_2 *)entry_buf;
2318         while (i < buf_size - csum_size) {
2319                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2320                                          bh->b_data, bh->b_size, i))
2321                         return -EFSCORRUPTED;
2322                 if (de == de_del)  {
2323                         if (pde)
2324                                 pde->rec_len = ext4_rec_len_to_disk(
2325                                         ext4_rec_len_from_disk(pde->rec_len,
2326                                                                blocksize) +
2327                                         ext4_rec_len_from_disk(de->rec_len,
2328                                                                blocksize),
2329                                         blocksize);
2330                         else
2331                                 de->inode = 0;
2332                         inode_inc_iversion(dir);
2333                         return 0;
2334                 }
2335                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2336                 pde = de;
2337                 de = ext4_next_entry(de, blocksize);
2338         }
2339         return -ENOENT;
2340 }
2341
2342 static int ext4_delete_entry(handle_t *handle,
2343                              struct inode *dir,
2344                              struct ext4_dir_entry_2 *de_del,
2345                              struct buffer_head *bh)
2346 {
2347         int err, csum_size = 0;
2348
2349         if (ext4_has_inline_data(dir)) {
2350                 int has_inline_data = 1;
2351                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2352                                                &has_inline_data);
2353                 if (has_inline_data)
2354                         return err;
2355         }
2356
2357         if (ext4_has_metadata_csum(dir->i_sb))
2358                 csum_size = sizeof(struct ext4_dir_entry_tail);
2359
2360         BUFFER_TRACE(bh, "get_write_access");
2361         err = ext4_journal_get_write_access(handle, bh);
2362         if (unlikely(err))
2363                 goto out;
2364
2365         err = ext4_generic_delete_entry(handle, dir, de_del,
2366                                         bh, bh->b_data,
2367                                         dir->i_sb->s_blocksize, csum_size);
2368         if (err)
2369                 goto out;
2370
2371         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2372         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2373         if (unlikely(err))
2374                 goto out;
2375
2376         return 0;
2377 out:
2378         if (err != -ENOENT)
2379                 ext4_std_error(dir->i_sb, err);
2380         return err;
2381 }
2382
2383 /*
2384  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2385  * since this indicates that nlinks count was previously 1 to avoid overflowing
2386  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2387  * that subdirectory link counts are not being maintained accurately.
2388  *
2389  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2390  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2391  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2392  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2393  */
2394 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2395 {
2396         inc_nlink(inode);
2397         if (is_dx(inode) &&
2398             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2399                 set_nlink(inode, 1);
2400 }
2401
2402 /*
2403  * If a directory had nlink == 1, then we should let it be 1. This indicates
2404  * directory has >EXT4_LINK_MAX subdirs.
2405  */
2406 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2407 {
2408         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2409                 drop_nlink(inode);
2410 }
2411
2412
2413 static int ext4_add_nondir(handle_t *handle,
2414                 struct dentry *dentry, struct inode *inode)
2415 {
2416         int err = ext4_add_entry(handle, dentry, inode);
2417         if (!err) {
2418                 ext4_mark_inode_dirty(handle, inode);
2419                 unlock_new_inode(inode);
2420                 d_instantiate(dentry, inode);
2421                 return 0;
2422         }
2423         drop_nlink(inode);
2424         unlock_new_inode(inode);
2425         iput(inode);
2426         return err;
2427 }
2428
2429 /*
2430  * By the time this is called, we already have created
2431  * the directory cache entry for the new file, but it
2432  * is so far negative - it has no inode.
2433  *
2434  * If the create succeeds, we fill in the inode information
2435  * with d_instantiate().
2436  */
2437 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2438                        bool excl)
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                 inode->i_op = &ext4_file_inode_operations;
2457                 inode->i_fop = &ext4_file_operations;
2458                 ext4_set_aops(inode);
2459                 err = ext4_add_nondir(handle, dentry, inode);
2460                 if (!err && IS_DIRSYNC(dir))
2461                         ext4_handle_sync(handle);
2462         }
2463         if (handle)
2464                 ext4_journal_stop(handle);
2465         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2466                 goto retry;
2467         return err;
2468 }
2469
2470 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2471                       umode_t mode, dev_t rdev)
2472 {
2473         handle_t *handle;
2474         struct inode *inode;
2475         int err, credits, retries = 0;
2476
2477         err = dquot_initialize(dir);
2478         if (err)
2479                 return err;
2480
2481         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2482                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2483 retry:
2484         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2485                                             NULL, EXT4_HT_DIR, credits);
2486         handle = ext4_journal_current_handle();
2487         err = PTR_ERR(inode);
2488         if (!IS_ERR(inode)) {
2489                 init_special_inode(inode, inode->i_mode, rdev);
2490                 inode->i_op = &ext4_special_inode_operations;
2491                 err = ext4_add_nondir(handle, dentry, inode);
2492                 if (!err && IS_DIRSYNC(dir))
2493                         ext4_handle_sync(handle);
2494         }
2495         if (handle)
2496                 ext4_journal_stop(handle);
2497         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2498                 goto retry;
2499         return err;
2500 }
2501
2502 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2503 {
2504         handle_t *handle;
2505         struct inode *inode;
2506         int err, retries = 0;
2507
2508         err = dquot_initialize(dir);
2509         if (err)
2510                 return err;
2511
2512 retry:
2513         inode = ext4_new_inode_start_handle(dir, mode,
2514                                             NULL, 0, NULL,
2515                                             EXT4_HT_DIR,
2516                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2517                           4 + EXT4_XATTR_TRANS_BLOCKS);
2518         handle = ext4_journal_current_handle();
2519         err = PTR_ERR(inode);
2520         if (!IS_ERR(inode)) {
2521                 inode->i_op = &ext4_file_inode_operations;
2522                 inode->i_fop = &ext4_file_operations;
2523                 ext4_set_aops(inode);
2524                 d_tmpfile(dentry, inode);
2525                 err = ext4_orphan_add(handle, inode);
2526                 if (err)
2527                         goto err_unlock_inode;
2528                 mark_inode_dirty(inode);
2529                 unlock_new_inode(inode);
2530         }
2531         if (handle)
2532                 ext4_journal_stop(handle);
2533         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2534                 goto retry;
2535         return err;
2536 err_unlock_inode:
2537         ext4_journal_stop(handle);
2538         unlock_new_inode(inode);
2539         return err;
2540 }
2541
2542 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2543                           struct ext4_dir_entry_2 *de,
2544                           int blocksize, int csum_size,
2545                           unsigned int parent_ino, int dotdot_real_len)
2546 {
2547         de->inode = cpu_to_le32(inode->i_ino);
2548         de->name_len = 1;
2549         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2550                                            blocksize);
2551         strcpy(de->name, ".");
2552         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2553
2554         de = ext4_next_entry(de, blocksize);
2555         de->inode = cpu_to_le32(parent_ino);
2556         de->name_len = 2;
2557         if (!dotdot_real_len)
2558                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2559                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2560                                         blocksize);
2561         else
2562                 de->rec_len = ext4_rec_len_to_disk(
2563                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2564         strcpy(de->name, "..");
2565         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2566
2567         return ext4_next_entry(de, blocksize);
2568 }
2569
2570 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2571                              struct inode *inode)
2572 {
2573         struct buffer_head *dir_block = NULL;
2574         struct ext4_dir_entry_2 *de;
2575         struct ext4_dir_entry_tail *t;
2576         ext4_lblk_t block = 0;
2577         unsigned int blocksize = dir->i_sb->s_blocksize;
2578         int csum_size = 0;
2579         int err;
2580
2581         if (ext4_has_metadata_csum(dir->i_sb))
2582                 csum_size = sizeof(struct ext4_dir_entry_tail);
2583
2584         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2585                 err = ext4_try_create_inline_dir(handle, dir, inode);
2586                 if (err < 0 && err != -ENOSPC)
2587                         goto out;
2588                 if (!err)
2589                         goto out;
2590         }
2591
2592         inode->i_size = 0;
2593         dir_block = ext4_append(handle, inode, &block);
2594         if (IS_ERR(dir_block))
2595                 return PTR_ERR(dir_block);
2596         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2597         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2598         set_nlink(inode, 2);
2599         if (csum_size) {
2600                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2601                 initialize_dirent_tail(t, blocksize);
2602         }
2603
2604         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2605         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2606         if (err)
2607                 goto out;
2608         set_buffer_verified(dir_block);
2609 out:
2610         brelse(dir_block);
2611         return err;
2612 }
2613
2614 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2615 {
2616         handle_t *handle;
2617         struct inode *inode;
2618         int err, credits, retries = 0;
2619
2620         if (EXT4_DIR_LINK_MAX(dir))
2621                 return -EMLINK;
2622
2623         err = dquot_initialize(dir);
2624         if (err)
2625                 return err;
2626
2627         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2628                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2629 retry:
2630         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2631                                             &dentry->d_name,
2632                                             0, NULL, EXT4_HT_DIR, credits);
2633         handle = ext4_journal_current_handle();
2634         err = PTR_ERR(inode);
2635         if (IS_ERR(inode))
2636                 goto out_stop;
2637
2638         inode->i_op = &ext4_dir_inode_operations;
2639         inode->i_fop = &ext4_dir_operations;
2640         err = ext4_init_new_dir(handle, dir, inode);
2641         if (err)
2642                 goto out_clear_inode;
2643         err = ext4_mark_inode_dirty(handle, inode);
2644         if (!err)
2645                 err = ext4_add_entry(handle, dentry, inode);
2646         if (err) {
2647 out_clear_inode:
2648                 clear_nlink(inode);
2649                 unlock_new_inode(inode);
2650                 ext4_mark_inode_dirty(handle, inode);
2651                 iput(inode);
2652                 goto out_stop;
2653         }
2654         ext4_inc_count(handle, dir);
2655         ext4_update_dx_flag(dir);
2656         err = ext4_mark_inode_dirty(handle, dir);
2657         if (err)
2658                 goto out_clear_inode;
2659         unlock_new_inode(inode);
2660         d_instantiate(dentry, inode);
2661         if (IS_DIRSYNC(dir))
2662                 ext4_handle_sync(handle);
2663
2664 out_stop:
2665         if (handle)
2666                 ext4_journal_stop(handle);
2667         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2668                 goto retry;
2669         return err;
2670 }
2671
2672 /*
2673  * routine to check that the specified directory is empty (for rmdir)
2674  */
2675 bool ext4_empty_dir(struct inode *inode)
2676 {
2677         unsigned int offset;
2678         struct buffer_head *bh;
2679         struct ext4_dir_entry_2 *de, *de1;
2680         struct super_block *sb;
2681
2682         if (ext4_has_inline_data(inode)) {
2683                 int has_inline_data = 1;
2684                 int ret;
2685
2686                 ret = empty_inline_dir(inode, &has_inline_data);
2687                 if (has_inline_data)
2688                         return ret;
2689         }
2690
2691         sb = inode->i_sb;
2692         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2693                 EXT4_ERROR_INODE(inode, "invalid size");
2694                 return true;
2695         }
2696         bh = ext4_read_dirblock(inode, 0, EITHER);
2697         if (IS_ERR(bh))
2698                 return true;
2699
2700         de = (struct ext4_dir_entry_2 *) bh->b_data;
2701         de1 = ext4_next_entry(de, sb->s_blocksize);
2702         if (le32_to_cpu(de->inode) != inode->i_ino ||
2703                         le32_to_cpu(de1->inode) == 0 ||
2704                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2705                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2706                 brelse(bh);
2707                 return true;
2708         }
2709         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2710                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2711         de = ext4_next_entry(de1, sb->s_blocksize);
2712         while (offset < inode->i_size) {
2713                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2714                         unsigned int lblock;
2715                         brelse(bh);
2716                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2717                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2718                         if (IS_ERR(bh))
2719                                 return true;
2720                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2721                 }
2722                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2723                                          bh->b_data, bh->b_size, offset)) {
2724                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2725                                                          sb->s_blocksize);
2726                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2727                         continue;
2728                 }
2729                 if (le32_to_cpu(de->inode)) {
2730                         brelse(bh);
2731                         return false;
2732                 }
2733                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2734                 de = ext4_next_entry(de, sb->s_blocksize);
2735         }
2736         brelse(bh);
2737         return true;
2738 }
2739
2740 /*
2741  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2742  * such inodes, starting at the superblock, in case we crash before the
2743  * file is closed/deleted, or in case the inode truncate spans multiple
2744  * transactions and the last transaction is not recovered after a crash.
2745  *
2746  * At filesystem recovery time, we walk this list deleting unlinked
2747  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2748  *
2749  * Orphan list manipulation functions must be called under i_mutex unless
2750  * we are just creating the inode or deleting it.
2751  */
2752 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2753 {
2754         struct super_block *sb = inode->i_sb;
2755         struct ext4_sb_info *sbi = EXT4_SB(sb);
2756         struct ext4_iloc iloc;
2757         int err = 0, rc;
2758         bool dirty = false;
2759
2760         if (!sbi->s_journal || is_bad_inode(inode))
2761                 return 0;
2762
2763         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2764                      !inode_is_locked(inode));
2765         /*
2766          * Exit early if inode already is on orphan list. This is a big speedup
2767          * since we don't have to contend on the global s_orphan_lock.
2768          */
2769         if (!list_empty(&EXT4_I(inode)->i_orphan))
2770                 return 0;
2771
2772         /*
2773          * Orphan handling is only valid for files with data blocks
2774          * being truncated, or files being unlinked. Note that we either
2775          * hold i_mutex, or the inode can not be referenced from outside,
2776          * so i_nlink should not be bumped due to race
2777          */
2778         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2779                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2780
2781         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2782         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2783         if (err)
2784                 goto out;
2785
2786         err = ext4_reserve_inode_write(handle, inode, &iloc);
2787         if (err)
2788                 goto out;
2789
2790         mutex_lock(&sbi->s_orphan_lock);
2791         /*
2792          * Due to previous errors inode may be already a part of on-disk
2793          * orphan list. If so skip on-disk list modification.
2794          */
2795         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2796             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2797                 /* Insert this inode at the head of the on-disk orphan list */
2798                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2799                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2800                 dirty = true;
2801         }
2802         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2803         mutex_unlock(&sbi->s_orphan_lock);
2804
2805         if (dirty) {
2806                 err = ext4_handle_dirty_super(handle, sb);
2807                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2808                 if (!err)
2809                         err = rc;
2810                 if (err) {
2811                         /*
2812                          * We have to remove inode from in-memory list if
2813                          * addition to on disk orphan list failed. Stray orphan
2814                          * list entries can cause panics at unmount time.
2815                          */
2816                         mutex_lock(&sbi->s_orphan_lock);
2817                         list_del_init(&EXT4_I(inode)->i_orphan);
2818                         mutex_unlock(&sbi->s_orphan_lock);
2819                 }
2820         }
2821         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2822         jbd_debug(4, "orphan inode %lu will point to %d\n",
2823                         inode->i_ino, NEXT_ORPHAN(inode));
2824 out:
2825         ext4_std_error(sb, err);
2826         return err;
2827 }
2828
2829 /*
2830  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2831  * of such inodes stored on disk, because it is finally being cleaned up.
2832  */
2833 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2834 {
2835         struct list_head *prev;
2836         struct ext4_inode_info *ei = EXT4_I(inode);
2837         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2838         __u32 ino_next;
2839         struct ext4_iloc iloc;
2840         int err = 0;
2841
2842         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2843                 return 0;
2844
2845         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2846                      !inode_is_locked(inode));
2847         /* Do this quick check before taking global s_orphan_lock. */
2848         if (list_empty(&ei->i_orphan))
2849                 return 0;
2850
2851         if (handle) {
2852                 /* Grab inode buffer early before taking global s_orphan_lock */
2853                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2854         }
2855
2856         mutex_lock(&sbi->s_orphan_lock);
2857         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2858
2859         prev = ei->i_orphan.prev;
2860         list_del_init(&ei->i_orphan);
2861
2862         /* If we're on an error path, we may not have a valid
2863          * transaction handle with which to update the orphan list on
2864          * disk, but we still need to remove the inode from the linked
2865          * list in memory. */
2866         if (!handle || err) {
2867                 mutex_unlock(&sbi->s_orphan_lock);
2868                 goto out_err;
2869         }
2870
2871         ino_next = NEXT_ORPHAN(inode);
2872         if (prev == &sbi->s_orphan) {
2873                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2874                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2875                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2876                 if (err) {
2877                         mutex_unlock(&sbi->s_orphan_lock);
2878                         goto out_brelse;
2879                 }
2880                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2881                 mutex_unlock(&sbi->s_orphan_lock);
2882                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2883         } else {
2884                 struct ext4_iloc iloc2;
2885                 struct inode *i_prev =
2886                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2887
2888                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2889                           i_prev->i_ino, ino_next);
2890                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2891                 if (err) {
2892                         mutex_unlock(&sbi->s_orphan_lock);
2893                         goto out_brelse;
2894                 }
2895                 NEXT_ORPHAN(i_prev) = ino_next;
2896                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2897                 mutex_unlock(&sbi->s_orphan_lock);
2898         }
2899         if (err)
2900                 goto out_brelse;
2901         NEXT_ORPHAN(inode) = 0;
2902         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2903 out_err:
2904         ext4_std_error(inode->i_sb, err);
2905         return err;
2906
2907 out_brelse:
2908         brelse(iloc.bh);
2909         goto out_err;
2910 }
2911
2912 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2913 {
2914         int retval;
2915         struct inode *inode;
2916         struct buffer_head *bh;
2917         struct ext4_dir_entry_2 *de;
2918         handle_t *handle = NULL;
2919
2920         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2921                 return -EIO;
2922
2923         /* Initialize quotas before so that eventual writes go in
2924          * separate transaction */
2925         retval = dquot_initialize(dir);
2926         if (retval)
2927                 return retval;
2928         retval = dquot_initialize(d_inode(dentry));
2929         if (retval)
2930                 return retval;
2931
2932         retval = -ENOENT;
2933         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2934         if (IS_ERR(bh))
2935                 return PTR_ERR(bh);
2936         if (!bh)
2937                 goto end_rmdir;
2938
2939         inode = d_inode(dentry);
2940
2941         retval = -EFSCORRUPTED;
2942         if (le32_to_cpu(de->inode) != inode->i_ino)
2943                 goto end_rmdir;
2944
2945         retval = -ENOTEMPTY;
2946         if (!ext4_empty_dir(inode))
2947                 goto end_rmdir;
2948
2949         handle = ext4_journal_start(dir, EXT4_HT_DIR,
2950                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2951         if (IS_ERR(handle)) {
2952                 retval = PTR_ERR(handle);
2953                 handle = NULL;
2954                 goto end_rmdir;
2955         }
2956
2957         if (IS_DIRSYNC(dir))
2958                 ext4_handle_sync(handle);
2959
2960         retval = ext4_delete_entry(handle, dir, de, bh);
2961         if (retval)
2962                 goto end_rmdir;
2963         if (!EXT4_DIR_LINK_EMPTY(inode))
2964                 ext4_warning_inode(inode,
2965                              "empty directory '%.*s' has too many links (%u)",
2966                              dentry->d_name.len, dentry->d_name.name,
2967                              inode->i_nlink);
2968         inode->i_version++;
2969         clear_nlink(inode);
2970         /* There's no need to set i_disksize: the fact that i_nlink is
2971          * zero will ensure that the right thing happens during any
2972          * recovery. */
2973         inode->i_size = 0;
2974         ext4_orphan_add(handle, inode);
2975         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2976         ext4_mark_inode_dirty(handle, inode);
2977         ext4_dec_count(handle, dir);
2978         ext4_update_dx_flag(dir);
2979         ext4_mark_inode_dirty(handle, dir);
2980
2981 end_rmdir:
2982         brelse(bh);
2983         if (handle)
2984                 ext4_journal_stop(handle);
2985         return retval;
2986 }
2987
2988 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2989 {
2990         int retval;
2991         struct inode *inode;
2992         struct buffer_head *bh;
2993         struct ext4_dir_entry_2 *de;
2994         handle_t *handle = NULL;
2995
2996         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2997                 return -EIO;
2998
2999         trace_ext4_unlink_enter(dir, dentry);
3000         /* Initialize quotas before so that eventual writes go
3001          * in separate transaction */
3002         retval = dquot_initialize(dir);
3003         if (retval)
3004                 return retval;
3005         retval = dquot_initialize(d_inode(dentry));
3006         if (retval)
3007                 return retval;
3008
3009         retval = -ENOENT;
3010         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3011         if (IS_ERR(bh))
3012                 return PTR_ERR(bh);
3013         if (!bh)
3014                 goto end_unlink;
3015
3016         inode = d_inode(dentry);
3017
3018         retval = -EFSCORRUPTED;
3019         if (le32_to_cpu(de->inode) != inode->i_ino)
3020                 goto end_unlink;
3021
3022         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3023                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3024         if (IS_ERR(handle)) {
3025                 retval = PTR_ERR(handle);
3026                 handle = NULL;
3027                 goto end_unlink;
3028         }
3029
3030         if (IS_DIRSYNC(dir))
3031                 ext4_handle_sync(handle);
3032
3033         if (inode->i_nlink == 0) {
3034                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3035                                    dentry->d_name.len, dentry->d_name.name);
3036                 set_nlink(inode, 1);
3037         }
3038         retval = ext4_delete_entry(handle, dir, de, bh);
3039         if (retval)
3040                 goto end_unlink;
3041         dir->i_ctime = dir->i_mtime = current_time(dir);
3042         ext4_update_dx_flag(dir);
3043         ext4_mark_inode_dirty(handle, dir);
3044         drop_nlink(inode);
3045         if (!inode->i_nlink)
3046                 ext4_orphan_add(handle, inode);
3047         inode->i_ctime = current_time(inode);
3048         ext4_mark_inode_dirty(handle, inode);
3049
3050 end_unlink:
3051         brelse(bh);
3052         if (handle)
3053                 ext4_journal_stop(handle);
3054         trace_ext4_unlink_exit(dentry, retval);
3055         return retval;
3056 }
3057
3058 static int ext4_symlink(struct inode *dir,
3059                         struct dentry *dentry, const char *symname)
3060 {
3061         handle_t *handle;
3062         struct inode *inode;
3063         int err, len = strlen(symname);
3064         int credits;
3065         bool encryption_required;
3066         struct fscrypt_str disk_link;
3067         struct fscrypt_symlink_data *sd = NULL;
3068
3069         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3070                 return -EIO;
3071
3072         disk_link.len = len + 1;
3073         disk_link.name = (char *) symname;
3074
3075         encryption_required = (ext4_encrypted_inode(dir) ||
3076                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3077         if (encryption_required) {
3078                 err = fscrypt_get_encryption_info(dir);
3079                 if (err)
3080                         return err;
3081                 if (!fscrypt_has_encryption_key(dir))
3082                         return -ENOKEY;
3083                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3084                                  sizeof(struct fscrypt_symlink_data));
3085                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3086                 if (!sd)
3087                         return -ENOMEM;
3088         }
3089
3090         if (disk_link.len > dir->i_sb->s_blocksize) {
3091                 err = -ENAMETOOLONG;
3092                 goto err_free_sd;
3093         }
3094
3095         err = dquot_initialize(dir);
3096         if (err)
3097                 goto err_free_sd;
3098
3099         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3100                 /*
3101                  * For non-fast symlinks, we just allocate inode and put it on
3102                  * orphan list in the first transaction => we need bitmap,
3103                  * group descriptor, sb, inode block, quota blocks, and
3104                  * possibly selinux xattr blocks.
3105                  */
3106                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3107                           EXT4_XATTR_TRANS_BLOCKS;
3108         } else {
3109                 /*
3110                  * Fast symlink. We have to add entry to directory
3111                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3112                  * allocate new inode (bitmap, group descriptor, inode block,
3113                  * quota blocks, sb is already counted in previous macros).
3114                  */
3115                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3116                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3117         }
3118
3119         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3120                                             &dentry->d_name, 0, NULL,
3121                                             EXT4_HT_DIR, credits);
3122         handle = ext4_journal_current_handle();
3123         if (IS_ERR(inode)) {
3124                 if (handle)
3125                         ext4_journal_stop(handle);
3126                 err = PTR_ERR(inode);
3127                 goto err_free_sd;
3128         }
3129
3130         if (encryption_required) {
3131                 struct qstr istr;
3132                 struct fscrypt_str ostr =
3133                         FSTR_INIT(sd->encrypted_path, disk_link.len);
3134
3135                 istr.name = (const unsigned char *) symname;
3136                 istr.len = len;
3137                 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3138                 if (err)
3139                         goto err_drop_inode;
3140                 sd->len = cpu_to_le16(ostr.len);
3141                 disk_link.name = (char *) sd;
3142                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3143         }
3144
3145         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3146                 if (!encryption_required)
3147                         inode->i_op = &ext4_symlink_inode_operations;
3148                 inode_nohighmem(inode);
3149                 ext4_set_aops(inode);
3150                 /*
3151                  * We cannot call page_symlink() with transaction started
3152                  * because it calls into ext4_write_begin() which can wait
3153                  * for transaction commit if we are running out of space
3154                  * and thus we deadlock. So we have to stop transaction now
3155                  * and restart it when symlink contents is written.
3156                  *
3157                  * To keep fs consistent in case of crash, we have to put inode
3158                  * to orphan list in the mean time.
3159                  */
3160                 drop_nlink(inode);
3161                 err = ext4_orphan_add(handle, inode);
3162                 ext4_journal_stop(handle);
3163                 handle = NULL;
3164                 if (err)
3165                         goto err_drop_inode;
3166                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3167                 if (err)
3168                         goto err_drop_inode;
3169                 /*
3170                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3171                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3172                  */
3173                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3174                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3175                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3176                 if (IS_ERR(handle)) {
3177                         err = PTR_ERR(handle);
3178                         handle = NULL;
3179                         goto err_drop_inode;
3180                 }
3181                 set_nlink(inode, 1);
3182                 err = ext4_orphan_del(handle, inode);
3183                 if (err)
3184                         goto err_drop_inode;
3185         } else {
3186                 /* clear the extent format for fast symlink */
3187                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3188                 if (!encryption_required) {
3189                         inode->i_op = &ext4_fast_symlink_inode_operations;
3190                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3191                 }
3192                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3193                        disk_link.len);
3194                 inode->i_size = disk_link.len - 1;
3195         }
3196         EXT4_I(inode)->i_disksize = inode->i_size;
3197         err = ext4_add_nondir(handle, dentry, inode);
3198         if (!err && IS_DIRSYNC(dir))
3199                 ext4_handle_sync(handle);
3200
3201         if (handle)
3202                 ext4_journal_stop(handle);
3203         kfree(sd);
3204         return err;
3205 err_drop_inode:
3206         if (handle)
3207                 ext4_journal_stop(handle);
3208         clear_nlink(inode);
3209         unlock_new_inode(inode);
3210         iput(inode);
3211 err_free_sd:
3212         kfree(sd);
3213         return err;
3214 }
3215
3216 static int ext4_link(struct dentry *old_dentry,
3217                      struct inode *dir, struct dentry *dentry)
3218 {
3219         handle_t *handle;
3220         struct inode *inode = d_inode(old_dentry);
3221         int err, retries = 0;
3222
3223         if (inode->i_nlink >= EXT4_LINK_MAX)
3224                 return -EMLINK;
3225         if (ext4_encrypted_inode(dir) &&
3226                         !fscrypt_has_permitted_context(dir, inode))
3227                 return -EPERM;
3228
3229        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3230            (!projid_eq(EXT4_I(dir)->i_projid,
3231                        EXT4_I(old_dentry->d_inode)->i_projid)))
3232                 return -EXDEV;
3233
3234         err = dquot_initialize(dir);
3235         if (err)
3236                 return err;
3237
3238 retry:
3239         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3240                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3241                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3242         if (IS_ERR(handle))
3243                 return PTR_ERR(handle);
3244
3245         if (IS_DIRSYNC(dir))
3246                 ext4_handle_sync(handle);
3247
3248         inode->i_ctime = current_time(inode);
3249         ext4_inc_count(handle, inode);
3250         ihold(inode);
3251
3252         err = ext4_add_entry(handle, dentry, inode);
3253         if (!err) {
3254                 ext4_mark_inode_dirty(handle, inode);
3255                 /* this can happen only for tmpfile being
3256                  * linked the first time
3257                  */
3258                 if (inode->i_nlink == 1)
3259                         ext4_orphan_del(handle, inode);
3260                 d_instantiate(dentry, inode);
3261         } else {
3262                 drop_nlink(inode);
3263                 iput(inode);
3264         }
3265         ext4_journal_stop(handle);
3266         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3267                 goto retry;
3268         return err;
3269 }
3270
3271
3272 /*
3273  * Try to find buffer head where contains the parent block.
3274  * It should be the inode block if it is inlined or the 1st block
3275  * if it is a normal dir.
3276  */
3277 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3278                                         struct inode *inode,
3279                                         int *retval,
3280                                         struct ext4_dir_entry_2 **parent_de,
3281                                         int *inlined)
3282 {
3283         struct buffer_head *bh;
3284
3285         if (!ext4_has_inline_data(inode)) {
3286                 bh = ext4_read_dirblock(inode, 0, EITHER);
3287                 if (IS_ERR(bh)) {
3288                         *retval = PTR_ERR(bh);
3289                         return NULL;
3290                 }
3291                 *parent_de = ext4_next_entry(
3292                                         (struct ext4_dir_entry_2 *)bh->b_data,
3293                                         inode->i_sb->s_blocksize);
3294                 return bh;
3295         }
3296
3297         *inlined = 1;
3298         return ext4_get_first_inline_block(inode, parent_de, retval);
3299 }
3300
3301 struct ext4_renament {
3302         struct inode *dir;
3303         struct dentry *dentry;
3304         struct inode *inode;
3305         bool is_dir;
3306         int dir_nlink_delta;
3307
3308         /* entry for "dentry" */
3309         struct buffer_head *bh;
3310         struct ext4_dir_entry_2 *de;
3311         int inlined;
3312
3313         /* entry for ".." in inode if it's a directory */
3314         struct buffer_head *dir_bh;
3315         struct ext4_dir_entry_2 *parent_de;
3316         int dir_inlined;
3317 };
3318
3319 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3320 {
3321         int retval;
3322
3323         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3324                                               &retval, &ent->parent_de,
3325                                               &ent->dir_inlined);
3326         if (!ent->dir_bh)
3327                 return retval;
3328         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3329                 return -EFSCORRUPTED;
3330         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3331         return ext4_journal_get_write_access(handle, ent->dir_bh);
3332 }
3333
3334 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3335                                   unsigned dir_ino)
3336 {
3337         int retval;
3338
3339         ent->parent_de->inode = cpu_to_le32(dir_ino);
3340         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3341         if (!ent->dir_inlined) {
3342                 if (is_dx(ent->inode)) {
3343                         retval = ext4_handle_dirty_dx_node(handle,
3344                                                            ent->inode,
3345                                                            ent->dir_bh);
3346                 } else {
3347                         retval = ext4_handle_dirty_dirent_node(handle,
3348                                                                ent->inode,
3349                                                                ent->dir_bh);
3350                 }
3351         } else {
3352                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3353         }
3354         if (retval) {
3355                 ext4_std_error(ent->dir->i_sb, retval);
3356                 return retval;
3357         }
3358         return 0;
3359 }
3360
3361 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3362                        unsigned ino, unsigned file_type)
3363 {
3364         int retval;
3365
3366         BUFFER_TRACE(ent->bh, "get write access");
3367         retval = ext4_journal_get_write_access(handle, ent->bh);
3368         if (retval)
3369                 return retval;
3370         ent->de->inode = cpu_to_le32(ino);
3371         if (ext4_has_feature_filetype(ent->dir->i_sb))
3372                 ent->de->file_type = file_type;
3373         ent->dir->i_version++;
3374         ent->dir->i_ctime = ent->dir->i_mtime =
3375                 current_time(ent->dir);
3376         ext4_mark_inode_dirty(handle, ent->dir);
3377         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3378         if (!ent->inlined) {
3379                 retval = ext4_handle_dirty_dirent_node(handle,
3380                                                        ent->dir, ent->bh);
3381                 if (unlikely(retval)) {
3382                         ext4_std_error(ent->dir->i_sb, retval);
3383                         return retval;
3384                 }
3385         }
3386         brelse(ent->bh);
3387         ent->bh = NULL;
3388
3389         return 0;
3390 }
3391
3392 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3393                                   const struct qstr *d_name)
3394 {
3395         int retval = -ENOENT;
3396         struct buffer_head *bh;
3397         struct ext4_dir_entry_2 *de;
3398
3399         bh = ext4_find_entry(dir, d_name, &de, NULL);
3400         if (IS_ERR(bh))
3401                 return PTR_ERR(bh);
3402         if (bh) {
3403                 retval = ext4_delete_entry(handle, dir, de, bh);
3404                 brelse(bh);
3405         }
3406         return retval;
3407 }
3408
3409 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3410                                int force_reread)
3411 {
3412         int retval;
3413         /*
3414          * ent->de could have moved from under us during htree split, so make
3415          * sure that we are deleting the right entry.  We might also be pointing
3416          * to a stale entry in the unused part of ent->bh so just checking inum
3417          * and the name isn't enough.
3418          */
3419         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3420             ent->de->name_len != ent->dentry->d_name.len ||
3421             strncmp(ent->de->name, ent->dentry->d_name.name,
3422                     ent->de->name_len) ||
3423             force_reread) {
3424                 retval = ext4_find_delete_entry(handle, ent->dir,
3425                                                 &ent->dentry->d_name);
3426         } else {
3427                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3428                 if (retval == -ENOENT) {
3429                         retval = ext4_find_delete_entry(handle, ent->dir,
3430                                                         &ent->dentry->d_name);
3431                 }
3432         }
3433
3434         if (retval) {
3435                 ext4_warning_inode(ent->dir,
3436                                    "Deleting old file: nlink %d, error=%d",
3437                                    ent->dir->i_nlink, retval);
3438         }
3439 }
3440
3441 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3442 {
3443         if (ent->dir_nlink_delta) {
3444                 if (ent->dir_nlink_delta == -1)
3445                         ext4_dec_count(handle, ent->dir);
3446                 else
3447                         ext4_inc_count(handle, ent->dir);
3448                 ext4_mark_inode_dirty(handle, ent->dir);
3449         }
3450 }
3451
3452 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3453                                               int credits, handle_t **h)
3454 {
3455         struct inode *wh;
3456         handle_t *handle;
3457         int retries = 0;
3458
3459         /*
3460          * for inode block, sb block, group summaries,
3461          * and inode bitmap
3462          */
3463         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3464                     EXT4_XATTR_TRANS_BLOCKS + 4);
3465 retry:
3466         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3467                                          &ent->dentry->d_name, 0, NULL,
3468                                          EXT4_HT_DIR, credits);
3469
3470         handle = ext4_journal_current_handle();
3471         if (IS_ERR(wh)) {
3472                 if (handle)
3473                         ext4_journal_stop(handle);
3474                 if (PTR_ERR(wh) == -ENOSPC &&
3475                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3476                         goto retry;
3477         } else {
3478                 *h = handle;
3479                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3480                 wh->i_op = &ext4_special_inode_operations;
3481         }
3482         return wh;
3483 }
3484
3485 /*
3486  * Anybody can rename anything with this: the permission checks are left to the
3487  * higher-level routines.
3488  *
3489  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3490  * while new_{dentry,inode) refers to the destination dentry/inode
3491  * This comes from rename(const char *oldpath, const char *newpath)
3492  */
3493 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3494                        struct inode *new_dir, struct dentry *new_dentry,
3495                        unsigned int flags)
3496 {
3497         handle_t *handle = NULL;
3498         struct ext4_renament old = {
3499                 .dir = old_dir,
3500                 .dentry = old_dentry,
3501                 .inode = d_inode(old_dentry),
3502         };
3503         struct ext4_renament new = {
3504                 .dir = new_dir,
3505                 .dentry = new_dentry,
3506                 .inode = d_inode(new_dentry),
3507         };
3508         int force_reread;
3509         int retval;
3510         struct inode *whiteout = NULL;
3511         int credits;
3512         u8 old_file_type;
3513
3514         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3515             (!projid_eq(EXT4_I(new_dir)->i_projid,
3516                         EXT4_I(old_dentry->d_inode)->i_projid)))
3517                 return -EXDEV;
3518
3519         if ((ext4_encrypted_inode(old_dir) &&
3520              !fscrypt_has_encryption_key(old_dir)) ||
3521             (ext4_encrypted_inode(new_dir) &&
3522              !fscrypt_has_encryption_key(new_dir)))
3523                 return -ENOKEY;
3524
3525         retval = dquot_initialize(old.dir);
3526         if (retval)
3527                 return retval;
3528         retval = dquot_initialize(new.dir);
3529         if (retval)
3530                 return retval;
3531
3532         /* Initialize quotas before so that eventual writes go
3533          * in separate transaction */
3534         if (new.inode) {
3535                 retval = dquot_initialize(new.inode);
3536                 if (retval)
3537                         return retval;
3538         }
3539
3540         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3541         if (IS_ERR(old.bh))
3542                 return PTR_ERR(old.bh);
3543         /*
3544          *  Check for inode number is _not_ due to possible IO errors.
3545          *  We might rmdir the source, keep it as pwd of some process
3546          *  and merrily kill the link to whatever was created under the
3547          *  same name. Goodbye sticky bit ;-<
3548          */
3549         retval = -ENOENT;
3550         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3551                 goto end_rename;
3552
3553         if ((old.dir != new.dir) &&
3554             ext4_encrypted_inode(new.dir) &&
3555             !fscrypt_has_permitted_context(new.dir, old.inode)) {
3556                 retval = -EPERM;
3557                 goto end_rename;
3558         }
3559
3560         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3561                                  &new.de, &new.inlined);
3562         if (IS_ERR(new.bh)) {
3563                 retval = PTR_ERR(new.bh);
3564                 new.bh = NULL;
3565                 goto end_rename;
3566         }
3567         if (new.bh) {
3568                 if (!new.inode) {
3569                         brelse(new.bh);
3570                         new.bh = NULL;
3571                 }
3572         }
3573         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3574                 ext4_alloc_da_blocks(old.inode);
3575
3576         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3577                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3578         if (!(flags & RENAME_WHITEOUT)) {
3579                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3580                 if (IS_ERR(handle)) {
3581                         retval = PTR_ERR(handle);
3582                         handle = NULL;
3583                         goto end_rename;
3584                 }
3585         } else {
3586                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3587                 if (IS_ERR(whiteout)) {
3588                         retval = PTR_ERR(whiteout);
3589                         whiteout = NULL;
3590                         goto end_rename;
3591                 }
3592         }
3593
3594         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3595                 ext4_handle_sync(handle);
3596
3597         if (S_ISDIR(old.inode->i_mode)) {
3598                 if (new.inode) {
3599                         retval = -ENOTEMPTY;
3600                         if (!ext4_empty_dir(new.inode))
3601                                 goto end_rename;
3602                 } else {
3603                         retval = -EMLINK;
3604                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3605                                 goto end_rename;
3606                 }
3607                 retval = ext4_rename_dir_prepare(handle, &old);
3608                 if (retval)
3609                         goto end_rename;
3610         }
3611         /*
3612          * If we're renaming a file within an inline_data dir and adding or
3613          * setting the new dirent causes a conversion from inline_data to
3614          * extents/blockmap, we need to force the dirent delete code to
3615          * re-read the directory, or else we end up trying to delete a dirent
3616          * from what is now the extent tree root (or a block map).
3617          */
3618         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3619                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3620
3621         old_file_type = old.de->file_type;
3622         if (whiteout) {
3623                 /*
3624                  * Do this before adding a new entry, so the old entry is sure
3625                  * to be still pointing to the valid old entry.
3626                  */
3627                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3628                                      EXT4_FT_CHRDEV);
3629                 if (retval)
3630                         goto end_rename;
3631                 ext4_mark_inode_dirty(handle, whiteout);
3632         }
3633         if (!new.bh) {
3634                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3635                 if (retval)
3636                         goto end_rename;
3637         } else {
3638                 retval = ext4_setent(handle, &new,
3639                                      old.inode->i_ino, old_file_type);
3640                 if (retval)
3641                         goto end_rename;
3642         }
3643         if (force_reread)
3644                 force_reread = !ext4_test_inode_flag(new.dir,
3645                                                      EXT4_INODE_INLINE_DATA);
3646
3647         /*
3648          * Like most other Unix systems, set the ctime for inodes on a
3649          * rename.
3650          */
3651         old.inode->i_ctime = current_time(old.inode);
3652         ext4_mark_inode_dirty(handle, old.inode);
3653
3654         if (!whiteout) {
3655                 /*
3656                  * ok, that's it
3657                  */
3658                 ext4_rename_delete(handle, &old, force_reread);
3659         }
3660
3661         if (new.inode) {
3662                 ext4_dec_count(handle, new.inode);
3663                 new.inode->i_ctime = current_time(new.inode);
3664         }
3665         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3666         ext4_update_dx_flag(old.dir);
3667         if (old.dir_bh) {
3668                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3669                 if (retval)
3670                         goto end_rename;
3671
3672                 ext4_dec_count(handle, old.dir);
3673                 if (new.inode) {
3674                         /* checked ext4_empty_dir above, can't have another
3675                          * parent, ext4_dec_count() won't work for many-linked
3676                          * dirs */
3677                         clear_nlink(new.inode);
3678                 } else {
3679                         ext4_inc_count(handle, new.dir);
3680                         ext4_update_dx_flag(new.dir);
3681                         ext4_mark_inode_dirty(handle, new.dir);
3682                 }
3683         }
3684         ext4_mark_inode_dirty(handle, old.dir);
3685         if (new.inode) {
3686                 ext4_mark_inode_dirty(handle, new.inode);
3687                 if (!new.inode->i_nlink)
3688                         ext4_orphan_add(handle, new.inode);
3689         }
3690         retval = 0;
3691
3692 end_rename:
3693         brelse(old.dir_bh);
3694         brelse(old.bh);
3695         brelse(new.bh);
3696         if (whiteout) {
3697                 if (retval)
3698                         drop_nlink(whiteout);
3699                 unlock_new_inode(whiteout);
3700                 iput(whiteout);
3701         }
3702         if (handle)
3703                 ext4_journal_stop(handle);
3704         return retval;
3705 }
3706
3707 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3708                              struct inode *new_dir, struct dentry *new_dentry)
3709 {
3710         handle_t *handle = NULL;
3711         struct ext4_renament old = {
3712                 .dir = old_dir,
3713                 .dentry = old_dentry,
3714                 .inode = d_inode(old_dentry),
3715         };
3716         struct ext4_renament new = {
3717                 .dir = new_dir,
3718                 .dentry = new_dentry,
3719                 .inode = d_inode(new_dentry),
3720         };
3721         u8 new_file_type;
3722         int retval;
3723         struct timespec ctime;
3724
3725         if ((ext4_encrypted_inode(old_dir) &&
3726              !fscrypt_has_encryption_key(old_dir)) ||
3727             (ext4_encrypted_inode(new_dir) &&
3728              !fscrypt_has_encryption_key(new_dir)))
3729                 return -ENOKEY;
3730
3731         if ((ext4_encrypted_inode(old_dir) ||
3732              ext4_encrypted_inode(new_dir)) &&
3733             (old_dir != new_dir) &&
3734             (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3735              !fscrypt_has_permitted_context(old_dir, new.inode)))
3736                 return -EPERM;
3737
3738         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3739              !projid_eq(EXT4_I(new_dir)->i_projid,
3740                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3741             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3742              !projid_eq(EXT4_I(old_dir)->i_projid,
3743                         EXT4_I(new_dentry->d_inode)->i_projid)))
3744                 return -EXDEV;
3745
3746         retval = dquot_initialize(old.dir);
3747         if (retval)
3748                 return retval;
3749         retval = dquot_initialize(new.dir);
3750         if (retval)
3751                 return retval;
3752
3753         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3754                                  &old.de, &old.inlined);
3755         if (IS_ERR(old.bh))
3756                 return PTR_ERR(old.bh);
3757         /*
3758          *  Check for inode number is _not_ due to possible IO errors.
3759          *  We might rmdir the source, keep it as pwd of some process
3760          *  and merrily kill the link to whatever was created under the
3761          *  same name. Goodbye sticky bit ;-<
3762          */
3763         retval = -ENOENT;
3764         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3765                 goto end_rename;
3766
3767         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3768                                  &new.de, &new.inlined);
3769         if (IS_ERR(new.bh)) {
3770                 retval = PTR_ERR(new.bh);
3771                 new.bh = NULL;
3772                 goto end_rename;
3773         }
3774
3775         /* RENAME_EXCHANGE case: old *and* new must both exist */
3776         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3777                 goto end_rename;
3778
3779         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3780                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3781                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3782         if (IS_ERR(handle)) {
3783                 retval = PTR_ERR(handle);
3784                 handle = NULL;
3785                 goto end_rename;
3786         }
3787
3788         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3789                 ext4_handle_sync(handle);
3790
3791         if (S_ISDIR(old.inode->i_mode)) {
3792                 old.is_dir = true;
3793                 retval = ext4_rename_dir_prepare(handle, &old);
3794                 if (retval)
3795                         goto end_rename;
3796         }
3797         if (S_ISDIR(new.inode->i_mode)) {
3798                 new.is_dir = true;
3799                 retval = ext4_rename_dir_prepare(handle, &new);
3800                 if (retval)
3801                         goto end_rename;
3802         }
3803
3804         /*
3805          * Other than the special case of overwriting a directory, parents'
3806          * nlink only needs to be modified if this is a cross directory rename.
3807          */
3808         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3809                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3810                 new.dir_nlink_delta = -old.dir_nlink_delta;
3811                 retval = -EMLINK;
3812                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3813                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3814                         goto end_rename;
3815         }
3816
3817         new_file_type = new.de->file_type;
3818         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3819         if (retval)
3820                 goto end_rename;
3821
3822         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3823         if (retval)
3824                 goto end_rename;
3825
3826         /*
3827          * Like most other Unix systems, set the ctime for inodes on a
3828          * rename.
3829          */
3830         ctime = current_time(old.inode);
3831         old.inode->i_ctime = ctime;
3832         new.inode->i_ctime = ctime;
3833         ext4_mark_inode_dirty(handle, old.inode);
3834         ext4_mark_inode_dirty(handle, new.inode);
3835
3836         if (old.dir_bh) {
3837                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3838                 if (retval)
3839                         goto end_rename;
3840         }
3841         if (new.dir_bh) {
3842                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3843                 if (retval)
3844                         goto end_rename;
3845         }
3846         ext4_update_dir_count(handle, &old);
3847         ext4_update_dir_count(handle, &new);
3848         retval = 0;
3849
3850 end_rename:
3851         brelse(old.dir_bh);
3852         brelse(new.dir_bh);
3853         brelse(old.bh);
3854         brelse(new.bh);
3855         if (handle)
3856                 ext4_journal_stop(handle);
3857         return retval;
3858 }
3859
3860 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3861                         struct inode *new_dir, struct dentry *new_dentry,
3862                         unsigned int flags)
3863 {
3864         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3865                 return -EIO;
3866
3867         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3868                 return -EINVAL;
3869
3870         if (flags & RENAME_EXCHANGE) {
3871                 return ext4_cross_rename(old_dir, old_dentry,
3872                                          new_dir, new_dentry);
3873         }
3874
3875         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3876 }
3877
3878 /*
3879  * directories can handle most operations...
3880  */
3881 const struct inode_operations ext4_dir_inode_operations = {
3882         .create         = ext4_create,
3883         .lookup         = ext4_lookup,
3884         .link           = ext4_link,
3885         .unlink         = ext4_unlink,
3886         .symlink        = ext4_symlink,
3887         .mkdir          = ext4_mkdir,
3888         .rmdir          = ext4_rmdir,
3889         .mknod          = ext4_mknod,
3890         .tmpfile        = ext4_tmpfile,
3891         .rename         = ext4_rename2,
3892         .setattr        = ext4_setattr,
3893         .getattr        = ext4_getattr,
3894         .listxattr      = ext4_listxattr,
3895         .get_acl        = ext4_get_acl,
3896         .set_acl        = ext4_set_acl,
3897         .fiemap         = ext4_fiemap,
3898 };
3899
3900 const struct inode_operations ext4_special_inode_operations = {
3901         .setattr        = ext4_setattr,
3902         .getattr        = ext4_getattr,
3903         .listxattr      = ext4_listxattr,
3904         .get_acl        = ext4_get_acl,
3905         .set_acl        = ext4_set_acl,
3906 };