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 /*
  84  * Hints to ext4_read_dirblock regarding whether we expect a directory
  85  * block being read to be an index block, or a block containing
  86  * directory entries (and if the latter, whether it was found via a
  87  * logical block in an htree index block).  This is used to control
  88  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  89  * directory block read from the storage device.  EITHER will means
  90  * the caller doesn't know what kind of directory block will be read,
  91  * so no specific verification will be done.
  92  */
  93 typedef enum {
  94         EITHER, INDEX, DIRENT, DIRENT_HTREE
  95 } dirblock_type_t;
  96
  97 #define ext4_read_dirblock(inode, block, type) \
  98         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
  99
 100 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 101                                                 ext4_lblk_t block,
 102                                                 dirblock_type_t type,
 103                                                 const char *func,
 104                                                 unsigned int line)
 105 {
 106         struct buffer_head *bh;
 107         struct ext4_dir_entry *dirent;
 108         int is_dx_block = 0;
 109
 110         bh = ext4_bread(NULL, inode, block, 0);
 111         if (IS_ERR(bh)) {
 112                 __ext4_warning(inode->i_sb, func, line,
 113                                "inode #%lu: lblock %lu: comm %s: "
 114                                "error %ld reading directory block",
 115                                inode->i_ino, (unsigned long)block,
 116                                current->comm, PTR_ERR(bh));
 117
 118                 return bh;
 119         }
 120         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 121                 ext4_error_inode(inode, func, line, block,
 122                                  "Directory hole found for htree %s block",
 123                                  (type == INDEX) ? "index" : "leaf");
 124                 return ERR_PTR(-EFSCORRUPTED);
 125         }
 126         if (!bh)
 127                 return NULL;
 128         dirent = (struct ext4_dir_entry *) bh->b_data;
 129         /* Determine whether or not we have an index block */
 130         if (is_dx(inode)) {
 131                 if (block == 0)
 132                         is_dx_block = 1;
 133                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 134                                                 inode->i_sb->s_blocksize) ==
 135                          inode->i_sb->s_blocksize)
 136                         is_dx_block = 1;
 137         }
 138         if (!is_dx_block && type == INDEX) {
 139                 ext4_error_inode(inode, func, line, block,
 140                        "directory leaf block found instead of index block");
 141                 brelse(bh);
 142                 return ERR_PTR(-EFSCORRUPTED);
 143         }
 144         if (!ext4_has_metadata_csum(inode->i_sb) ||
 145             buffer_verified(bh))
 146                 return bh;
 147
 148         /*
 149          * An empty leaf block can get mistaken for a index block; for
 150          * this reason, we can only check the index checksum when the
 151          * caller is sure it should be an index block.
 152          */
 153         if (is_dx_block && type == INDEX) {
 154                 if (ext4_dx_csum_verify(inode, dirent))
 155                         set_buffer_verified(bh);
 156                 else {
 157                         ext4_error_inode(inode, func, line, block,
 158                                          "Directory index failed checksum");
 159                         brelse(bh);
 160                         return ERR_PTR(-EFSBADCRC);
 161                 }
 162         }
 163         if (!is_dx_block) {
 164                 if (ext4_dirent_csum_verify(inode, dirent))
 165                         set_buffer_verified(bh);
 166                 else {
 167                         ext4_error_inode(inode, func, line, block,
 168                                          "Directory block failed checksum");
 169                         brelse(bh);
 170                         return ERR_PTR(-EFSBADCRC);
 171                 }
 172         }
 173         return bh;
 174 }
 175
 176 #ifndef assert
 177 #define assert(test) J_ASSERT(test)
 178 #endif
 179
 180 #ifdef DX_DEBUG
 181 #define dxtrace(command) command
 182 #else
 183 #define dxtrace(command)
 184 #endif
 185
 186 struct fake_dirent
 187 {
 188         __le32 inode;
 189         __le16 rec_len;
 190         u8 name_len;
 191         u8 file_type;
 192 };
 193
 194 struct dx_countlimit
 195 {
 196         __le16 limit;
 197         __le16 count;
 198 };
 199
 200 struct dx_entry
 201 {
 202         __le32 hash;
 203         __le32 block;
 204 };
 205
 206 /*
 207  * dx_root_info is laid out so that if it should somehow get overlaid by a
 208  * dirent the two low bits of the hash version will be zero.  Therefore, the
 209  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 210  */
 211
 212 struct dx_root
 213 {
 214         struct fake_dirent dot;
 215         char dot_name[4];
 216         struct fake_dirent dotdot;
 217         char dotdot_name[4];
 218         struct dx_root_info
 219         {
 220                 __le32 reserved_zero;
 221                 u8 hash_version;
 222                 u8 info_length; /* 8 */
 223                 u8 indirect_levels;
 224                 u8 unused_flags;
 225         }
 226         info;
 227         struct dx_entry entries[0];
 228 };
 229
 230 struct dx_node
 231 {
 232         struct fake_dirent fake;
 233         struct dx_entry entries[0];
 234 };
 235
 236
 237 struct dx_frame
 238 {
 239         struct buffer_head *bh;
 240         struct dx_entry *entries;
 241         struct dx_entry *at;
 242 };
 243
 244 struct dx_map_entry
 245 {
 246         u32 hash;
 247         u16 offs;
 248         u16 size;
 249 };
 250
 251 /*
 252  * This goes at the end of each htree block.
 253  */
 254 struct dx_tail {
 255         u32 dt_reserved;
 256         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 257 };
 258
 259 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 260 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 261 static inline unsigned dx_get_hash(struct dx_entry *entry);
 262 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 263 static unsigned dx_get_count(struct dx_entry *entries);
 264 static unsigned dx_get_limit(struct dx_entry *entries);
 265 static void dx_set_count(struct dx_entry *entries, unsigned value);
 266 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 267 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 268 static unsigned dx_node_limit(struct inode *dir);
 269 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 270                                  struct inode *dir,
 271                                  struct dx_hash_info *hinfo,
 272                                  struct dx_frame *frame);
 273 static void dx_release(struct dx_frame *frames);
 274 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 275                        unsigned blocksize, struct dx_hash_info *hinfo,
 276                        struct dx_map_entry map[]);
 277 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 278 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 279                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 280 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 281 static void dx_insert_block(struct dx_frame *frame,
 282                                         u32 hash, ext4_lblk_t block);
 283 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 284                                  struct dx_frame *frame,
 285                                  struct dx_frame *frames,
 286                                  __u32 *start_hash);
 287 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 288                 struct ext4_filename *fname,
 289                 struct ext4_dir_entry_2 **res_dir);
 290 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 291                              struct inode *dir, struct inode *inode);
 292
 293 /* checksumming functions */
 294 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 295                             unsigned int blocksize)
 296 {
 297         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 298         t->det_rec_len = ext4_rec_len_to_disk(
 299                         sizeof(struct ext4_dir_entry_tail), blocksize);
 300         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 301 }
 302
 303 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 304 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 305                                                    struct ext4_dir_entry *de)
 306 {
 307         struct ext4_dir_entry_tail *t;
 308
 309 #ifdef PARANOID
 310         struct ext4_dir_entry *d, *top;
 311
 312         d = de;
 313         top = (struct ext4_dir_entry *)(((void *)de) +
 314                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 315                 sizeof(struct ext4_dir_entry_tail)));
 316         while (d < top && d->rec_len)
 317                 d = (struct ext4_dir_entry *)(((void *)d) +
 318                     le16_to_cpu(d->rec_len));
 319
 320         if (d != top)
 321                 return NULL;
 322
 323         t = (struct ext4_dir_entry_tail *)d;
 324 #else
 325         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 326 #endif
 327
 328         if (t->det_reserved_zero1 ||
 329             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 330             t->det_reserved_zero2 ||
 331             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 332                 return NULL;
 333
 334         return t;
 335 }
 336
 337 static __le32 ext4_dirent_csum(struct inode *inode,
 338                                struct ext4_dir_entry *dirent, int size)
 339 {
 340         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 341         struct ext4_inode_info *ei = EXT4_I(inode);
 342         __u32 csum;
 343
 344         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 345         return cpu_to_le32(csum);
 346 }
 347
 348 #define warn_no_space_for_csum(inode)                                   \
 349         __warn_no_space_for_csum((inode), __func__, __LINE__)
 350
 351 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 352                                      unsigned int line)
 353 {
 354         __ext4_warning_inode(inode, func, line,
 355                 "No space for directory leaf checksum. Please run e2fsck -D.");
 356 }
 357
 358 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 359 {
 360         struct ext4_dir_entry_tail *t;
 361
 362         if (!ext4_has_metadata_csum(inode->i_sb))
 363                 return 1;
 364
 365         t = get_dirent_tail(inode, dirent);
 366         if (!t) {
 367                 warn_no_space_for_csum(inode);
 368                 return 0;
 369         }
 370
 371         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 372                                                 (void *)t - (void *)dirent))
 373                 return 0;
 374
 375         return 1;
 376 }
 377
 378 static void ext4_dirent_csum_set(struct inode *inode,
 379                                  struct ext4_dir_entry *dirent)
 380 {
 381         struct ext4_dir_entry_tail *t;
 382
 383         if (!ext4_has_metadata_csum(inode->i_sb))
 384                 return;
 385
 386         t = get_dirent_tail(inode, dirent);
 387         if (!t) {
 388                 warn_no_space_for_csum(inode);
 389                 return;
 390         }
 391
 392         t->det_checksum = ext4_dirent_csum(inode, dirent,
 393                                            (void *)t - (void *)dirent);
 394 }
 395
 396 int ext4_handle_dirty_dirent_node(handle_t *handle,
 397                                   struct inode *inode,
 398                                   struct buffer_head *bh)
 399 {
 400         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 401         return ext4_handle_dirty_metadata(handle, inode, bh);
 402 }
 403
 404 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 405                                                struct ext4_dir_entry *dirent,
 406                                                int *offset)
 407 {
 408         struct ext4_dir_entry *dp;
 409         struct dx_root_info *root;
 410         int count_offset;
 411
 412         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 413                 count_offset = 8;
 414         else if (le16_to_cpu(dirent->rec_len) == 12) {
 415                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 416                 if (le16_to_cpu(dp->rec_len) !=
 417                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 418                         return NULL;
 419                 root = (struct dx_root_info *)(((void *)dp + 12));
 420                 if (root->reserved_zero ||
 421                     root->info_length != sizeof(struct dx_root_info))
 422                         return NULL;
 423                 count_offset = 32;
 424         } else
 425                 return NULL;
 426
 427         if (offset)
 428                 *offset = count_offset;
 429         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 430 }
 431
 432 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 433                            int count_offset, int count, struct dx_tail *t)
 434 {
 435         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 436         struct ext4_inode_info *ei = EXT4_I(inode);
 437         __u32 csum;
 438         int size;
 439         __u32 dummy_csum = 0;
 440         int offset = offsetof(struct dx_tail, dt_checksum);
 441
 442         size = count_offset + (count * sizeof(struct dx_entry));
 443         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 444         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 445         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 446
 447         return cpu_to_le32(csum);
 448 }
 449
 450 static int ext4_dx_csum_verify(struct inode *inode,
 451                                struct ext4_dir_entry *dirent)
 452 {
 453         struct dx_countlimit *c;
 454         struct dx_tail *t;
 455         int count_offset, limit, count;
 456
 457         if (!ext4_has_metadata_csum(inode->i_sb))
 458                 return 1;
 459
 460         c = get_dx_countlimit(inode, dirent, &count_offset);
 461         if (!c) {
 462                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 463                 return 0;
 464         }
 465         limit = le16_to_cpu(c->limit);
 466         count = le16_to_cpu(c->count);
 467         if (count_offset + (limit * sizeof(struct dx_entry)) >
 468             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 469                 warn_no_space_for_csum(inode);
 470                 return 0;
 471         }
 472         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 473
 474         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 475                                             count, t))
 476                 return 0;
 477         return 1;
 478 }
 479
 480 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 481 {
 482         struct dx_countlimit *c;
 483         struct dx_tail *t;
 484         int count_offset, limit, count;
 485
 486         if (!ext4_has_metadata_csum(inode->i_sb))
 487                 return;
 488
 489         c = get_dx_countlimit(inode, dirent, &count_offset);
 490         if (!c) {
 491                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 492                 return;
 493         }
 494         limit = le16_to_cpu(c->limit);
 495         count = le16_to_cpu(c->count);
 496         if (count_offset + (limit * sizeof(struct dx_entry)) >
 497             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 498                 warn_no_space_for_csum(inode);
 499                 return;
 500         }
 501         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 502
 503         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 504 }
 505
 506 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 507                                             struct inode *inode,
 508                                             struct buffer_head *bh)
 509 {
 510         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 511         return ext4_handle_dirty_metadata(handle, inode, bh);
 512 }
 513
 514 /*
 515  * p is at least 6 bytes before the end of page
 516  */
 517 static inline struct ext4_dir_entry_2 *
 518 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 519 {
 520         return (struct ext4_dir_entry_2 *)((char *)p +
 521                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 522 }
 523
 524 /*
 525  * Future: use high four bits of block for coalesce-on-delete flags
 526  * Mask them off for now.
 527  */
 528
 529 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 530 {
 531         return le32_to_cpu(entry->block) & 0x0fffffff;
 532 }
 533
 534 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 535 {
 536         entry->block = cpu_to_le32(value);
 537 }
 538
 539 static inline unsigned dx_get_hash(struct dx_entry *entry)
 540 {
 541         return le32_to_cpu(entry->hash);
 542 }
 543
 544 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 545 {
 546         entry->hash = cpu_to_le32(value);
 547 }
 548
 549 static inline unsigned dx_get_count(struct dx_entry *entries)
 550 {
 551         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 552 }
 553
 554 static inline unsigned dx_get_limit(struct dx_entry *entries)
 555 {
 556         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 557 }
 558
 559 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 560 {
 561         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 562 }
 563
 564 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 565 {
 566         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 567 }
 568
 569 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 570 {
 571         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 572                 EXT4_DIR_REC_LEN(2) - infosize;
 573
 574         if (ext4_has_metadata_csum(dir->i_sb))
 575                 entry_space -= sizeof(struct dx_tail);
 576         return entry_space / sizeof(struct dx_entry);
 577 }
 578
 579 static inline unsigned dx_node_limit(struct inode *dir)
 580 {
 581         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 582
 583         if (ext4_has_metadata_csum(dir->i_sb))
 584                 entry_space -= sizeof(struct dx_tail);
 585         return entry_space / sizeof(struct dx_entry);
 586 }
 587
 588 /*
 589  * Debug
 590  */
 591 #ifdef DX_DEBUG
 592 static void dx_show_index(char * label, struct dx_entry *entries)
 593 {
 594         int i, n = dx_get_count (entries);
 595         printk(KERN_DEBUG "%s index", label);
 596         for (i = 0; i < n; i++) {
 597                 printk(KERN_CONT " %x->%lu",
 598                        i ? dx_get_hash(entries + i) : 0,
 599                        (unsigned long)dx_get_block(entries + i));
 600         }
 601         printk(KERN_CONT "\n");
 602 }
 603
 604 struct stats
 605 {
 606         unsigned names;
 607         unsigned space;
 608         unsigned bcount;
 609 };
 610
 611 static struct stats dx_show_leaf(struct inode *dir,
 612                                 struct dx_hash_info *hinfo,
 613                                 struct ext4_dir_entry_2 *de,
 614                                 int size, int show_names)
 615 {
 616         unsigned names = 0, space = 0;
 617         char *base = (char *) de;
 618         struct dx_hash_info h = *hinfo;
 619
 620         printk("names: ");
 621         while ((char *) de < base + size)
 622         {
 623                 if (de->inode)
 624                 {
 625                         if (show_names)
 626                         {
 627 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 628                                 int len;
 629                                 char *name;
 630                                 struct fscrypt_str fname_crypto_str =
 631                                         FSTR_INIT(NULL, 0);
 632                                 int res = 0;
 633
 634                                 name  = de->name;
 635                                 len = de->name_len;
 636                                 if (ext4_encrypted_inode(dir))
 637                                         res = fscrypt_get_encryption_info(dir);
 638                                 if (res) {
 639                                         printk(KERN_WARNING "Error setting up"
 640                                                " fname crypto: %d\n", res);
 641                                 }
 642                                 if (!fscrypt_has_encryption_key(dir)) {
 643                                         /* Directory is not encrypted */
 644                                         ext4fs_dirhash(de->name,
 645                                                 de->name_len, &h);
 646                                         printk("%*.s:(U)%x.%u ", len,
 647                                                name, h.hash,
 648                                                (unsigned) ((char *) de
 649                                                            - base));
 650                                 } else {
 651                                         struct fscrypt_str de_name =
 652                                                 FSTR_INIT(name, len);
 653
 654                                         /* Directory is encrypted */
 655                                         res = fscrypt_fname_alloc_buffer(
 656                                                 dir, len,
 657                                                 &fname_crypto_str);
 658                                         if (res)
 659                                                 printk(KERN_WARNING "Error "
 660                                                         "allocating crypto "
 661                                                         "buffer--skipping "
 662                                                         "crypto\n");
 663                                         res = fscrypt_fname_disk_to_usr(dir,
 664                                                 0, 0, &de_name,
 665                                                 &fname_crypto_str);
 666                                         if (res) {
 667                                                 printk(KERN_WARNING "Error "
 668                                                         "converting filename "
 669                                                         "from disk to usr"
 670                                                         "\n");
 671                                                 name = "??";
 672                                                 len = 2;
 673                                         } else {
 674                                                 name = fname_crypto_str.name;
 675                                                 len = fname_crypto_str.len;
 676                                         }
 677                                         ext4fs_dirhash(de->name, de->name_len,
 678                                                        &h);
 679                                         printk("%*.s:(E)%x.%u ", len, name,
 680                                                h.hash, (unsigned) ((char *) de
 681                                                                    - base));
 682                                         fscrypt_fname_free_buffer(
 683                                                         &fname_crypto_str);
 684                                 }
 685 #else
 686                                 int len = de->name_len;
 687                                 char *name = de->name;
 688                                 ext4fs_dirhash(de->name, de->name_len, &h);
 689                                 printk("%*.s:%x.%u ", len, name, h.hash,
 690                                        (unsigned) ((char *) de - base));
 691 #endif
 692                         }
 693                         space += EXT4_DIR_REC_LEN(de->name_len);
 694                         names++;
 695                 }
 696                 de = ext4_next_entry(de, size);
 697         }
 698         printk(KERN_CONT "(%i)\n", names);
 699         return (struct stats) { names, space, 1 };
 700 }
 701
 702 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 703                              struct dx_entry *entries, int levels)
 704 {
 705         unsigned blocksize = dir->i_sb->s_blocksize;
 706         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 707         unsigned bcount = 0;
 708         struct buffer_head *bh;
 709         printk("%i indexed blocks...\n", count);
 710         for (i = 0; i < count; i++, entries++)
 711         {
 712                 ext4_lblk_t block = dx_get_block(entries);
 713                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 714                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 715                 struct stats stats;
 716                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 717                 bh = ext4_bread(NULL,dir, block, 0);
 718                 if (!bh || IS_ERR(bh))
 719                         continue;
 720                 stats = levels?
 721                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 722                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 723                         bh->b_data, blocksize, 0);
 724                 names += stats.names;
 725                 space += stats.space;
 726                 bcount += stats.bcount;
 727                 brelse(bh);
 728         }
 729         if (bcount)
 730                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 731                        levels ? "" : "   ", names, space/bcount,
 732                        (space/bcount)*100/blocksize);
 733         return (struct stats) { names, space, bcount};
 734 }
 735 #endif /* DX_DEBUG */
 736
 737 /*
 738  * Probe for a directory leaf block to search.
 739  *
 740  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 741  * error in the directory index, and the caller should fall back to
 742  * searching the directory normally.  The callers of dx_probe **MUST**
 743  * check for this error code, and make sure it never gets reflected
 744  * back to userspace.
 745  */
 746 static struct dx_frame *
 747 dx_probe(struct ext4_filename *fname, struct inode *dir,
 748          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 749 {
 750         unsigned count, indirect;
 751         struct dx_entry *at, *entries, *p, *q, *m;
 752         struct dx_root *root;
 753         struct dx_frame *frame = frame_in;
 754         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 755         u32 hash;
 756
 757         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 758         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 759         if (IS_ERR(frame->bh))
 760                 return (struct dx_frame *) frame->bh;
 761
 762         root = (struct dx_root *) frame->bh->b_data;
 763         if (root->info.hash_version != DX_HASH_TEA &&
 764             root->info.hash_version != DX_HASH_HALF_MD4 &&
 765             root->info.hash_version != DX_HASH_LEGACY) {
 766                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 767                                    root->info.hash_version);
 768                 goto fail;
 769         }
 770         if (fname)
 771                 hinfo = &fname->hinfo;
 772         hinfo->hash_version = root->info.hash_version;
 773         if (hinfo->hash_version <= DX_HASH_TEA)
 774                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 775         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 776         if (fname && fname_name(fname))
 777                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 778         hash = hinfo->hash;
 779
 780         if (root->info.unused_flags & 1) {
 781                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 782                                    root->info.unused_flags);
 783                 goto fail;
 784         }
 785
 786         indirect = root->info.indirect_levels;
 787         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 788                 ext4_warning(dir->i_sb,
 789                              "Directory (ino: %lu) htree depth %#06x exceed"
 790                              "supported value", dir->i_ino,
 791                              ext4_dir_htree_level(dir->i_sb));
 792                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 793                         ext4_warning(dir->i_sb, "Enable large directory "
 794                                                 "feature to access it");
 795                 }
 796                 goto fail;
 797         }
 798
 799         entries = (struct dx_entry *)(((char *)&root->info) +
 800                                       root->info.info_length);
 801
 802         if (dx_get_limit(entries) != dx_root_limit(dir,
 803                                                    root->info.info_length)) {
 804                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 805                                    dx_get_limit(entries),
 806                                    dx_root_limit(dir, root->info.info_length));
 807                 goto fail;
 808         }
 809
 810         dxtrace(printk("Look up %x", hash));
 811         while (1) {
 812                 count = dx_get_count(entries);
 813                 if (!count || count > dx_get_limit(entries)) {
 814                         ext4_warning_inode(dir,
 815                                            "dx entry: count %u beyond limit %u",
 816                                            count, dx_get_limit(entries));
 817                         goto fail;
 818                 }
 819
 820                 p = entries + 1;
 821                 q = entries + count - 1;
 822                 while (p <= q) {
 823                         m = p + (q - p) / 2;
 824                         dxtrace(printk(KERN_CONT "."));
 825                         if (dx_get_hash(m) > hash)
 826                                 q = m - 1;
 827                         else
 828                                 p = m + 1;
 829                 }
 830
 831                 if (0) { // linear search cross check
 832                         unsigned n = count - 1;
 833                         at = entries;
 834                         while (n--)
 835                         {
 836                                 dxtrace(printk(KERN_CONT ","));
 837                                 if (dx_get_hash(++at) > hash)
 838                                 {
 839                                         at--;
 840                                         break;
 841                                 }
 842                         }
 843                         assert (at == p - 1);
 844                 }
 845
 846                 at = p - 1;
 847                 dxtrace(printk(KERN_CONT " %x->%u\n",
 848                                at == entries ? 0 : dx_get_hash(at),
 849                                dx_get_block(at)));
 850                 frame->entries = entries;
 851                 frame->at = at;
 852                 if (!indirect--)
 853                         return frame;
 854                 frame++;
 855                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 856                 if (IS_ERR(frame->bh)) {
 857                         ret_err = (struct dx_frame *) frame->bh;
 858                         frame->bh = NULL;
 859                         goto fail;
 860                 }
 861                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 862
 863                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 864                         ext4_warning_inode(dir,
 865                                 "dx entry: limit %u != node limit %u",
 866                                 dx_get_limit(entries), dx_node_limit(dir));
 867                         goto fail;
 868                 }
 869         }
 870 fail:
 871         while (frame >= frame_in) {
 872                 brelse(frame->bh);
 873                 frame--;
 874         }
 875
 876         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 877                 ext4_warning_inode(dir,
 878                         "Corrupt directory, running e2fsck is recommended");
 879         return ret_err;
 880 }
 881
 882 static void dx_release(struct dx_frame *frames)
 883 {
 884         struct dx_root_info *info;
 885         int i;
 886         unsigned int indirect_levels;
 887
 888         if (frames[0].bh == NULL)
 889                 return;
 890
 891         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 892         /* save local copy, "info" may be freed after brelse() */
 893         indirect_levels = info->indirect_levels;
 894         for (i = 0; i <= indirect_levels; i++) {
 895                 if (frames[i].bh == NULL)
 896                         break;
 897                 brelse(frames[i].bh);
 898                 frames[i].bh = NULL;
 899         }
 900 }
 901
 902 /*
 903  * This function increments the frame pointer to search the next leaf
 904  * block, and reads in the necessary intervening nodes if the search
 905  * should be necessary.  Whether or not the search is necessary is
 906  * controlled by the hash parameter.  If the hash value is even, then
 907  * the search is only continued if the next block starts with that
 908  * hash value.  This is used if we are searching for a specific file.
 909  *
 910  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 911  *
 912  * This function returns 1 if the caller should continue to search,
 913  * or 0 if it should not.  If there is an error reading one of the
 914  * index blocks, it will a negative error code.
 915  *
 916  * If start_hash is non-null, it will be filled in with the starting
 917  * hash of the next page.
 918  */
 919 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 920                                  struct dx_frame *frame,
 921                                  struct dx_frame *frames,
 922                                  __u32 *start_hash)
 923 {
 924         struct dx_frame *p;
 925         struct buffer_head *bh;
 926         int num_frames = 0;
 927         __u32 bhash;
 928
 929         p = frame;
 930         /*
 931          * Find the next leaf page by incrementing the frame pointer.
 932          * If we run out of entries in the interior node, loop around and
 933          * increment pointer in the parent node.  When we break out of
 934          * this loop, num_frames indicates the number of interior
 935          * nodes need to be read.
 936          */
 937         while (1) {
 938                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 939                         break;
 940                 if (p == frames)
 941                         return 0;
 942                 num_frames++;
 943                 p--;
 944         }
 945
 946         /*
 947          * If the hash is 1, then continue only if the next page has a
 948          * continuation hash of any value.  This is used for readdir
 949          * handling.  Otherwise, check to see if the hash matches the
 950          * desired contiuation hash.  If it doesn't, return since
 951          * there's no point to read in the successive index pages.
 952          */
 953         bhash = dx_get_hash(p->at);
 954         if (start_hash)
 955                 *start_hash = bhash;
 956         if ((hash & 1) == 0) {
 957                 if ((bhash & ~1) != hash)
 958                         return 0;
 959         }
 960         /*
 961          * If the hash is HASH_NB_ALWAYS, we always go to the next
 962          * block so no check is necessary
 963          */
 964         while (num_frames--) {
 965                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 966                 if (IS_ERR(bh))
 967                         return PTR_ERR(bh);
 968                 p++;
 969                 brelse(p->bh);
 970                 p->bh = bh;
 971                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 972         }
 973         return 1;
 974 }
 975
 976
 977 /*
 978  * This function fills a red-black tree with information from a
 979  * directory block.  It returns the number directory entries loaded
 980  * into the tree.  If there is an error it is returned in err.
 981  */
 982 static int htree_dirblock_to_tree(struct file *dir_file,
 983                                   struct inode *dir, ext4_lblk_t block,
 984                                   struct dx_hash_info *hinfo,
 985                                   __u32 start_hash, __u32 start_minor_hash)
 986 {
 987         struct buffer_head *bh;
 988         struct ext4_dir_entry_2 *de, *top;
 989         int err = 0, count = 0;
 990         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 991
 992         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 993                                                         (unsigned long)block));
 994         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
 995         if (IS_ERR(bh))
 996                 return PTR_ERR(bh);
 997
 998         de = (struct ext4_dir_entry_2 *) bh->b_data;
 999         top = (struct ext4_dir_entry_2 *) ((char *) de +
1000                                            dir->i_sb->s_blocksize -
1001                                            EXT4_DIR_REC_LEN(0));
1002 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1003         /* Check if the directory is encrypted */
1004         if (ext4_encrypted_inode(dir)) {
1005                 err = fscrypt_get_encryption_info(dir);
1006                 if (err < 0) {
1007                         brelse(bh);
1008                         return err;
1009                 }
1010                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1011                                                      &fname_crypto_str);
1012                 if (err < 0) {
1013                         brelse(bh);
1014                         return err;
1015                 }
1016         }
1017 #endif
1018         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1019                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1020                                 bh->b_data, bh->b_size,
1021                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1022                                          + ((char *)de - bh->b_data))) {
1023                         /* silently ignore the rest of the block */
1024                         break;
1025                 }
1026                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1027                 if ((hinfo->hash < start_hash) ||
1028                     ((hinfo->hash == start_hash) &&
1029                      (hinfo->minor_hash < start_minor_hash)))
1030                         continue;
1031                 if (de->inode == 0)
1032                         continue;
1033                 if (!ext4_encrypted_inode(dir)) {
1034                         tmp_str.name = de->name;
1035                         tmp_str.len = de->name_len;
1036                         err = ext4_htree_store_dirent(dir_file,
1037                                    hinfo->hash, hinfo->minor_hash, de,
1038                                    &tmp_str);
1039                 } else {
1040                         int save_len = fname_crypto_str.len;
1041                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1042                                                                 de->name_len);
1043
1044                         /* Directory is encrypted */
1045                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1046                                         hinfo->minor_hash, &de_name,
1047                                         &fname_crypto_str);
1048                         if (err) {
1049                                 count = err;
1050                                 goto errout;
1051                         }
1052                         err = ext4_htree_store_dirent(dir_file,
1053                                    hinfo->hash, hinfo->minor_hash, de,
1054                                         &fname_crypto_str);
1055                         fname_crypto_str.len = save_len;
1056                 }
1057                 if (err != 0) {
1058                         count = err;
1059                         goto errout;
1060                 }
1061                 count++;
1062         }
1063 errout:
1064         brelse(bh);
1065 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1066         fscrypt_fname_free_buffer(&fname_crypto_str);
1067 #endif
1068         return count;
1069 }
1070
1071
1072 /*
1073  * This function fills a red-black tree with information from a
1074  * directory.  We start scanning the directory in hash order, starting
1075  * at start_hash and start_minor_hash.
1076  *
1077  * This function returns the number of entries inserted into the tree,
1078  * or a negative error code.
1079  */
1080 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1081                          __u32 start_minor_hash, __u32 *next_hash)
1082 {
1083         struct dx_hash_info hinfo;
1084         struct ext4_dir_entry_2 *de;
1085         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1086         struct inode *dir;
1087         ext4_lblk_t block;
1088         int count = 0;
1089         int ret, err;
1090         __u32 hashval;
1091         struct fscrypt_str tmp_str;
1092
1093         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1094                        start_hash, start_minor_hash));
1095         dir = file_inode(dir_file);
1096         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1097                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1098                 if (hinfo.hash_version <= DX_HASH_TEA)
1099                         hinfo.hash_version +=
1100                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1101                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1102                 if (ext4_has_inline_data(dir)) {
1103                         int has_inline_data = 1;
1104                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1105                                                         &hinfo, start_hash,
1106                                                         start_minor_hash,
1107                                                         &has_inline_data);
1108                         if (has_inline_data) {
1109                                 *next_hash = ~0;
1110                                 return count;
1111                         }
1112                 }
1113                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1114                                                start_hash, start_minor_hash);
1115                 *next_hash = ~0;
1116                 return count;
1117         }
1118         hinfo.hash = start_hash;
1119         hinfo.minor_hash = 0;
1120         frame = dx_probe(NULL, dir, &hinfo, frames);
1121         if (IS_ERR(frame))
1122                 return PTR_ERR(frame);
1123
1124         /* Add '.' and '..' from the htree header */
1125         if (!start_hash && !start_minor_hash) {
1126                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1127                 tmp_str.name = de->name;
1128                 tmp_str.len = de->name_len;
1129                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1130                                               de, &tmp_str);
1131                 if (err != 0)
1132                         goto errout;
1133                 count++;
1134         }
1135         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1136                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1137                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1138                 tmp_str.name = de->name;
1139                 tmp_str.len = de->name_len;
1140                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1141                                               de, &tmp_str);
1142                 if (err != 0)
1143                         goto errout;
1144                 count++;
1145         }
1146
1147         while (1) {
1148                 if (fatal_signal_pending(current)) {
1149                         err = -ERESTARTSYS;
1150                         goto errout;
1151                 }
1152                 cond_resched();
1153                 block = dx_get_block(frame->at);
1154                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1155                                              start_hash, start_minor_hash);
1156                 if (ret < 0) {
1157                         err = ret;
1158                         goto errout;
1159                 }
1160                 count += ret;
1161                 hashval = ~0;
1162                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1163                                             frame, frames, &hashval);
1164                 *next_hash = hashval;
1165                 if (ret < 0) {
1166                         err = ret;
1167                         goto errout;
1168                 }
1169                 /*
1170                  * Stop if:  (a) there are no more entries, or
1171                  * (b) we have inserted at least one entry and the
1172                  * next hash value is not a continuation
1173                  */
1174                 if ((ret == 0) ||
1175                     (count && ((hashval & 1) == 0)))
1176                         break;
1177         }
1178         dx_release(frames);
1179         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1180                        "next hash: %x\n", count, *next_hash));
1181         return count;
1182 errout:
1183         dx_release(frames);
1184         return (err);
1185 }
1186
1187 static inline int search_dirblock(struct buffer_head *bh,
1188                                   struct inode *dir,
1189                                   struct ext4_filename *fname,
1190                                   unsigned int offset,
1191                                   struct ext4_dir_entry_2 **res_dir)
1192 {
1193         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1194                                fname, offset, res_dir);
1195 }
1196
1197 /*
1198  * Directory block splitting, compacting
1199  */
1200
1201 /*
1202  * Create map of hash values, offsets, and sizes, stored at end of block.
1203  * Returns number of entries mapped.
1204  */
1205 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1206                        unsigned blocksize, struct dx_hash_info *hinfo,
1207                        struct dx_map_entry *map_tail)
1208 {
1209         int count = 0;
1210         char *base = (char *) de;
1211         struct dx_hash_info h = *hinfo;
1212
1213         while ((char *) de < base + blocksize) {
1214                 if (de->name_len && de->inode) {
1215                         ext4fs_dirhash(de->name, de->name_len, &h);
1216                         map_tail--;
1217                         map_tail->hash = h.hash;
1218                         map_tail->offs = ((char *) de - base)>>2;
1219                         map_tail->size = le16_to_cpu(de->rec_len);
1220                         count++;
1221                         cond_resched();
1222                 }
1223                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1224                 de = ext4_next_entry(de, blocksize);
1225         }
1226         return count;
1227 }
1228
1229 /* Sort map by hash value */
1230 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1231 {
1232         struct dx_map_entry *p, *q, *top = map + count - 1;
1233         int more;
1234         /* Combsort until bubble sort doesn't suck */
1235         while (count > 2) {
1236                 count = count*10/13;
1237                 if (count - 9 < 2) /* 9, 10 -> 11 */
1238                         count = 11;
1239                 for (p = top, q = p - count; q >= map; p--, q--)
1240                         if (p->hash < q->hash)
1241                                 swap(*p, *q);
1242         }
1243         /* Garden variety bubble sort */
1244         do {
1245                 more = 0;
1246                 q = top;
1247                 while (q-- > map) {
1248                         if (q[1].hash >= q[0].hash)
1249                                 continue;
1250                         swap(*(q+1), *q);
1251                         more = 1;
1252                 }
1253         } while(more);
1254 }
1255
1256 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1257 {
1258         struct dx_entry *entries = frame->entries;
1259         struct dx_entry *old = frame->at, *new = old + 1;
1260         int count = dx_get_count(entries);
1261
1262         assert(count < dx_get_limit(entries));
1263         assert(old < entries + count);
1264         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1265         dx_set_hash(new, hash);
1266         dx_set_block(new, block);
1267         dx_set_count(entries, count + 1);
1268 }
1269
1270 /*
1271  * Test whether a directory entry matches the filename being searched for.
1272  *
1273  * Return: %true if the directory entry matches, otherwise %false.
1274  */
1275 static inline bool ext4_match(const struct ext4_filename *fname,
1276                               const struct ext4_dir_entry_2 *de)
1277 {
1278         struct fscrypt_name f;
1279
1280         if (!de->inode)
1281                 return false;
1282
1283         f.usr_fname = fname->usr_fname;
1284         f.disk_name = fname->disk_name;
1285 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1286         f.crypto_buf = fname->crypto_buf;
1287 #endif
1288         return fscrypt_match_name(&f, de->name, de->name_len);
1289 }
1290
1291 /*
1292  * Returns 0 if not found, -1 on failure, and 1 on success
1293  */
1294 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1295                     struct inode *dir, struct ext4_filename *fname,
1296                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1297 {
1298         struct ext4_dir_entry_2 * de;
1299         char * dlimit;
1300         int de_len;
1301
1302         de = (struct ext4_dir_entry_2 *)search_buf;
1303         dlimit = search_buf + buf_size;
1304         while ((char *) de < dlimit) {
1305                 /* this code is executed quadratically often */
1306                 /* do minimal checking `by hand' */
1307                 if ((char *) de + de->name_len <= dlimit &&
1308                     ext4_match(fname, de)) {
1309                         /* found a match - just to be sure, do
1310                          * a full check */
1311                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1312                                                  buf_size, offset))
1313                                 return -1;
1314                         *res_dir = de;
1315                         return 1;
1316                 }
1317                 /* prevent looping on a bad block */
1318                 de_len = ext4_rec_len_from_disk(de->rec_len,
1319                                                 dir->i_sb->s_blocksize);
1320                 if (de_len <= 0)
1321                         return -1;
1322                 offset += de_len;
1323                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1324         }
1325         return 0;
1326 }
1327
1328 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329                                struct ext4_dir_entry *de)
1330 {
1331         struct super_block *sb = dir->i_sb;
1332
1333         if (!is_dx(dir))
1334                 return 0;
1335         if (block == 0)
1336                 return 1;
1337         if (de->inode == 0 &&
1338             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339                         sb->s_blocksize)
1340                 return 1;
1341         return 0;
1342 }
1343
1344 /*
1345  *      ext4_find_entry()
1346  *
1347  * finds an entry in the specified directory with the wanted name. It
1348  * returns the cache buffer in which the entry was found, and the entry
1349  * itself (as a parameter - res_dir). It does NOT read the inode of the
1350  * entry - you'll have to do that yourself if you want to.
1351  *
1352  * The returned buffer_head has ->b_count elevated.  The caller is expected
1353  * to brelse() it when appropriate.
1354  */
1355 static struct buffer_head * ext4_find_entry (struct inode *dir,
1356                                         const struct qstr *d_name,
1357                                         struct ext4_dir_entry_2 **res_dir,
1358                                         int *inlined)
1359 {
1360         struct super_block *sb;
1361         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362         struct buffer_head *bh, *ret = NULL;
1363         ext4_lblk_t start, block;
1364         const u8 *name = d_name->name;
1365         size_t ra_max = 0;      /* Number of bh's in the readahead
1366                                    buffer, bh_use[] */
1367         size_t ra_ptr = 0;      /* Current index into readahead
1368                                    buffer */
1369         ext4_lblk_t  nblocks;
1370         int i, namelen, retval;
1371         struct ext4_filename fname;
1372
1373         *res_dir = NULL;
1374         sb = dir->i_sb;
1375         namelen = d_name->len;
1376         if (namelen > EXT4_NAME_LEN)
1377                 return NULL;
1378
1379         retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1380         if (retval == -ENOENT)
1381                 return NULL;
1382         if (retval)
1383                 return ERR_PTR(retval);
1384
1385         if (ext4_has_inline_data(dir)) {
1386                 int has_inline_data = 1;
1387                 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1388                                              &has_inline_data);
1389                 if (has_inline_data) {
1390                         if (inlined)
1391                                 *inlined = 1;
1392                         goto cleanup_and_exit;
1393                 }
1394         }
1395
1396         if ((namelen <= 2) && (name[0] == '.') &&
1397             (name[1] == '.' || name[1] == '\0')) {
1398                 /*
1399                  * "." or ".." will only be in the first block
1400                  * NFS may look up ".."; "." should be handled by the VFS
1401                  */
1402                 block = start = 0;
1403                 nblocks = 1;
1404                 goto restart;
1405         }
1406         if (is_dx(dir)) {
1407                 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1408                 /*
1409                  * On success, or if the error was file not found,
1410                  * return.  Otherwise, fall back to doing a search the
1411                  * old fashioned way.
1412                  */
1413                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1414                         goto cleanup_and_exit;
1415                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1416                                "falling back\n"));
1417                 ret = NULL;
1418         }
1419         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1420         if (!nblocks) {
1421                 ret = NULL;
1422                 goto cleanup_and_exit;
1423         }
1424         start = EXT4_I(dir)->i_dir_start_lookup;
1425         if (start >= nblocks)
1426                 start = 0;
1427         block = start;
1428 restart:
1429         do {
1430                 /*
1431                  * We deal with the read-ahead logic here.
1432                  */
1433                 cond_resched();
1434                 if (ra_ptr >= ra_max) {
1435                         /* Refill the readahead buffer */
1436                         ra_ptr = 0;
1437                         if (block < start)
1438                                 ra_max = start - block;
1439                         else
1440                                 ra_max = nblocks - block;
1441                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1442                         retval = ext4_bread_batch(dir, block, ra_max,
1443                                                   false /* wait */, bh_use);
1444                         if (retval) {
1445                                 ret = ERR_PTR(retval);
1446                                 ra_max = 0;
1447                                 goto cleanup_and_exit;
1448                         }
1449                 }
1450                 if ((bh = bh_use[ra_ptr++]) == NULL)
1451                         goto next;
1452                 wait_on_buffer(bh);
1453                 if (!buffer_uptodate(bh)) {
1454                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1455                                          (unsigned long) block);
1456                         brelse(bh);
1457                         ret = ERR_PTR(-EIO);
1458                         goto cleanup_and_exit;
1459                 }
1460                 if (!buffer_verified(bh) &&
1461                     !is_dx_internal_node(dir, block,
1462                                          (struct ext4_dir_entry *)bh->b_data) &&
1463                     !ext4_dirent_csum_verify(dir,
1464                                 (struct ext4_dir_entry *)bh->b_data)) {
1465                         EXT4_ERROR_INODE(dir, "checksumming directory "
1466                                          "block %lu", (unsigned long)block);
1467                         brelse(bh);
1468                         ret = ERR_PTR(-EFSBADCRC);
1469                         goto cleanup_and_exit;
1470                 }
1471                 set_buffer_verified(bh);
1472                 i = search_dirblock(bh, dir, &fname,
1473                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1474                 if (i == 1) {
1475                         EXT4_I(dir)->i_dir_start_lookup = block;
1476                         ret = bh;
1477                         goto cleanup_and_exit;
1478                 } else {
1479                         brelse(bh);
1480                         if (i < 0)
1481                                 goto cleanup_and_exit;
1482                 }
1483         next:
1484                 if (++block >= nblocks)
1485                         block = 0;
1486         } while (block != start);
1487
1488         /*
1489          * If the directory has grown while we were searching, then
1490          * search the last part of the directory before giving up.
1491          */
1492         block = nblocks;
1493         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1494         if (block < nblocks) {
1495                 start = 0;
1496                 goto restart;
1497         }
1498
1499 cleanup_and_exit:
1500         /* Clean up the read-ahead blocks */
1501         for (; ra_ptr < ra_max; ra_ptr++)
1502                 brelse(bh_use[ra_ptr]);
1503         ext4_fname_free_filename(&fname);
1504         return ret;
1505 }
1506
1507 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1508                         struct ext4_filename *fname,
1509                         struct ext4_dir_entry_2 **res_dir)
1510 {
1511         struct super_block * sb = dir->i_sb;
1512         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1513         struct buffer_head *bh;
1514         ext4_lblk_t block;
1515         int retval;
1516
1517 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1518         *res_dir = NULL;
1519 #endif
1520         frame = dx_probe(fname, dir, NULL, frames);
1521         if (IS_ERR(frame))
1522                 return (struct buffer_head *) frame;
1523         do {
1524                 block = dx_get_block(frame->at);
1525                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1526                 if (IS_ERR(bh))
1527                         goto errout;
1528
1529                 retval = search_dirblock(bh, dir, fname,
1530                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1531                                          res_dir);
1532                 if (retval == 1)
1533                         goto success;
1534                 brelse(bh);
1535                 if (retval == -1) {
1536                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1537                         goto errout;
1538                 }
1539
1540                 /* Check to see if we should continue to search */
1541                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1542                                                frames, NULL);
1543                 if (retval < 0) {
1544                         ext4_warning_inode(dir,
1545                                 "error %d reading directory index block",
1546                                 retval);
1547                         bh = ERR_PTR(retval);
1548                         goto errout;
1549                 }
1550         } while (retval == 1);
1551
1552         bh = NULL;
1553 errout:
1554         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1555 success:
1556         dx_release(frames);
1557         return bh;
1558 }
1559
1560 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1561 {
1562         struct inode *inode;
1563         struct ext4_dir_entry_2 *de;
1564         struct buffer_head *bh;
1565
1566         if (ext4_encrypted_inode(dir)) {
1567                 int res = fscrypt_get_encryption_info(dir);
1568
1569                 /*
1570                  * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1571                  * created while the directory was encrypted and we
1572                  * have access to the key.
1573                  */
1574                 if (fscrypt_has_encryption_key(dir))
1575                         fscrypt_set_encrypted_dentry(dentry);
1576                 fscrypt_set_d_op(dentry);
1577                 if (res && res != -ENOKEY)
1578                         return ERR_PTR(res);
1579         }
1580
1581        if (dentry->d_name.len > EXT4_NAME_LEN)
1582                return ERR_PTR(-ENAMETOOLONG);
1583
1584         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1585         if (IS_ERR(bh))
1586                 return (struct dentry *) bh;
1587         inode = NULL;
1588         if (bh) {
1589                 __u32 ino = le32_to_cpu(de->inode);
1590                 brelse(bh);
1591                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1592                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1593                         return ERR_PTR(-EFSCORRUPTED);
1594                 }
1595                 if (unlikely(ino == dir->i_ino)) {
1596                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1597                                          dentry);
1598                         return ERR_PTR(-EFSCORRUPTED);
1599                 }
1600                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1601                 if (inode == ERR_PTR(-ESTALE)) {
1602                         EXT4_ERROR_INODE(dir,
1603                                          "deleted inode referenced: %u",
1604                                          ino);
1605                         return ERR_PTR(-EFSCORRUPTED);
1606                 }
1607                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1608                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1609                     !fscrypt_has_permitted_context(dir, inode)) {
1610                         ext4_warning(inode->i_sb,
1611                                      "Inconsistent encryption contexts: %lu/%lu",
1612                                      dir->i_ino, inode->i_ino);
1613                         iput(inode);
1614                         return ERR_PTR(-EPERM);
1615                 }
1616         }
1617         return d_splice_alias(inode, dentry);
1618 }
1619
1620
1621 struct dentry *ext4_get_parent(struct dentry *child)
1622 {
1623         __u32 ino;
1624         static const struct qstr dotdot = QSTR_INIT("..", 2);
1625         struct ext4_dir_entry_2 * de;
1626         struct buffer_head *bh;
1627
1628         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1629         if (IS_ERR(bh))
1630                 return (struct dentry *) bh;
1631         if (!bh)
1632                 return ERR_PTR(-ENOENT);
1633         ino = le32_to_cpu(de->inode);
1634         brelse(bh);
1635
1636         if (!ext4_valid_inum(child->d_sb, ino)) {
1637                 EXT4_ERROR_INODE(d_inode(child),
1638                                  "bad parent inode number: %u", ino);
1639                 return ERR_PTR(-EFSCORRUPTED);
1640         }
1641
1642         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1643 }
1644
1645 /*
1646  * Move count entries from end of map between two memory locations.
1647  * Returns pointer to last entry moved.
1648  */
1649 static struct ext4_dir_entry_2 *
1650 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1651                 unsigned blocksize)
1652 {
1653         unsigned rec_len = 0;
1654
1655         while (count--) {
1656                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1657                                                 (from + (map->offs<<2));
1658                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1659                 memcpy (to, de, rec_len);
1660                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1661                                 ext4_rec_len_to_disk(rec_len, blocksize);
1662                 de->inode = 0;
1663                 map++;
1664                 to += rec_len;
1665         }
1666         return (struct ext4_dir_entry_2 *) (to - rec_len);
1667 }
1668
1669 /*
1670  * Compact each dir entry in the range to the minimal rec_len.
1671  * Returns pointer to last entry in range.
1672  */
1673 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1674 {
1675         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1676         unsigned rec_len = 0;
1677
1678         prev = to = de;
1679         while ((char*)de < base + blocksize) {
1680                 next = ext4_next_entry(de, blocksize);
1681                 if (de->inode && de->name_len) {
1682                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1683                         if (de > to)
1684                                 memmove(to, de, rec_len);
1685                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1686                         prev = to;
1687                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1688                 }
1689                 de = next;
1690         }
1691         return prev;
1692 }
1693
1694 /*
1695  * Split a full leaf block to make room for a new dir entry.
1696  * Allocate a new block, and move entries so that they are approx. equally full.
1697  * Returns pointer to de in block into which the new entry will be inserted.
1698  */
1699 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1700                         struct buffer_head **bh,struct dx_frame *frame,
1701                         struct dx_hash_info *hinfo)
1702 {
1703         unsigned blocksize = dir->i_sb->s_blocksize;
1704         unsigned count, continued;
1705         struct buffer_head *bh2;
1706         ext4_lblk_t newblock;
1707         u32 hash2;
1708         struct dx_map_entry *map;
1709         char *data1 = (*bh)->b_data, *data2;
1710         unsigned split, move, size;
1711         struct ext4_dir_entry_2 *de = NULL, *de2;
1712         struct ext4_dir_entry_tail *t;
1713         int     csum_size = 0;
1714         int     err = 0, i;
1715
1716         if (ext4_has_metadata_csum(dir->i_sb))
1717                 csum_size = sizeof(struct ext4_dir_entry_tail);
1718
1719         bh2 = ext4_append(handle, dir, &newblock);
1720         if (IS_ERR(bh2)) {
1721                 brelse(*bh);
1722                 *bh = NULL;
1723                 return (struct ext4_dir_entry_2 *) bh2;
1724         }
1725
1726         BUFFER_TRACE(*bh, "get_write_access");
1727         err = ext4_journal_get_write_access(handle, *bh);
1728         if (err)
1729                 goto journal_error;
1730
1731         BUFFER_TRACE(frame->bh, "get_write_access");
1732         err = ext4_journal_get_write_access(handle, frame->bh);
1733         if (err)
1734                 goto journal_error;
1735
1736         data2 = bh2->b_data;
1737
1738         /* create map in the end of data2 block */
1739         map = (struct dx_map_entry *) (data2 + blocksize);
1740         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1741                              blocksize, hinfo, map);
1742         map -= count;
1743         dx_sort_map(map, count);
1744         /* Ensure that neither split block is over half full */
1745         size = 0;
1746         move = 0;
1747         for (i = count-1; i >= 0; i--) {
1748                 /* is more than half of this entry in 2nd half of the block? */
1749                 if (size + map[i].size/2 > blocksize/2)
1750                         break;
1751                 size += map[i].size;
1752                 move++;
1753         }
1754         /*
1755          * map index at which we will split
1756          *
1757          * If the sum of active entries didn't exceed half the block size, just
1758          * split it in half by count; each resulting block will have at least
1759          * half the space free.
1760          */
1761         if (i > 0)
1762                 split = count - move;
1763         else
1764                 split = count/2;
1765
1766         hash2 = map[split].hash;
1767         continued = hash2 == map[split - 1].hash;
1768         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1769                         (unsigned long)dx_get_block(frame->at),
1770                                         hash2, split, count-split));
1771
1772         /* Fancy dance to stay within two buffers */
1773         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1774                               blocksize);
1775         de = dx_pack_dirents(data1, blocksize);
1776         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1777                                            (char *) de,
1778                                            blocksize);
1779         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1780                                             (char *) de2,
1781                                             blocksize);
1782         if (csum_size) {
1783                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1784                 initialize_dirent_tail(t, blocksize);
1785
1786                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1787                 initialize_dirent_tail(t, blocksize);
1788         }
1789
1790         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1791                         blocksize, 1));
1792         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1793                         blocksize, 1));
1794
1795         /* Which block gets the new entry? */
1796         if (hinfo->hash >= hash2) {
1797                 swap(*bh, bh2);
1798                 de = de2;
1799         }
1800         dx_insert_block(frame, hash2 + continued, newblock);
1801         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1802         if (err)
1803                 goto journal_error;
1804         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1805         if (err)
1806                 goto journal_error;
1807         brelse(bh2);
1808         dxtrace(dx_show_index("frame", frame->entries));
1809         return de;
1810
1811 journal_error:
1812         brelse(*bh);
1813         brelse(bh2);
1814         *bh = NULL;
1815         ext4_std_error(dir->i_sb, err);
1816         return ERR_PTR(err);
1817 }
1818
1819 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1820                       struct buffer_head *bh,
1821                       void *buf, int buf_size,
1822                       struct ext4_filename *fname,
1823                       struct ext4_dir_entry_2 **dest_de)
1824 {
1825         struct ext4_dir_entry_2 *de;
1826         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1827         int nlen, rlen;
1828         unsigned int offset = 0;
1829         char *top;
1830
1831         de = (struct ext4_dir_entry_2 *)buf;
1832         top = buf + buf_size - reclen;
1833         while ((char *) de <= top) {
1834                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1835                                          buf, buf_size, offset))
1836                         return -EFSCORRUPTED;
1837                 if (ext4_match(fname, de))
1838                         return -EEXIST;
1839                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1840                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1841                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1842                         break;
1843                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1844                 offset += rlen;
1845         }
1846         if ((char *) de > top)
1847                 return -ENOSPC;
1848
1849         *dest_de = de;
1850         return 0;
1851 }
1852
1853 void ext4_insert_dentry(struct inode *inode,
1854                         struct ext4_dir_entry_2 *de,
1855                         int buf_size,
1856                         struct ext4_filename *fname)
1857 {
1858
1859         int nlen, rlen;
1860
1861         nlen = EXT4_DIR_REC_LEN(de->name_len);
1862         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1863         if (de->inode) {
1864                 struct ext4_dir_entry_2 *de1 =
1865                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1866                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1867                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1868                 de = de1;
1869         }
1870         de->file_type = EXT4_FT_UNKNOWN;
1871         de->inode = cpu_to_le32(inode->i_ino);
1872         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1873         de->name_len = fname_len(fname);
1874         memcpy(de->name, fname_name(fname), fname_len(fname));
1875 }
1876
1877 /*
1878  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1879  * it points to a directory entry which is guaranteed to be large
1880  * enough for new directory entry.  If de is NULL, then
1881  * add_dirent_to_buf will attempt search the directory block for
1882  * space.  It will return -ENOSPC if no space is available, and -EIO
1883  * and -EEXIST if directory entry already exists.
1884  */
1885 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1886                              struct inode *dir,
1887                              struct inode *inode, struct ext4_dir_entry_2 *de,
1888                              struct buffer_head *bh)
1889 {
1890         unsigned int    blocksize = dir->i_sb->s_blocksize;
1891         int             csum_size = 0;
1892         int             err;
1893
1894         if (ext4_has_metadata_csum(inode->i_sb))
1895                 csum_size = sizeof(struct ext4_dir_entry_tail);
1896
1897         if (!de) {
1898                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1899                                         blocksize - csum_size, fname, &de);
1900                 if (err)
1901                         return err;
1902         }
1903         BUFFER_TRACE(bh, "get_write_access");
1904         err = ext4_journal_get_write_access(handle, bh);
1905         if (err) {
1906                 ext4_std_error(dir->i_sb, err);
1907                 return err;
1908         }
1909
1910         /* By now the buffer is marked for journaling */
1911         ext4_insert_dentry(inode, de, blocksize, fname);
1912
1913         /*
1914          * XXX shouldn't update any times until successful
1915          * completion of syscall, but too many callers depend
1916          * on this.
1917          *
1918          * XXX similarly, too many callers depend on
1919          * ext4_new_inode() setting the times, but error
1920          * recovery deletes the inode, so the worst that can
1921          * happen is that the times are slightly out of date
1922          * and/or different from the directory change time.
1923          */
1924         dir->i_mtime = dir->i_ctime = current_time(dir);
1925         ext4_update_dx_flag(dir);
1926         inode_inc_iversion(dir);
1927         ext4_mark_inode_dirty(handle, dir);
1928         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1929         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1930         if (err)
1931                 ext4_std_error(dir->i_sb, err);
1932         return 0;
1933 }
1934
1935 /*
1936  * This converts a one block unindexed directory to a 3 block indexed
1937  * directory, and adds the dentry to the indexed directory.
1938  */
1939 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1940                             struct inode *dir,
1941                             struct inode *inode, struct buffer_head *bh)
1942 {
1943         struct buffer_head *bh2;
1944         struct dx_root  *root;
1945         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1946         struct dx_entry *entries;
1947         struct ext4_dir_entry_2 *de, *de2;
1948         struct ext4_dir_entry_tail *t;
1949         char            *data1, *top;
1950         unsigned        len;
1951         int             retval;
1952         unsigned        blocksize;
1953         ext4_lblk_t  block;
1954         struct fake_dirent *fde;
1955         int csum_size = 0;
1956
1957         if (ext4_has_metadata_csum(inode->i_sb))
1958                 csum_size = sizeof(struct ext4_dir_entry_tail);
1959
1960         blocksize =  dir->i_sb->s_blocksize;
1961         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1962         BUFFER_TRACE(bh, "get_write_access");
1963         retval = ext4_journal_get_write_access(handle, bh);
1964         if (retval) {
1965                 ext4_std_error(dir->i_sb, retval);
1966                 brelse(bh);
1967                 return retval;
1968         }
1969         root = (struct dx_root *) bh->b_data;
1970
1971         /* The 0th block becomes the root, move the dirents out */
1972         fde = &root->dotdot;
1973         de = (struct ext4_dir_entry_2 *)((char *)fde +
1974                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1975         if ((char *) de >= (((char *) root) + blocksize)) {
1976                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1977                 brelse(bh);
1978                 return -EFSCORRUPTED;
1979         }
1980         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1981
1982         /* Allocate new block for the 0th block's dirents */
1983         bh2 = ext4_append(handle, dir, &block);
1984         if (IS_ERR(bh2)) {
1985                 brelse(bh);
1986                 return PTR_ERR(bh2);
1987         }
1988         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1989         data1 = bh2->b_data;
1990
1991         memcpy (data1, de, len);
1992         de = (struct ext4_dir_entry_2 *) data1;
1993         top = data1 + len;
1994         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1995                 de = de2;
1996         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1997                                            (char *) de,
1998                                            blocksize);
1999
2000         if (csum_size) {
2001                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2002                 initialize_dirent_tail(t, blocksize);
2003         }
2004
2005         /* Initialize the root; the dot dirents already exist */
2006         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2007         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2008                                            blocksize);
2009         memset (&root->info, 0, sizeof(root->info));
2010         root->info.info_length = sizeof(root->info);
2011         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2012         entries = root->entries;
2013         dx_set_block(entries, 1);
2014         dx_set_count(entries, 1);
2015         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2016
2017         /* Initialize as for dx_probe */
2018         fname->hinfo.hash_version = root->info.hash_version;
2019         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2020                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2021         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2022         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2023
2024         memset(frames, 0, sizeof(frames));
2025         frame = frames;
2026         frame->entries = entries;
2027         frame->at = entries;
2028         frame->bh = bh;
2029
2030         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2031         if (retval)
2032                 goto out_frames;
2033         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2034         if (retval)
2035                 goto out_frames;
2036
2037         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2038         if (IS_ERR(de)) {
2039                 retval = PTR_ERR(de);
2040                 goto out_frames;
2041         }
2042
2043         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2044 out_frames:
2045         /*
2046          * Even if the block split failed, we have to properly write
2047          * out all the changes we did so far. Otherwise we can end up
2048          * with corrupted filesystem.
2049          */
2050         if (retval)
2051                 ext4_mark_inode_dirty(handle, dir);
2052         dx_release(frames);
2053         brelse(bh2);
2054         return retval;
2055 }
2056
2057 /*
2058  *      ext4_add_entry()
2059  *
2060  * adds a file entry to the specified directory, using the same
2061  * semantics as ext4_find_entry(). It returns NULL if it failed.
2062  *
2063  * NOTE!! The inode part of 'de' is left at 0 - which means you
2064  * may not sleep between calling this and putting something into
2065  * the entry, as someone else might have used it while you slept.
2066  */
2067 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2068                           struct inode *inode)
2069 {
2070         struct inode *dir = d_inode(dentry->d_parent);
2071         struct buffer_head *bh = NULL;
2072         struct ext4_dir_entry_2 *de;
2073         struct ext4_dir_entry_tail *t;
2074         struct super_block *sb;
2075         struct ext4_filename fname;
2076         int     retval;
2077         int     dx_fallback=0;
2078         unsigned blocksize;
2079         ext4_lblk_t block, blocks;
2080         int     csum_size = 0;
2081
2082         if (ext4_has_metadata_csum(inode->i_sb))
2083                 csum_size = sizeof(struct ext4_dir_entry_tail);
2084
2085         sb = dir->i_sb;
2086         blocksize = sb->s_blocksize;
2087         if (!dentry->d_name.len)
2088                 return -EINVAL;
2089
2090         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2091         if (retval)
2092                 return retval;
2093
2094         if (ext4_has_inline_data(dir)) {
2095                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2096                 if (retval < 0)
2097                         goto out;
2098                 if (retval == 1) {
2099                         retval = 0;
2100                         goto out;
2101                 }
2102         }
2103
2104         if (is_dx(dir)) {
2105                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2106                 if (!retval || (retval != ERR_BAD_DX_DIR))
2107                         goto out;
2108                 /* Can we just ignore htree data? */
2109                 if (ext4_has_metadata_csum(sb)) {
2110                         EXT4_ERROR_INODE(dir,
2111                                 "Directory has corrupted htree index.");
2112                         retval = -EFSCORRUPTED;
2113                         goto out;
2114                 }
2115                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2116                 dx_fallback++;
2117                 ext4_mark_inode_dirty(handle, dir);
2118         }
2119         blocks = dir->i_size >> sb->s_blocksize_bits;
2120         for (block = 0; block < blocks; block++) {
2121                 bh = ext4_read_dirblock(dir, block, DIRENT);
2122                 if (bh == NULL) {
2123                         bh = ext4_bread(handle, dir, block,
2124                                         EXT4_GET_BLOCKS_CREATE);
2125                         goto add_to_new_block;
2126                 }
2127                 if (IS_ERR(bh)) {
2128                         retval = PTR_ERR(bh);
2129                         bh = NULL;
2130                         goto out;
2131                 }
2132                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2133                                            NULL, bh);
2134                 if (retval != -ENOSPC)
2135                         goto out;
2136
2137                 if (blocks == 1 && !dx_fallback &&
2138                     ext4_has_feature_dir_index(sb)) {
2139                         retval = make_indexed_dir(handle, &fname, dir,
2140                                                   inode, bh);
2141                         bh = NULL; /* make_indexed_dir releases bh */
2142                         goto out;
2143                 }
2144                 brelse(bh);
2145         }
2146         bh = ext4_append(handle, dir, &block);
2147 add_to_new_block:
2148         if (IS_ERR(bh)) {
2149                 retval = PTR_ERR(bh);
2150                 bh = NULL;
2151                 goto out;
2152         }
2153         de = (struct ext4_dir_entry_2 *) bh->b_data;
2154         de->inode = 0;
2155         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2156
2157         if (csum_size) {
2158                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2159                 initialize_dirent_tail(t, blocksize);
2160         }
2161
2162         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2163 out:
2164         ext4_fname_free_filename(&fname);
2165         brelse(bh);
2166         if (retval == 0)
2167                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2168         return retval;
2169 }
2170
2171 /*
2172  * Returns 0 for success, or a negative error value
2173  */
2174 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2175                              struct inode *dir, struct inode *inode)
2176 {
2177         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2178         struct dx_entry *entries, *at;
2179         struct buffer_head *bh;
2180         struct super_block *sb = dir->i_sb;
2181         struct ext4_dir_entry_2 *de;
2182         int restart;
2183         int err;
2184
2185 again:
2186         restart = 0;
2187         frame = dx_probe(fname, dir, NULL, frames);
2188         if (IS_ERR(frame))
2189                 return PTR_ERR(frame);
2190         entries = frame->entries;
2191         at = frame->at;
2192         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2193         if (IS_ERR(bh)) {
2194                 err = PTR_ERR(bh);
2195                 bh = NULL;
2196                 goto cleanup;
2197         }
2198
2199         BUFFER_TRACE(bh, "get_write_access");
2200         err = ext4_journal_get_write_access(handle, bh);
2201         if (err)
2202                 goto journal_error;
2203
2204         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2205         if (err != -ENOSPC)
2206                 goto cleanup;
2207
2208         err = 0;
2209         /* Block full, should compress but for now just split */
2210         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2211                        dx_get_count(entries), dx_get_limit(entries)));
2212         /* Need to split index? */
2213         if (dx_get_count(entries) == dx_get_limit(entries)) {
2214                 ext4_lblk_t newblock;
2215                 int levels = frame - frames + 1;
2216                 unsigned int icount;
2217                 int add_level = 1;
2218                 struct dx_entry *entries2;
2219                 struct dx_node *node2;
2220                 struct buffer_head *bh2;
2221
2222                 while (frame > frames) {
2223                         if (dx_get_count((frame - 1)->entries) <
2224                             dx_get_limit((frame - 1)->entries)) {
2225                                 add_level = 0;
2226                                 break;
2227                         }
2228                         frame--; /* split higher index block */
2229                         at = frame->at;
2230                         entries = frame->entries;
2231                         restart = 1;
2232                 }
2233                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2234                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2235                                          "reach max htree level :%d",
2236                                          dir->i_ino, levels);
2237                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2238                                 ext4_warning(sb, "Large directory feature is "
2239                                                  "not enabled on this "
2240                                                  "filesystem");
2241                         }
2242                         err = -ENOSPC;
2243                         goto cleanup;
2244                 }
2245                 icount = dx_get_count(entries);
2246                 bh2 = ext4_append(handle, dir, &newblock);
2247                 if (IS_ERR(bh2)) {
2248                         err = PTR_ERR(bh2);
2249                         goto cleanup;
2250                 }
2251                 node2 = (struct dx_node *)(bh2->b_data);
2252                 entries2 = node2->entries;
2253                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2254                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2255                                                            sb->s_blocksize);
2256                 BUFFER_TRACE(frame->bh, "get_write_access");
2257                 err = ext4_journal_get_write_access(handle, frame->bh);
2258                 if (err)
2259                         goto journal_error;
2260                 if (!add_level) {
2261                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2262                         unsigned hash2 = dx_get_hash(entries + icount1);
2263                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2264                                        icount1, icount2));
2265
2266                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2267                         err = ext4_journal_get_write_access(handle,
2268                                                              (frame - 1)->bh);
2269                         if (err)
2270                                 goto journal_error;
2271
2272                         memcpy((char *) entries2, (char *) (entries + icount1),
2273                                icount2 * sizeof(struct dx_entry));
2274                         dx_set_count(entries, icount1);
2275                         dx_set_count(entries2, icount2);
2276                         dx_set_limit(entries2, dx_node_limit(dir));
2277
2278                         /* Which index block gets the new entry? */
2279                         if (at - entries >= icount1) {
2280                                 frame->at = at = at - entries - icount1 + entries2;
2281                                 frame->entries = entries = entries2;
2282                                 swap(frame->bh, bh2);
2283                         }
2284                         dx_insert_block((frame - 1), hash2, newblock);
2285                         dxtrace(dx_show_index("node", frame->entries));
2286                         dxtrace(dx_show_index("node",
2287                                ((struct dx_node *) bh2->b_data)->entries));
2288                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2289                         if (err)
2290                                 goto journal_error;
2291                         brelse (bh2);
2292                         err = ext4_handle_dirty_dx_node(handle, dir,
2293                                                    (frame - 1)->bh);
2294                         if (err)
2295                                 goto journal_error;
2296                         err = ext4_handle_dirty_dx_node(handle, dir,
2297                                                         frame->bh);
2298                         if (restart || err)
2299                                 goto journal_error;
2300                 } else {
2301                         struct dx_root *dxroot;
2302                         memcpy((char *) entries2, (char *) entries,
2303                                icount * sizeof(struct dx_entry));
2304                         dx_set_limit(entries2, dx_node_limit(dir));
2305
2306                         /* Set up root */
2307                         dx_set_count(entries, 1);
2308                         dx_set_block(entries + 0, newblock);
2309                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2310                         dxroot->info.indirect_levels += 1;
2311                         dxtrace(printk(KERN_DEBUG
2312                                        "Creating %d level index...\n",
2313                                        dxroot->info.indirect_levels));
2314                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2315                         if (err)
2316                                 goto journal_error;
2317                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2318                         brelse(bh2);
2319                         restart = 1;
2320                         goto journal_error;
2321                 }
2322         }
2323         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2324         if (IS_ERR(de)) {
2325                 err = PTR_ERR(de);
2326                 goto cleanup;
2327         }
2328         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2329         goto cleanup;
2330
2331 journal_error:
2332         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2333 cleanup:
2334         brelse(bh);
2335         dx_release(frames);
2336         /* @restart is true means htree-path has been changed, we need to
2337          * repeat dx_probe() to find out valid htree-path
2338          */
2339         if (restart && err == 0)
2340                 goto again;
2341         return err;
2342 }
2343
2344 /*
2345  * ext4_generic_delete_entry deletes a directory entry by merging it
2346  * with the previous entry
2347  */
2348 int ext4_generic_delete_entry(handle_t *handle,
2349                               struct inode *dir,
2350                               struct ext4_dir_entry_2 *de_del,
2351                               struct buffer_head *bh,
2352                               void *entry_buf,
2353                               int buf_size,
2354                               int csum_size)
2355 {
2356         struct ext4_dir_entry_2 *de, *pde;
2357         unsigned int blocksize = dir->i_sb->s_blocksize;
2358         int i;
2359
2360         i = 0;
2361         pde = NULL;
2362         de = (struct ext4_dir_entry_2 *)entry_buf;
2363         while (i < buf_size - csum_size) {
2364                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2365                                          entry_buf, buf_size, i))
2366                         return -EFSCORRUPTED;
2367                 if (de == de_del)  {
2368                         if (pde)
2369                                 pde->rec_len = ext4_rec_len_to_disk(
2370                                         ext4_rec_len_from_disk(pde->rec_len,
2371                                                                blocksize) +
2372                                         ext4_rec_len_from_disk(de->rec_len,
2373                                                                blocksize),
2374                                         blocksize);
2375                         else
2376                                 de->inode = 0;
2377                         inode_inc_iversion(dir);
2378                         return 0;
2379                 }
2380                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2381                 pde = de;
2382                 de = ext4_next_entry(de, blocksize);
2383         }
2384         return -ENOENT;
2385 }
2386
2387 static int ext4_delete_entry(handle_t *handle,
2388                              struct inode *dir,
2389                              struct ext4_dir_entry_2 *de_del,
2390                              struct buffer_head *bh)
2391 {
2392         int err, csum_size = 0;
2393
2394         if (ext4_has_inline_data(dir)) {
2395                 int has_inline_data = 1;
2396                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2397                                                &has_inline_data);
2398                 if (has_inline_data)
2399                         return err;
2400         }
2401
2402         if (ext4_has_metadata_csum(dir->i_sb))
2403                 csum_size = sizeof(struct ext4_dir_entry_tail);
2404
2405         BUFFER_TRACE(bh, "get_write_access");
2406         err = ext4_journal_get_write_access(handle, bh);
2407         if (unlikely(err))
2408                 goto out;
2409
2410         err = ext4_generic_delete_entry(handle, dir, de_del,
2411                                         bh, bh->b_data,
2412                                         dir->i_sb->s_blocksize, csum_size);
2413         if (err)
2414                 goto out;
2415
2416         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2417         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2418         if (unlikely(err))
2419                 goto out;
2420
2421         return 0;
2422 out:
2423         if (err != -ENOENT)
2424                 ext4_std_error(dir->i_sb, err);
2425         return err;
2426 }
2427
2428 /*
2429  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2430  * since this indicates that nlinks count was previously 1 to avoid overflowing
2431  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2432  * that subdirectory link counts are not being maintained accurately.
2433  *
2434  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2435  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2436  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2437  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2438  */
2439 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2440 {
2441         inc_nlink(inode);
2442         if (is_dx(inode) &&
2443             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2444                 set_nlink(inode, 1);
2445 }
2446
2447 /*
2448  * If a directory had nlink == 1, then we should let it be 1. This indicates
2449  * directory has >EXT4_LINK_MAX subdirs.
2450  */
2451 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2452 {
2453         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2454                 drop_nlink(inode);
2455 }
2456
2457
2458 static int ext4_add_nondir(handle_t *handle,
2459                 struct dentry *dentry, struct inode *inode)
2460 {
2461         int err = ext4_add_entry(handle, dentry, inode);
2462         if (!err) {
2463                 ext4_mark_inode_dirty(handle, inode);
2464                 d_instantiate_new(dentry, inode);
2465                 return 0;
2466         }
2467         drop_nlink(inode);
2468         unlock_new_inode(inode);
2469         iput(inode);
2470         return err;
2471 }
2472
2473 /*
2474  * By the time this is called, we already have created
2475  * the directory cache entry for the new file, but it
2476  * is so far negative - it has no inode.
2477  *
2478  * If the create succeeds, we fill in the inode information
2479  * with d_instantiate().
2480  */
2481 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2482                        bool excl)
2483 {
2484         handle_t *handle;
2485         struct inode *inode;
2486         int err, credits, retries = 0;
2487
2488         err = dquot_initialize(dir);
2489         if (err)
2490                 return err;
2491
2492         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2493                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2494 retry:
2495         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2496                                             NULL, EXT4_HT_DIR, credits);
2497         handle = ext4_journal_current_handle();
2498         err = PTR_ERR(inode);
2499         if (!IS_ERR(inode)) {
2500                 inode->i_op = &ext4_file_inode_operations;
2501                 inode->i_fop = &ext4_file_operations;
2502                 ext4_set_aops(inode);
2503                 err = ext4_add_nondir(handle, dentry, inode);
2504                 if (!err && IS_DIRSYNC(dir))
2505                         ext4_handle_sync(handle);
2506         }
2507         if (handle)
2508                 ext4_journal_stop(handle);
2509         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2510                 goto retry;
2511         return err;
2512 }
2513
2514 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2515                       umode_t mode, dev_t rdev)
2516 {
2517         handle_t *handle;
2518         struct inode *inode;
2519         int err, credits, retries = 0;
2520
2521         err = dquot_initialize(dir);
2522         if (err)
2523                 return err;
2524
2525         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2526                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2527 retry:
2528         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2529                                             NULL, EXT4_HT_DIR, credits);
2530         handle = ext4_journal_current_handle();
2531         err = PTR_ERR(inode);
2532         if (!IS_ERR(inode)) {
2533                 init_special_inode(inode, inode->i_mode, rdev);
2534                 inode->i_op = &ext4_special_inode_operations;
2535                 err = ext4_add_nondir(handle, dentry, inode);
2536                 if (!err && IS_DIRSYNC(dir))
2537                         ext4_handle_sync(handle);
2538         }
2539         if (handle)
2540                 ext4_journal_stop(handle);
2541         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2542                 goto retry;
2543         return err;
2544 }
2545
2546 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2547 {
2548         handle_t *handle;
2549         struct inode *inode;
2550         int err, retries = 0;
2551
2552         err = dquot_initialize(dir);
2553         if (err)
2554                 return err;
2555
2556 retry:
2557         inode = ext4_new_inode_start_handle(dir, mode,
2558                                             NULL, 0, NULL,
2559                                             EXT4_HT_DIR,
2560                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2561                           4 + EXT4_XATTR_TRANS_BLOCKS);
2562         handle = ext4_journal_current_handle();
2563         err = PTR_ERR(inode);
2564         if (!IS_ERR(inode)) {
2565                 inode->i_op = &ext4_file_inode_operations;
2566                 inode->i_fop = &ext4_file_operations;
2567                 ext4_set_aops(inode);
2568                 d_tmpfile(dentry, inode);
2569                 err = ext4_orphan_add(handle, inode);
2570                 if (err)
2571                         goto err_unlock_inode;
2572                 mark_inode_dirty(inode);
2573                 unlock_new_inode(inode);
2574         }
2575         if (handle)
2576                 ext4_journal_stop(handle);
2577         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2578                 goto retry;
2579         return err;
2580 err_unlock_inode:
2581         ext4_journal_stop(handle);
2582         unlock_new_inode(inode);
2583         return err;
2584 }
2585
2586 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2587                           struct ext4_dir_entry_2 *de,
2588                           int blocksize, int csum_size,
2589                           unsigned int parent_ino, int dotdot_real_len)
2590 {
2591         de->inode = cpu_to_le32(inode->i_ino);
2592         de->name_len = 1;
2593         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2594                                            blocksize);
2595         strcpy(de->name, ".");
2596         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2597
2598         de = ext4_next_entry(de, blocksize);
2599         de->inode = cpu_to_le32(parent_ino);
2600         de->name_len = 2;
2601         if (!dotdot_real_len)
2602                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2603                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2604                                         blocksize);
2605         else
2606                 de->rec_len = ext4_rec_len_to_disk(
2607                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2608         strcpy(de->name, "..");
2609         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2610
2611         return ext4_next_entry(de, blocksize);
2612 }
2613
2614 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2615                              struct inode *inode)
2616 {
2617         struct buffer_head *dir_block = NULL;
2618         struct ext4_dir_entry_2 *de;
2619         struct ext4_dir_entry_tail *t;
2620         ext4_lblk_t block = 0;
2621         unsigned int blocksize = dir->i_sb->s_blocksize;
2622         int csum_size = 0;
2623         int err;
2624
2625         if (ext4_has_metadata_csum(dir->i_sb))
2626                 csum_size = sizeof(struct ext4_dir_entry_tail);
2627
2628         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2629                 err = ext4_try_create_inline_dir(handle, dir, inode);
2630                 if (err < 0 && err != -ENOSPC)
2631                         goto out;
2632                 if (!err)
2633                         goto out;
2634         }
2635
2636         inode->i_size = 0;
2637         dir_block = ext4_append(handle, inode, &block);
2638         if (IS_ERR(dir_block))
2639                 return PTR_ERR(dir_block);
2640         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2641         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2642         set_nlink(inode, 2);
2643         if (csum_size) {
2644                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2645                 initialize_dirent_tail(t, blocksize);
2646         }
2647
2648         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2649         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2650         if (err)
2651                 goto out;
2652         set_buffer_verified(dir_block);
2653 out:
2654         brelse(dir_block);
2655         return err;
2656 }
2657
2658 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2659 {
2660         handle_t *handle;
2661         struct inode *inode;
2662         int err, credits, retries = 0;
2663
2664         if (EXT4_DIR_LINK_MAX(dir))
2665                 return -EMLINK;
2666
2667         err = dquot_initialize(dir);
2668         if (err)
2669                 return err;
2670
2671         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2672                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2673 retry:
2674         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2675                                             &dentry->d_name,
2676                                             0, NULL, EXT4_HT_DIR, credits);
2677         handle = ext4_journal_current_handle();
2678         err = PTR_ERR(inode);
2679         if (IS_ERR(inode))
2680                 goto out_stop;
2681
2682         inode->i_op = &ext4_dir_inode_operations;
2683         inode->i_fop = &ext4_dir_operations;
2684         err = ext4_init_new_dir(handle, dir, inode);
2685         if (err)
2686                 goto out_clear_inode;
2687         err = ext4_mark_inode_dirty(handle, inode);
2688         if (!err)
2689                 err = ext4_add_entry(handle, dentry, inode);
2690         if (err) {
2691 out_clear_inode:
2692                 clear_nlink(inode);
2693                 unlock_new_inode(inode);
2694                 ext4_mark_inode_dirty(handle, inode);
2695                 iput(inode);
2696                 goto out_stop;
2697         }
2698         ext4_inc_count(handle, dir);
2699         ext4_update_dx_flag(dir);
2700         err = ext4_mark_inode_dirty(handle, dir);
2701         if (err)
2702                 goto out_clear_inode;
2703         d_instantiate_new(dentry, inode);
2704         if (IS_DIRSYNC(dir))
2705                 ext4_handle_sync(handle);
2706
2707 out_stop:
2708         if (handle)
2709                 ext4_journal_stop(handle);
2710         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2711                 goto retry;
2712         return err;
2713 }
2714
2715 /*
2716  * routine to check that the specified directory is empty (for rmdir)
2717  */
2718 bool ext4_empty_dir(struct inode *inode)
2719 {
2720         unsigned int offset;
2721         struct buffer_head *bh;
2722         struct ext4_dir_entry_2 *de;
2723         struct super_block *sb;
2724
2725         if (ext4_has_inline_data(inode)) {
2726                 int has_inline_data = 1;
2727                 int ret;
2728
2729                 ret = empty_inline_dir(inode, &has_inline_data);
2730                 if (has_inline_data)
2731                         return ret;
2732         }
2733
2734         sb = inode->i_sb;
2735         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2736                 EXT4_ERROR_INODE(inode, "invalid size");
2737                 return true;
2738         }
2739         /* The first directory block must not be a hole,
2740          * so treat it as DIRENT_HTREE
2741          */
2742         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2743         if (IS_ERR(bh))
2744                 return true;
2745
2746         de = (struct ext4_dir_entry_2 *) bh->b_data;
2747         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2748                                  0) ||
2749             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2750                 ext4_warning_inode(inode, "directory missing '.'");
2751                 brelse(bh);
2752                 return true;
2753         }
2754         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2755         de = ext4_next_entry(de, sb->s_blocksize);
2756         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2757                                  offset) ||
2758             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2759                 ext4_warning_inode(inode, "directory missing '..'");
2760                 brelse(bh);
2761                 return true;
2762         }
2763         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2764         while (offset < inode->i_size) {
2765                 if (!(offset & (sb->s_blocksize - 1))) {
2766                         unsigned int lblock;
2767                         brelse(bh);
2768                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2769                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2770                         if (bh == NULL) {
2771                                 offset += sb->s_blocksize;
2772                                 continue;
2773                         }
2774                         if (IS_ERR(bh))
2775                                 return true;
2776                 }
2777                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2778                                         (offset & (sb->s_blocksize - 1)));
2779                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2780                                          bh->b_data, bh->b_size, offset)) {
2781                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2782                         continue;
2783                 }
2784                 if (le32_to_cpu(de->inode)) {
2785                         brelse(bh);
2786                         return false;
2787                 }
2788                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2789         }
2790         brelse(bh);
2791         return true;
2792 }
2793
2794 /*
2795  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2796  * such inodes, starting at the superblock, in case we crash before the
2797  * file is closed/deleted, or in case the inode truncate spans multiple
2798  * transactions and the last transaction is not recovered after a crash.
2799  *
2800  * At filesystem recovery time, we walk this list deleting unlinked
2801  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2802  *
2803  * Orphan list manipulation functions must be called under i_mutex unless
2804  * we are just creating the inode or deleting it.
2805  */
2806 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2807 {
2808         struct super_block *sb = inode->i_sb;
2809         struct ext4_sb_info *sbi = EXT4_SB(sb);
2810         struct ext4_iloc iloc;
2811         int err = 0, rc;
2812         bool dirty = false;
2813
2814         if (!sbi->s_journal || is_bad_inode(inode))
2815                 return 0;
2816
2817         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2818                      !inode_is_locked(inode));
2819         /*
2820          * Exit early if inode already is on orphan list. This is a big speedup
2821          * since we don't have to contend on the global s_orphan_lock.
2822          */
2823         if (!list_empty(&EXT4_I(inode)->i_orphan))
2824                 return 0;
2825
2826         /*
2827          * Orphan handling is only valid for files with data blocks
2828          * being truncated, or files being unlinked. Note that we either
2829          * hold i_mutex, or the inode can not be referenced from outside,
2830          * so i_nlink should not be bumped due to race
2831          */
2832         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2833                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2834
2835         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2836         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2837         if (err)
2838                 goto out;
2839
2840         err = ext4_reserve_inode_write(handle, inode, &iloc);
2841         if (err)
2842                 goto out;
2843
2844         mutex_lock(&sbi->s_orphan_lock);
2845         /*
2846          * Due to previous errors inode may be already a part of on-disk
2847          * orphan list. If so skip on-disk list modification.
2848          */
2849         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2850             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2851                 /* Insert this inode at the head of the on-disk orphan list */
2852                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2853                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2854                 dirty = true;
2855         }
2856         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2857         mutex_unlock(&sbi->s_orphan_lock);
2858
2859         if (dirty) {
2860                 err = ext4_handle_dirty_super(handle, sb);
2861                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2862                 if (!err)
2863                         err = rc;
2864                 if (err) {
2865                         /*
2866                          * We have to remove inode from in-memory list if
2867                          * addition to on disk orphan list failed. Stray orphan
2868                          * list entries can cause panics at unmount time.
2869                          */
2870                         mutex_lock(&sbi->s_orphan_lock);
2871                         list_del_init(&EXT4_I(inode)->i_orphan);
2872                         mutex_unlock(&sbi->s_orphan_lock);
2873                 }
2874         } else
2875                 brelse(iloc.bh);
2876
2877         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2878         jbd_debug(4, "orphan inode %lu will point to %d\n",
2879                         inode->i_ino, NEXT_ORPHAN(inode));
2880 out:
2881         ext4_std_error(sb, err);
2882         return err;
2883 }
2884
2885 /*
2886  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2887  * of such inodes stored on disk, because it is finally being cleaned up.
2888  */
2889 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2890 {
2891         struct list_head *prev;
2892         struct ext4_inode_info *ei = EXT4_I(inode);
2893         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2894         __u32 ino_next;
2895         struct ext4_iloc iloc;
2896         int err = 0;
2897
2898         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2899                 return 0;
2900
2901         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2902                      !inode_is_locked(inode));
2903         /* Do this quick check before taking global s_orphan_lock. */
2904         if (list_empty(&ei->i_orphan))
2905                 return 0;
2906
2907         if (handle) {
2908                 /* Grab inode buffer early before taking global s_orphan_lock */
2909                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2910         }
2911
2912         mutex_lock(&sbi->s_orphan_lock);
2913         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2914
2915         prev = ei->i_orphan.prev;
2916         list_del_init(&ei->i_orphan);
2917
2918         /* If we're on an error path, we may not have a valid
2919          * transaction handle with which to update the orphan list on
2920          * disk, but we still need to remove the inode from the linked
2921          * list in memory. */
2922         if (!handle || err) {
2923                 mutex_unlock(&sbi->s_orphan_lock);
2924                 goto out_err;
2925         }
2926
2927         ino_next = NEXT_ORPHAN(inode);
2928         if (prev == &sbi->s_orphan) {
2929                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2930                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2931                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2932                 if (err) {
2933                         mutex_unlock(&sbi->s_orphan_lock);
2934                         goto out_brelse;
2935                 }
2936                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2937                 mutex_unlock(&sbi->s_orphan_lock);
2938                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2939         } else {
2940                 struct ext4_iloc iloc2;
2941                 struct inode *i_prev =
2942                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2943
2944                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2945                           i_prev->i_ino, ino_next);
2946                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2947                 if (err) {
2948                         mutex_unlock(&sbi->s_orphan_lock);
2949                         goto out_brelse;
2950                 }
2951                 NEXT_ORPHAN(i_prev) = ino_next;
2952                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2953                 mutex_unlock(&sbi->s_orphan_lock);
2954         }
2955         if (err)
2956                 goto out_brelse;
2957         NEXT_ORPHAN(inode) = 0;
2958         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2959 out_err:
2960         ext4_std_error(inode->i_sb, err);
2961         return err;
2962
2963 out_brelse:
2964         brelse(iloc.bh);
2965         goto out_err;
2966 }
2967
2968 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2969 {
2970         int retval;
2971         struct inode *inode;
2972         struct buffer_head *bh;
2973         struct ext4_dir_entry_2 *de;
2974         handle_t *handle = NULL;
2975
2976         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2977                 return -EIO;
2978
2979         /* Initialize quotas before so that eventual writes go in
2980          * separate transaction */
2981         retval = dquot_initialize(dir);
2982         if (retval)
2983                 return retval;
2984         retval = dquot_initialize(d_inode(dentry));
2985         if (retval)
2986                 return retval;
2987
2988         retval = -ENOENT;
2989         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2990         if (IS_ERR(bh))
2991                 return PTR_ERR(bh);
2992         if (!bh)
2993                 goto end_rmdir;
2994
2995         inode = d_inode(dentry);
2996
2997         retval = -EFSCORRUPTED;
2998         if (le32_to_cpu(de->inode) != inode->i_ino)
2999                 goto end_rmdir;
3000
3001         retval = -ENOTEMPTY;
3002         if (!ext4_empty_dir(inode))
3003                 goto end_rmdir;
3004
3005         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3006                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3007         if (IS_ERR(handle)) {
3008                 retval = PTR_ERR(handle);
3009                 handle = NULL;
3010                 goto end_rmdir;
3011         }
3012
3013         if (IS_DIRSYNC(dir))
3014                 ext4_handle_sync(handle);
3015
3016         retval = ext4_delete_entry(handle, dir, de, bh);
3017         if (retval)
3018                 goto end_rmdir;
3019         if (!EXT4_DIR_LINK_EMPTY(inode))
3020                 ext4_warning_inode(inode,
3021                              "empty directory '%.*s' has too many links (%u)",
3022                              dentry->d_name.len, dentry->d_name.name,
3023                              inode->i_nlink);
3024         inode->i_version++;
3025         clear_nlink(inode);
3026         /* There's no need to set i_disksize: the fact that i_nlink is
3027          * zero will ensure that the right thing happens during any
3028          * recovery. */
3029         inode->i_size = 0;
3030         ext4_orphan_add(handle, inode);
3031         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3032         ext4_mark_inode_dirty(handle, inode);
3033         ext4_dec_count(handle, dir);
3034         ext4_update_dx_flag(dir);
3035         ext4_mark_inode_dirty(handle, dir);
3036
3037 end_rmdir:
3038         brelse(bh);
3039         if (handle)
3040                 ext4_journal_stop(handle);
3041         return retval;
3042 }
3043
3044 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3045 {
3046         int retval;
3047         struct inode *inode;
3048         struct buffer_head *bh;
3049         struct ext4_dir_entry_2 *de;
3050         handle_t *handle = NULL;
3051
3052         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3053                 return -EIO;
3054
3055         trace_ext4_unlink_enter(dir, dentry);
3056         /* Initialize quotas before so that eventual writes go
3057          * in separate transaction */
3058         retval = dquot_initialize(dir);
3059         if (retval)
3060                 return retval;
3061         retval = dquot_initialize(d_inode(dentry));
3062         if (retval)
3063                 return retval;
3064
3065         retval = -ENOENT;
3066         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3067         if (IS_ERR(bh))
3068                 return PTR_ERR(bh);
3069         if (!bh)
3070                 goto end_unlink;
3071
3072         inode = d_inode(dentry);
3073
3074         retval = -EFSCORRUPTED;
3075         if (le32_to_cpu(de->inode) != inode->i_ino)
3076                 goto end_unlink;
3077
3078         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3079                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3080         if (IS_ERR(handle)) {
3081                 retval = PTR_ERR(handle);
3082                 handle = NULL;
3083                 goto end_unlink;
3084         }
3085
3086         if (IS_DIRSYNC(dir))
3087                 ext4_handle_sync(handle);
3088
3089         retval = ext4_delete_entry(handle, dir, de, bh);
3090         if (retval)
3091                 goto end_unlink;
3092         dir->i_ctime = dir->i_mtime = current_time(dir);
3093         ext4_update_dx_flag(dir);
3094         ext4_mark_inode_dirty(handle, dir);
3095         if (inode->i_nlink == 0)
3096                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3097                                    dentry->d_name.len, dentry->d_name.name);
3098         else
3099                 drop_nlink(inode);
3100         if (!inode->i_nlink)
3101                 ext4_orphan_add(handle, inode);
3102         inode->i_ctime = current_time(inode);
3103         ext4_mark_inode_dirty(handle, inode);
3104
3105 end_unlink:
3106         brelse(bh);
3107         if (handle)
3108                 ext4_journal_stop(handle);
3109         trace_ext4_unlink_exit(dentry, retval);
3110         return retval;
3111 }
3112
3113 static int ext4_symlink(struct inode *dir,
3114                         struct dentry *dentry, const char *symname)
3115 {
3116         handle_t *handle;
3117         struct inode *inode;
3118         int err, len = strlen(symname);
3119         int credits;
3120         bool encryption_required;
3121         struct fscrypt_str disk_link;
3122         struct fscrypt_symlink_data *sd = NULL;
3123
3124         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3125                 return -EIO;
3126
3127         disk_link.len = len + 1;
3128         disk_link.name = (char *) symname;
3129
3130         encryption_required = (ext4_encrypted_inode(dir) ||
3131                                DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3132         if (encryption_required) {
3133                 err = fscrypt_get_encryption_info(dir);
3134                 if (err)
3135                         return err;
3136                 if (!fscrypt_has_encryption_key(dir))
3137                         return -ENOKEY;
3138                 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3139                                  sizeof(struct fscrypt_symlink_data));
3140                 sd = kzalloc(disk_link.len, GFP_KERNEL);
3141                 if (!sd)
3142                         return -ENOMEM;
3143         }
3144
3145         if (disk_link.len > dir->i_sb->s_blocksize) {
3146                 err = -ENAMETOOLONG;
3147                 goto err_free_sd;
3148         }
3149
3150         err = dquot_initialize(dir);
3151         if (err)
3152                 goto err_free_sd;
3153
3154         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3155                 /*
3156                  * For non-fast symlinks, we just allocate inode and put it on
3157                  * orphan list in the first transaction => we need bitmap,
3158                  * group descriptor, sb, inode block, quota blocks, and
3159                  * possibly selinux xattr blocks.
3160                  */
3161                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3162                           EXT4_XATTR_TRANS_BLOCKS;
3163         } else {
3164                 /*
3165                  * Fast symlink. We have to add entry to directory
3166                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3167                  * allocate new inode (bitmap, group descriptor, inode block,
3168                  * quota blocks, sb is already counted in previous macros).
3169                  */
3170                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3171                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3172         }
3173
3174         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3175                                             &dentry->d_name, 0, NULL,
3176                                             EXT4_HT_DIR, credits);
3177         handle = ext4_journal_current_handle();
3178         if (IS_ERR(inode)) {
3179                 if (handle)
3180                         ext4_journal_stop(handle);
3181                 err = PTR_ERR(inode);
3182                 goto err_free_sd;
3183         }
3184
3185         if (encryption_required) {
3186                 struct qstr istr;
3187                 struct fscrypt_str ostr =
3188                         FSTR_INIT(sd->encrypted_path, disk_link.len);
3189
3190                 istr.name = (const unsigned char *) symname;
3191                 istr.len = len;
3192                 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3193                 if (err)
3194                         goto err_drop_inode;
3195                 sd->len = cpu_to_le16(ostr.len);
3196                 disk_link.name = (char *) sd;
3197                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3198         }
3199
3200         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3201                 if (!encryption_required)
3202                         inode->i_op = &ext4_symlink_inode_operations;
3203                 inode_nohighmem(inode);
3204                 ext4_set_aops(inode);
3205                 /*
3206                  * We cannot call page_symlink() with transaction started
3207                  * because it calls into ext4_write_begin() which can wait
3208                  * for transaction commit if we are running out of space
3209                  * and thus we deadlock. So we have to stop transaction now
3210                  * and restart it when symlink contents is written.
3211                  *
3212                  * To keep fs consistent in case of crash, we have to put inode
3213                  * to orphan list in the mean time.
3214                  */
3215                 drop_nlink(inode);
3216                 err = ext4_orphan_add(handle, inode);
3217                 ext4_journal_stop(handle);
3218                 handle = NULL;
3219                 if (err)
3220                         goto err_drop_inode;
3221                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3222                 if (err)
3223                         goto err_drop_inode;
3224                 /*
3225                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3226                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3227                  */
3228                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3229                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3230                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3231                 if (IS_ERR(handle)) {
3232                         err = PTR_ERR(handle);
3233                         handle = NULL;
3234                         goto err_drop_inode;
3235                 }
3236                 set_nlink(inode, 1);
3237                 err = ext4_orphan_del(handle, inode);
3238                 if (err)
3239                         goto err_drop_inode;
3240         } else {
3241                 /* clear the extent format for fast symlink */
3242                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3243                 if (!encryption_required) {
3244                         inode->i_op = &ext4_fast_symlink_inode_operations;
3245                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3246                 }
3247                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3248                        disk_link.len);
3249                 inode->i_size = disk_link.len - 1;
3250         }
3251         EXT4_I(inode)->i_disksize = inode->i_size;
3252         err = ext4_add_nondir(handle, dentry, inode);
3253         if (!err && IS_DIRSYNC(dir))
3254                 ext4_handle_sync(handle);
3255
3256         if (handle)
3257                 ext4_journal_stop(handle);
3258         kfree(sd);
3259         return err;
3260 err_drop_inode:
3261         if (handle)
3262                 ext4_journal_stop(handle);
3263         clear_nlink(inode);
3264         unlock_new_inode(inode);
3265         iput(inode);
3266 err_free_sd:
3267         kfree(sd);
3268         return err;
3269 }
3270
3271 static int ext4_link(struct dentry *old_dentry,
3272                      struct inode *dir, struct dentry *dentry)
3273 {
3274         handle_t *handle;
3275         struct inode *inode = d_inode(old_dentry);
3276         int err, retries = 0;
3277
3278         if (inode->i_nlink >= EXT4_LINK_MAX)
3279                 return -EMLINK;
3280         if (ext4_encrypted_inode(dir) &&
3281                         !fscrypt_has_permitted_context(dir, inode))
3282                 return -EXDEV;
3283
3284        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3285            (!projid_eq(EXT4_I(dir)->i_projid,
3286                        EXT4_I(old_dentry->d_inode)->i_projid)))
3287                 return -EXDEV;
3288
3289         err = dquot_initialize(dir);
3290         if (err)
3291                 return err;
3292
3293 retry:
3294         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3295                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3296                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3297         if (IS_ERR(handle))
3298                 return PTR_ERR(handle);
3299
3300         if (IS_DIRSYNC(dir))
3301                 ext4_handle_sync(handle);
3302
3303         inode->i_ctime = current_time(inode);
3304         ext4_inc_count(handle, inode);
3305         ihold(inode);
3306
3307         err = ext4_add_entry(handle, dentry, inode);
3308         if (!err) {
3309                 ext4_mark_inode_dirty(handle, inode);
3310                 /* this can happen only for tmpfile being
3311                  * linked the first time
3312                  */
3313                 if (inode->i_nlink == 1)
3314                         ext4_orphan_del(handle, inode);
3315                 d_instantiate(dentry, inode);
3316         } else {
3317                 drop_nlink(inode);
3318                 iput(inode);
3319         }
3320         ext4_journal_stop(handle);
3321         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3322                 goto retry;
3323         return err;
3324 }
3325
3326
3327 /*
3328  * Try to find buffer head where contains the parent block.
3329  * It should be the inode block if it is inlined or the 1st block
3330  * if it is a normal dir.
3331  */
3332 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3333                                         struct inode *inode,
3334                                         int *retval,
3335                                         struct ext4_dir_entry_2 **parent_de,
3336                                         int *inlined)
3337 {
3338         struct buffer_head *bh;
3339
3340         if (!ext4_has_inline_data(inode)) {
3341                 /* The first directory block must not be a hole, so
3342                  * treat it as DIRENT_HTREE
3343                  */
3344                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3345                 if (IS_ERR(bh)) {
3346                         *retval = PTR_ERR(bh);
3347                         return NULL;
3348                 }
3349                 *parent_de = ext4_next_entry(
3350                                         (struct ext4_dir_entry_2 *)bh->b_data,
3351                                         inode->i_sb->s_blocksize);
3352                 return bh;
3353         }
3354
3355         *inlined = 1;
3356         return ext4_get_first_inline_block(inode, parent_de, retval);
3357 }
3358
3359 struct ext4_renament {
3360         struct inode *dir;
3361         struct dentry *dentry;
3362         struct inode *inode;
3363         bool is_dir;
3364         int dir_nlink_delta;
3365
3366         /* entry for "dentry" */
3367         struct buffer_head *bh;
3368         struct ext4_dir_entry_2 *de;
3369         int inlined;
3370
3371         /* entry for ".." in inode if it's a directory */
3372         struct buffer_head *dir_bh;
3373         struct ext4_dir_entry_2 *parent_de;
3374         int dir_inlined;
3375 };
3376
3377 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3378 {
3379         int retval;
3380
3381         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3382                                               &retval, &ent->parent_de,
3383                                               &ent->dir_inlined);
3384         if (!ent->dir_bh)
3385                 return retval;
3386         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3387                 return -EFSCORRUPTED;
3388         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3389         return ext4_journal_get_write_access(handle, ent->dir_bh);
3390 }
3391
3392 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3393                                   unsigned dir_ino)
3394 {
3395         int retval;
3396
3397         ent->parent_de->inode = cpu_to_le32(dir_ino);
3398         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3399         if (!ent->dir_inlined) {
3400                 if (is_dx(ent->inode)) {
3401                         retval = ext4_handle_dirty_dx_node(handle,
3402                                                            ent->inode,
3403                                                            ent->dir_bh);
3404                 } else {
3405                         retval = ext4_handle_dirty_dirent_node(handle,
3406                                                                ent->inode,
3407                                                                ent->dir_bh);
3408                 }
3409         } else {
3410                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3411         }
3412         if (retval) {
3413                 ext4_std_error(ent->dir->i_sb, retval);
3414                 return retval;
3415         }
3416         return 0;
3417 }
3418
3419 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3420                        unsigned ino, unsigned file_type)
3421 {
3422         int retval;
3423
3424         BUFFER_TRACE(ent->bh, "get write access");
3425         retval = ext4_journal_get_write_access(handle, ent->bh);
3426         if (retval)
3427                 return retval;
3428         ent->de->inode = cpu_to_le32(ino);
3429         if (ext4_has_feature_filetype(ent->dir->i_sb))
3430                 ent->de->file_type = file_type;
3431         ent->dir->i_version++;
3432         ent->dir->i_ctime = ent->dir->i_mtime =
3433                 current_time(ent->dir);
3434         ext4_mark_inode_dirty(handle, ent->dir);
3435         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3436         if (!ent->inlined) {
3437                 retval = ext4_handle_dirty_dirent_node(handle,
3438                                                        ent->dir, ent->bh);
3439                 if (unlikely(retval)) {
3440                         ext4_std_error(ent->dir->i_sb, retval);
3441                         return retval;
3442                 }
3443         }
3444
3445         return 0;
3446 }
3447
3448 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3449                           unsigned ino, unsigned file_type)
3450 {
3451         struct ext4_renament old = *ent;
3452         int retval = 0;
3453
3454         /*
3455          * old->de could have moved from under us during make indexed dir,
3456          * so the old->de may no longer valid and need to find it again
3457          * before reset old inode info.
3458          */
3459         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3460         if (IS_ERR(old.bh))
3461                 retval = PTR_ERR(old.bh);
3462         if (!old.bh)
3463                 retval = -ENOENT;
3464         if (retval) {
3465                 ext4_std_error(old.dir->i_sb, retval);
3466                 return;
3467         }
3468
3469         ext4_setent(handle, &old, ino, file_type);
3470         brelse(old.bh);
3471 }
3472
3473 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3474                                   const struct qstr *d_name)
3475 {
3476         int retval = -ENOENT;
3477         struct buffer_head *bh;
3478         struct ext4_dir_entry_2 *de;
3479
3480         bh = ext4_find_entry(dir, d_name, &de, NULL);
3481         if (IS_ERR(bh))
3482                 return PTR_ERR(bh);
3483         if (bh) {
3484                 retval = ext4_delete_entry(handle, dir, de, bh);
3485                 brelse(bh);
3486         }
3487         return retval;
3488 }
3489
3490 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3491                                int force_reread)
3492 {
3493         int retval;
3494         /*
3495          * ent->de could have moved from under us during htree split, so make
3496          * sure that we are deleting the right entry.  We might also be pointing
3497          * to a stale entry in the unused part of ent->bh so just checking inum
3498          * and the name isn't enough.
3499          */
3500         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3501             ent->de->name_len != ent->dentry->d_name.len ||
3502             strncmp(ent->de->name, ent->dentry->d_name.name,
3503                     ent->de->name_len) ||
3504             force_reread) {
3505                 retval = ext4_find_delete_entry(handle, ent->dir,
3506                                                 &ent->dentry->d_name);
3507         } else {
3508                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3509                 if (retval == -ENOENT) {
3510                         retval = ext4_find_delete_entry(handle, ent->dir,
3511                                                         &ent->dentry->d_name);
3512                 }
3513         }
3514
3515         if (retval) {
3516                 ext4_warning_inode(ent->dir,
3517                                    "Deleting old file: nlink %d, error=%d",
3518                                    ent->dir->i_nlink, retval);
3519         }
3520 }
3521
3522 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3523 {
3524         if (ent->dir_nlink_delta) {
3525                 if (ent->dir_nlink_delta == -1)
3526                         ext4_dec_count(handle, ent->dir);
3527                 else
3528                         ext4_inc_count(handle, ent->dir);
3529                 ext4_mark_inode_dirty(handle, ent->dir);
3530         }
3531 }
3532
3533 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3534                                               int credits, handle_t **h)
3535 {
3536         struct inode *wh;
3537         handle_t *handle;
3538         int retries = 0;
3539
3540         /*
3541          * for inode block, sb block, group summaries,
3542          * and inode bitmap
3543          */
3544         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3545                     EXT4_XATTR_TRANS_BLOCKS + 4);
3546 retry:
3547         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3548                                          &ent->dentry->d_name, 0, NULL,
3549                                          EXT4_HT_DIR, credits);
3550
3551         handle = ext4_journal_current_handle();
3552         if (IS_ERR(wh)) {
3553                 if (handle)
3554                         ext4_journal_stop(handle);
3555                 if (PTR_ERR(wh) == -ENOSPC &&
3556                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3557                         goto retry;
3558         } else {
3559                 *h = handle;
3560                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3561                 wh->i_op = &ext4_special_inode_operations;
3562         }
3563         return wh;
3564 }
3565
3566 /*
3567  * Anybody can rename anything with this: the permission checks are left to the
3568  * higher-level routines.
3569  *
3570  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3571  * while new_{dentry,inode) refers to the destination dentry/inode
3572  * This comes from rename(const char *oldpath, const char *newpath)
3573  */
3574 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3575                        struct inode *new_dir, struct dentry *new_dentry,
3576                        unsigned int flags)
3577 {
3578         handle_t *handle = NULL;
3579         struct ext4_renament old = {
3580                 .dir = old_dir,
3581                 .dentry = old_dentry,
3582                 .inode = d_inode(old_dentry),
3583         };
3584         struct ext4_renament new = {
3585                 .dir = new_dir,
3586                 .dentry = new_dentry,
3587                 .inode = d_inode(new_dentry),
3588         };
3589         int force_reread;
3590         int retval;
3591         struct inode *whiteout = NULL;
3592         int credits;
3593         u8 old_file_type;
3594
3595         if (new.inode && new.inode->i_nlink == 0) {
3596                 EXT4_ERROR_INODE(new.inode,
3597                                  "target of rename is already freed");
3598                 return -EFSCORRUPTED;
3599         }
3600
3601         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3602             (!projid_eq(EXT4_I(new_dir)->i_projid,
3603                         EXT4_I(old_dentry->d_inode)->i_projid)))
3604                 return -EXDEV;
3605
3606         if ((ext4_encrypted_inode(old_dir) &&
3607              !fscrypt_has_encryption_key(old_dir)) ||
3608             (ext4_encrypted_inode(new_dir) &&
3609              !fscrypt_has_encryption_key(new_dir)))
3610                 return -ENOKEY;
3611
3612         retval = dquot_initialize(old.dir);
3613         if (retval)
3614                 return retval;
3615         retval = dquot_initialize(new.dir);
3616         if (retval)
3617                 return retval;
3618
3619         /* Initialize quotas before so that eventual writes go
3620          * in separate transaction */
3621         if (new.inode) {
3622                 retval = dquot_initialize(new.inode);
3623                 if (retval)
3624                         return retval;
3625         }
3626
3627         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3628         if (IS_ERR(old.bh))
3629                 return PTR_ERR(old.bh);
3630         /*
3631          *  Check for inode number is _not_ due to possible IO errors.
3632          *  We might rmdir the source, keep it as pwd of some process
3633          *  and merrily kill the link to whatever was created under the
3634          *  same name. Goodbye sticky bit ;-<
3635          */
3636         retval = -ENOENT;
3637         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3638                 goto release_bh;
3639
3640         if ((old.dir != new.dir) &&
3641             ext4_encrypted_inode(new.dir) &&
3642             !fscrypt_has_permitted_context(new.dir, old.inode)) {
3643                 retval = -EXDEV;
3644                 goto release_bh;
3645         }
3646
3647         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3648                                  &new.de, &new.inlined);
3649         if (IS_ERR(new.bh)) {
3650                 retval = PTR_ERR(new.bh);
3651                 new.bh = NULL;
3652                 goto release_bh;
3653         }
3654         if (new.bh) {
3655                 if (!new.inode) {
3656                         brelse(new.bh);
3657                         new.bh = NULL;
3658                 }
3659         }
3660         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3661                 ext4_alloc_da_blocks(old.inode);
3662
3663         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3664                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3665         if (!(flags & RENAME_WHITEOUT)) {
3666                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3667                 if (IS_ERR(handle)) {
3668                         retval = PTR_ERR(handle);
3669                         goto release_bh;
3670                 }
3671         } else {
3672                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3673                 if (IS_ERR(whiteout)) {
3674                         retval = PTR_ERR(whiteout);
3675                         goto release_bh;
3676                 }
3677         }
3678
3679         old_file_type = old.de->file_type;
3680         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3681                 ext4_handle_sync(handle);
3682
3683         if (S_ISDIR(old.inode->i_mode)) {
3684                 if (new.inode) {
3685                         retval = -ENOTEMPTY;
3686                         if (!ext4_empty_dir(new.inode))
3687                                 goto end_rename;
3688                 } else {
3689                         retval = -EMLINK;
3690                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3691                                 goto end_rename;
3692                 }
3693                 retval = ext4_rename_dir_prepare(handle, &old);
3694                 if (retval)
3695                         goto end_rename;
3696         }
3697         /*
3698          * If we're renaming a file within an inline_data dir and adding or
3699          * setting the new dirent causes a conversion from inline_data to
3700          * extents/blockmap, we need to force the dirent delete code to
3701          * re-read the directory, or else we end up trying to delete a dirent
3702          * from what is now the extent tree root (or a block map).
3703          */
3704         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3705                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3706
3707         if (whiteout) {
3708                 /*
3709                  * Do this before adding a new entry, so the old entry is sure
3710                  * to be still pointing to the valid old entry.
3711                  */
3712                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3713                                      EXT4_FT_CHRDEV);
3714                 if (retval)
3715                         goto end_rename;
3716                 ext4_mark_inode_dirty(handle, whiteout);
3717         }
3718         if (!new.bh) {
3719                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3720                 if (retval)
3721                         goto end_rename;
3722         } else {
3723                 retval = ext4_setent(handle, &new,
3724                                      old.inode->i_ino, old_file_type);
3725                 if (retval)
3726                         goto end_rename;
3727         }
3728         if (force_reread)
3729                 force_reread = !ext4_test_inode_flag(new.dir,
3730                                                      EXT4_INODE_INLINE_DATA);
3731
3732         /*
3733          * Like most other Unix systems, set the ctime for inodes on a
3734          * rename.
3735          */
3736         old.inode->i_ctime = current_time(old.inode);
3737         ext4_mark_inode_dirty(handle, old.inode);
3738
3739         if (!whiteout) {
3740                 /*
3741                  * ok, that's it
3742                  */
3743                 ext4_rename_delete(handle, &old, force_reread);
3744         }
3745
3746         if (new.inode) {
3747                 ext4_dec_count(handle, new.inode);
3748                 new.inode->i_ctime = current_time(new.inode);
3749         }
3750         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3751         ext4_update_dx_flag(old.dir);
3752         if (old.dir_bh) {
3753                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3754                 if (retval)
3755                         goto end_rename;
3756
3757                 ext4_dec_count(handle, old.dir);
3758                 if (new.inode) {
3759                         /* checked ext4_empty_dir above, can't have another
3760                          * parent, ext4_dec_count() won't work for many-linked
3761                          * dirs */
3762                         clear_nlink(new.inode);
3763                 } else {
3764                         ext4_inc_count(handle, new.dir);
3765                         ext4_update_dx_flag(new.dir);
3766                         ext4_mark_inode_dirty(handle, new.dir);
3767                 }
3768         }
3769         ext4_mark_inode_dirty(handle, old.dir);
3770         if (new.inode) {
3771                 ext4_mark_inode_dirty(handle, new.inode);
3772                 if (!new.inode->i_nlink)
3773                         ext4_orphan_add(handle, new.inode);
3774         }
3775         retval = 0;
3776
3777 end_rename:
3778         if (whiteout) {
3779                 if (retval) {
3780                         ext4_resetent(handle, &old,
3781                                       old.inode->i_ino, old_file_type);
3782                         drop_nlink(whiteout);
3783                         ext4_orphan_add(handle, whiteout);
3784                 }
3785                 unlock_new_inode(whiteout);
3786                 ext4_journal_stop(handle);
3787                 iput(whiteout);
3788         } else {
3789                 ext4_journal_stop(handle);
3790         }
3791 release_bh:
3792         brelse(old.dir_bh);
3793         brelse(old.bh);
3794         brelse(new.bh);
3795         return retval;
3796 }
3797
3798 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3799                              struct inode *new_dir, struct dentry *new_dentry)
3800 {
3801         handle_t *handle = NULL;
3802         struct ext4_renament old = {
3803                 .dir = old_dir,
3804                 .dentry = old_dentry,
3805                 .inode = d_inode(old_dentry),
3806         };
3807         struct ext4_renament new = {
3808                 .dir = new_dir,
3809                 .dentry = new_dentry,
3810                 .inode = d_inode(new_dentry),
3811         };
3812         u8 new_file_type;
3813         int retval;
3814         struct timespec ctime;
3815
3816         if ((ext4_encrypted_inode(old_dir) &&
3817              !fscrypt_has_encryption_key(old_dir)) ||
3818             (ext4_encrypted_inode(new_dir) &&
3819              !fscrypt_has_encryption_key(new_dir)))
3820                 return -ENOKEY;
3821
3822         if ((ext4_encrypted_inode(old_dir) ||
3823              ext4_encrypted_inode(new_dir)) &&
3824             (old_dir != new_dir) &&
3825             (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3826              !fscrypt_has_permitted_context(old_dir, new.inode)))
3827                 return -EXDEV;
3828
3829         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3830              !projid_eq(EXT4_I(new_dir)->i_projid,
3831                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3832             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3833              !projid_eq(EXT4_I(old_dir)->i_projid,
3834                         EXT4_I(new_dentry->d_inode)->i_projid)))
3835                 return -EXDEV;
3836
3837         retval = dquot_initialize(old.dir);
3838         if (retval)
3839                 return retval;
3840         retval = dquot_initialize(new.dir);
3841         if (retval)
3842                 return retval;
3843
3844         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3845                                  &old.de, &old.inlined);
3846         if (IS_ERR(old.bh))
3847                 return PTR_ERR(old.bh);
3848         /*
3849          *  Check for inode number is _not_ due to possible IO errors.
3850          *  We might rmdir the source, keep it as pwd of some process
3851          *  and merrily kill the link to whatever was created under the
3852          *  same name. Goodbye sticky bit ;-<
3853          */
3854         retval = -ENOENT;
3855         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3856                 goto end_rename;
3857
3858         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3859                                  &new.de, &new.inlined);
3860         if (IS_ERR(new.bh)) {
3861                 retval = PTR_ERR(new.bh);
3862                 new.bh = NULL;
3863                 goto end_rename;
3864         }
3865
3866         /* RENAME_EXCHANGE case: old *and* new must both exist */
3867         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3868                 goto end_rename;
3869
3870         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3871                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3872                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3873         if (IS_ERR(handle)) {
3874                 retval = PTR_ERR(handle);
3875                 handle = NULL;
3876                 goto end_rename;
3877         }
3878
3879         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3880                 ext4_handle_sync(handle);
3881
3882         if (S_ISDIR(old.inode->i_mode)) {
3883                 old.is_dir = true;
3884                 retval = ext4_rename_dir_prepare(handle, &old);
3885                 if (retval)
3886                         goto end_rename;
3887         }
3888         if (S_ISDIR(new.inode->i_mode)) {
3889                 new.is_dir = true;
3890                 retval = ext4_rename_dir_prepare(handle, &new);
3891                 if (retval)
3892                         goto end_rename;
3893         }
3894
3895         /*
3896          * Other than the special case of overwriting a directory, parents'
3897          * nlink only needs to be modified if this is a cross directory rename.
3898          */
3899         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3900                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3901                 new.dir_nlink_delta = -old.dir_nlink_delta;
3902                 retval = -EMLINK;
3903                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3904                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3905                         goto end_rename;
3906         }
3907
3908         new_file_type = new.de->file_type;
3909         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3910         if (retval)
3911                 goto end_rename;
3912
3913         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3914         if (retval)
3915                 goto end_rename;
3916
3917         /*
3918          * Like most other Unix systems, set the ctime for inodes on a
3919          * rename.
3920          */
3921         ctime = current_time(old.inode);
3922         old.inode->i_ctime = ctime;
3923         new.inode->i_ctime = ctime;
3924         ext4_mark_inode_dirty(handle, old.inode);
3925         ext4_mark_inode_dirty(handle, new.inode);
3926
3927         if (old.dir_bh) {
3928                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3929                 if (retval)
3930                         goto end_rename;
3931         }
3932         if (new.dir_bh) {
3933                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3934                 if (retval)
3935                         goto end_rename;
3936         }
3937         ext4_update_dir_count(handle, &old);
3938         ext4_update_dir_count(handle, &new);
3939         retval = 0;
3940
3941 end_rename:
3942         brelse(old.dir_bh);
3943         brelse(new.dir_bh);
3944         brelse(old.bh);
3945         brelse(new.bh);
3946         if (handle)
3947                 ext4_journal_stop(handle);
3948         return retval;
3949 }
3950
3951 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3952                         struct inode *new_dir, struct dentry *new_dentry,
3953                         unsigned int flags)
3954 {
3955         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3956                 return -EIO;
3957
3958         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3959                 return -EINVAL;
3960
3961         if (flags & RENAME_EXCHANGE) {
3962                 return ext4_cross_rename(old_dir, old_dentry,
3963                                          new_dir, new_dentry);
3964         }
3965
3966         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3967 }
3968
3969 /*
3970  * directories can handle most operations...
3971  */
3972 const struct inode_operations ext4_dir_inode_operations = {
3973         .create         = ext4_create,
3974         .lookup         = ext4_lookup,
3975         .link           = ext4_link,
3976         .unlink         = ext4_unlink,
3977         .symlink        = ext4_symlink,
3978         .mkdir          = ext4_mkdir,
3979         .rmdir          = ext4_rmdir,
3980         .mknod          = ext4_mknod,
3981         .tmpfile        = ext4_tmpfile,
3982         .rename         = ext4_rename2,
3983         .setattr        = ext4_setattr,
3984         .getattr        = ext4_getattr,
3985         .listxattr      = ext4_listxattr,
3986         .get_acl        = ext4_get_acl,
3987         .set_acl        = ext4_set_acl,
3988         .fiemap         = ext4_fiemap,
3989 };
3990
3991 const struct inode_operations ext4_special_inode_operations = {
3992         .setattr        = ext4_setattr,
3993         .getattr        = ext4_getattr,
3994         .listxattr      = ext4_listxattr,
3995         .get_acl        = ext4_get_acl,
3996         .set_acl        = ext4_set_acl,
3997 };