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 <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 113                 bh = ERR_PTR(-EIO);
 114         else
 115                 bh = ext4_bread(NULL, inode, block, 0);
 116         if (IS_ERR(bh)) {
 117                 __ext4_warning(inode->i_sb, func, line,
 118                                "inode #%lu: lblock %lu: comm %s: "
 119                                "error %ld reading directory block",
 120                                inode->i_ino, (unsigned long)block,
 121                                current->comm, PTR_ERR(bh));
 122
 123                 return bh;
 124         }
 125         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 126                 ext4_error_inode(inode, func, line, block,
 127                                  "Directory hole found for htree %s block",
 128                                  (type == INDEX) ? "index" : "leaf");
 129                 return ERR_PTR(-EFSCORRUPTED);
 130         }
 131         if (!bh)
 132                 return NULL;
 133         dirent = (struct ext4_dir_entry *) bh->b_data;
 134         /* Determine whether or not we have an index block */
 135         if (is_dx(inode)) {
 136                 if (block == 0)
 137                         is_dx_block = 1;
 138                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 139                                                 inode->i_sb->s_blocksize) ==
 140                          inode->i_sb->s_blocksize)
 141                         is_dx_block = 1;
 142         }
 143         if (!is_dx_block && type == INDEX) {
 144                 ext4_error_inode(inode, func, line, block,
 145                        "directory leaf block found instead of index block");
 146                 brelse(bh);
 147                 return ERR_PTR(-EFSCORRUPTED);
 148         }
 149         if (!ext4_has_metadata_csum(inode->i_sb) ||
 150             buffer_verified(bh))
 151                 return bh;
 152
 153         /*
 154          * An empty leaf block can get mistaken for a index block; for
 155          * this reason, we can only check the index checksum when the
 156          * caller is sure it should be an index block.
 157          */
 158         if (is_dx_block && type == INDEX) {
 159                 if (ext4_dx_csum_verify(inode, dirent) &&
 160                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 161                         set_buffer_verified(bh);
 162                 else {
 163                         ext4_error_inode_err(inode, func, line, block,
 164                                              EFSBADCRC,
 165                                              "Directory index failed checksum");
 166                         brelse(bh);
 167                         return ERR_PTR(-EFSBADCRC);
 168                 }
 169         }
 170         if (!is_dx_block) {
 171                 if (ext4_dirblock_csum_verify(inode, bh) &&
 172                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 173                         set_buffer_verified(bh);
 174                 else {
 175                         ext4_error_inode_err(inode, func, line, block,
 176                                              EFSBADCRC,
 177                                              "Directory block failed checksum");
 178                         brelse(bh);
 179                         return ERR_PTR(-EFSBADCRC);
 180                 }
 181         }
 182         return bh;
 183 }
 184
 185 #ifndef assert
 186 #define assert(test) J_ASSERT(test)
 187 #endif
 188
 189 #ifdef DX_DEBUG
 190 #define dxtrace(command) command
 191 #else
 192 #define dxtrace(command)
 193 #endif
 194
 195 struct fake_dirent
 196 {
 197         __le32 inode;
 198         __le16 rec_len;
 199         u8 name_len;
 200         u8 file_type;
 201 };
 202
 203 struct dx_countlimit
 204 {
 205         __le16 limit;
 206         __le16 count;
 207 };
 208
 209 struct dx_entry
 210 {
 211         __le32 hash;
 212         __le32 block;
 213 };
 214
 215 /*
 216  * dx_root_info is laid out so that if it should somehow get overlaid by a
 217  * dirent the two low bits of the hash version will be zero.  Therefore, the
 218  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 219  */
 220
 221 struct dx_root
 222 {
 223         struct fake_dirent dot;
 224         char dot_name[4];
 225         struct fake_dirent dotdot;
 226         char dotdot_name[4];
 227         struct dx_root_info
 228         {
 229                 __le32 reserved_zero;
 230                 u8 hash_version;
 231                 u8 info_length; /* 8 */
 232                 u8 indirect_levels;
 233                 u8 unused_flags;
 234         }
 235         info;
 236         struct dx_entry entries[];
 237 };
 238
 239 struct dx_node
 240 {
 241         struct fake_dirent fake;
 242         struct dx_entry entries[];
 243 };
 244
 245
 246 struct dx_frame
 247 {
 248         struct buffer_head *bh;
 249         struct dx_entry *entries;
 250         struct dx_entry *at;
 251 };
 252
 253 struct dx_map_entry
 254 {
 255         u32 hash;
 256         u16 offs;
 257         u16 size;
 258 };
 259
 260 /*
 261  * This goes at the end of each htree block.
 262  */
 263 struct dx_tail {
 264         u32 dt_reserved;
 265         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 266 };
 267
 268 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 269 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 270 static inline unsigned dx_get_hash(struct dx_entry *entry);
 271 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 272 static unsigned dx_get_count(struct dx_entry *entries);
 273 static unsigned dx_get_limit(struct dx_entry *entries);
 274 static void dx_set_count(struct dx_entry *entries, unsigned value);
 275 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 276 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 277 static unsigned dx_node_limit(struct inode *dir);
 278 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 279                                  struct inode *dir,
 280                                  struct dx_hash_info *hinfo,
 281                                  struct dx_frame *frame);
 282 static void dx_release(struct dx_frame *frames);
 283 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 284                        unsigned blocksize, struct dx_hash_info *hinfo,
 285                        struct dx_map_entry map[]);
 286 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 287 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 288                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 289 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 290 static void dx_insert_block(struct dx_frame *frame,
 291                                         u32 hash, ext4_lblk_t block);
 292 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 293                                  struct dx_frame *frame,
 294                                  struct dx_frame *frames,
 295                                  __u32 *start_hash);
 296 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 297                 struct ext4_filename *fname,
 298                 struct ext4_dir_entry_2 **res_dir);
 299 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 300                              struct inode *dir, struct inode *inode);
 301
 302 /* checksumming functions */
 303 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 304                                  unsigned int blocksize)
 305 {
 306         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 307
 308         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 309         t->det_rec_len = ext4_rec_len_to_disk(
 310                         sizeof(struct ext4_dir_entry_tail), blocksize);
 311         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 312 }
 313
 314 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 315 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 316                                                    struct buffer_head *bh)
 317 {
 318         struct ext4_dir_entry_tail *t;
 319
 320 #ifdef PARANOID
 321         struct ext4_dir_entry *d, *top;
 322
 323         d = (struct ext4_dir_entry *)bh->b_data;
 324         top = (struct ext4_dir_entry *)(bh->b_data +
 325                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 326                  sizeof(struct ext4_dir_entry_tail)));
 327         while (d < top && d->rec_len)
 328                 d = (struct ext4_dir_entry *)(((void *)d) +
 329                     le16_to_cpu(d->rec_len));
 330
 331         if (d != top)
 332                 return NULL;
 333
 334         t = (struct ext4_dir_entry_tail *)d;
 335 #else
 336         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 337 #endif
 338
 339         if (t->det_reserved_zero1 ||
 340             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 341             t->det_reserved_zero2 ||
 342             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 343                 return NULL;
 344
 345         return t;
 346 }
 347
 348 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 349 {
 350         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 351         struct ext4_inode_info *ei = EXT4_I(inode);
 352         __u32 csum;
 353
 354         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 355         return cpu_to_le32(csum);
 356 }
 357
 358 #define warn_no_space_for_csum(inode)                                   \
 359         __warn_no_space_for_csum((inode), __func__, __LINE__)
 360
 361 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 362                                      unsigned int line)
 363 {
 364         __ext4_warning_inode(inode, func, line,
 365                 "No space for directory leaf checksum. Please run e2fsck -D.");
 366 }
 367
 368 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 369 {
 370         struct ext4_dir_entry_tail *t;
 371
 372         if (!ext4_has_metadata_csum(inode->i_sb))
 373                 return 1;
 374
 375         t = get_dirent_tail(inode, bh);
 376         if (!t) {
 377                 warn_no_space_for_csum(inode);
 378                 return 0;
 379         }
 380
 381         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 382                                                   (char *)t - bh->b_data))
 383                 return 0;
 384
 385         return 1;
 386 }
 387
 388 static void ext4_dirblock_csum_set(struct inode *inode,
 389                                  struct buffer_head *bh)
 390 {
 391         struct ext4_dir_entry_tail *t;
 392
 393         if (!ext4_has_metadata_csum(inode->i_sb))
 394                 return;
 395
 396         t = get_dirent_tail(inode, bh);
 397         if (!t) {
 398                 warn_no_space_for_csum(inode);
 399                 return;
 400         }
 401
 402         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 403                                              (char *)t - bh->b_data);
 404 }
 405
 406 int ext4_handle_dirty_dirblock(handle_t *handle,
 407                                struct inode *inode,
 408                                struct buffer_head *bh)
 409 {
 410         ext4_dirblock_csum_set(inode, bh);
 411         return ext4_handle_dirty_metadata(handle, inode, bh);
 412 }
 413
 414 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 415                                                struct ext4_dir_entry *dirent,
 416                                                int *offset)
 417 {
 418         struct ext4_dir_entry *dp;
 419         struct dx_root_info *root;
 420         int count_offset;
 421
 422         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 423                 count_offset = 8;
 424         else if (le16_to_cpu(dirent->rec_len) == 12) {
 425                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 426                 if (le16_to_cpu(dp->rec_len) !=
 427                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 428                         return NULL;
 429                 root = (struct dx_root_info *)(((void *)dp + 12));
 430                 if (root->reserved_zero ||
 431                     root->info_length != sizeof(struct dx_root_info))
 432                         return NULL;
 433                 count_offset = 32;
 434         } else
 435                 return NULL;
 436
 437         if (offset)
 438                 *offset = count_offset;
 439         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 440 }
 441
 442 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 443                            int count_offset, int count, struct dx_tail *t)
 444 {
 445         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 446         struct ext4_inode_info *ei = EXT4_I(inode);
 447         __u32 csum;
 448         int size;
 449         __u32 dummy_csum = 0;
 450         int offset = offsetof(struct dx_tail, dt_checksum);
 451
 452         size = count_offset + (count * sizeof(struct dx_entry));
 453         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 454         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 455         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 456
 457         return cpu_to_le32(csum);
 458 }
 459
 460 static int ext4_dx_csum_verify(struct inode *inode,
 461                                struct ext4_dir_entry *dirent)
 462 {
 463         struct dx_countlimit *c;
 464         struct dx_tail *t;
 465         int count_offset, limit, count;
 466
 467         if (!ext4_has_metadata_csum(inode->i_sb))
 468                 return 1;
 469
 470         c = get_dx_countlimit(inode, dirent, &count_offset);
 471         if (!c) {
 472                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 473                 return 0;
 474         }
 475         limit = le16_to_cpu(c->limit);
 476         count = le16_to_cpu(c->count);
 477         if (count_offset + (limit * sizeof(struct dx_entry)) >
 478             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 479                 warn_no_space_for_csum(inode);
 480                 return 0;
 481         }
 482         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 483
 484         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 485                                             count, t))
 486                 return 0;
 487         return 1;
 488 }
 489
 490 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 491 {
 492         struct dx_countlimit *c;
 493         struct dx_tail *t;
 494         int count_offset, limit, count;
 495
 496         if (!ext4_has_metadata_csum(inode->i_sb))
 497                 return;
 498
 499         c = get_dx_countlimit(inode, dirent, &count_offset);
 500         if (!c) {
 501                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 502                 return;
 503         }
 504         limit = le16_to_cpu(c->limit);
 505         count = le16_to_cpu(c->count);
 506         if (count_offset + (limit * sizeof(struct dx_entry)) >
 507             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 508                 warn_no_space_for_csum(inode);
 509                 return;
 510         }
 511         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 512
 513         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 514 }
 515
 516 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 517                                             struct inode *inode,
 518                                             struct buffer_head *bh)
 519 {
 520         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 521         return ext4_handle_dirty_metadata(handle, inode, bh);
 522 }
 523
 524 /*
 525  * p is at least 6 bytes before the end of page
 526  */
 527 static inline struct ext4_dir_entry_2 *
 528 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 529 {
 530         return (struct ext4_dir_entry_2 *)((char *)p +
 531                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 532 }
 533
 534 /*
 535  * Future: use high four bits of block for coalesce-on-delete flags
 536  * Mask them off for now.
 537  */
 538
 539 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 540 {
 541         return le32_to_cpu(entry->block) & 0x0fffffff;
 542 }
 543
 544 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 545 {
 546         entry->block = cpu_to_le32(value);
 547 }
 548
 549 static inline unsigned dx_get_hash(struct dx_entry *entry)
 550 {
 551         return le32_to_cpu(entry->hash);
 552 }
 553
 554 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 555 {
 556         entry->hash = cpu_to_le32(value);
 557 }
 558
 559 static inline unsigned dx_get_count(struct dx_entry *entries)
 560 {
 561         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 562 }
 563
 564 static inline unsigned dx_get_limit(struct dx_entry *entries)
 565 {
 566         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 567 }
 568
 569 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 570 {
 571         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 572 }
 573
 574 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 575 {
 576         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 577 }
 578
 579 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 580 {
 581         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 582                 EXT4_DIR_REC_LEN(2) - infosize;
 583
 584         if (ext4_has_metadata_csum(dir->i_sb))
 585                 entry_space -= sizeof(struct dx_tail);
 586         return entry_space / sizeof(struct dx_entry);
 587 }
 588
 589 static inline unsigned dx_node_limit(struct inode *dir)
 590 {
 591         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 592
 593         if (ext4_has_metadata_csum(dir->i_sb))
 594                 entry_space -= sizeof(struct dx_tail);
 595         return entry_space / sizeof(struct dx_entry);
 596 }
 597
 598 /*
 599  * Debug
 600  */
 601 #ifdef DX_DEBUG
 602 static void dx_show_index(char * label, struct dx_entry *entries)
 603 {
 604         int i, n = dx_get_count (entries);
 605         printk(KERN_DEBUG "%s index", label);
 606         for (i = 0; i < n; i++) {
 607                 printk(KERN_CONT " %x->%lu",
 608                        i ? dx_get_hash(entries + i) : 0,
 609                        (unsigned long)dx_get_block(entries + i));
 610         }
 611         printk(KERN_CONT "\n");
 612 }
 613
 614 struct stats
 615 {
 616         unsigned names;
 617         unsigned space;
 618         unsigned bcount;
 619 };
 620
 621 static struct stats dx_show_leaf(struct inode *dir,
 622                                 struct dx_hash_info *hinfo,
 623                                 struct ext4_dir_entry_2 *de,
 624                                 int size, int show_names)
 625 {
 626         unsigned names = 0, space = 0;
 627         char *base = (char *) de;
 628         struct dx_hash_info h = *hinfo;
 629
 630         printk("names: ");
 631         while ((char *) de < base + size)
 632         {
 633                 if (de->inode)
 634                 {
 635                         if (show_names)
 636                         {
 637 #ifdef CONFIG_FS_ENCRYPTION
 638                                 int len;
 639                                 char *name;
 640                                 struct fscrypt_str fname_crypto_str =
 641                                         FSTR_INIT(NULL, 0);
 642                                 int res = 0;
 643
 644                                 name  = de->name;
 645                                 len = de->name_len;
 646                                 if (IS_ENCRYPTED(dir))
 647                                         res = fscrypt_get_encryption_info(dir);
 648                                 if (res) {
 649                                         printk(KERN_WARNING "Error setting up"
 650                                                " fname crypto: %d\n", res);
 651                                 }
 652                                 if (!fscrypt_has_encryption_key(dir)) {
 653                                         /* Directory is not encrypted */
 654                                         ext4fs_dirhash(dir, de->name,
 655                                                 de->name_len, &h);
 656                                         printk("%*.s:(U)%x.%u ", len,
 657                                                name, h.hash,
 658                                                (unsigned) ((char *) de
 659                                                            - base));
 660                                 } else {
 661                                         struct fscrypt_str de_name =
 662                                                 FSTR_INIT(name, len);
 663
 664                                         /* Directory is encrypted */
 665                                         res = fscrypt_fname_alloc_buffer(
 666                                                 len, &fname_crypto_str);
 667                                         if (res)
 668                                                 printk(KERN_WARNING "Error "
 669                                                         "allocating crypto "
 670                                                         "buffer--skipping "
 671                                                         "crypto\n");
 672                                         res = fscrypt_fname_disk_to_usr(dir,
 673                                                 0, 0, &de_name,
 674                                                 &fname_crypto_str);
 675                                         if (res) {
 676                                                 printk(KERN_WARNING "Error "
 677                                                         "converting filename "
 678                                                         "from disk to usr"
 679                                                         "\n");
 680                                                 name = "??";
 681                                                 len = 2;
 682                                         } else {
 683                                                 name = fname_crypto_str.name;
 684                                                 len = fname_crypto_str.len;
 685                                         }
 686                                         ext4fs_dirhash(dir, de->name,
 687                                                        de->name_len, &h);
 688                                         printk("%*.s:(E)%x.%u ", len, name,
 689                                                h.hash, (unsigned) ((char *) de
 690                                                                    - base));
 691                                         fscrypt_fname_free_buffer(
 692                                                         &fname_crypto_str);
 693                                 }
 694 #else
 695                                 int len = de->name_len;
 696                                 char *name = de->name;
 697                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 698                                 printk("%*.s:%x.%u ", len, name, h.hash,
 699                                        (unsigned) ((char *) de - base));
 700 #endif
 701                         }
 702                         space += EXT4_DIR_REC_LEN(de->name_len);
 703                         names++;
 704                 }
 705                 de = ext4_next_entry(de, size);
 706         }
 707         printk(KERN_CONT "(%i)\n", names);
 708         return (struct stats) { names, space, 1 };
 709 }
 710
 711 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 712                              struct dx_entry *entries, int levels)
 713 {
 714         unsigned blocksize = dir->i_sb->s_blocksize;
 715         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 716         unsigned bcount = 0;
 717         struct buffer_head *bh;
 718         printk("%i indexed blocks...\n", count);
 719         for (i = 0; i < count; i++, entries++)
 720         {
 721                 ext4_lblk_t block = dx_get_block(entries);
 722                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 723                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 724                 struct stats stats;
 725                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 726                 bh = ext4_bread(NULL,dir, block, 0);
 727                 if (!bh || IS_ERR(bh))
 728                         continue;
 729                 stats = levels?
 730                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 731                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 732                         bh->b_data, blocksize, 0);
 733                 names += stats.names;
 734                 space += stats.space;
 735                 bcount += stats.bcount;
 736                 brelse(bh);
 737         }
 738         if (bcount)
 739                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 740                        levels ? "" : "   ", names, space/bcount,
 741                        (space/bcount)*100/blocksize);
 742         return (struct stats) { names, space, bcount};
 743 }
 744 #endif /* DX_DEBUG */
 745
 746 /*
 747  * Probe for a directory leaf block to search.
 748  *
 749  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 750  * error in the directory index, and the caller should fall back to
 751  * searching the directory normally.  The callers of dx_probe **MUST**
 752  * check for this error code, and make sure it never gets reflected
 753  * back to userspace.
 754  */
 755 static struct dx_frame *
 756 dx_probe(struct ext4_filename *fname, struct inode *dir,
 757          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 758 {
 759         unsigned count, indirect;
 760         struct dx_entry *at, *entries, *p, *q, *m;
 761         struct dx_root *root;
 762         struct dx_frame *frame = frame_in;
 763         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 764         u32 hash;
 765
 766         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 767         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 768         if (IS_ERR(frame->bh))
 769                 return (struct dx_frame *) frame->bh;
 770
 771         root = (struct dx_root *) frame->bh->b_data;
 772         if (root->info.hash_version != DX_HASH_TEA &&
 773             root->info.hash_version != DX_HASH_HALF_MD4 &&
 774             root->info.hash_version != DX_HASH_LEGACY) {
 775                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 776                                    root->info.hash_version);
 777                 goto fail;
 778         }
 779         if (fname)
 780                 hinfo = &fname->hinfo;
 781         hinfo->hash_version = root->info.hash_version;
 782         if (hinfo->hash_version <= DX_HASH_TEA)
 783                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 784         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 785         if (fname && fname_name(fname))
 786                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 787         hash = hinfo->hash;
 788
 789         if (root->info.unused_flags & 1) {
 790                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 791                                    root->info.unused_flags);
 792                 goto fail;
 793         }
 794
 795         indirect = root->info.indirect_levels;
 796         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 797                 ext4_warning(dir->i_sb,
 798                              "Directory (ino: %lu) htree depth %#06x exceed"
 799                              "supported value", dir->i_ino,
 800                              ext4_dir_htree_level(dir->i_sb));
 801                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 802                         ext4_warning(dir->i_sb, "Enable large directory "
 803                                                 "feature to access it");
 804                 }
 805                 goto fail;
 806         }
 807
 808         entries = (struct dx_entry *)(((char *)&root->info) +
 809                                       root->info.info_length);
 810
 811         if (dx_get_limit(entries) != dx_root_limit(dir,
 812                                                    root->info.info_length)) {
 813                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 814                                    dx_get_limit(entries),
 815                                    dx_root_limit(dir, root->info.info_length));
 816                 goto fail;
 817         }
 818
 819         dxtrace(printk("Look up %x", hash));
 820         while (1) {
 821                 count = dx_get_count(entries);
 822                 if (!count || count > dx_get_limit(entries)) {
 823                         ext4_warning_inode(dir,
 824                                            "dx entry: count %u beyond limit %u",
 825                                            count, dx_get_limit(entries));
 826                         goto fail;
 827                 }
 828
 829                 p = entries + 1;
 830                 q = entries + count - 1;
 831                 while (p <= q) {
 832                         m = p + (q - p) / 2;
 833                         dxtrace(printk(KERN_CONT "."));
 834                         if (dx_get_hash(m) > hash)
 835                                 q = m - 1;
 836                         else
 837                                 p = m + 1;
 838                 }
 839
 840                 if (0) { // linear search cross check
 841                         unsigned n = count - 1;
 842                         at = entries;
 843                         while (n--)
 844                         {
 845                                 dxtrace(printk(KERN_CONT ","));
 846                                 if (dx_get_hash(++at) > hash)
 847                                 {
 848                                         at--;
 849                                         break;
 850                                 }
 851                         }
 852                         assert (at == p - 1);
 853                 }
 854
 855                 at = p - 1;
 856                 dxtrace(printk(KERN_CONT " %x->%u\n",
 857                                at == entries ? 0 : dx_get_hash(at),
 858                                dx_get_block(at)));
 859                 frame->entries = entries;
 860                 frame->at = at;
 861                 if (!indirect--)
 862                         return frame;
 863                 frame++;
 864                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 865                 if (IS_ERR(frame->bh)) {
 866                         ret_err = (struct dx_frame *) frame->bh;
 867                         frame->bh = NULL;
 868                         goto fail;
 869                 }
 870                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 871
 872                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 873                         ext4_warning_inode(dir,
 874                                 "dx entry: limit %u != node limit %u",
 875                                 dx_get_limit(entries), dx_node_limit(dir));
 876                         goto fail;
 877                 }
 878         }
 879 fail:
 880         while (frame >= frame_in) {
 881                 brelse(frame->bh);
 882                 frame--;
 883         }
 884
 885         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 886                 ext4_warning_inode(dir,
 887                         "Corrupt directory, running e2fsck is recommended");
 888         return ret_err;
 889 }
 890
 891 static void dx_release(struct dx_frame *frames)
 892 {
 893         struct dx_root_info *info;
 894         int i;
 895         unsigned int indirect_levels;
 896
 897         if (frames[0].bh == NULL)
 898                 return;
 899
 900         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 901         /* save local copy, "info" may be freed after brelse() */
 902         indirect_levels = info->indirect_levels;
 903         for (i = 0; i <= indirect_levels; i++) {
 904                 if (frames[i].bh == NULL)
 905                         break;
 906                 brelse(frames[i].bh);
 907                 frames[i].bh = NULL;
 908         }
 909 }
 910
 911 /*
 912  * This function increments the frame pointer to search the next leaf
 913  * block, and reads in the necessary intervening nodes if the search
 914  * should be necessary.  Whether or not the search is necessary is
 915  * controlled by the hash parameter.  If the hash value is even, then
 916  * the search is only continued if the next block starts with that
 917  * hash value.  This is used if we are searching for a specific file.
 918  *
 919  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 920  *
 921  * This function returns 1 if the caller should continue to search,
 922  * or 0 if it should not.  If there is an error reading one of the
 923  * index blocks, it will a negative error code.
 924  *
 925  * If start_hash is non-null, it will be filled in with the starting
 926  * hash of the next page.
 927  */
 928 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 929                                  struct dx_frame *frame,
 930                                  struct dx_frame *frames,
 931                                  __u32 *start_hash)
 932 {
 933         struct dx_frame *p;
 934         struct buffer_head *bh;
 935         int num_frames = 0;
 936         __u32 bhash;
 937
 938         p = frame;
 939         /*
 940          * Find the next leaf page by incrementing the frame pointer.
 941          * If we run out of entries in the interior node, loop around and
 942          * increment pointer in the parent node.  When we break out of
 943          * this loop, num_frames indicates the number of interior
 944          * nodes need to be read.
 945          */
 946         while (1) {
 947                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 948                         break;
 949                 if (p == frames)
 950                         return 0;
 951                 num_frames++;
 952                 p--;
 953         }
 954
 955         /*
 956          * If the hash is 1, then continue only if the next page has a
 957          * continuation hash of any value.  This is used for readdir
 958          * handling.  Otherwise, check to see if the hash matches the
 959          * desired contiuation hash.  If it doesn't, return since
 960          * there's no point to read in the successive index pages.
 961          */
 962         bhash = dx_get_hash(p->at);
 963         if (start_hash)
 964                 *start_hash = bhash;
 965         if ((hash & 1) == 0) {
 966                 if ((bhash & ~1) != hash)
 967                         return 0;
 968         }
 969         /*
 970          * If the hash is HASH_NB_ALWAYS, we always go to the next
 971          * block so no check is necessary
 972          */
 973         while (num_frames--) {
 974                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 975                 if (IS_ERR(bh))
 976                         return PTR_ERR(bh);
 977                 p++;
 978                 brelse(p->bh);
 979                 p->bh = bh;
 980                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 981         }
 982         return 1;
 983 }
 984
 985
 986 /*
 987  * This function fills a red-black tree with information from a
 988  * directory block.  It returns the number directory entries loaded
 989  * into the tree.  If there is an error it is returned in err.
 990  */
 991 static int htree_dirblock_to_tree(struct file *dir_file,
 992                                   struct inode *dir, ext4_lblk_t block,
 993                                   struct dx_hash_info *hinfo,
 994                                   __u32 start_hash, __u32 start_minor_hash)
 995 {
 996         struct buffer_head *bh;
 997         struct ext4_dir_entry_2 *de, *top;
 998         int err = 0, count = 0;
 999         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1000
1001         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1002                                                         (unsigned long)block));
1003         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1004         if (IS_ERR(bh))
1005                 return PTR_ERR(bh);
1006
1007         de = (struct ext4_dir_entry_2 *) bh->b_data;
1008         top = (struct ext4_dir_entry_2 *) ((char *) de +
1009                                            dir->i_sb->s_blocksize -
1010                                            EXT4_DIR_REC_LEN(0));
1011         /* Check if the directory is encrypted */
1012         if (IS_ENCRYPTED(dir)) {
1013                 err = fscrypt_get_encryption_info(dir);
1014                 if (err < 0) {
1015                         brelse(bh);
1016                         return err;
1017                 }
1018                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1019                                                  &fname_crypto_str);
1020                 if (err < 0) {
1021                         brelse(bh);
1022                         return err;
1023                 }
1024         }
1025
1026         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1027                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1028                                 bh->b_data, bh->b_size,
1029                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1030                                          + ((char *)de - bh->b_data))) {
1031                         /* silently ignore the rest of the block */
1032                         break;
1033                 }
1034                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1035                 if ((hinfo->hash < start_hash) ||
1036                     ((hinfo->hash == start_hash) &&
1037                      (hinfo->minor_hash < start_minor_hash)))
1038                         continue;
1039                 if (de->inode == 0)
1040                         continue;
1041                 if (!IS_ENCRYPTED(dir)) {
1042                         tmp_str.name = de->name;
1043                         tmp_str.len = de->name_len;
1044                         err = ext4_htree_store_dirent(dir_file,
1045                                    hinfo->hash, hinfo->minor_hash, de,
1046                                    &tmp_str);
1047                 } else {
1048                         int save_len = fname_crypto_str.len;
1049                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1050                                                                 de->name_len);
1051
1052                         /* Directory is encrypted */
1053                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1054                                         hinfo->minor_hash, &de_name,
1055                                         &fname_crypto_str);
1056                         if (err) {
1057                                 count = err;
1058                                 goto errout;
1059                         }
1060                         err = ext4_htree_store_dirent(dir_file,
1061                                    hinfo->hash, hinfo->minor_hash, de,
1062                                         &fname_crypto_str);
1063                         fname_crypto_str.len = save_len;
1064                 }
1065                 if (err != 0) {
1066                         count = err;
1067                         goto errout;
1068                 }
1069                 count++;
1070         }
1071 errout:
1072         brelse(bh);
1073         fscrypt_fname_free_buffer(&fname_crypto_str);
1074         return count;
1075 }
1076
1077
1078 /*
1079  * This function fills a red-black tree with information from a
1080  * directory.  We start scanning the directory in hash order, starting
1081  * at start_hash and start_minor_hash.
1082  *
1083  * This function returns the number of entries inserted into the tree,
1084  * or a negative error code.
1085  */
1086 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1087                          __u32 start_minor_hash, __u32 *next_hash)
1088 {
1089         struct dx_hash_info hinfo;
1090         struct ext4_dir_entry_2 *de;
1091         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1092         struct inode *dir;
1093         ext4_lblk_t block;
1094         int count = 0;
1095         int ret, err;
1096         __u32 hashval;
1097         struct fscrypt_str tmp_str;
1098
1099         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1100                        start_hash, start_minor_hash));
1101         dir = file_inode(dir_file);
1102         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1103                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1104                 if (hinfo.hash_version <= DX_HASH_TEA)
1105                         hinfo.hash_version +=
1106                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1107                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1108                 if (ext4_has_inline_data(dir)) {
1109                         int has_inline_data = 1;
1110                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1111                                                        &hinfo, start_hash,
1112                                                        start_minor_hash,
1113                                                        &has_inline_data);
1114                         if (has_inline_data) {
1115                                 *next_hash = ~0;
1116                                 return count;
1117                         }
1118                 }
1119                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1120                                                start_hash, start_minor_hash);
1121                 *next_hash = ~0;
1122                 return count;
1123         }
1124         hinfo.hash = start_hash;
1125         hinfo.minor_hash = 0;
1126         frame = dx_probe(NULL, dir, &hinfo, frames);
1127         if (IS_ERR(frame))
1128                 return PTR_ERR(frame);
1129
1130         /* Add '.' and '..' from the htree header */
1131         if (!start_hash && !start_minor_hash) {
1132                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1133                 tmp_str.name = de->name;
1134                 tmp_str.len = de->name_len;
1135                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1136                                               de, &tmp_str);
1137                 if (err != 0)
1138                         goto errout;
1139                 count++;
1140         }
1141         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1142                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1143                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1144                 tmp_str.name = de->name;
1145                 tmp_str.len = de->name_len;
1146                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1147                                               de, &tmp_str);
1148                 if (err != 0)
1149                         goto errout;
1150                 count++;
1151         }
1152
1153         while (1) {
1154                 if (fatal_signal_pending(current)) {
1155                         err = -ERESTARTSYS;
1156                         goto errout;
1157                 }
1158                 cond_resched();
1159                 block = dx_get_block(frame->at);
1160                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1161                                              start_hash, start_minor_hash);
1162                 if (ret < 0) {
1163                         err = ret;
1164                         goto errout;
1165                 }
1166                 count += ret;
1167                 hashval = ~0;
1168                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1169                                             frame, frames, &hashval);
1170                 *next_hash = hashval;
1171                 if (ret < 0) {
1172                         err = ret;
1173                         goto errout;
1174                 }
1175                 /*
1176                  * Stop if:  (a) there are no more entries, or
1177                  * (b) we have inserted at least one entry and the
1178                  * next hash value is not a continuation
1179                  */
1180                 if ((ret == 0) ||
1181                     (count && ((hashval & 1) == 0)))
1182                         break;
1183         }
1184         dx_release(frames);
1185         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1186                        "next hash: %x\n", count, *next_hash));
1187         return count;
1188 errout:
1189         dx_release(frames);
1190         return (err);
1191 }
1192
1193 static inline int search_dirblock(struct buffer_head *bh,
1194                                   struct inode *dir,
1195                                   struct ext4_filename *fname,
1196                                   unsigned int offset,
1197                                   struct ext4_dir_entry_2 **res_dir)
1198 {
1199         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1200                                fname, offset, res_dir);
1201 }
1202
1203 /*
1204  * Directory block splitting, compacting
1205  */
1206
1207 /*
1208  * Create map of hash values, offsets, and sizes, stored at end of block.
1209  * Returns number of entries mapped.
1210  */
1211 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1212                        unsigned blocksize, struct dx_hash_info *hinfo,
1213                        struct dx_map_entry *map_tail)
1214 {
1215         int count = 0;
1216         char *base = (char *) de;
1217         struct dx_hash_info h = *hinfo;
1218
1219         while ((char *) de < base + blocksize) {
1220                 if (de->name_len && de->inode) {
1221                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1222                         map_tail--;
1223                         map_tail->hash = h.hash;
1224                         map_tail->offs = ((char *) de - base)>>2;
1225                         map_tail->size = le16_to_cpu(de->rec_len);
1226                         count++;
1227                         cond_resched();
1228                 }
1229                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1230                 de = ext4_next_entry(de, blocksize);
1231         }
1232         return count;
1233 }
1234
1235 /* Sort map by hash value */
1236 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1237 {
1238         struct dx_map_entry *p, *q, *top = map + count - 1;
1239         int more;
1240         /* Combsort until bubble sort doesn't suck */
1241         while (count > 2) {
1242                 count = count*10/13;
1243                 if (count - 9 < 2) /* 9, 10 -> 11 */
1244                         count = 11;
1245                 for (p = top, q = p - count; q >= map; p--, q--)
1246                         if (p->hash < q->hash)
1247                                 swap(*p, *q);
1248         }
1249         /* Garden variety bubble sort */
1250         do {
1251                 more = 0;
1252                 q = top;
1253                 while (q-- > map) {
1254                         if (q[1].hash >= q[0].hash)
1255                                 continue;
1256                         swap(*(q+1), *q);
1257                         more = 1;
1258                 }
1259         } while(more);
1260 }
1261
1262 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1263 {
1264         struct dx_entry *entries = frame->entries;
1265         struct dx_entry *old = frame->at, *new = old + 1;
1266         int count = dx_get_count(entries);
1267
1268         assert(count < dx_get_limit(entries));
1269         assert(old < entries + count);
1270         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1271         dx_set_hash(new, hash);
1272         dx_set_block(new, block);
1273         dx_set_count(entries, count + 1);
1274 }
1275
1276 #ifdef CONFIG_UNICODE
1277 /*
1278  * Test whether a case-insensitive directory entry matches the filename
1279  * being searched for.  If quick is set, assume the name being looked up
1280  * is already in the casefolded form.
1281  *
1282  * Returns: 0 if the directory entry matches, more than 0 if it
1283  * doesn't match or less than zero on error.
1284  */
1285 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1286                     const struct qstr *entry, bool quick)
1287 {
1288         const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1289         const struct unicode_map *um = sbi->s_encoding;
1290         int ret;
1291
1292         if (quick)
1293                 ret = utf8_strncasecmp_folded(um, name, entry);
1294         else
1295                 ret = utf8_strncasecmp(um, name, entry);
1296
1297         if (ret < 0) {
1298                 /* Handle invalid character sequence as either an error
1299                  * or as an opaque byte sequence.
1300                  */
1301                 if (ext4_has_strict_mode(sbi))
1302                         return -EINVAL;
1303
1304                 if (name->len != entry->len)
1305                         return 1;
1306
1307                 return !!memcmp(name->name, entry->name, name->len);
1308         }
1309
1310         return ret;
1311 }
1312
1313 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1314                                   struct fscrypt_str *cf_name)
1315 {
1316         int len;
1317
1318         if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1319                 cf_name->name = NULL;
1320                 return;
1321         }
1322
1323         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1324         if (!cf_name->name)
1325                 return;
1326
1327         len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1328                             iname, cf_name->name,
1329                             EXT4_NAME_LEN);
1330         if (len <= 0) {
1331                 kfree(cf_name->name);
1332                 cf_name->name = NULL;
1333                 return;
1334         }
1335         cf_name->len = (unsigned) len;
1336
1337 }
1338 #endif
1339
1340 /*
1341  * Test whether a directory entry matches the filename being searched for.
1342  *
1343  * Return: %true if the directory entry matches, otherwise %false.
1344  */
1345 static inline bool ext4_match(const struct inode *parent,
1346                               const struct ext4_filename *fname,
1347                               const struct ext4_dir_entry_2 *de)
1348 {
1349         struct fscrypt_name f;
1350 #ifdef CONFIG_UNICODE
1351         const struct qstr entry = {.name = de->name, .len = de->name_len};
1352 #endif
1353
1354         if (!de->inode)
1355                 return false;
1356
1357         f.usr_fname = fname->usr_fname;
1358         f.disk_name = fname->disk_name;
1359 #ifdef CONFIG_FS_ENCRYPTION
1360         f.crypto_buf = fname->crypto_buf;
1361 #endif
1362
1363 #ifdef CONFIG_UNICODE
1364         if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1365                 if (fname->cf_name.name) {
1366                         struct qstr cf = {.name = fname->cf_name.name,
1367                                           .len = fname->cf_name.len};
1368                         return !ext4_ci_compare(parent, &cf, &entry, true);
1369                 }
1370                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1371                                         false);
1372         }
1373 #endif
1374
1375         return fscrypt_match_name(&f, de->name, de->name_len);
1376 }
1377
1378 /*
1379  * Returns 0 if not found, -1 on failure, and 1 on success
1380  */
1381 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1382                     struct inode *dir, struct ext4_filename *fname,
1383                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1384 {
1385         struct ext4_dir_entry_2 * de;
1386         char * dlimit;
1387         int de_len;
1388
1389         de = (struct ext4_dir_entry_2 *)search_buf;
1390         dlimit = search_buf + buf_size;
1391         while ((char *) de < dlimit) {
1392                 /* this code is executed quadratically often */
1393                 /* do minimal checking `by hand' */
1394                 if ((char *) de + de->name_len <= dlimit &&
1395                     ext4_match(dir, fname, de)) {
1396                         /* found a match - just to be sure, do
1397                          * a full check */
1398                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1399                                                  buf_size, offset))
1400                                 return -1;
1401                         *res_dir = de;
1402                         return 1;
1403                 }
1404                 /* prevent looping on a bad block */
1405                 de_len = ext4_rec_len_from_disk(de->rec_len,
1406                                                 dir->i_sb->s_blocksize);
1407                 if (de_len <= 0)
1408                         return -1;
1409                 offset += de_len;
1410                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1411         }
1412         return 0;
1413 }
1414
1415 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1416                                struct ext4_dir_entry *de)
1417 {
1418         struct super_block *sb = dir->i_sb;
1419
1420         if (!is_dx(dir))
1421                 return 0;
1422         if (block == 0)
1423                 return 1;
1424         if (de->inode == 0 &&
1425             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1426                         sb->s_blocksize)
1427                 return 1;
1428         return 0;
1429 }
1430
1431 /*
1432  *      __ext4_find_entry()
1433  *
1434  * finds an entry in the specified directory with the wanted name. It
1435  * returns the cache buffer in which the entry was found, and the entry
1436  * itself (as a parameter - res_dir). It does NOT read the inode of the
1437  * entry - you'll have to do that yourself if you want to.
1438  *
1439  * The returned buffer_head has ->b_count elevated.  The caller is expected
1440  * to brelse() it when appropriate.
1441  */
1442 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1443                                              struct ext4_filename *fname,
1444                                              struct ext4_dir_entry_2 **res_dir,
1445                                              int *inlined)
1446 {
1447         struct super_block *sb;
1448         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1449         struct buffer_head *bh, *ret = NULL;
1450         ext4_lblk_t start, block;
1451         const u8 *name = fname->usr_fname->name;
1452         size_t ra_max = 0;      /* Number of bh's in the readahead
1453                                    buffer, bh_use[] */
1454         size_t ra_ptr = 0;      /* Current index into readahead
1455                                    buffer */
1456         ext4_lblk_t  nblocks;
1457         int i, namelen, retval;
1458
1459         *res_dir = NULL;
1460         sb = dir->i_sb;
1461         namelen = fname->usr_fname->len;
1462         if (namelen > EXT4_NAME_LEN)
1463                 return NULL;
1464
1465         if (ext4_has_inline_data(dir)) {
1466                 int has_inline_data = 1;
1467                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1468                                              &has_inline_data);
1469                 if (has_inline_data) {
1470                         if (inlined)
1471                                 *inlined = 1;
1472                         goto cleanup_and_exit;
1473                 }
1474         }
1475
1476         if ((namelen <= 2) && (name[0] == '.') &&
1477             (name[1] == '.' || name[1] == '\0')) {
1478                 /*
1479                  * "." or ".." will only be in the first block
1480                  * NFS may look up ".."; "." should be handled by the VFS
1481                  */
1482                 block = start = 0;
1483                 nblocks = 1;
1484                 goto restart;
1485         }
1486         if (is_dx(dir)) {
1487                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1488                 /*
1489                  * On success, or if the error was file not found,
1490                  * return.  Otherwise, fall back to doing a search the
1491                  * old fashioned way.
1492                  */
1493                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1494                         goto cleanup_and_exit;
1495                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1496                                "falling back\n"));
1497                 ret = NULL;
1498         }
1499         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1500         if (!nblocks) {
1501                 ret = NULL;
1502                 goto cleanup_and_exit;
1503         }
1504         start = EXT4_I(dir)->i_dir_start_lookup;
1505         if (start >= nblocks)
1506                 start = 0;
1507         block = start;
1508 restart:
1509         do {
1510                 /*
1511                  * We deal with the read-ahead logic here.
1512                  */
1513                 cond_resched();
1514                 if (ra_ptr >= ra_max) {
1515                         /* Refill the readahead buffer */
1516                         ra_ptr = 0;
1517                         if (block < start)
1518                                 ra_max = start - block;
1519                         else
1520                                 ra_max = nblocks - block;
1521                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1522                         retval = ext4_bread_batch(dir, block, ra_max,
1523                                                   false /* wait */, bh_use);
1524                         if (retval) {
1525                                 ret = ERR_PTR(retval);
1526                                 ra_max = 0;
1527                                 goto cleanup_and_exit;
1528                         }
1529                 }
1530                 if ((bh = bh_use[ra_ptr++]) == NULL)
1531                         goto next;
1532                 wait_on_buffer(bh);
1533                 if (!buffer_uptodate(bh)) {
1534                         EXT4_ERROR_INODE_ERR(dir, EIO,
1535                                              "reading directory lblock %lu",
1536                                              (unsigned long) block);
1537                         brelse(bh);
1538                         ret = ERR_PTR(-EIO);
1539                         goto cleanup_and_exit;
1540                 }
1541                 if (!buffer_verified(bh) &&
1542                     !is_dx_internal_node(dir, block,
1543                                          (struct ext4_dir_entry *)bh->b_data) &&
1544                     !ext4_dirblock_csum_verify(dir, bh)) {
1545                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1546                                              "checksumming directory "
1547                                              "block %lu", (unsigned long)block);
1548                         brelse(bh);
1549                         ret = ERR_PTR(-EFSBADCRC);
1550                         goto cleanup_and_exit;
1551                 }
1552                 set_buffer_verified(bh);
1553                 i = search_dirblock(bh, dir, fname,
1554                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1555                 if (i == 1) {
1556                         EXT4_I(dir)->i_dir_start_lookup = block;
1557                         ret = bh;
1558                         goto cleanup_and_exit;
1559                 } else {
1560                         brelse(bh);
1561                         if (i < 0)
1562                                 goto cleanup_and_exit;
1563                 }
1564         next:
1565                 if (++block >= nblocks)
1566                         block = 0;
1567         } while (block != start);
1568
1569         /*
1570          * If the directory has grown while we were searching, then
1571          * search the last part of the directory before giving up.
1572          */
1573         block = nblocks;
1574         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1575         if (block < nblocks) {
1576                 start = 0;
1577                 goto restart;
1578         }
1579
1580 cleanup_and_exit:
1581         /* Clean up the read-ahead blocks */
1582         for (; ra_ptr < ra_max; ra_ptr++)
1583                 brelse(bh_use[ra_ptr]);
1584         return ret;
1585 }
1586
1587 static struct buffer_head *ext4_find_entry(struct inode *dir,
1588                                            const struct qstr *d_name,
1589                                            struct ext4_dir_entry_2 **res_dir,
1590                                            int *inlined)
1591 {
1592         int err;
1593         struct ext4_filename fname;
1594         struct buffer_head *bh;
1595
1596         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1597         if (err == -ENOENT)
1598                 return NULL;
1599         if (err)
1600                 return ERR_PTR(err);
1601
1602         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1603
1604         ext4_fname_free_filename(&fname);
1605         return bh;
1606 }
1607
1608 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1609                                              struct dentry *dentry,
1610                                              struct ext4_dir_entry_2 **res_dir)
1611 {
1612         int err;
1613         struct ext4_filename fname;
1614         struct buffer_head *bh;
1615
1616         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1617         if (err == -ENOENT)
1618                 return NULL;
1619         if (err)
1620                 return ERR_PTR(err);
1621
1622         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1623
1624         ext4_fname_free_filename(&fname);
1625         return bh;
1626 }
1627
1628 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1629                         struct ext4_filename *fname,
1630                         struct ext4_dir_entry_2 **res_dir)
1631 {
1632         struct super_block * sb = dir->i_sb;
1633         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1634         struct buffer_head *bh;
1635         ext4_lblk_t block;
1636         int retval;
1637
1638 #ifdef CONFIG_FS_ENCRYPTION
1639         *res_dir = NULL;
1640 #endif
1641         frame = dx_probe(fname, dir, NULL, frames);
1642         if (IS_ERR(frame))
1643                 return (struct buffer_head *) frame;
1644         do {
1645                 block = dx_get_block(frame->at);
1646                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1647                 if (IS_ERR(bh))
1648                         goto errout;
1649
1650                 retval = search_dirblock(bh, dir, fname,
1651                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1652                                          res_dir);
1653                 if (retval == 1)
1654                         goto success;
1655                 brelse(bh);
1656                 if (retval == -1) {
1657                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1658                         goto errout;
1659                 }
1660
1661                 /* Check to see if we should continue to search */
1662                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1663                                                frames, NULL);
1664                 if (retval < 0) {
1665                         ext4_warning_inode(dir,
1666                                 "error %d reading directory index block",
1667                                 retval);
1668                         bh = ERR_PTR(retval);
1669                         goto errout;
1670                 }
1671         } while (retval == 1);
1672
1673         bh = NULL;
1674 errout:
1675         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1676 success:
1677         dx_release(frames);
1678         return bh;
1679 }
1680
1681 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1682 {
1683         struct inode *inode;
1684         struct ext4_dir_entry_2 *de;
1685         struct buffer_head *bh;
1686
1687         if (dentry->d_name.len > EXT4_NAME_LEN)
1688                 return ERR_PTR(-ENAMETOOLONG);
1689
1690         bh = ext4_lookup_entry(dir, dentry, &de);
1691         if (IS_ERR(bh))
1692                 return ERR_CAST(bh);
1693         inode = NULL;
1694         if (bh) {
1695                 __u32 ino = le32_to_cpu(de->inode);
1696                 brelse(bh);
1697                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1698                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1699                         return ERR_PTR(-EFSCORRUPTED);
1700                 }
1701                 if (unlikely(ino == dir->i_ino)) {
1702                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1703                                          dentry);
1704                         return ERR_PTR(-EFSCORRUPTED);
1705                 }
1706                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1707                 if (inode == ERR_PTR(-ESTALE)) {
1708                         EXT4_ERROR_INODE(dir,
1709                                          "deleted inode referenced: %u",
1710                                          ino);
1711                         return ERR_PTR(-EFSCORRUPTED);
1712                 }
1713                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1714                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1715                     !fscrypt_has_permitted_context(dir, inode)) {
1716                         ext4_warning(inode->i_sb,
1717                                      "Inconsistent encryption contexts: %lu/%lu",
1718                                      dir->i_ino, inode->i_ino);
1719                         iput(inode);
1720                         return ERR_PTR(-EPERM);
1721                 }
1722         }
1723
1724 #ifdef CONFIG_UNICODE
1725         if (!inode && IS_CASEFOLDED(dir)) {
1726                 /* Eventually we want to call d_add_ci(dentry, NULL)
1727                  * for negative dentries in the encoding case as
1728                  * well.  For now, prevent the negative dentry
1729                  * from being cached.
1730                  */
1731                 return NULL;
1732         }
1733 #endif
1734         return d_splice_alias(inode, dentry);
1735 }
1736
1737
1738 struct dentry *ext4_get_parent(struct dentry *child)
1739 {
1740         __u32 ino;
1741         static const struct qstr dotdot = QSTR_INIT("..", 2);
1742         struct ext4_dir_entry_2 * de;
1743         struct buffer_head *bh;
1744
1745         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1746         if (IS_ERR(bh))
1747                 return ERR_CAST(bh);
1748         if (!bh)
1749                 return ERR_PTR(-ENOENT);
1750         ino = le32_to_cpu(de->inode);
1751         brelse(bh);
1752
1753         if (!ext4_valid_inum(child->d_sb, ino)) {
1754                 EXT4_ERROR_INODE(d_inode(child),
1755                                  "bad parent inode number: %u", ino);
1756                 return ERR_PTR(-EFSCORRUPTED);
1757         }
1758
1759         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1760 }
1761
1762 /*
1763  * Move count entries from end of map between two memory locations.
1764  * Returns pointer to last entry moved.
1765  */
1766 static struct ext4_dir_entry_2 *
1767 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1768                 unsigned blocksize)
1769 {
1770         unsigned rec_len = 0;
1771
1772         while (count--) {
1773                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1774                                                 (from + (map->offs<<2));
1775                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1776                 memcpy (to, de, rec_len);
1777                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1778                                 ext4_rec_len_to_disk(rec_len, blocksize);
1779                 de->inode = 0;
1780                 map++;
1781                 to += rec_len;
1782         }
1783         return (struct ext4_dir_entry_2 *) (to - rec_len);
1784 }
1785
1786 /*
1787  * Compact each dir entry in the range to the minimal rec_len.
1788  * Returns pointer to last entry in range.
1789  */
1790 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1791 {
1792         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1793         unsigned rec_len = 0;
1794
1795         prev = to = de;
1796         while ((char*)de < base + blocksize) {
1797                 next = ext4_next_entry(de, blocksize);
1798                 if (de->inode && de->name_len) {
1799                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1800                         if (de > to)
1801                                 memmove(to, de, rec_len);
1802                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1803                         prev = to;
1804                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1805                 }
1806                 de = next;
1807         }
1808         return prev;
1809 }
1810
1811 /*
1812  * Split a full leaf block to make room for a new dir entry.
1813  * Allocate a new block, and move entries so that they are approx. equally full.
1814  * Returns pointer to de in block into which the new entry will be inserted.
1815  */
1816 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1817                         struct buffer_head **bh,struct dx_frame *frame,
1818                         struct dx_hash_info *hinfo)
1819 {
1820         unsigned blocksize = dir->i_sb->s_blocksize;
1821         unsigned count, continued;
1822         struct buffer_head *bh2;
1823         ext4_lblk_t newblock;
1824         u32 hash2;
1825         struct dx_map_entry *map;
1826         char *data1 = (*bh)->b_data, *data2;
1827         unsigned split, move, size;
1828         struct ext4_dir_entry_2 *de = NULL, *de2;
1829         int     csum_size = 0;
1830         int     err = 0, i;
1831
1832         if (ext4_has_metadata_csum(dir->i_sb))
1833                 csum_size = sizeof(struct ext4_dir_entry_tail);
1834
1835         bh2 = ext4_append(handle, dir, &newblock);
1836         if (IS_ERR(bh2)) {
1837                 brelse(*bh);
1838                 *bh = NULL;
1839                 return (struct ext4_dir_entry_2 *) bh2;
1840         }
1841
1842         BUFFER_TRACE(*bh, "get_write_access");
1843         err = ext4_journal_get_write_access(handle, *bh);
1844         if (err)
1845                 goto journal_error;
1846
1847         BUFFER_TRACE(frame->bh, "get_write_access");
1848         err = ext4_journal_get_write_access(handle, frame->bh);
1849         if (err)
1850                 goto journal_error;
1851
1852         data2 = bh2->b_data;
1853
1854         /* create map in the end of data2 block */
1855         map = (struct dx_map_entry *) (data2 + blocksize);
1856         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1857                              blocksize, hinfo, map);
1858         map -= count;
1859         dx_sort_map(map, count);
1860         /* Ensure that neither split block is over half full */
1861         size = 0;
1862         move = 0;
1863         for (i = count-1; i >= 0; i--) {
1864                 /* is more than half of this entry in 2nd half of the block? */
1865                 if (size + map[i].size/2 > blocksize/2)
1866                         break;
1867                 size += map[i].size;
1868                 move++;
1869         }
1870         /*
1871          * map index at which we will split
1872          *
1873          * If the sum of active entries didn't exceed half the block size, just
1874          * split it in half by count; each resulting block will have at least
1875          * half the space free.
1876          */
1877         if (i > 0)
1878                 split = count - move;
1879         else
1880                 split = count/2;
1881
1882         hash2 = map[split].hash;
1883         continued = hash2 == map[split - 1].hash;
1884         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1885                         (unsigned long)dx_get_block(frame->at),
1886                                         hash2, split, count-split));
1887
1888         /* Fancy dance to stay within two buffers */
1889         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1890                               blocksize);
1891         de = dx_pack_dirents(data1, blocksize);
1892         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1893                                            (char *) de,
1894                                            blocksize);
1895         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1896                                             (char *) de2,
1897                                             blocksize);
1898         if (csum_size) {
1899                 ext4_initialize_dirent_tail(*bh, blocksize);
1900                 ext4_initialize_dirent_tail(bh2, blocksize);
1901         }
1902
1903         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1904                         blocksize, 1));
1905         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1906                         blocksize, 1));
1907
1908         /* Which block gets the new entry? */
1909         if (hinfo->hash >= hash2) {
1910                 swap(*bh, bh2);
1911                 de = de2;
1912         }
1913         dx_insert_block(frame, hash2 + continued, newblock);
1914         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1915         if (err)
1916                 goto journal_error;
1917         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1918         if (err)
1919                 goto journal_error;
1920         brelse(bh2);
1921         dxtrace(dx_show_index("frame", frame->entries));
1922         return de;
1923
1924 journal_error:
1925         brelse(*bh);
1926         brelse(bh2);
1927         *bh = NULL;
1928         ext4_std_error(dir->i_sb, err);
1929         return ERR_PTR(err);
1930 }
1931
1932 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1933                       struct buffer_head *bh,
1934                       void *buf, int buf_size,
1935                       struct ext4_filename *fname,
1936                       struct ext4_dir_entry_2 **dest_de)
1937 {
1938         struct ext4_dir_entry_2 *de;
1939         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1940         int nlen, rlen;
1941         unsigned int offset = 0;
1942         char *top;
1943
1944         de = (struct ext4_dir_entry_2 *)buf;
1945         top = buf + buf_size - reclen;
1946         while ((char *) de <= top) {
1947                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1948                                          buf, buf_size, offset))
1949                         return -EFSCORRUPTED;
1950                 if (ext4_match(dir, fname, de))
1951                         return -EEXIST;
1952                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1953                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1954                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1955                         break;
1956                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1957                 offset += rlen;
1958         }
1959         if ((char *) de > top)
1960                 return -ENOSPC;
1961
1962         *dest_de = de;
1963         return 0;
1964 }
1965
1966 void ext4_insert_dentry(struct inode *inode,
1967                         struct ext4_dir_entry_2 *de,
1968                         int buf_size,
1969                         struct ext4_filename *fname)
1970 {
1971
1972         int nlen, rlen;
1973
1974         nlen = EXT4_DIR_REC_LEN(de->name_len);
1975         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1976         if (de->inode) {
1977                 struct ext4_dir_entry_2 *de1 =
1978                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1979                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1980                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1981                 de = de1;
1982         }
1983         de->file_type = EXT4_FT_UNKNOWN;
1984         de->inode = cpu_to_le32(inode->i_ino);
1985         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1986         de->name_len = fname_len(fname);
1987         memcpy(de->name, fname_name(fname), fname_len(fname));
1988 }
1989
1990 /*
1991  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1992  * it points to a directory entry which is guaranteed to be large
1993  * enough for new directory entry.  If de is NULL, then
1994  * add_dirent_to_buf will attempt search the directory block for
1995  * space.  It will return -ENOSPC if no space is available, and -EIO
1996  * and -EEXIST if directory entry already exists.
1997  */
1998 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1999                              struct inode *dir,
2000                              struct inode *inode, struct ext4_dir_entry_2 *de,
2001                              struct buffer_head *bh)
2002 {
2003         unsigned int    blocksize = dir->i_sb->s_blocksize;
2004         int             csum_size = 0;
2005         int             err, err2;
2006
2007         if (ext4_has_metadata_csum(inode->i_sb))
2008                 csum_size = sizeof(struct ext4_dir_entry_tail);
2009
2010         if (!de) {
2011                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2012                                         blocksize - csum_size, fname, &de);
2013                 if (err)
2014                         return err;
2015         }
2016         BUFFER_TRACE(bh, "get_write_access");
2017         err = ext4_journal_get_write_access(handle, bh);
2018         if (err) {
2019                 ext4_std_error(dir->i_sb, err);
2020                 return err;
2021         }
2022
2023         /* By now the buffer is marked for journaling */
2024         ext4_insert_dentry(inode, de, blocksize, fname);
2025
2026         /*
2027          * XXX shouldn't update any times until successful
2028          * completion of syscall, but too many callers depend
2029          * on this.
2030          *
2031          * XXX similarly, too many callers depend on
2032          * ext4_new_inode() setting the times, but error
2033          * recovery deletes the inode, so the worst that can
2034          * happen is that the times are slightly out of date
2035          * and/or different from the directory change time.
2036          */
2037         dir->i_mtime = dir->i_ctime = current_time(dir);
2038         ext4_update_dx_flag(dir);
2039         inode_inc_iversion(dir);
2040         err2 = ext4_mark_inode_dirty(handle, dir);
2041         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2042         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2043         if (err)
2044                 ext4_std_error(dir->i_sb, err);
2045         return err ? err : err2;
2046 }
2047
2048 /*
2049  * This converts a one block unindexed directory to a 3 block indexed
2050  * directory, and adds the dentry to the indexed directory.
2051  */
2052 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2053                             struct inode *dir,
2054                             struct inode *inode, struct buffer_head *bh)
2055 {
2056         struct buffer_head *bh2;
2057         struct dx_root  *root;
2058         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2059         struct dx_entry *entries;
2060         struct ext4_dir_entry_2 *de, *de2;
2061         char            *data2, *top;
2062         unsigned        len;
2063         int             retval;
2064         unsigned        blocksize;
2065         ext4_lblk_t  block;
2066         struct fake_dirent *fde;
2067         int csum_size = 0;
2068
2069         if (ext4_has_metadata_csum(inode->i_sb))
2070                 csum_size = sizeof(struct ext4_dir_entry_tail);
2071
2072         blocksize =  dir->i_sb->s_blocksize;
2073         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2074         BUFFER_TRACE(bh, "get_write_access");
2075         retval = ext4_journal_get_write_access(handle, bh);
2076         if (retval) {
2077                 ext4_std_error(dir->i_sb, retval);
2078                 brelse(bh);
2079                 return retval;
2080         }
2081         root = (struct dx_root *) bh->b_data;
2082
2083         /* The 0th block becomes the root, move the dirents out */
2084         fde = &root->dotdot;
2085         de = (struct ext4_dir_entry_2 *)((char *)fde +
2086                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2087         if ((char *) de >= (((char *) root) + blocksize)) {
2088                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2089                 brelse(bh);
2090                 return -EFSCORRUPTED;
2091         }
2092         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2093
2094         /* Allocate new block for the 0th block's dirents */
2095         bh2 = ext4_append(handle, dir, &block);
2096         if (IS_ERR(bh2)) {
2097                 brelse(bh);
2098                 return PTR_ERR(bh2);
2099         }
2100         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2101         data2 = bh2->b_data;
2102
2103         memcpy(data2, de, len);
2104         de = (struct ext4_dir_entry_2 *) data2;
2105         top = data2 + len;
2106         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2107                 de = de2;
2108         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2109                                            (char *) de, blocksize);
2110
2111         if (csum_size)
2112                 ext4_initialize_dirent_tail(bh2, blocksize);
2113
2114         /* Initialize the root; the dot dirents already exist */
2115         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2116         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2117                                            blocksize);
2118         memset (&root->info, 0, sizeof(root->info));
2119         root->info.info_length = sizeof(root->info);
2120         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2121         entries = root->entries;
2122         dx_set_block(entries, 1);
2123         dx_set_count(entries, 1);
2124         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2125
2126         /* Initialize as for dx_probe */
2127         fname->hinfo.hash_version = root->info.hash_version;
2128         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2129                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2130         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2131         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2132
2133         memset(frames, 0, sizeof(frames));
2134         frame = frames;
2135         frame->entries = entries;
2136         frame->at = entries;
2137         frame->bh = bh;
2138
2139         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2140         if (retval)
2141                 goto out_frames;
2142         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2143         if (retval)
2144                 goto out_frames;
2145
2146         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2147         if (IS_ERR(de)) {
2148                 retval = PTR_ERR(de);
2149                 goto out_frames;
2150         }
2151
2152         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2153 out_frames:
2154         /*
2155          * Even if the block split failed, we have to properly write
2156          * out all the changes we did so far. Otherwise we can end up
2157          * with corrupted filesystem.
2158          */
2159         if (retval)
2160                 ext4_mark_inode_dirty(handle, dir);
2161         dx_release(frames);
2162         brelse(bh2);
2163         return retval;
2164 }
2165
2166 /*
2167  *      ext4_add_entry()
2168  *
2169  * adds a file entry to the specified directory, using the same
2170  * semantics as ext4_find_entry(). It returns NULL if it failed.
2171  *
2172  * NOTE!! The inode part of 'de' is left at 0 - which means you
2173  * may not sleep between calling this and putting something into
2174  * the entry, as someone else might have used it while you slept.
2175  */
2176 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2177                           struct inode *inode)
2178 {
2179         struct inode *dir = d_inode(dentry->d_parent);
2180         struct buffer_head *bh = NULL;
2181         struct ext4_dir_entry_2 *de;
2182         struct super_block *sb;
2183 #ifdef CONFIG_UNICODE
2184         struct ext4_sb_info *sbi;
2185 #endif
2186         struct ext4_filename fname;
2187         int     retval;
2188         int     dx_fallback=0;
2189         unsigned blocksize;
2190         ext4_lblk_t block, blocks;
2191         int     csum_size = 0;
2192
2193         if (ext4_has_metadata_csum(inode->i_sb))
2194                 csum_size = sizeof(struct ext4_dir_entry_tail);
2195
2196         sb = dir->i_sb;
2197         blocksize = sb->s_blocksize;
2198         if (!dentry->d_name.len)
2199                 return -EINVAL;
2200
2201 #ifdef CONFIG_UNICODE
2202         sbi = EXT4_SB(sb);
2203         if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2204             sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2205                 return -EINVAL;
2206 #endif
2207
2208         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2209         if (retval)
2210                 return retval;
2211
2212         if (ext4_has_inline_data(dir)) {
2213                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2214                 if (retval < 0)
2215                         goto out;
2216                 if (retval == 1) {
2217                         retval = 0;
2218                         goto out;
2219                 }
2220         }
2221
2222         if (is_dx(dir)) {
2223                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2224                 if (!retval || (retval != ERR_BAD_DX_DIR))
2225                         goto out;
2226                 /* Can we just ignore htree data? */
2227                 if (ext4_has_metadata_csum(sb)) {
2228                         EXT4_ERROR_INODE(dir,
2229                                 "Directory has corrupted htree index.");
2230                         retval = -EFSCORRUPTED;
2231                         goto out;
2232                 }
2233                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2234                 dx_fallback++;
2235                 retval = ext4_mark_inode_dirty(handle, dir);
2236                 if (unlikely(retval))
2237                         goto out;
2238         }
2239         blocks = dir->i_size >> sb->s_blocksize_bits;
2240         for (block = 0; block < blocks; block++) {
2241                 bh = ext4_read_dirblock(dir, block, DIRENT);
2242                 if (bh == NULL) {
2243                         bh = ext4_bread(handle, dir, block,
2244                                         EXT4_GET_BLOCKS_CREATE);
2245                         goto add_to_new_block;
2246                 }
2247                 if (IS_ERR(bh)) {
2248                         retval = PTR_ERR(bh);
2249                         bh = NULL;
2250                         goto out;
2251                 }
2252                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2253                                            NULL, bh);
2254                 if (retval != -ENOSPC)
2255                         goto out;
2256
2257                 if (blocks == 1 && !dx_fallback &&
2258                     ext4_has_feature_dir_index(sb)) {
2259                         retval = make_indexed_dir(handle, &fname, dir,
2260                                                   inode, bh);
2261                         bh = NULL; /* make_indexed_dir releases bh */
2262                         goto out;
2263                 }
2264                 brelse(bh);
2265         }
2266         bh = ext4_append(handle, dir, &block);
2267 add_to_new_block:
2268         if (IS_ERR(bh)) {
2269                 retval = PTR_ERR(bh);
2270                 bh = NULL;
2271                 goto out;
2272         }
2273         de = (struct ext4_dir_entry_2 *) bh->b_data;
2274         de->inode = 0;
2275         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2276
2277         if (csum_size)
2278                 ext4_initialize_dirent_tail(bh, blocksize);
2279
2280         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2281 out:
2282         ext4_fname_free_filename(&fname);
2283         brelse(bh);
2284         if (retval == 0)
2285                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2286         return retval;
2287 }
2288
2289 /*
2290  * Returns 0 for success, or a negative error value
2291  */
2292 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2293                              struct inode *dir, struct inode *inode)
2294 {
2295         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2296         struct dx_entry *entries, *at;
2297         struct buffer_head *bh;
2298         struct super_block *sb = dir->i_sb;
2299         struct ext4_dir_entry_2 *de;
2300         int restart;
2301         int err;
2302
2303 again:
2304         restart = 0;
2305         frame = dx_probe(fname, dir, NULL, frames);
2306         if (IS_ERR(frame))
2307                 return PTR_ERR(frame);
2308         entries = frame->entries;
2309         at = frame->at;
2310         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2311         if (IS_ERR(bh)) {
2312                 err = PTR_ERR(bh);
2313                 bh = NULL;
2314                 goto cleanup;
2315         }
2316
2317         BUFFER_TRACE(bh, "get_write_access");
2318         err = ext4_journal_get_write_access(handle, bh);
2319         if (err)
2320                 goto journal_error;
2321
2322         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2323         if (err != -ENOSPC)
2324                 goto cleanup;
2325
2326         err = 0;
2327         /* Block full, should compress but for now just split */
2328         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2329                        dx_get_count(entries), dx_get_limit(entries)));
2330         /* Need to split index? */
2331         if (dx_get_count(entries) == dx_get_limit(entries)) {
2332                 ext4_lblk_t newblock;
2333                 int levels = frame - frames + 1;
2334                 unsigned int icount;
2335                 int add_level = 1;
2336                 struct dx_entry *entries2;
2337                 struct dx_node *node2;
2338                 struct buffer_head *bh2;
2339
2340                 while (frame > frames) {
2341                         if (dx_get_count((frame - 1)->entries) <
2342                             dx_get_limit((frame - 1)->entries)) {
2343                                 add_level = 0;
2344                                 break;
2345                         }
2346                         frame--; /* split higher index block */
2347                         at = frame->at;
2348                         entries = frame->entries;
2349                         restart = 1;
2350                 }
2351                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2352                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2353                                          "reach max htree level :%d",
2354                                          dir->i_ino, levels);
2355                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2356                                 ext4_warning(sb, "Large directory feature is "
2357                                                  "not enabled on this "
2358                                                  "filesystem");
2359                         }
2360                         err = -ENOSPC;
2361                         goto cleanup;
2362                 }
2363                 icount = dx_get_count(entries);
2364                 bh2 = ext4_append(handle, dir, &newblock);
2365                 if (IS_ERR(bh2)) {
2366                         err = PTR_ERR(bh2);
2367                         goto cleanup;
2368                 }
2369                 node2 = (struct dx_node *)(bh2->b_data);
2370                 entries2 = node2->entries;
2371                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2372                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2373                                                            sb->s_blocksize);
2374                 BUFFER_TRACE(frame->bh, "get_write_access");
2375                 err = ext4_journal_get_write_access(handle, frame->bh);
2376                 if (err)
2377                         goto journal_error;
2378                 if (!add_level) {
2379                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2380                         unsigned hash2 = dx_get_hash(entries + icount1);
2381                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2382                                        icount1, icount2));
2383
2384                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2385                         err = ext4_journal_get_write_access(handle,
2386                                                              (frame - 1)->bh);
2387                         if (err)
2388                                 goto journal_error;
2389
2390                         memcpy((char *) entries2, (char *) (entries + icount1),
2391                                icount2 * sizeof(struct dx_entry));
2392                         dx_set_count(entries, icount1);
2393                         dx_set_count(entries2, icount2);
2394                         dx_set_limit(entries2, dx_node_limit(dir));
2395
2396                         /* Which index block gets the new entry? */
2397                         if (at - entries >= icount1) {
2398                                 frame->at = at = at - entries - icount1 + entries2;
2399                                 frame->entries = entries = entries2;
2400                                 swap(frame->bh, bh2);
2401                         }
2402                         dx_insert_block((frame - 1), hash2, newblock);
2403                         dxtrace(dx_show_index("node", frame->entries));
2404                         dxtrace(dx_show_index("node",
2405                                ((struct dx_node *) bh2->b_data)->entries));
2406                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2407                         if (err)
2408                                 goto journal_error;
2409                         brelse (bh2);
2410                         err = ext4_handle_dirty_dx_node(handle, dir,
2411                                                    (frame - 1)->bh);
2412                         if (err)
2413                                 goto journal_error;
2414                         if (restart) {
2415                                 err = ext4_handle_dirty_dx_node(handle, dir,
2416                                                            frame->bh);
2417                                 goto journal_error;
2418                         }
2419                 } else {
2420                         struct dx_root *dxroot;
2421                         memcpy((char *) entries2, (char *) entries,
2422                                icount * sizeof(struct dx_entry));
2423                         dx_set_limit(entries2, dx_node_limit(dir));
2424
2425                         /* Set up root */
2426                         dx_set_count(entries, 1);
2427                         dx_set_block(entries + 0, newblock);
2428                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2429                         dxroot->info.indirect_levels += 1;
2430                         dxtrace(printk(KERN_DEBUG
2431                                        "Creating %d level index...\n",
2432                                        dxroot->info.indirect_levels));
2433                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2434                         if (err)
2435                                 goto journal_error;
2436                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2437                         brelse(bh2);
2438                         restart = 1;
2439                         goto journal_error;
2440                 }
2441         }
2442         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2443         if (IS_ERR(de)) {
2444                 err = PTR_ERR(de);
2445                 goto cleanup;
2446         }
2447         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2448         goto cleanup;
2449
2450 journal_error:
2451         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2452 cleanup:
2453         brelse(bh);
2454         dx_release(frames);
2455         /* @restart is true means htree-path has been changed, we need to
2456          * repeat dx_probe() to find out valid htree-path
2457          */
2458         if (restart && err == 0)
2459                 goto again;
2460         return err;
2461 }
2462
2463 /*
2464  * ext4_generic_delete_entry deletes a directory entry by merging it
2465  * with the previous entry
2466  */
2467 int ext4_generic_delete_entry(struct inode *dir,
2468                               struct ext4_dir_entry_2 *de_del,
2469                               struct buffer_head *bh,
2470                               void *entry_buf,
2471                               int buf_size,
2472                               int csum_size)
2473 {
2474         struct ext4_dir_entry_2 *de, *pde;
2475         unsigned int blocksize = dir->i_sb->s_blocksize;
2476         int i;
2477
2478         i = 0;
2479         pde = NULL;
2480         de = (struct ext4_dir_entry_2 *)entry_buf;
2481         while (i < buf_size - csum_size) {
2482                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2483                                          entry_buf, buf_size, i))
2484                         return -EFSCORRUPTED;
2485                 if (de == de_del)  {
2486                         if (pde)
2487                                 pde->rec_len = ext4_rec_len_to_disk(
2488                                         ext4_rec_len_from_disk(pde->rec_len,
2489                                                                blocksize) +
2490                                         ext4_rec_len_from_disk(de->rec_len,
2491                                                                blocksize),
2492                                         blocksize);
2493                         else
2494                                 de->inode = 0;
2495                         inode_inc_iversion(dir);
2496                         return 0;
2497                 }
2498                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2499                 pde = de;
2500                 de = ext4_next_entry(de, blocksize);
2501         }
2502         return -ENOENT;
2503 }
2504
2505 static int ext4_delete_entry(handle_t *handle,
2506                              struct inode *dir,
2507                              struct ext4_dir_entry_2 *de_del,
2508                              struct buffer_head *bh)
2509 {
2510         int err, csum_size = 0;
2511
2512         if (ext4_has_inline_data(dir)) {
2513                 int has_inline_data = 1;
2514                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2515                                                &has_inline_data);
2516                 if (has_inline_data)
2517                         return err;
2518         }
2519
2520         if (ext4_has_metadata_csum(dir->i_sb))
2521                 csum_size = sizeof(struct ext4_dir_entry_tail);
2522
2523         BUFFER_TRACE(bh, "get_write_access");
2524         err = ext4_journal_get_write_access(handle, bh);
2525         if (unlikely(err))
2526                 goto out;
2527
2528         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2529                                         dir->i_sb->s_blocksize, csum_size);
2530         if (err)
2531                 goto out;
2532
2533         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2534         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2535         if (unlikely(err))
2536                 goto out;
2537
2538         return 0;
2539 out:
2540         if (err != -ENOENT)
2541                 ext4_std_error(dir->i_sb, err);
2542         return err;
2543 }
2544
2545 /*
2546  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2547  * since this indicates that nlinks count was previously 1 to avoid overflowing
2548  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2549  * that subdirectory link counts are not being maintained accurately.
2550  *
2551  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2552  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2553  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2554  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2555  */
2556 static void ext4_inc_count(struct inode *inode)
2557 {
2558         inc_nlink(inode);
2559         if (is_dx(inode) &&
2560             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2561                 set_nlink(inode, 1);
2562 }
2563
2564 /*
2565  * If a directory had nlink == 1, then we should let it be 1. This indicates
2566  * directory has >EXT4_LINK_MAX subdirs.
2567  */
2568 static void ext4_dec_count(struct inode *inode)
2569 {
2570         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2571                 drop_nlink(inode);
2572 }
2573
2574
2575 /*
2576  * Add non-directory inode to a directory. On success, the inode reference is
2577  * consumed by dentry is instantiation. This is also indicated by clearing of
2578  * *inodep pointer. On failure, the caller is responsible for dropping the
2579  * inode reference in the safe context.
2580  */
2581 static int ext4_add_nondir(handle_t *handle,
2582                 struct dentry *dentry, struct inode **inodep)
2583 {
2584         struct inode *dir = d_inode(dentry->d_parent);
2585         struct inode *inode = *inodep;
2586         int err = ext4_add_entry(handle, dentry, inode);
2587         if (!err) {
2588                 err = ext4_mark_inode_dirty(handle, inode);
2589                 if (IS_DIRSYNC(dir))
2590                         ext4_handle_sync(handle);
2591                 d_instantiate_new(dentry, inode);
2592                 *inodep = NULL;
2593                 return err;
2594         }
2595         drop_nlink(inode);
2596         ext4_orphan_add(handle, inode);
2597         unlock_new_inode(inode);
2598         return err;
2599 }
2600
2601 /*
2602  * By the time this is called, we already have created
2603  * the directory cache entry for the new file, but it
2604  * is so far negative - it has no inode.
2605  *
2606  * If the create succeeds, we fill in the inode information
2607  * with d_instantiate().
2608  */
2609 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2610                        bool excl)
2611 {
2612         handle_t *handle;
2613         struct inode *inode, *inode_save;
2614         int err, credits, retries = 0;
2615
2616         err = dquot_initialize(dir);
2617         if (err)
2618                 return err;
2619
2620         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2621                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2622 retry:
2623         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2624                                             NULL, EXT4_HT_DIR, credits);
2625         handle = ext4_journal_current_handle();
2626         err = PTR_ERR(inode);
2627         if (!IS_ERR(inode)) {
2628                 inode->i_op = &ext4_file_inode_operations;
2629                 inode->i_fop = &ext4_file_operations;
2630                 ext4_set_aops(inode);
2631                 inode_save = inode;
2632                 ihold(inode_save);
2633                 err = ext4_add_nondir(handle, dentry, &inode);
2634                 ext4_fc_track_create(inode_save, dentry);
2635                 iput(inode_save);
2636         }
2637         if (handle)
2638                 ext4_journal_stop(handle);
2639         if (!IS_ERR_OR_NULL(inode))
2640                 iput(inode);
2641         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2642                 goto retry;
2643         return err;
2644 }
2645
2646 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2647                       umode_t mode, dev_t rdev)
2648 {
2649         handle_t *handle;
2650         struct inode *inode, *inode_save;
2651         int err, credits, retries = 0;
2652
2653         err = dquot_initialize(dir);
2654         if (err)
2655                 return err;
2656
2657         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2658                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2659 retry:
2660         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2661                                             NULL, EXT4_HT_DIR, credits);
2662         handle = ext4_journal_current_handle();
2663         err = PTR_ERR(inode);
2664         if (!IS_ERR(inode)) {
2665                 init_special_inode(inode, inode->i_mode, rdev);
2666                 inode->i_op = &ext4_special_inode_operations;
2667                 inode_save = inode;
2668                 ihold(inode_save);
2669                 err = ext4_add_nondir(handle, dentry, &inode);
2670                 if (!err)
2671                         ext4_fc_track_create(inode_save, dentry);
2672                 iput(inode_save);
2673         }
2674         if (handle)
2675                 ext4_journal_stop(handle);
2676         if (!IS_ERR_OR_NULL(inode))
2677                 iput(inode);
2678         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2679                 goto retry;
2680         return err;
2681 }
2682
2683 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2684 {
2685         handle_t *handle;
2686         struct inode *inode;
2687         int err, retries = 0;
2688
2689         err = dquot_initialize(dir);
2690         if (err)
2691                 return err;
2692
2693 retry:
2694         inode = ext4_new_inode_start_handle(dir, mode,
2695                                             NULL, 0, NULL,
2696                                             EXT4_HT_DIR,
2697                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2698                           4 + EXT4_XATTR_TRANS_BLOCKS);
2699         handle = ext4_journal_current_handle();
2700         err = PTR_ERR(inode);
2701         if (!IS_ERR(inode)) {
2702                 inode->i_op = &ext4_file_inode_operations;
2703                 inode->i_fop = &ext4_file_operations;
2704                 ext4_set_aops(inode);
2705                 d_tmpfile(dentry, inode);
2706                 err = ext4_orphan_add(handle, inode);
2707                 if (err)
2708                         goto err_unlock_inode;
2709                 mark_inode_dirty(inode);
2710                 unlock_new_inode(inode);
2711         }
2712         if (handle)
2713                 ext4_journal_stop(handle);
2714         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2715                 goto retry;
2716         return err;
2717 err_unlock_inode:
2718         ext4_journal_stop(handle);
2719         unlock_new_inode(inode);
2720         return err;
2721 }
2722
2723 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2724                           struct ext4_dir_entry_2 *de,
2725                           int blocksize, int csum_size,
2726                           unsigned int parent_ino, int dotdot_real_len)
2727 {
2728         de->inode = cpu_to_le32(inode->i_ino);
2729         de->name_len = 1;
2730         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2731                                            blocksize);
2732         strcpy(de->name, ".");
2733         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2734
2735         de = ext4_next_entry(de, blocksize);
2736         de->inode = cpu_to_le32(parent_ino);
2737         de->name_len = 2;
2738         if (!dotdot_real_len)
2739                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2740                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2741                                         blocksize);
2742         else
2743                 de->rec_len = ext4_rec_len_to_disk(
2744                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2745         strcpy(de->name, "..");
2746         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2747
2748         return ext4_next_entry(de, blocksize);
2749 }
2750
2751 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2752                              struct inode *inode)
2753 {
2754         struct buffer_head *dir_block = NULL;
2755         struct ext4_dir_entry_2 *de;
2756         ext4_lblk_t block = 0;
2757         unsigned int blocksize = dir->i_sb->s_blocksize;
2758         int csum_size = 0;
2759         int err;
2760
2761         if (ext4_has_metadata_csum(dir->i_sb))
2762                 csum_size = sizeof(struct ext4_dir_entry_tail);
2763
2764         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2765                 err = ext4_try_create_inline_dir(handle, dir, inode);
2766                 if (err < 0 && err != -ENOSPC)
2767                         goto out;
2768                 if (!err)
2769                         goto out;
2770         }
2771
2772         inode->i_size = 0;
2773         dir_block = ext4_append(handle, inode, &block);
2774         if (IS_ERR(dir_block))
2775                 return PTR_ERR(dir_block);
2776         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2777         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2778         set_nlink(inode, 2);
2779         if (csum_size)
2780                 ext4_initialize_dirent_tail(dir_block, blocksize);
2781
2782         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2783         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2784         if (err)
2785                 goto out;
2786         set_buffer_verified(dir_block);
2787 out:
2788         brelse(dir_block);
2789         return err;
2790 }
2791
2792 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2793 {
2794         handle_t *handle;
2795         struct inode *inode;
2796         int err, err2 = 0, credits, retries = 0;
2797
2798         if (EXT4_DIR_LINK_MAX(dir))
2799                 return -EMLINK;
2800
2801         err = dquot_initialize(dir);
2802         if (err)
2803                 return err;
2804
2805         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2806                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2807 retry:
2808         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2809                                             &dentry->d_name,
2810                                             0, NULL, EXT4_HT_DIR, credits);
2811         handle = ext4_journal_current_handle();
2812         err = PTR_ERR(inode);
2813         if (IS_ERR(inode))
2814                 goto out_stop;
2815
2816         inode->i_op = &ext4_dir_inode_operations;
2817         inode->i_fop = &ext4_dir_operations;
2818         err = ext4_init_new_dir(handle, dir, inode);
2819         if (err)
2820                 goto out_clear_inode;
2821         err = ext4_mark_inode_dirty(handle, inode);
2822         if (!err)
2823                 err = ext4_add_entry(handle, dentry, inode);
2824         if (err) {
2825 out_clear_inode:
2826                 clear_nlink(inode);
2827                 ext4_orphan_add(handle, inode);
2828                 unlock_new_inode(inode);
2829                 err2 = ext4_mark_inode_dirty(handle, inode);
2830                 if (unlikely(err2))
2831                         err = err2;
2832                 ext4_journal_stop(handle);
2833                 iput(inode);
2834                 goto out_retry;
2835         }
2836         ext4_fc_track_create(inode, dentry);
2837         ext4_inc_count(dir);
2838
2839         ext4_update_dx_flag(dir);
2840         err = ext4_mark_inode_dirty(handle, dir);
2841         if (err)
2842                 goto out_clear_inode;
2843         d_instantiate_new(dentry, inode);
2844         if (IS_DIRSYNC(dir))
2845                 ext4_handle_sync(handle);
2846
2847 out_stop:
2848         if (handle)
2849                 ext4_journal_stop(handle);
2850 out_retry:
2851         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2852                 goto retry;
2853         return err;
2854 }
2855
2856 /*
2857  * routine to check that the specified directory is empty (for rmdir)
2858  */
2859 bool ext4_empty_dir(struct inode *inode)
2860 {
2861         unsigned int offset;
2862         struct buffer_head *bh;
2863         struct ext4_dir_entry_2 *de;
2864         struct super_block *sb;
2865
2866         if (ext4_has_inline_data(inode)) {
2867                 int has_inline_data = 1;
2868                 int ret;
2869
2870                 ret = empty_inline_dir(inode, &has_inline_data);
2871                 if (has_inline_data)
2872                         return ret;
2873         }
2874
2875         sb = inode->i_sb;
2876         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2877                 EXT4_ERROR_INODE(inode, "invalid size");
2878                 return true;
2879         }
2880         /* The first directory block must not be a hole,
2881          * so treat it as DIRENT_HTREE
2882          */
2883         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2884         if (IS_ERR(bh))
2885                 return true;
2886
2887         de = (struct ext4_dir_entry_2 *) bh->b_data;
2888         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2889                                  0) ||
2890             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2891                 ext4_warning_inode(inode, "directory missing '.'");
2892                 brelse(bh);
2893                 return true;
2894         }
2895         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2896         de = ext4_next_entry(de, sb->s_blocksize);
2897         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2898                                  offset) ||
2899             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2900                 ext4_warning_inode(inode, "directory missing '..'");
2901                 brelse(bh);
2902                 return true;
2903         }
2904         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2905         while (offset < inode->i_size) {
2906                 if (!(offset & (sb->s_blocksize - 1))) {
2907                         unsigned int lblock;
2908                         brelse(bh);
2909                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2910                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2911                         if (bh == NULL) {
2912                                 offset += sb->s_blocksize;
2913                                 continue;
2914                         }
2915                         if (IS_ERR(bh))
2916                                 return true;
2917                 }
2918                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2919                                         (offset & (sb->s_blocksize - 1)));
2920                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2921                                          bh->b_data, bh->b_size, offset)) {
2922                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2923                         continue;
2924                 }
2925                 if (le32_to_cpu(de->inode)) {
2926                         brelse(bh);
2927                         return false;
2928                 }
2929                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2930         }
2931         brelse(bh);
2932         return true;
2933 }
2934
2935 /*
2936  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2937  * such inodes, starting at the superblock, in case we crash before the
2938  * file is closed/deleted, or in case the inode truncate spans multiple
2939  * transactions and the last transaction is not recovered after a crash.
2940  *
2941  * At filesystem recovery time, we walk this list deleting unlinked
2942  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2943  *
2944  * Orphan list manipulation functions must be called under i_mutex unless
2945  * we are just creating the inode or deleting it.
2946  */
2947 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2948 {
2949         struct super_block *sb = inode->i_sb;
2950         struct ext4_sb_info *sbi = EXT4_SB(sb);
2951         struct ext4_iloc iloc;
2952         int err = 0, rc;
2953         bool dirty = false;
2954
2955         if (!sbi->s_journal || is_bad_inode(inode))
2956                 return 0;
2957
2958         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2959                      !inode_is_locked(inode));
2960         /*
2961          * Exit early if inode already is on orphan list. This is a big speedup
2962          * since we don't have to contend on the global s_orphan_lock.
2963          */
2964         if (!list_empty(&EXT4_I(inode)->i_orphan))
2965                 return 0;
2966
2967         /*
2968          * Orphan handling is only valid for files with data blocks
2969          * being truncated, or files being unlinked. Note that we either
2970          * hold i_mutex, or the inode can not be referenced from outside,
2971          * so i_nlink should not be bumped due to race
2972          */
2973         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2974                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2975
2976         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2977         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2978         if (err)
2979                 goto out;
2980
2981         err = ext4_reserve_inode_write(handle, inode, &iloc);
2982         if (err)
2983                 goto out;
2984
2985         mutex_lock(&sbi->s_orphan_lock);
2986         /*
2987          * Due to previous errors inode may be already a part of on-disk
2988          * orphan list. If so skip on-disk list modification.
2989          */
2990         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2991             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2992                 /* Insert this inode at the head of the on-disk orphan list */
2993                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2994                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2995                 dirty = true;
2996         }
2997         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2998         mutex_unlock(&sbi->s_orphan_lock);
2999
3000         if (dirty) {
3001                 err = ext4_handle_dirty_super(handle, sb);
3002                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3003                 if (!err)
3004                         err = rc;
3005                 if (err) {
3006                         /*
3007                          * We have to remove inode from in-memory list if
3008                          * addition to on disk orphan list failed. Stray orphan
3009                          * list entries can cause panics at unmount time.
3010                          */
3011                         mutex_lock(&sbi->s_orphan_lock);
3012                         list_del_init(&EXT4_I(inode)->i_orphan);
3013                         mutex_unlock(&sbi->s_orphan_lock);
3014                 }
3015         } else
3016                 brelse(iloc.bh);
3017
3018         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3019         jbd_debug(4, "orphan inode %lu will point to %d\n",
3020                         inode->i_ino, NEXT_ORPHAN(inode));
3021 out:
3022         ext4_std_error(sb, err);
3023         return err;
3024 }
3025
3026 /*
3027  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3028  * of such inodes stored on disk, because it is finally being cleaned up.
3029  */
3030 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3031 {
3032         struct list_head *prev;
3033         struct ext4_inode_info *ei = EXT4_I(inode);
3034         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3035         __u32 ino_next;
3036         struct ext4_iloc iloc;
3037         int err = 0;
3038
3039         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3040                 return 0;
3041
3042         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3043                      !inode_is_locked(inode));
3044         /* Do this quick check before taking global s_orphan_lock. */
3045         if (list_empty(&ei->i_orphan))
3046                 return 0;
3047
3048         if (handle) {
3049                 /* Grab inode buffer early before taking global s_orphan_lock */
3050                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3051         }
3052
3053         mutex_lock(&sbi->s_orphan_lock);
3054         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3055
3056         prev = ei->i_orphan.prev;
3057         list_del_init(&ei->i_orphan);
3058
3059         /* If we're on an error path, we may not have a valid
3060          * transaction handle with which to update the orphan list on
3061          * disk, but we still need to remove the inode from the linked
3062          * list in memory. */
3063         if (!handle || err) {
3064                 mutex_unlock(&sbi->s_orphan_lock);
3065                 goto out_err;
3066         }
3067
3068         ino_next = NEXT_ORPHAN(inode);
3069         if (prev == &sbi->s_orphan) {
3070                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3071                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3072                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3073                 if (err) {
3074                         mutex_unlock(&sbi->s_orphan_lock);
3075                         goto out_brelse;
3076                 }
3077                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3078                 mutex_unlock(&sbi->s_orphan_lock);
3079                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3080         } else {
3081                 struct ext4_iloc iloc2;
3082                 struct inode *i_prev =
3083                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3084
3085                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3086                           i_prev->i_ino, ino_next);
3087                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3088                 if (err) {
3089                         mutex_unlock(&sbi->s_orphan_lock);
3090                         goto out_brelse;
3091                 }
3092                 NEXT_ORPHAN(i_prev) = ino_next;
3093                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3094                 mutex_unlock(&sbi->s_orphan_lock);
3095         }
3096         if (err)
3097                 goto out_brelse;
3098         NEXT_ORPHAN(inode) = 0;
3099         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3100 out_err:
3101         ext4_std_error(inode->i_sb, err);
3102         return err;
3103
3104 out_brelse:
3105         brelse(iloc.bh);
3106         goto out_err;
3107 }
3108
3109 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3110 {
3111         int retval;
3112         struct inode *inode;
3113         struct buffer_head *bh;
3114         struct ext4_dir_entry_2 *de;
3115         handle_t *handle = NULL;
3116
3117         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3118                 return -EIO;
3119
3120         /* Initialize quotas before so that eventual writes go in
3121          * separate transaction */
3122         retval = dquot_initialize(dir);
3123         if (retval)
3124                 return retval;
3125         retval = dquot_initialize(d_inode(dentry));
3126         if (retval)
3127                 return retval;
3128
3129         retval = -ENOENT;
3130         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3131         if (IS_ERR(bh))
3132                 return PTR_ERR(bh);
3133         if (!bh)
3134                 goto end_rmdir;
3135
3136         inode = d_inode(dentry);
3137
3138         retval = -EFSCORRUPTED;
3139         if (le32_to_cpu(de->inode) != inode->i_ino)
3140                 goto end_rmdir;
3141
3142         retval = -ENOTEMPTY;
3143         if (!ext4_empty_dir(inode))
3144                 goto end_rmdir;
3145
3146         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3147                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3148         if (IS_ERR(handle)) {
3149                 retval = PTR_ERR(handle);
3150                 handle = NULL;
3151                 goto end_rmdir;
3152         }
3153
3154         if (IS_DIRSYNC(dir))
3155                 ext4_handle_sync(handle);
3156
3157         retval = ext4_delete_entry(handle, dir, de, bh);
3158         if (retval)
3159                 goto end_rmdir;
3160         if (!EXT4_DIR_LINK_EMPTY(inode))
3161                 ext4_warning_inode(inode,
3162                              "empty directory '%.*s' has too many links (%u)",
3163                              dentry->d_name.len, dentry->d_name.name,
3164                              inode->i_nlink);
3165         inode_inc_iversion(inode);
3166         clear_nlink(inode);
3167         /* There's no need to set i_disksize: the fact that i_nlink is
3168          * zero will ensure that the right thing happens during any
3169          * recovery. */
3170         inode->i_size = 0;
3171         ext4_orphan_add(handle, inode);
3172         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3173         retval = ext4_mark_inode_dirty(handle, inode);
3174         if (retval)
3175                 goto end_rmdir;
3176         ext4_dec_count(dir);
3177         ext4_update_dx_flag(dir);
3178         ext4_fc_track_unlink(inode, dentry);
3179         retval = ext4_mark_inode_dirty(handle, dir);
3180
3181 #ifdef CONFIG_UNICODE
3182         /* VFS negative dentries are incompatible with Encoding and
3183          * Case-insensitiveness. Eventually we'll want avoid
3184          * invalidating the dentries here, alongside with returning the
3185          * negative dentries at ext4_lookup(), when it is better
3186          * supported by the VFS for the CI case.
3187          */
3188         if (IS_CASEFOLDED(dir))
3189                 d_invalidate(dentry);
3190 #endif
3191
3192 end_rmdir:
3193         brelse(bh);
3194         if (handle)
3195                 ext4_journal_stop(handle);
3196         return retval;
3197 }
3198
3199 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3200                   struct inode *inode)
3201 {
3202         int retval = -ENOENT;
3203         struct buffer_head *bh;
3204         struct ext4_dir_entry_2 *de;
3205         handle_t *handle = NULL;
3206         int skip_remove_dentry = 0;
3207
3208         bh = ext4_find_entry(dir, d_name, &de, NULL);
3209         if (IS_ERR(bh))
3210                 return PTR_ERR(bh);
3211
3212         if (!bh)
3213                 return -ENOENT;
3214
3215         if (le32_to_cpu(de->inode) != inode->i_ino) {
3216                 /*
3217                  * It's okay if we find dont find dentry which matches
3218                  * the inode. That's because it might have gotten
3219                  * renamed to a different inode number
3220                  */
3221                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3222                         skip_remove_dentry = 1;
3223                 else
3224                         goto out_bh;
3225         }
3226
3227         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3228                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3229         if (IS_ERR(handle)) {
3230                 retval = PTR_ERR(handle);
3231                 goto out_bh;
3232         }
3233
3234         if (IS_DIRSYNC(dir))
3235                 ext4_handle_sync(handle);
3236
3237         if (!skip_remove_dentry) {
3238                 retval = ext4_delete_entry(handle, dir, de, bh);
3239                 if (retval)
3240                         goto out_handle;
3241                 dir->i_ctime = dir->i_mtime = current_time(dir);
3242                 ext4_update_dx_flag(dir);
3243                 retval = ext4_mark_inode_dirty(handle, dir);
3244                 if (retval)
3245                         goto out_handle;
3246         } else {
3247                 retval = 0;
3248         }
3249         if (inode->i_nlink == 0)
3250                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3251                                    d_name->len, d_name->name);
3252         else
3253                 drop_nlink(inode);
3254         if (!inode->i_nlink)
3255                 ext4_orphan_add(handle, inode);
3256         inode->i_ctime = current_time(inode);
3257         retval = ext4_mark_inode_dirty(handle, inode);
3258
3259 out_handle:
3260         ext4_journal_stop(handle);
3261 out_bh:
3262         brelse(bh);
3263         return retval;
3264 }
3265
3266 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3267 {
3268         int retval;
3269
3270         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3271                 return -EIO;
3272
3273         trace_ext4_unlink_enter(dir, dentry);
3274         /*
3275          * Initialize quotas before so that eventual writes go
3276          * in separate transaction
3277          */
3278         retval = dquot_initialize(dir);
3279         if (retval)
3280                 goto out_trace;
3281         retval = dquot_initialize(d_inode(dentry));
3282         if (retval)
3283                 goto out_trace;
3284
3285         retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry));
3286         if (!retval)
3287                 ext4_fc_track_unlink(d_inode(dentry), dentry);
3288 #ifdef CONFIG_UNICODE
3289         /* VFS negative dentries are incompatible with Encoding and
3290          * Case-insensitiveness. Eventually we'll want avoid
3291          * invalidating the dentries here, alongside with returning the
3292          * negative dentries at ext4_lookup(), when it is  better
3293          * supported by the VFS for the CI case.
3294          */
3295         if (IS_CASEFOLDED(dir))
3296                 d_invalidate(dentry);
3297 #endif
3298
3299 out_trace:
3300         trace_ext4_unlink_exit(dentry, retval);
3301         return retval;
3302 }
3303
3304 static int ext4_symlink(struct inode *dir,
3305                         struct dentry *dentry, const char *symname)
3306 {
3307         handle_t *handle;
3308         struct inode *inode;
3309         int err, len = strlen(symname);
3310         int credits;
3311         struct fscrypt_str disk_link;
3312
3313         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3314                 return -EIO;
3315
3316         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3317                                       &disk_link);
3318         if (err)
3319                 return err;
3320
3321         err = dquot_initialize(dir);
3322         if (err)
3323                 return err;
3324
3325         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3326                 /*
3327                  * For non-fast symlinks, we just allocate inode and put it on
3328                  * orphan list in the first transaction => we need bitmap,
3329                  * group descriptor, sb, inode block, quota blocks, and
3330                  * possibly selinux xattr blocks.
3331                  */
3332                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3333                           EXT4_XATTR_TRANS_BLOCKS;
3334         } else {
3335                 /*
3336                  * Fast symlink. We have to add entry to directory
3337                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3338                  * allocate new inode (bitmap, group descriptor, inode block,
3339                  * quota blocks, sb is already counted in previous macros).
3340                  */
3341                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3342                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3343         }
3344
3345         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3346                                             &dentry->d_name, 0, NULL,
3347                                             EXT4_HT_DIR, credits);
3348         handle = ext4_journal_current_handle();
3349         if (IS_ERR(inode)) {
3350                 if (handle)
3351                         ext4_journal_stop(handle);
3352                 return PTR_ERR(inode);
3353         }
3354
3355         if (IS_ENCRYPTED(inode)) {
3356                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3357                 if (err)
3358                         goto err_drop_inode;
3359                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3360         }
3361
3362         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3363                 if (!IS_ENCRYPTED(inode))
3364                         inode->i_op = &ext4_symlink_inode_operations;
3365                 inode_nohighmem(inode);
3366                 ext4_set_aops(inode);
3367                 /*
3368                  * We cannot call page_symlink() with transaction started
3369                  * because it calls into ext4_write_begin() which can wait
3370                  * for transaction commit if we are running out of space
3371                  * and thus we deadlock. So we have to stop transaction now
3372                  * and restart it when symlink contents is written.
3373                  *
3374                  * To keep fs consistent in case of crash, we have to put inode
3375                  * to orphan list in the mean time.
3376                  */
3377                 drop_nlink(inode);
3378                 err = ext4_orphan_add(handle, inode);
3379                 if (handle)
3380                         ext4_journal_stop(handle);
3381                 handle = NULL;
3382                 if (err)
3383                         goto err_drop_inode;
3384                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3385                 if (err)
3386                         goto err_drop_inode;
3387                 /*
3388                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3389                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3390                  */
3391                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3392                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3393                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3394                 if (IS_ERR(handle)) {
3395                         err = PTR_ERR(handle);
3396                         handle = NULL;
3397                         goto err_drop_inode;
3398                 }
3399                 set_nlink(inode, 1);
3400                 err = ext4_orphan_del(handle, inode);
3401                 if (err)
3402                         goto err_drop_inode;
3403         } else {
3404                 /* clear the extent format for fast symlink */
3405                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3406                 if (!IS_ENCRYPTED(inode)) {
3407                         inode->i_op = &ext4_fast_symlink_inode_operations;
3408                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3409                 }
3410                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3411                        disk_link.len);
3412                 inode->i_size = disk_link.len - 1;
3413         }
3414         EXT4_I(inode)->i_disksize = inode->i_size;
3415         err = ext4_add_nondir(handle, dentry, &inode);
3416         if (handle)
3417                 ext4_journal_stop(handle);
3418         if (inode)
3419                 iput(inode);
3420         goto out_free_encrypted_link;
3421
3422 err_drop_inode:
3423         if (handle)
3424                 ext4_journal_stop(handle);
3425         clear_nlink(inode);
3426         unlock_new_inode(inode);
3427         iput(inode);
3428 out_free_encrypted_link:
3429         if (disk_link.name != (unsigned char *)symname)
3430                 kfree(disk_link.name);
3431         return err;
3432 }
3433
3434 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3435 {
3436         handle_t *handle;
3437         int err, retries = 0;
3438 retry:
3439         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3440                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3441                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3442         if (IS_ERR(handle))
3443                 return PTR_ERR(handle);
3444
3445         if (IS_DIRSYNC(dir))
3446                 ext4_handle_sync(handle);
3447
3448         inode->i_ctime = current_time(inode);
3449         ext4_inc_count(inode);
3450         ihold(inode);
3451
3452         err = ext4_add_entry(handle, dentry, inode);
3453         if (!err) {
3454                 ext4_fc_track_link(inode, dentry);
3455                 err = ext4_mark_inode_dirty(handle, inode);
3456                 /* this can happen only for tmpfile being
3457                  * linked the first time
3458                  */
3459                 if (inode->i_nlink == 1)
3460                         ext4_orphan_del(handle, inode);
3461                 d_instantiate(dentry, inode);
3462         } else {
3463                 drop_nlink(inode);
3464                 iput(inode);
3465         }
3466         ext4_journal_stop(handle);
3467         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3468                 goto retry;
3469         return err;
3470 }
3471
3472 static int ext4_link(struct dentry *old_dentry,
3473                      struct inode *dir, struct dentry *dentry)
3474 {
3475         struct inode *inode = d_inode(old_dentry);
3476         int err;
3477
3478         if (inode->i_nlink >= EXT4_LINK_MAX)
3479                 return -EMLINK;
3480
3481         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3482         if (err)
3483                 return err;
3484
3485         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3486             (!projid_eq(EXT4_I(dir)->i_projid,
3487                         EXT4_I(old_dentry->d_inode)->i_projid)))
3488                 return -EXDEV;
3489
3490         err = dquot_initialize(dir);
3491         if (err)
3492                 return err;
3493         return __ext4_link(dir, inode, dentry);
3494 }
3495
3496 /*
3497  * Try to find buffer head where contains the parent block.
3498  * It should be the inode block if it is inlined or the 1st block
3499  * if it is a normal dir.
3500  */
3501 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3502                                         struct inode *inode,
3503                                         int *retval,
3504                                         struct ext4_dir_entry_2 **parent_de,
3505                                         int *inlined)
3506 {
3507         struct buffer_head *bh;
3508
3509         if (!ext4_has_inline_data(inode)) {
3510                 /* The first directory block must not be a hole, so
3511                  * treat it as DIRENT_HTREE
3512                  */
3513                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3514                 if (IS_ERR(bh)) {
3515                         *retval = PTR_ERR(bh);
3516                         return NULL;
3517                 }
3518                 *parent_de = ext4_next_entry(
3519                                         (struct ext4_dir_entry_2 *)bh->b_data,
3520                                         inode->i_sb->s_blocksize);
3521                 return bh;
3522         }
3523
3524         *inlined = 1;
3525         return ext4_get_first_inline_block(inode, parent_de, retval);
3526 }
3527
3528 struct ext4_renament {
3529         struct inode *dir;
3530         struct dentry *dentry;
3531         struct inode *inode;
3532         bool is_dir;
3533         int dir_nlink_delta;
3534
3535         /* entry for "dentry" */
3536         struct buffer_head *bh;
3537         struct ext4_dir_entry_2 *de;
3538         int inlined;
3539
3540         /* entry for ".." in inode if it's a directory */
3541         struct buffer_head *dir_bh;
3542         struct ext4_dir_entry_2 *parent_de;
3543         int dir_inlined;
3544 };
3545
3546 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3547 {
3548         int retval;
3549
3550         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3551                                               &retval, &ent->parent_de,
3552                                               &ent->dir_inlined);
3553         if (!ent->dir_bh)
3554                 return retval;
3555         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3556                 return -EFSCORRUPTED;
3557         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3558         return ext4_journal_get_write_access(handle, ent->dir_bh);
3559 }
3560
3561 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3562                                   unsigned dir_ino)
3563 {
3564         int retval;
3565
3566         ent->parent_de->inode = cpu_to_le32(dir_ino);
3567         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3568         if (!ent->dir_inlined) {
3569                 if (is_dx(ent->inode)) {
3570                         retval = ext4_handle_dirty_dx_node(handle,
3571                                                            ent->inode,
3572                                                            ent->dir_bh);
3573                 } else {
3574                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3575                                                             ent->dir_bh);
3576                 }
3577         } else {
3578                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3579         }
3580         if (retval) {
3581                 ext4_std_error(ent->dir->i_sb, retval);
3582                 return retval;
3583         }
3584         return 0;
3585 }
3586
3587 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3588                        unsigned ino, unsigned file_type)
3589 {
3590         int retval, retval2;
3591
3592         BUFFER_TRACE(ent->bh, "get write access");
3593         retval = ext4_journal_get_write_access(handle, ent->bh);
3594         if (retval)
3595                 return retval;
3596         ent->de->inode = cpu_to_le32(ino);
3597         if (ext4_has_feature_filetype(ent->dir->i_sb))
3598                 ent->de->file_type = file_type;
3599         inode_inc_iversion(ent->dir);
3600         ent->dir->i_ctime = ent->dir->i_mtime =
3601                 current_time(ent->dir);
3602         retval = ext4_mark_inode_dirty(handle, ent->dir);
3603         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3604         if (!ent->inlined) {
3605                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3606                 if (unlikely(retval2)) {
3607                         ext4_std_error(ent->dir->i_sb, retval2);
3608                         return retval2;
3609                 }
3610         }
3611         brelse(ent->bh);
3612         ent->bh = NULL;
3613
3614         return retval;
3615 }
3616
3617 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3618                                   const struct qstr *d_name)
3619 {
3620         int retval = -ENOENT;
3621         struct buffer_head *bh;
3622         struct ext4_dir_entry_2 *de;
3623
3624         bh = ext4_find_entry(dir, d_name, &de, NULL);
3625         if (IS_ERR(bh))
3626                 return PTR_ERR(bh);
3627         if (bh) {
3628                 retval = ext4_delete_entry(handle, dir, de, bh);
3629                 brelse(bh);
3630         }
3631         return retval;
3632 }
3633
3634 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3635                                int force_reread)
3636 {
3637         int retval;
3638         /*
3639          * ent->de could have moved from under us during htree split, so make
3640          * sure that we are deleting the right entry.  We might also be pointing
3641          * to a stale entry in the unused part of ent->bh so just checking inum
3642          * and the name isn't enough.
3643          */
3644         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3645             ent->de->name_len != ent->dentry->d_name.len ||
3646             strncmp(ent->de->name, ent->dentry->d_name.name,
3647                     ent->de->name_len) ||
3648             force_reread) {
3649                 retval = ext4_find_delete_entry(handle, ent->dir,
3650                                                 &ent->dentry->d_name);
3651         } else {
3652                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3653                 if (retval == -ENOENT) {
3654                         retval = ext4_find_delete_entry(handle, ent->dir,
3655                                                         &ent->dentry->d_name);
3656                 }
3657         }
3658
3659         if (retval) {
3660                 ext4_warning_inode(ent->dir,
3661                                    "Deleting old file: nlink %d, error=%d",
3662                                    ent->dir->i_nlink, retval);
3663         }
3664 }
3665
3666 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3667 {
3668         if (ent->dir_nlink_delta) {
3669                 if (ent->dir_nlink_delta == -1)
3670                         ext4_dec_count(ent->dir);
3671                 else
3672                         ext4_inc_count(ent->dir);
3673                 ext4_mark_inode_dirty(handle, ent->dir);
3674         }
3675 }
3676
3677 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3678                                               int credits, handle_t **h)
3679 {
3680         struct inode *wh;
3681         handle_t *handle;
3682         int retries = 0;
3683
3684         /*
3685          * for inode block, sb block, group summaries,
3686          * and inode bitmap
3687          */
3688         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3689                     EXT4_XATTR_TRANS_BLOCKS + 4);
3690 retry:
3691         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3692                                          &ent->dentry->d_name, 0, NULL,
3693                                          EXT4_HT_DIR, credits);
3694
3695         handle = ext4_journal_current_handle();
3696         if (IS_ERR(wh)) {
3697                 if (handle)
3698                         ext4_journal_stop(handle);
3699                 if (PTR_ERR(wh) == -ENOSPC &&
3700                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3701                         goto retry;
3702         } else {
3703                 *h = handle;
3704                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3705                 wh->i_op = &ext4_special_inode_operations;
3706         }
3707         return wh;
3708 }
3709
3710 /*
3711  * Anybody can rename anything with this: the permission checks are left to the
3712  * higher-level routines.
3713  *
3714  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3715  * while new_{dentry,inode) refers to the destination dentry/inode
3716  * This comes from rename(const char *oldpath, const char *newpath)
3717  */
3718 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3719                        struct inode *new_dir, struct dentry *new_dentry,
3720                        unsigned int flags)
3721 {
3722         handle_t *handle = NULL;
3723         struct ext4_renament old = {
3724                 .dir = old_dir,
3725                 .dentry = old_dentry,
3726                 .inode = d_inode(old_dentry),
3727         };
3728         struct ext4_renament new = {
3729                 .dir = new_dir,
3730                 .dentry = new_dentry,
3731                 .inode = d_inode(new_dentry),
3732         };
3733         int force_reread;
3734         int retval;
3735         struct inode *whiteout = NULL;
3736         int credits;
3737         u8 old_file_type;
3738
3739         if (new.inode && new.inode->i_nlink == 0) {
3740                 EXT4_ERROR_INODE(new.inode,
3741                                  "target of rename is already freed");
3742                 return -EFSCORRUPTED;
3743         }
3744
3745         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3746             (!projid_eq(EXT4_I(new_dir)->i_projid,
3747                         EXT4_I(old_dentry->d_inode)->i_projid)))
3748                 return -EXDEV;
3749
3750         retval = dquot_initialize(old.dir);
3751         if (retval)
3752                 return retval;
3753         retval = dquot_initialize(new.dir);
3754         if (retval)
3755                 return retval;
3756
3757         /* Initialize quotas before so that eventual writes go
3758          * in separate transaction */
3759         if (new.inode) {
3760                 retval = dquot_initialize(new.inode);
3761                 if (retval)
3762                         return retval;
3763         }
3764
3765         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3766         if (IS_ERR(old.bh))
3767                 return PTR_ERR(old.bh);
3768         /*
3769          *  Check for inode number is _not_ due to possible IO errors.
3770          *  We might rmdir the source, keep it as pwd of some process
3771          *  and merrily kill the link to whatever was created under the
3772          *  same name. Goodbye sticky bit ;-<
3773          */
3774         retval = -ENOENT;
3775         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3776                 goto end_rename;
3777
3778         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3779                                  &new.de, &new.inlined);
3780         if (IS_ERR(new.bh)) {
3781                 retval = PTR_ERR(new.bh);
3782                 new.bh = NULL;
3783                 goto end_rename;
3784         }
3785         if (new.bh) {
3786                 if (!new.inode) {
3787                         brelse(new.bh);
3788                         new.bh = NULL;
3789                 }
3790         }
3791         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3792                 ext4_alloc_da_blocks(old.inode);
3793
3794         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3795                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3796         if (!(flags & RENAME_WHITEOUT)) {
3797                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3798                 if (IS_ERR(handle)) {
3799                         retval = PTR_ERR(handle);
3800                         handle = NULL;
3801                         goto end_rename;
3802                 }
3803         } else {
3804                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3805                 if (IS_ERR(whiteout)) {
3806                         retval = PTR_ERR(whiteout);
3807                         whiteout = NULL;
3808                         goto end_rename;
3809                 }
3810         }
3811
3812         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3813                 ext4_handle_sync(handle);
3814
3815         if (S_ISDIR(old.inode->i_mode)) {
3816                 if (new.inode) {
3817                         retval = -ENOTEMPTY;
3818                         if (!ext4_empty_dir(new.inode))
3819                                 goto end_rename;
3820                 } else {
3821                         retval = -EMLINK;
3822                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3823                                 goto end_rename;
3824                 }
3825                 retval = ext4_rename_dir_prepare(handle, &old);
3826                 if (retval)
3827                         goto end_rename;
3828         }
3829         /*
3830          * If we're renaming a file within an inline_data dir and adding or
3831          * setting the new dirent causes a conversion from inline_data to
3832          * extents/blockmap, we need to force the dirent delete code to
3833          * re-read the directory, or else we end up trying to delete a dirent
3834          * from what is now the extent tree root (or a block map).
3835          */
3836         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3837                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3838
3839         old_file_type = old.de->file_type;
3840         if (whiteout) {
3841                 /*
3842                  * Do this before adding a new entry, so the old entry is sure
3843                  * to be still pointing to the valid old entry.
3844                  */
3845                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3846                                      EXT4_FT_CHRDEV);
3847                 if (retval)
3848                         goto end_rename;
3849                 retval = ext4_mark_inode_dirty(handle, whiteout);
3850                 if (unlikely(retval))
3851                         goto end_rename;
3852         }
3853         if (!new.bh) {
3854                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3855                 if (retval)
3856                         goto end_rename;
3857         } else {
3858                 retval = ext4_setent(handle, &new,
3859                                      old.inode->i_ino, old_file_type);
3860                 if (retval)
3861                         goto end_rename;
3862         }
3863         if (force_reread)
3864                 force_reread = !ext4_test_inode_flag(new.dir,
3865                                                      EXT4_INODE_INLINE_DATA);
3866
3867         /*
3868          * Like most other Unix systems, set the ctime for inodes on a
3869          * rename.
3870          */
3871         old.inode->i_ctime = current_time(old.inode);
3872         retval = ext4_mark_inode_dirty(handle, old.inode);
3873         if (unlikely(retval))
3874                 goto end_rename;
3875
3876         if (!whiteout) {
3877                 /*
3878                  * ok, that's it
3879                  */
3880                 ext4_rename_delete(handle, &old, force_reread);
3881         }
3882
3883         if (new.inode) {
3884                 ext4_dec_count(new.inode);
3885                 new.inode->i_ctime = current_time(new.inode);
3886         }
3887         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3888         ext4_update_dx_flag(old.dir);
3889         if (old.dir_bh) {
3890                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3891                 if (retval)
3892                         goto end_rename;
3893
3894                 ext4_dec_count(old.dir);
3895                 if (new.inode) {
3896                         /* checked ext4_empty_dir above, can't have another
3897                          * parent, ext4_dec_count() won't work for many-linked
3898                          * dirs */
3899                         clear_nlink(new.inode);
3900                 } else {
3901                         ext4_inc_count(new.dir);
3902                         ext4_update_dx_flag(new.dir);
3903                         retval = ext4_mark_inode_dirty(handle, new.dir);
3904                         if (unlikely(retval))
3905                                 goto end_rename;
3906                 }
3907         }
3908         retval = ext4_mark_inode_dirty(handle, old.dir);
3909         if (unlikely(retval))
3910                 goto end_rename;
3911
3912         if (S_ISDIR(old.inode->i_mode)) {
3913                 /*
3914                  * We disable fast commits here that's because the
3915                  * replay code is not yet capable of changing dot dot
3916                  * dirents in directories.
3917                  */
3918                 ext4_fc_mark_ineligible(old.inode->i_sb,
3919                         EXT4_FC_REASON_RENAME_DIR);
3920         } else {
3921                 if (new.inode)
3922                         ext4_fc_track_unlink(new.inode, new.dentry);
3923                 ext4_fc_track_link(old.inode, new.dentry);
3924                 ext4_fc_track_unlink(old.inode, old.dentry);
3925         }
3926
3927         if (new.inode) {
3928                 retval = ext4_mark_inode_dirty(handle, new.inode);
3929                 if (unlikely(retval))
3930                         goto end_rename;
3931                 if (!new.inode->i_nlink)
3932                         ext4_orphan_add(handle, new.inode);
3933         }
3934         retval = 0;
3935
3936 end_rename:
3937         brelse(old.dir_bh);
3938         brelse(old.bh);
3939         brelse(new.bh);
3940         if (whiteout) {
3941                 if (retval)
3942                         drop_nlink(whiteout);
3943                 unlock_new_inode(whiteout);
3944                 iput(whiteout);
3945         }
3946         if (handle)
3947                 ext4_journal_stop(handle);
3948         return retval;
3949 }
3950
3951 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3952                              struct inode *new_dir, struct dentry *new_dentry)
3953 {
3954         handle_t *handle = NULL;
3955         struct ext4_renament old = {
3956                 .dir = old_dir,
3957                 .dentry = old_dentry,
3958                 .inode = d_inode(old_dentry),
3959         };
3960         struct ext4_renament new = {
3961                 .dir = new_dir,
3962                 .dentry = new_dentry,
3963                 .inode = d_inode(new_dentry),
3964         };
3965         u8 new_file_type;
3966         int retval;
3967         struct timespec64 ctime;
3968
3969         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3970              !projid_eq(EXT4_I(new_dir)->i_projid,
3971                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3972             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3973              !projid_eq(EXT4_I(old_dir)->i_projid,
3974                         EXT4_I(new_dentry->d_inode)->i_projid)))
3975                 return -EXDEV;
3976
3977         retval = dquot_initialize(old.dir);
3978         if (retval)
3979                 return retval;
3980         retval = dquot_initialize(new.dir);
3981         if (retval)
3982                 return retval;
3983
3984         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3985                                  &old.de, &old.inlined);
3986         if (IS_ERR(old.bh))
3987                 return PTR_ERR(old.bh);
3988         /*
3989          *  Check for inode number is _not_ due to possible IO errors.
3990          *  We might rmdir the source, keep it as pwd of some process
3991          *  and merrily kill the link to whatever was created under the
3992          *  same name. Goodbye sticky bit ;-<
3993          */
3994         retval = -ENOENT;
3995         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3996                 goto end_rename;
3997
3998         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3999                                  &new.de, &new.inlined);
4000         if (IS_ERR(new.bh)) {
4001                 retval = PTR_ERR(new.bh);
4002                 new.bh = NULL;
4003                 goto end_rename;
4004         }
4005
4006         /* RENAME_EXCHANGE case: old *and* new must both exist */
4007         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4008                 goto end_rename;
4009
4010         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4011                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4012                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4013         if (IS_ERR(handle)) {
4014                 retval = PTR_ERR(handle);
4015                 handle = NULL;
4016                 goto end_rename;
4017         }
4018
4019         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4020                 ext4_handle_sync(handle);
4021
4022         if (S_ISDIR(old.inode->i_mode)) {
4023                 old.is_dir = true;
4024                 retval = ext4_rename_dir_prepare(handle, &old);
4025                 if (retval)
4026                         goto end_rename;
4027         }
4028         if (S_ISDIR(new.inode->i_mode)) {
4029                 new.is_dir = true;
4030                 retval = ext4_rename_dir_prepare(handle, &new);
4031                 if (retval)
4032                         goto end_rename;
4033         }
4034
4035         /*
4036          * Other than the special case of overwriting a directory, parents'
4037          * nlink only needs to be modified if this is a cross directory rename.
4038          */
4039         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4040                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4041                 new.dir_nlink_delta = -old.dir_nlink_delta;
4042                 retval = -EMLINK;
4043                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4044                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4045                         goto end_rename;
4046         }
4047
4048         new_file_type = new.de->file_type;
4049         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4050         if (retval)
4051                 goto end_rename;
4052
4053         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4054         if (retval)
4055                 goto end_rename;
4056
4057         /*
4058          * Like most other Unix systems, set the ctime for inodes on a
4059          * rename.
4060          */
4061         ctime = current_time(old.inode);
4062         old.inode->i_ctime = ctime;
4063         new.inode->i_ctime = ctime;
4064         retval = ext4_mark_inode_dirty(handle, old.inode);
4065         if (unlikely(retval))
4066                 goto end_rename;
4067         retval = ext4_mark_inode_dirty(handle, new.inode);
4068         if (unlikely(retval))
4069                 goto end_rename;
4070         ext4_fc_mark_ineligible(new.inode->i_sb,
4071                                 EXT4_FC_REASON_CROSS_RENAME);
4072         if (old.dir_bh) {
4073                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4074                 if (retval)
4075                         goto end_rename;
4076         }
4077         if (new.dir_bh) {
4078                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4079                 if (retval)
4080                         goto end_rename;
4081         }
4082         ext4_update_dir_count(handle, &old);
4083         ext4_update_dir_count(handle, &new);
4084         retval = 0;
4085
4086 end_rename:
4087         brelse(old.dir_bh);
4088         brelse(new.dir_bh);
4089         brelse(old.bh);
4090         brelse(new.bh);
4091         if (handle)
4092                 ext4_journal_stop(handle);
4093         return retval;
4094 }
4095
4096 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4097                         struct inode *new_dir, struct dentry *new_dentry,
4098                         unsigned int flags)
4099 {
4100         int err;
4101
4102         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4103                 return -EIO;
4104
4105         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4106                 return -EINVAL;
4107
4108         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4109                                      flags);
4110         if (err)
4111                 return err;
4112
4113         if (flags & RENAME_EXCHANGE) {
4114                 return ext4_cross_rename(old_dir, old_dentry,
4115                                          new_dir, new_dentry);
4116         }
4117
4118         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4119 }
4120
4121 /*
4122  * directories can handle most operations...
4123  */
4124 const struct inode_operations ext4_dir_inode_operations = {
4125         .create         = ext4_create,
4126         .lookup         = ext4_lookup,
4127         .link           = ext4_link,
4128         .unlink         = ext4_unlink,
4129         .symlink        = ext4_symlink,
4130         .mkdir          = ext4_mkdir,
4131         .rmdir          = ext4_rmdir,
4132         .mknod          = ext4_mknod,
4133         .tmpfile        = ext4_tmpfile,
4134         .rename         = ext4_rename2,
4135         .setattr        = ext4_setattr,
4136         .getattr        = ext4_getattr,
4137         .listxattr      = ext4_listxattr,
4138         .get_acl        = ext4_get_acl,
4139         .set_acl        = ext4_set_acl,
4140         .fiemap         = ext4_fiemap,
4141 };
4142
4143 const struct inode_operations ext4_special_inode_operations = {
4144         .setattr        = ext4_setattr,
4145         .getattr        = ext4_getattr,
4146         .listxattr      = ext4_listxattr,
4147         .get_acl        = ext4_get_acl,
4148         .set_acl        = ext4_set_acl,
4149 };