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 ext4_map_blocks map;
  58         struct buffer_head *bh;
  59         int err;
  60
  61         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  62                      ((inode->i_size >> 10) >=
  63                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  64                 return ERR_PTR(-ENOSPC);
  65
  66         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  67         map.m_lblk = *block;
  68         map.m_len = 1;
  69
  70         /*
  71          * We're appending new directory block. Make sure the block is not
  72          * allocated yet, otherwise we will end up corrupting the
  73          * directory.
  74          */
  75         err = ext4_map_blocks(NULL, inode, &map, 0);
  76         if (err < 0)
  77                 return ERR_PTR(err);
  78         if (err) {
  79                 EXT4_ERROR_INODE(inode, "Logical block already allocated");
  80                 return ERR_PTR(-EFSCORRUPTED);
  81         }
  82
  83         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  84         if (IS_ERR(bh))
  85                 return bh;
  86         inode->i_size += inode->i_sb->s_blocksize;
  87         EXT4_I(inode)->i_disksize = inode->i_size;
  88         err = ext4_mark_inode_dirty(handle, inode);
  89         if (err)
  90                 goto out;
  91         BUFFER_TRACE(bh, "get_write_access");
  92         err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
  93                                             EXT4_JTR_NONE);
  94         if (err)
  95                 goto out;
  96         return bh;
  97
  98 out:
  99         brelse(bh);
 100         ext4_std_error(inode->i_sb, err);
 101         return ERR_PTR(err);
 102 }
 103
 104 static int ext4_dx_csum_verify(struct inode *inode,
 105                                struct ext4_dir_entry *dirent);
 106
 107 /*
 108  * Hints to ext4_read_dirblock regarding whether we expect a directory
 109  * block being read to be an index block, or a block containing
 110  * directory entries (and if the latter, whether it was found via a
 111  * logical block in an htree index block).  This is used to control
 112  * what sort of sanity checkinig ext4_read_dirblock() will do on the
 113  * directory block read from the storage device.  EITHER will means
 114  * the caller doesn't know what kind of directory block will be read,
 115  * so no specific verification will be done.
 116  */
 117 typedef enum {
 118         EITHER, INDEX, DIRENT, DIRENT_HTREE
 119 } dirblock_type_t;
 120
 121 #define ext4_read_dirblock(inode, block, type) \
 122         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 123
 124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 125                                                 ext4_lblk_t block,
 126                                                 dirblock_type_t type,
 127                                                 const char *func,
 128                                                 unsigned int line)
 129 {
 130         struct buffer_head *bh;
 131         struct ext4_dir_entry *dirent;
 132         int is_dx_block = 0;
 133
 134         if (block >= inode->i_size >> inode->i_blkbits) {
 135                 ext4_error_inode(inode, func, line, block,
 136                        "Attempting to read directory block (%u) that is past i_size (%llu)",
 137                        block, inode->i_size);
 138                 return ERR_PTR(-EFSCORRUPTED);
 139         }
 140
 141         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 142                 bh = ERR_PTR(-EIO);
 143         else
 144                 bh = ext4_bread(NULL, inode, block, 0);
 145         if (IS_ERR(bh)) {
 146                 __ext4_warning(inode->i_sb, func, line,
 147                                "inode #%lu: lblock %lu: comm %s: "
 148                                "error %ld reading directory block",
 149                                inode->i_ino, (unsigned long)block,
 150                                current->comm, PTR_ERR(bh));
 151
 152                 return bh;
 153         }
 154         /* The first directory block must not be a hole. */
 155         if (!bh && (type == INDEX || type == DIRENT_HTREE || block == 0)) {
 156                 ext4_error_inode(inode, func, line, block,
 157                                  "Directory hole found for htree %s block %u",
 158                                  (type == INDEX) ? "index" : "leaf", block);
 159                 return ERR_PTR(-EFSCORRUPTED);
 160         }
 161         if (!bh)
 162                 return NULL;
 163         dirent = (struct ext4_dir_entry *) bh->b_data;
 164         /* Determine whether or not we have an index block */
 165         if (is_dx(inode)) {
 166                 if (block == 0)
 167                         is_dx_block = 1;
 168                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 169                                                 inode->i_sb->s_blocksize) ==
 170                          inode->i_sb->s_blocksize)
 171                         is_dx_block = 1;
 172         }
 173         if (!is_dx_block && type == INDEX) {
 174                 ext4_error_inode(inode, func, line, block,
 175                        "directory leaf block found instead of index block");
 176                 brelse(bh);
 177                 return ERR_PTR(-EFSCORRUPTED);
 178         }
 179         if (!ext4_has_feature_metadata_csum(inode->i_sb) ||
 180             buffer_verified(bh))
 181                 return bh;
 182
 183         /*
 184          * An empty leaf block can get mistaken for a index block; for
 185          * this reason, we can only check the index checksum when the
 186          * caller is sure it should be an index block.
 187          */
 188         if (is_dx_block && type == INDEX) {
 189                 if (ext4_dx_csum_verify(inode, dirent) &&
 190                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 191                         set_buffer_verified(bh);
 192                 else {
 193                         ext4_error_inode_err(inode, func, line, block,
 194                                              EFSBADCRC,
 195                                              "Directory index failed checksum");
 196                         brelse(bh);
 197                         return ERR_PTR(-EFSBADCRC);
 198                 }
 199         }
 200         if (!is_dx_block) {
 201                 if (ext4_dirblock_csum_verify(inode, bh) &&
 202                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 203                         set_buffer_verified(bh);
 204                 else {
 205                         ext4_error_inode_err(inode, func, line, block,
 206                                              EFSBADCRC,
 207                                              "Directory block failed checksum");
 208                         brelse(bh);
 209                         return ERR_PTR(-EFSBADCRC);
 210                 }
 211         }
 212         return bh;
 213 }
 214
 215 #ifdef DX_DEBUG
 216 #define dxtrace(command) command
 217 #else
 218 #define dxtrace(command)
 219 #endif
 220
 221 struct fake_dirent
 222 {
 223         __le32 inode;
 224         __le16 rec_len;
 225         u8 name_len;
 226         u8 file_type;
 227 };
 228
 229 struct dx_countlimit
 230 {
 231         __le16 limit;
 232         __le16 count;
 233 };
 234
 235 struct dx_entry
 236 {
 237         __le32 hash;
 238         __le32 block;
 239 };
 240
 241 /*
 242  * dx_root_info is laid out so that if it should somehow get overlaid by a
 243  * dirent the two low bits of the hash version will be zero.  Therefore, the
 244  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 245  */
 246
 247 struct dx_root
 248 {
 249         struct fake_dirent dot;
 250         char dot_name[4];
 251         struct fake_dirent dotdot;
 252         char dotdot_name[4];
 253         struct dx_root_info
 254         {
 255                 __le32 reserved_zero;
 256                 u8 hash_version;
 257                 u8 info_length; /* 8 */
 258                 u8 indirect_levels;
 259                 u8 unused_flags;
 260         }
 261         info;
 262         struct dx_entry entries[];
 263 };
 264
 265 struct dx_node
 266 {
 267         struct fake_dirent fake;
 268         struct dx_entry entries[];
 269 };
 270
 271
 272 struct dx_frame
 273 {
 274         struct buffer_head *bh;
 275         struct dx_entry *entries;
 276         struct dx_entry *at;
 277 };
 278
 279 struct dx_map_entry
 280 {
 281         u32 hash;
 282         u16 offs;
 283         u16 size;
 284 };
 285
 286 /*
 287  * This goes at the end of each htree block.
 288  */
 289 struct dx_tail {
 290         u32 dt_reserved;
 291         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 292 };
 293
 294 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 295                 struct ext4_filename *fname,
 296                 struct ext4_dir_entry_2 **res_dir);
 297 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 298                              struct inode *dir, struct inode *inode);
 299
 300 /* checksumming functions */
 301 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 302                                  unsigned int blocksize)
 303 {
 304         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 305
 306         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 307         t->det_rec_len = ext4_rec_len_to_disk(
 308                         sizeof(struct ext4_dir_entry_tail), blocksize);
 309         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 310 }
 311
 312 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 313 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 314                                                    struct buffer_head *bh)
 315 {
 316         struct ext4_dir_entry_tail *t;
 317         int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
 318
 319 #ifdef PARANOID
 320         struct ext4_dir_entry *d, *top;
 321
 322         d = (struct ext4_dir_entry *)bh->b_data;
 323         top = (struct ext4_dir_entry *)(bh->b_data +
 324                 (blocksize - sizeof(struct ext4_dir_entry_tail)));
 325         while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
 326                 d = (struct ext4_dir_entry *)(((void *)d) +
 327                     ext4_rec_len_from_disk(d->rec_len, blocksize));
 328
 329         if (d != top)
 330                 return NULL;
 331
 332         t = (struct ext4_dir_entry_tail *)d;
 333 #else
 334         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 335 #endif
 336
 337         if (t->det_reserved_zero1 ||
 338             (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
 339              sizeof(struct ext4_dir_entry_tail)) ||
 340             t->det_reserved_zero2 ||
 341             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 342                 return NULL;
 343
 344         return t;
 345 }
 346
 347 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 348 {
 349         struct ext4_inode_info *ei = EXT4_I(inode);
 350         __u32 csum;
 351
 352         csum = ext4_chksum(ei->i_csum_seed, (__u8 *)dirent, size);
 353         return cpu_to_le32(csum);
 354 }
 355
 356 #define warn_no_space_for_csum(inode)                                   \
 357         __warn_no_space_for_csum((inode), __func__, __LINE__)
 358
 359 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 360                                      unsigned int line)
 361 {
 362         __ext4_warning_inode(inode, func, line,
 363                 "No space for directory leaf checksum. Please run e2fsck -D.");
 364 }
 365
 366 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 367 {
 368         struct ext4_dir_entry_tail *t;
 369
 370         if (!ext4_has_feature_metadata_csum(inode->i_sb))
 371                 return 1;
 372
 373         t = get_dirent_tail(inode, bh);
 374         if (!t) {
 375                 warn_no_space_for_csum(inode);
 376                 return 0;
 377         }
 378
 379         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 380                                                   (char *)t - bh->b_data))
 381                 return 0;
 382
 383         return 1;
 384 }
 385
 386 static void ext4_dirblock_csum_set(struct inode *inode,
 387                                  struct buffer_head *bh)
 388 {
 389         struct ext4_dir_entry_tail *t;
 390
 391         if (!ext4_has_feature_metadata_csum(inode->i_sb))
 392                 return;
 393
 394         t = get_dirent_tail(inode, bh);
 395         if (!t) {
 396                 warn_no_space_for_csum(inode);
 397                 return;
 398         }
 399
 400         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 401                                              (char *)t - bh->b_data);
 402 }
 403
 404 int ext4_handle_dirty_dirblock(handle_t *handle,
 405                                struct inode *inode,
 406                                struct buffer_head *bh)
 407 {
 408         ext4_dirblock_csum_set(inode, bh);
 409         return ext4_handle_dirty_metadata(handle, inode, bh);
 410 }
 411
 412 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 413                                                struct ext4_dir_entry *dirent,
 414                                                int *offset)
 415 {
 416         struct ext4_dir_entry *dp;
 417         struct dx_root_info *root;
 418         int count_offset;
 419         int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
 420         unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
 421
 422         if (rlen == blocksize)
 423                 count_offset = 8;
 424         else if (rlen == 12) {
 425                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 426                 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
 427                         return NULL;
 428                 root = (struct dx_root_info *)(((void *)dp + 12));
 429                 if (root->reserved_zero ||
 430                     root->info_length != sizeof(struct dx_root_info))
 431                         return NULL;
 432                 count_offset = 32;
 433         } else
 434                 return NULL;
 435
 436         if (offset)
 437                 *offset = count_offset;
 438         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 439 }
 440
 441 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 442                            int count_offset, int count, struct dx_tail *t)
 443 {
 444         struct ext4_inode_info *ei = EXT4_I(inode);
 445         __u32 csum;
 446         int size;
 447         __u32 dummy_csum = 0;
 448         int offset = offsetof(struct dx_tail, dt_checksum);
 449
 450         size = count_offset + (count * sizeof(struct dx_entry));
 451         csum = ext4_chksum(ei->i_csum_seed, (__u8 *)dirent, size);
 452         csum = ext4_chksum(csum, (__u8 *)t, offset);
 453         csum = ext4_chksum(csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 454
 455         return cpu_to_le32(csum);
 456 }
 457
 458 static int ext4_dx_csum_verify(struct inode *inode,
 459                                struct ext4_dir_entry *dirent)
 460 {
 461         struct dx_countlimit *c;
 462         struct dx_tail *t;
 463         int count_offset, limit, count;
 464
 465         if (!ext4_has_feature_metadata_csum(inode->i_sb))
 466                 return 1;
 467
 468         c = get_dx_countlimit(inode, dirent, &count_offset);
 469         if (!c) {
 470                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 471                 return 0;
 472         }
 473         limit = le16_to_cpu(c->limit);
 474         count = le16_to_cpu(c->count);
 475         if (count_offset + (limit * sizeof(struct dx_entry)) >
 476             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 477                 warn_no_space_for_csum(inode);
 478                 return 0;
 479         }
 480         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 481
 482         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 483                                             count, t))
 484                 return 0;
 485         return 1;
 486 }
 487
 488 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 489 {
 490         struct dx_countlimit *c;
 491         struct dx_tail *t;
 492         int count_offset, limit, count;
 493
 494         if (!ext4_has_feature_metadata_csum(inode->i_sb))
 495                 return;
 496
 497         c = get_dx_countlimit(inode, dirent, &count_offset);
 498         if (!c) {
 499                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 500                 return;
 501         }
 502         limit = le16_to_cpu(c->limit);
 503         count = le16_to_cpu(c->count);
 504         if (count_offset + (limit * sizeof(struct dx_entry)) >
 505             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 506                 warn_no_space_for_csum(inode);
 507                 return;
 508         }
 509         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 510
 511         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 512 }
 513
 514 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 515                                             struct inode *inode,
 516                                             struct buffer_head *bh)
 517 {
 518         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 519         return ext4_handle_dirty_metadata(handle, inode, bh);
 520 }
 521
 522 /*
 523  * p is at least 6 bytes before the end of page
 524  */
 525 static inline struct ext4_dir_entry_2 *
 526 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 527 {
 528         return (struct ext4_dir_entry_2 *)((char *)p +
 529                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 530 }
 531
 532 /*
 533  * Future: use high four bits of block for coalesce-on-delete flags
 534  * Mask them off for now.
 535  */
 536
 537 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 538 {
 539         return le32_to_cpu(entry->block) & 0x0fffffff;
 540 }
 541
 542 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 543 {
 544         entry->block = cpu_to_le32(value);
 545 }
 546
 547 static inline unsigned dx_get_hash(struct dx_entry *entry)
 548 {
 549         return le32_to_cpu(entry->hash);
 550 }
 551
 552 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 553 {
 554         entry->hash = cpu_to_le32(value);
 555 }
 556
 557 static inline unsigned dx_get_count(struct dx_entry *entries)
 558 {
 559         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 560 }
 561
 562 static inline unsigned dx_get_limit(struct dx_entry *entries)
 563 {
 564         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 565 }
 566
 567 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 568 {
 569         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 570 }
 571
 572 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 573 {
 574         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 575 }
 576
 577 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 578 {
 579         unsigned int entry_space = dir->i_sb->s_blocksize -
 580                         ext4_dir_rec_len(1, NULL) -
 581                         ext4_dir_rec_len(2, NULL) - infosize;
 582
 583         if (ext4_has_feature_metadata_csum(dir->i_sb))
 584                 entry_space -= sizeof(struct dx_tail);
 585         return entry_space / sizeof(struct dx_entry);
 586 }
 587
 588 static inline unsigned dx_node_limit(struct inode *dir)
 589 {
 590         unsigned int entry_space = dir->i_sb->s_blocksize -
 591                         ext4_dir_rec_len(0, dir);
 592
 593         if (ext4_has_feature_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                                         /* Directory is not encrypted */
 648                                         (void) ext4fs_dirhash(dir, de->name,
 649                                                 de->name_len, &h);
 650                                         printk("%*.s:(U)%x.%u ", len,
 651                                                name, h.hash,
 652                                                (unsigned) ((char *) de
 653                                                            - base));
 654                                 } else {
 655                                         struct fscrypt_str de_name =
 656                                                 FSTR_INIT(name, len);
 657
 658                                         /* Directory is encrypted */
 659                                         res = fscrypt_fname_alloc_buffer(
 660                                                 len, &fname_crypto_str);
 661                                         if (res)
 662                                                 printk(KERN_WARNING "Error "
 663                                                         "allocating crypto "
 664                                                         "buffer--skipping "
 665                                                         "crypto\n");
 666                                         res = fscrypt_fname_disk_to_usr(dir,
 667                                                 0, 0, &de_name,
 668                                                 &fname_crypto_str);
 669                                         if (res) {
 670                                                 printk(KERN_WARNING "Error "
 671                                                         "converting filename "
 672                                                         "from disk to usr"
 673                                                         "\n");
 674                                                 name = "??";
 675                                                 len = 2;
 676                                         } else {
 677                                                 name = fname_crypto_str.name;
 678                                                 len = fname_crypto_str.len;
 679                                         }
 680                                         if (IS_CASEFOLDED(dir))
 681                                                 h.hash = EXT4_DIRENT_HASH(de);
 682                                         else
 683                                                 (void) ext4fs_dirhash(dir,
 684                                                         de->name,
 685                                                         de->name_len, &h);
 686                                         printk("%*.s:(E)%x.%u ", len, name,
 687                                                h.hash, (unsigned) ((char *) de
 688                                                                    - base));
 689                                         fscrypt_fname_free_buffer(
 690                                                         &fname_crypto_str);
 691                                 }
 692 #else
 693                                 int len = de->name_len;
 694                                 char *name = de->name;
 695                                 (void) ext4fs_dirhash(dir, de->name,
 696                                                       de->name_len, &h);
 697                                 printk("%*.s:%x.%u ", len, name, h.hash,
 698                                        (unsigned) ((char *) de - base));
 699 #endif
 700                         }
 701                         space += ext4_dir_rec_len(de->name_len, dir);
 702                         names++;
 703                 }
 704                 de = ext4_next_entry(de, size);
 705         }
 706         printk(KERN_CONT "(%i)\n", names);
 707         return (struct stats) { names, space, 1 };
 708 }
 709
 710 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 711                              struct dx_entry *entries, int levels)
 712 {
 713         unsigned blocksize = dir->i_sb->s_blocksize;
 714         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 715         unsigned bcount = 0;
 716         struct buffer_head *bh;
 717         printk("%i indexed blocks...\n", count);
 718         for (i = 0; i < count; i++, entries++)
 719         {
 720                 ext4_lblk_t block = dx_get_block(entries);
 721                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 722                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 723                 struct stats stats;
 724                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 725                 bh = ext4_bread(NULL,dir, block, 0);
 726                 if (!bh || IS_ERR(bh))
 727                         continue;
 728                 stats = levels?
 729                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 730                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 731                         bh->b_data, blocksize, 0);
 732                 names += stats.names;
 733                 space += stats.space;
 734                 bcount += stats.bcount;
 735                 brelse(bh);
 736         }
 737         if (bcount)
 738                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 739                        levels ? "" : "   ", names, space/bcount,
 740                        (space/bcount)*100/blocksize);
 741         return (struct stats) { names, space, bcount};
 742 }
 743
 744 /*
 745  * Linear search cross check
 746  */
 747 static inline void htree_rep_invariant_check(struct dx_entry *at,
 748                                              struct dx_entry *target,
 749                                              u32 hash, unsigned int n)
 750 {
 751         while (n--) {
 752                 dxtrace(printk(KERN_CONT ","));
 753                 if (dx_get_hash(++at) > hash) {
 754                         at--;
 755                         break;
 756                 }
 757         }
 758         ASSERT(at == target - 1);
 759 }
 760 #else /* DX_DEBUG */
 761 static inline void htree_rep_invariant_check(struct dx_entry *at,
 762                                              struct dx_entry *target,
 763                                              u32 hash, unsigned int n)
 764 {
 765 }
 766 #endif /* DX_DEBUG */
 767
 768 /*
 769  * Probe for a directory leaf block to search.
 770  *
 771  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 772  * error in the directory index, and the caller should fall back to
 773  * searching the directory normally.  The callers of dx_probe **MUST**
 774  * check for this error code, and make sure it never gets reflected
 775  * back to userspace.
 776  */
 777 static struct dx_frame *
 778 dx_probe(struct ext4_filename *fname, struct inode *dir,
 779          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 780 {
 781         unsigned count, indirect, level, i;
 782         struct dx_entry *at, *entries, *p, *q, *m;
 783         struct dx_root *root;
 784         struct dx_frame *frame = frame_in;
 785         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 786         u32 hash;
 787         ext4_lblk_t block;
 788         ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
 789
 790         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 791         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 792         if (IS_ERR(frame->bh))
 793                 return (struct dx_frame *) frame->bh;
 794
 795         root = (struct dx_root *) frame->bh->b_data;
 796         if (root->info.hash_version != DX_HASH_TEA &&
 797             root->info.hash_version != DX_HASH_HALF_MD4 &&
 798             root->info.hash_version != DX_HASH_LEGACY &&
 799             root->info.hash_version != DX_HASH_SIPHASH) {
 800                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 801                                    root->info.hash_version);
 802                 goto fail;
 803         }
 804         if (ext4_hash_in_dirent(dir)) {
 805                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 806                         ext4_warning_inode(dir,
 807                                 "Hash in dirent, but hash is not SIPHASH");
 808                         goto fail;
 809                 }
 810         } else {
 811                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 812                         ext4_warning_inode(dir,
 813                                 "Hash code is SIPHASH, but hash not in dirent");
 814                         goto fail;
 815                 }
 816         }
 817         if (fname)
 818                 hinfo = &fname->hinfo;
 819         hinfo->hash_version = root->info.hash_version;
 820         if (hinfo->hash_version <= DX_HASH_TEA)
 821                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 822         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 823         /* hash is already computed for encrypted casefolded directory */
 824         if (fname && fname_name(fname) &&
 825             !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
 826                 int ret = ext4fs_dirhash(dir, fname_name(fname),
 827                                          fname_len(fname), hinfo);
 828                 if (ret < 0) {
 829                         ret_err = ERR_PTR(ret);
 830                         goto fail;
 831                 }
 832         }
 833         hash = hinfo->hash;
 834
 835         if (root->info.unused_flags & 1) {
 836                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 837                                    root->info.unused_flags);
 838                 goto fail;
 839         }
 840
 841         indirect = root->info.indirect_levels;
 842         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 843                 ext4_warning(dir->i_sb,
 844                              "Directory (ino: %lu) htree depth %#06x exceed"
 845                              "supported value", dir->i_ino,
 846                              ext4_dir_htree_level(dir->i_sb));
 847                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 848                         ext4_warning(dir->i_sb, "Enable large directory "
 849                                                 "feature to access it");
 850                 }
 851                 goto fail;
 852         }
 853
 854         entries = (struct dx_entry *)(((char *)&root->info) +
 855                                       root->info.info_length);
 856
 857         if (dx_get_limit(entries) != dx_root_limit(dir,
 858                                                    root->info.info_length)) {
 859                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 860                                    dx_get_limit(entries),
 861                                    dx_root_limit(dir, root->info.info_length));
 862                 goto fail;
 863         }
 864
 865         dxtrace(printk("Look up %x", hash));
 866         level = 0;
 867         blocks[0] = 0;
 868         while (1) {
 869                 count = dx_get_count(entries);
 870                 if (!count || count > dx_get_limit(entries)) {
 871                         ext4_warning_inode(dir,
 872                                            "dx entry: count %u beyond limit %u",
 873                                            count, dx_get_limit(entries));
 874                         goto fail;
 875                 }
 876
 877                 p = entries + 1;
 878                 q = entries + count - 1;
 879                 while (p <= q) {
 880                         m = p + (q - p) / 2;
 881                         dxtrace(printk(KERN_CONT "."));
 882                         if (dx_get_hash(m) > hash)
 883                                 q = m - 1;
 884                         else
 885                                 p = m + 1;
 886                 }
 887
 888                 htree_rep_invariant_check(entries, p, hash, count - 1);
 889
 890                 at = p - 1;
 891                 dxtrace(printk(KERN_CONT " %x->%u\n",
 892                                at == entries ? 0 : dx_get_hash(at),
 893                                dx_get_block(at)));
 894                 frame->entries = entries;
 895                 frame->at = at;
 896
 897                 block = dx_get_block(at);
 898                 for (i = 0; i <= level; i++) {
 899                         if (blocks[i] == block) {
 900                                 ext4_warning_inode(dir,
 901                                         "dx entry: tree cycle block %u points back to block %u",
 902                                         blocks[level], block);
 903                                 goto fail;
 904                         }
 905                 }
 906                 if (++level > indirect)
 907                         return frame;
 908                 blocks[level] = block;
 909                 frame++;
 910                 frame->bh = ext4_read_dirblock(dir, block, INDEX);
 911                 if (IS_ERR(frame->bh)) {
 912                         ret_err = (struct dx_frame *) frame->bh;
 913                         frame->bh = NULL;
 914                         goto fail;
 915                 }
 916
 917                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 918
 919                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 920                         ext4_warning_inode(dir,
 921                                 "dx entry: limit %u != node limit %u",
 922                                 dx_get_limit(entries), dx_node_limit(dir));
 923                         goto fail;
 924                 }
 925         }
 926 fail:
 927         while (frame >= frame_in) {
 928                 brelse(frame->bh);
 929                 frame--;
 930         }
 931
 932         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 933                 ext4_warning_inode(dir,
 934                         "Corrupt directory, running e2fsck is recommended");
 935         return ret_err;
 936 }
 937
 938 static void dx_release(struct dx_frame *frames)
 939 {
 940         struct dx_root_info *info;
 941         int i;
 942         unsigned int indirect_levels;
 943
 944         if (frames[0].bh == NULL)
 945                 return;
 946
 947         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 948         /* save local copy, "info" may be freed after brelse() */
 949         indirect_levels = info->indirect_levels;
 950         for (i = 0; i <= indirect_levels; i++) {
 951                 if (frames[i].bh == NULL)
 952                         break;
 953                 brelse(frames[i].bh);
 954                 frames[i].bh = NULL;
 955         }
 956 }
 957
 958 /*
 959  * This function increments the frame pointer to search the next leaf
 960  * block, and reads in the necessary intervening nodes if the search
 961  * should be necessary.  Whether or not the search is necessary is
 962  * controlled by the hash parameter.  If the hash value is even, then
 963  * the search is only continued if the next block starts with that
 964  * hash value.  This is used if we are searching for a specific file.
 965  *
 966  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 967  *
 968  * This function returns 1 if the caller should continue to search,
 969  * or 0 if it should not.  If there is an error reading one of the
 970  * index blocks, it will a negative error code.
 971  *
 972  * If start_hash is non-null, it will be filled in with the starting
 973  * hash of the next page.
 974  */
 975 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 976                                  struct dx_frame *frame,
 977                                  struct dx_frame *frames,
 978                                  __u32 *start_hash)
 979 {
 980         struct dx_frame *p;
 981         struct buffer_head *bh;
 982         int num_frames = 0;
 983         __u32 bhash;
 984
 985         p = frame;
 986         /*
 987          * Find the next leaf page by incrementing the frame pointer.
 988          * If we run out of entries in the interior node, loop around and
 989          * increment pointer in the parent node.  When we break out of
 990          * this loop, num_frames indicates the number of interior
 991          * nodes need to be read.
 992          */
 993         while (1) {
 994                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 995                         break;
 996                 if (p == frames)
 997                         return 0;
 998                 num_frames++;
 999                 p--;
1000         }
1001
1002         /*
1003          * If the hash is 1, then continue only if the next page has a
1004          * continuation hash of any value.  This is used for readdir
1005          * handling.  Otherwise, check to see if the hash matches the
1006          * desired continuation hash.  If it doesn't, return since
1007          * there's no point to read in the successive index pages.
1008          */
1009         bhash = dx_get_hash(p->at);
1010         if (start_hash)
1011                 *start_hash = bhash;
1012         if ((hash & 1) == 0) {
1013                 if ((bhash & ~1) != hash)
1014                         return 0;
1015         }
1016         /*
1017          * If the hash is HASH_NB_ALWAYS, we always go to the next
1018          * block so no check is necessary
1019          */
1020         while (num_frames--) {
1021                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1022                 if (IS_ERR(bh))
1023                         return PTR_ERR(bh);
1024                 p++;
1025                 brelse(p->bh);
1026                 p->bh = bh;
1027                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1028         }
1029         return 1;
1030 }
1031
1032
1033 /*
1034  * This function fills a red-black tree with information from a
1035  * directory block.  It returns the number directory entries loaded
1036  * into the tree.  If there is an error it is returned in err.
1037  */
1038 static int htree_dirblock_to_tree(struct file *dir_file,
1039                                   struct inode *dir, ext4_lblk_t block,
1040                                   struct dx_hash_info *hinfo,
1041                                   __u32 start_hash, __u32 start_minor_hash)
1042 {
1043         struct buffer_head *bh;
1044         struct ext4_dir_entry_2 *de, *top;
1045         int err = 0, count = 0;
1046         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1047         int csum = ext4_has_feature_metadata_csum(dir->i_sb);
1048
1049         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1050                                                         (unsigned long)block));
1051         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1052         if (IS_ERR(bh))
1053                 return PTR_ERR(bh);
1054
1055         de = (struct ext4_dir_entry_2 *) bh->b_data;
1056         /* csum entries are not larger in the casefolded encrypted case */
1057         top = (struct ext4_dir_entry_2 *) ((char *) de +
1058                                            dir->i_sb->s_blocksize -
1059                                            ext4_dir_rec_len(0,
1060                                                            csum ? NULL : dir));
1061         /* Check if the directory is encrypted */
1062         if (IS_ENCRYPTED(dir)) {
1063                 err = fscrypt_prepare_readdir(dir);
1064                 if (err < 0) {
1065                         brelse(bh);
1066                         return err;
1067                 }
1068                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1069                                                  &fname_crypto_str);
1070                 if (err < 0) {
1071                         brelse(bh);
1072                         return err;
1073                 }
1074         }
1075
1076         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1077                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1078                                 bh->b_data, bh->b_size,
1079                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1080                                          + ((char *)de - bh->b_data))) {
1081                         /* silently ignore the rest of the block */
1082                         break;
1083                 }
1084                 if (ext4_hash_in_dirent(dir)) {
1085                         if (de->name_len && de->inode) {
1086                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1087                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1088                         } else {
1089                                 hinfo->hash = 0;
1090                                 hinfo->minor_hash = 0;
1091                         }
1092                 } else {
1093                         err = ext4fs_dirhash(dir, de->name,
1094                                              de->name_len, hinfo);
1095                         if (err < 0) {
1096                                 count = err;
1097                                 goto errout;
1098                         }
1099                 }
1100                 if ((hinfo->hash < start_hash) ||
1101                     ((hinfo->hash == start_hash) &&
1102                      (hinfo->minor_hash < start_minor_hash)))
1103                         continue;
1104                 if (de->inode == 0)
1105                         continue;
1106                 if (!IS_ENCRYPTED(dir)) {
1107                         tmp_str.name = de->name;
1108                         tmp_str.len = de->name_len;
1109                         err = ext4_htree_store_dirent(dir_file,
1110                                    hinfo->hash, hinfo->minor_hash, de,
1111                                    &tmp_str);
1112                 } else {
1113                         int save_len = fname_crypto_str.len;
1114                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1115                                                                 de->name_len);
1116
1117                         /* Directory is encrypted */
1118                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1119                                         hinfo->minor_hash, &de_name,
1120                                         &fname_crypto_str);
1121                         if (err) {
1122                                 count = err;
1123                                 goto errout;
1124                         }
1125                         err = ext4_htree_store_dirent(dir_file,
1126                                    hinfo->hash, hinfo->minor_hash, de,
1127                                         &fname_crypto_str);
1128                         fname_crypto_str.len = save_len;
1129                 }
1130                 if (err != 0) {
1131                         count = err;
1132                         goto errout;
1133                 }
1134                 count++;
1135         }
1136 errout:
1137         brelse(bh);
1138         fscrypt_fname_free_buffer(&fname_crypto_str);
1139         return count;
1140 }
1141
1142
1143 /*
1144  * This function fills a red-black tree with information from a
1145  * directory.  We start scanning the directory in hash order, starting
1146  * at start_hash and start_minor_hash.
1147  *
1148  * This function returns the number of entries inserted into the tree,
1149  * or a negative error code.
1150  */
1151 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1152                          __u32 start_minor_hash, __u32 *next_hash)
1153 {
1154         struct dx_hash_info hinfo;
1155         struct ext4_dir_entry_2 *de;
1156         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1157         struct inode *dir;
1158         ext4_lblk_t block;
1159         int count = 0;
1160         int ret, err;
1161         __u32 hashval;
1162         struct fscrypt_str tmp_str;
1163
1164         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1165                        start_hash, start_minor_hash));
1166         dir = file_inode(dir_file);
1167         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1168                 if (ext4_hash_in_dirent(dir))
1169                         hinfo.hash_version = DX_HASH_SIPHASH;
1170                 else
1171                         hinfo.hash_version =
1172                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1173                 if (hinfo.hash_version <= DX_HASH_TEA)
1174                         hinfo.hash_version +=
1175                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1176                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1177                 if (ext4_has_inline_data(dir)) {
1178                         int has_inline_data = 1;
1179                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1180                                                        &hinfo, start_hash,
1181                                                        start_minor_hash,
1182                                                        &has_inline_data);
1183                         if (has_inline_data) {
1184                                 *next_hash = ~0;
1185                                 return count;
1186                         }
1187                 }
1188                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1189                                                start_hash, start_minor_hash);
1190                 *next_hash = ~0;
1191                 return count;
1192         }
1193         hinfo.hash = start_hash;
1194         hinfo.minor_hash = 0;
1195         frame = dx_probe(NULL, dir, &hinfo, frames);
1196         if (IS_ERR(frame))
1197                 return PTR_ERR(frame);
1198
1199         /* Add '.' and '..' from the htree header */
1200         if (!start_hash && !start_minor_hash) {
1201                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1202                 tmp_str.name = de->name;
1203                 tmp_str.len = de->name_len;
1204                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1205                                               de, &tmp_str);
1206                 if (err != 0)
1207                         goto errout;
1208                 count++;
1209         }
1210         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1211                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1212                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1213                 tmp_str.name = de->name;
1214                 tmp_str.len = de->name_len;
1215                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1216                                               de, &tmp_str);
1217                 if (err != 0)
1218                         goto errout;
1219                 count++;
1220         }
1221
1222         while (1) {
1223                 if (fatal_signal_pending(current)) {
1224                         err = -ERESTARTSYS;
1225                         goto errout;
1226                 }
1227                 cond_resched();
1228                 block = dx_get_block(frame->at);
1229                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1230                                              start_hash, start_minor_hash);
1231                 if (ret < 0) {
1232                         err = ret;
1233                         goto errout;
1234                 }
1235                 count += ret;
1236                 hashval = ~0;
1237                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1238                                             frame, frames, &hashval);
1239                 *next_hash = hashval;
1240                 if (ret < 0) {
1241                         err = ret;
1242                         goto errout;
1243                 }
1244                 /*
1245                  * Stop if:  (a) there are no more entries, or
1246                  * (b) we have inserted at least one entry and the
1247                  * next hash value is not a continuation
1248                  */
1249                 if ((ret == 0) ||
1250                     (count && ((hashval & 1) == 0)))
1251                         break;
1252         }
1253         dx_release(frames);
1254         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1255                        "next hash: %x\n", count, *next_hash));
1256         return count;
1257 errout:
1258         dx_release(frames);
1259         return (err);
1260 }
1261
1262 static inline int search_dirblock(struct buffer_head *bh,
1263                                   struct inode *dir,
1264                                   struct ext4_filename *fname,
1265                                   unsigned int offset,
1266                                   struct ext4_dir_entry_2 **res_dir)
1267 {
1268         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1269                                fname, offset, res_dir);
1270 }
1271
1272 /*
1273  * Directory block splitting, compacting
1274  */
1275
1276 /*
1277  * Create map of hash values, offsets, and sizes, stored at end of block.
1278  * Returns number of entries mapped.
1279  */
1280 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1281                        struct dx_hash_info *hinfo,
1282                        struct dx_map_entry *map_tail)
1283 {
1284         int count = 0;
1285         struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1286         unsigned int buflen = bh->b_size;
1287         char *base = bh->b_data;
1288         struct dx_hash_info h = *hinfo;
1289         int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1290
1291         if (ext4_has_feature_metadata_csum(dir->i_sb))
1292                 buflen -= sizeof(struct ext4_dir_entry_tail);
1293
1294         while ((char *) de < base + buflen) {
1295                 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1296                                          ((char *)de) - base))
1297                         return -EFSCORRUPTED;
1298                 if (de->name_len && de->inode) {
1299                         if (ext4_hash_in_dirent(dir))
1300                                 h.hash = EXT4_DIRENT_HASH(de);
1301                         else {
1302                                 int err = ext4fs_dirhash(dir, de->name,
1303                                                      de->name_len, &h);
1304                                 if (err < 0)
1305                                         return err;
1306                         }
1307                         map_tail--;
1308                         map_tail->hash = h.hash;
1309                         map_tail->offs = ((char *) de - base)>>2;
1310                         map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1311                                                                 blocksize);
1312                         count++;
1313                         cond_resched();
1314                 }
1315                 de = ext4_next_entry(de, blocksize);
1316         }
1317         return count;
1318 }
1319
1320 /* Sort map by hash value */
1321 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1322 {
1323         struct dx_map_entry *p, *q, *top = map + count - 1;
1324         int more;
1325         /* Combsort until bubble sort doesn't suck */
1326         while (count > 2) {
1327                 count = count*10/13;
1328                 if (count - 9 < 2) /* 9, 10 -> 11 */
1329                         count = 11;
1330                 for (p = top, q = p - count; q >= map; p--, q--)
1331                         if (p->hash < q->hash)
1332                                 swap(*p, *q);
1333         }
1334         /* Garden variety bubble sort */
1335         do {
1336                 more = 0;
1337                 q = top;
1338                 while (q-- > map) {
1339                         if (q[1].hash >= q[0].hash)
1340                                 continue;
1341                         swap(*(q+1), *q);
1342                         more = 1;
1343                 }
1344         } while(more);
1345 }
1346
1347 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1348 {
1349         struct dx_entry *entries = frame->entries;
1350         struct dx_entry *old = frame->at, *new = old + 1;
1351         int count = dx_get_count(entries);
1352
1353         ASSERT(count < dx_get_limit(entries));
1354         ASSERT(old < entries + count);
1355         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1356         dx_set_hash(new, hash);
1357         dx_set_block(new, block);
1358         dx_set_count(entries, count + 1);
1359 }
1360
1361 #if IS_ENABLED(CONFIG_UNICODE)
1362 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1363                                   struct ext4_filename *name)
1364 {
1365         struct qstr *cf_name = &name->cf_name;
1366         unsigned char *buf;
1367         struct dx_hash_info *hinfo = &name->hinfo;
1368         int len;
1369
1370         if (!IS_CASEFOLDED(dir) ||
1371             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1372                 cf_name->name = NULL;
1373                 return 0;
1374         }
1375
1376         buf = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1377         if (!buf)
1378                 return -ENOMEM;
1379
1380         len = utf8_casefold(dir->i_sb->s_encoding, iname, buf, EXT4_NAME_LEN);
1381         if (len <= 0) {
1382                 kfree(buf);
1383                 buf = NULL;
1384         }
1385         cf_name->name = buf;
1386         cf_name->len = (unsigned) len;
1387
1388         if (!IS_ENCRYPTED(dir))
1389                 return 0;
1390
1391         hinfo->hash_version = DX_HASH_SIPHASH;
1392         hinfo->seed = NULL;
1393         if (cf_name->name)
1394                 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1395         else
1396                 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1397 }
1398 #endif
1399
1400 /*
1401  * Test whether a directory entry matches the filename being searched for.
1402  *
1403  * Return: %true if the directory entry matches, otherwise %false.
1404  */
1405 static bool ext4_match(struct inode *parent,
1406                               const struct ext4_filename *fname,
1407                               struct ext4_dir_entry_2 *de)
1408 {
1409         struct fscrypt_name f;
1410
1411         if (!de->inode)
1412                 return false;
1413
1414         f.usr_fname = fname->usr_fname;
1415         f.disk_name = fname->disk_name;
1416 #ifdef CONFIG_FS_ENCRYPTION
1417         f.crypto_buf = fname->crypto_buf;
1418 #endif
1419
1420 #if IS_ENABLED(CONFIG_UNICODE)
1421         if (IS_CASEFOLDED(parent) &&
1422             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1423                 /*
1424                  * Just checking IS_ENCRYPTED(parent) below is not
1425                  * sufficient to decide whether one can use the hash for
1426                  * skipping the string comparison, because the key might
1427                  * have been added right after
1428                  * ext4_fname_setup_ci_filename().  In this case, a hash
1429                  * mismatch will be a false negative.  Therefore, make
1430                  * sure cf_name was properly initialized before
1431                  * considering the calculated hash.
1432                  */
1433                 if (sb_no_casefold_compat_fallback(parent->i_sb) &&
1434                     IS_ENCRYPTED(parent) && fname->cf_name.name &&
1435                     (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1436                      fname->hinfo.minor_hash != EXT4_DIRENT_MINOR_HASH(de)))
1437                         return false;
1438                 /*
1439                  * Treat comparison errors as not a match.  The
1440                  * only case where it happens is on a disk
1441                  * corruption or ENOMEM.
1442                  */
1443
1444                 return generic_ci_match(parent, fname->usr_fname,
1445                                         &fname->cf_name, de->name,
1446                                         de->name_len) > 0;
1447         }
1448 #endif
1449
1450         return fscrypt_match_name(&f, de->name, de->name_len);
1451 }
1452
1453 /*
1454  * Returns 0 if not found, -EFSCORRUPTED on failure, and 1 on success
1455  */
1456 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1457                     struct inode *dir, struct ext4_filename *fname,
1458                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1459 {
1460         struct ext4_dir_entry_2 * de;
1461         char * dlimit;
1462         int de_len;
1463
1464         de = (struct ext4_dir_entry_2 *)search_buf;
1465         dlimit = search_buf + buf_size;
1466         while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1467                 /* this code is executed quadratically often */
1468                 /* do minimal checking `by hand' */
1469                 if (de->name + de->name_len <= dlimit &&
1470                     ext4_match(dir, fname, de)) {
1471                         /* found a match - just to be sure, do
1472                          * a full check */
1473                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1474                                                  buf_size, offset))
1475                                 return -EFSCORRUPTED;
1476                         *res_dir = de;
1477                         return 1;
1478                 }
1479                 /* prevent looping on a bad block */
1480                 de_len = ext4_rec_len_from_disk(de->rec_len,
1481                                                 dir->i_sb->s_blocksize);
1482                 if (de_len <= 0)
1483                         return -EFSCORRUPTED;
1484                 offset += de_len;
1485                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1486         }
1487         return 0;
1488 }
1489
1490 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1491                                struct ext4_dir_entry *de)
1492 {
1493         struct super_block *sb = dir->i_sb;
1494
1495         if (!is_dx(dir))
1496                 return 0;
1497         if (block == 0)
1498                 return 1;
1499         if (de->inode == 0 &&
1500             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1501                         sb->s_blocksize)
1502                 return 1;
1503         return 0;
1504 }
1505
1506 /*
1507  *      __ext4_find_entry()
1508  *
1509  * finds an entry in the specified directory with the wanted name. It
1510  * returns the cache buffer in which the entry was found, and the entry
1511  * itself (as a parameter - res_dir). It does NOT read the inode of the
1512  * entry - you'll have to do that yourself if you want to.
1513  *
1514  * The returned buffer_head has ->b_count elevated.  The caller is expected
1515  * to brelse() it when appropriate.
1516  */
1517 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1518                                              struct ext4_filename *fname,
1519                                              struct ext4_dir_entry_2 **res_dir,
1520                                              int *inlined)
1521 {
1522         struct super_block *sb;
1523         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1524         struct buffer_head *bh, *ret = NULL;
1525         ext4_lblk_t start, block;
1526         const u8 *name = fname->usr_fname->name;
1527         size_t ra_max = 0;      /* Number of bh's in the readahead
1528                                    buffer, bh_use[] */
1529         size_t ra_ptr = 0;      /* Current index into readahead
1530                                    buffer */
1531         ext4_lblk_t  nblocks;
1532         int i, namelen, retval;
1533
1534         *res_dir = NULL;
1535         sb = dir->i_sb;
1536         namelen = fname->usr_fname->len;
1537         if (namelen > EXT4_NAME_LEN)
1538                 return NULL;
1539
1540         if (ext4_has_inline_data(dir)) {
1541                 int has_inline_data = 1;
1542                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1543                                              &has_inline_data);
1544                 if (inlined)
1545                         *inlined = has_inline_data;
1546                 if (has_inline_data || IS_ERR(ret))
1547                         goto cleanup_and_exit;
1548         }
1549
1550         if ((namelen <= 2) && (name[0] == '.') &&
1551             (name[1] == '.' || name[1] == '\0')) {
1552                 /*
1553                  * "." or ".." will only be in the first block
1554                  * NFS may look up ".."; "." should be handled by the VFS
1555                  */
1556                 block = start = 0;
1557                 nblocks = 1;
1558                 goto restart;
1559         }
1560         if (is_dx(dir)) {
1561                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1562                 /*
1563                  * On success, or if the error was file not found,
1564                  * return.  Otherwise, fall back to doing a search the
1565                  * old fashioned way.
1566                  */
1567                 if (IS_ERR(ret) && PTR_ERR(ret) == ERR_BAD_DX_DIR)
1568                         dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1569                                        "falling back\n"));
1570                 else if (!sb_no_casefold_compat_fallback(dir->i_sb) &&
1571                          *res_dir == NULL && IS_CASEFOLDED(dir))
1572                         dxtrace(printk(KERN_DEBUG "ext4_find_entry: casefold "
1573                                        "failed, falling back\n"));
1574                 else
1575                         goto cleanup_and_exit;
1576                 ret = NULL;
1577         }
1578         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1579         if (!nblocks) {
1580                 ret = NULL;
1581                 goto cleanup_and_exit;
1582         }
1583         start = EXT4_I(dir)->i_dir_start_lookup;
1584         if (start >= nblocks)
1585                 start = 0;
1586         block = start;
1587 restart:
1588         do {
1589                 /*
1590                  * We deal with the read-ahead logic here.
1591                  */
1592                 cond_resched();
1593                 if (ra_ptr >= ra_max) {
1594                         /* Refill the readahead buffer */
1595                         ra_ptr = 0;
1596                         if (block < start)
1597                                 ra_max = start - block;
1598                         else
1599                                 ra_max = nblocks - block;
1600                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1601                         retval = ext4_bread_batch(dir, block, ra_max,
1602                                                   false /* wait */, bh_use);
1603                         if (retval) {
1604                                 ret = ERR_PTR(retval);
1605                                 ra_max = 0;
1606                                 goto cleanup_and_exit;
1607                         }
1608                 }
1609                 if ((bh = bh_use[ra_ptr++]) == NULL)
1610                         goto next;
1611                 wait_on_buffer(bh);
1612                 if (!buffer_uptodate(bh)) {
1613                         EXT4_ERROR_INODE_ERR(dir, EIO,
1614                                              "reading directory lblock %lu",
1615                                              (unsigned long) block);
1616                         brelse(bh);
1617                         ret = ERR_PTR(-EIO);
1618                         goto cleanup_and_exit;
1619                 }
1620                 if (!buffer_verified(bh) &&
1621                     !is_dx_internal_node(dir, block,
1622                                          (struct ext4_dir_entry *)bh->b_data) &&
1623                     !ext4_dirblock_csum_verify(dir, bh)) {
1624                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1625                                              "checksumming directory "
1626                                              "block %lu", (unsigned long)block);
1627                         brelse(bh);
1628                         ret = ERR_PTR(-EFSBADCRC);
1629                         goto cleanup_and_exit;
1630                 }
1631                 set_buffer_verified(bh);
1632                 i = search_dirblock(bh, dir, fname,
1633                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1634                 if (i == 1) {
1635                         EXT4_I(dir)->i_dir_start_lookup = block;
1636                         ret = bh;
1637                         goto cleanup_and_exit;
1638                 } else {
1639                         brelse(bh);
1640                         if (i < 0) {
1641                                 ret = ERR_PTR(i);
1642                                 goto cleanup_and_exit;
1643                         }
1644                 }
1645         next:
1646                 if (++block >= nblocks)
1647                         block = 0;
1648         } while (block != start);
1649
1650         /*
1651          * If the directory has grown while we were searching, then
1652          * search the last part of the directory before giving up.
1653          */
1654         block = nblocks;
1655         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1656         if (block < nblocks) {
1657                 start = 0;
1658                 goto restart;
1659         }
1660
1661 cleanup_and_exit:
1662         /* Clean up the read-ahead blocks */
1663         for (; ra_ptr < ra_max; ra_ptr++)
1664                 brelse(bh_use[ra_ptr]);
1665         return ret;
1666 }
1667
1668 static struct buffer_head *ext4_find_entry(struct inode *dir,
1669                                            const struct qstr *d_name,
1670                                            struct ext4_dir_entry_2 **res_dir,
1671                                            int *inlined)
1672 {
1673         int err;
1674         struct ext4_filename fname;
1675         struct buffer_head *bh;
1676
1677         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1678         if (err == -ENOENT)
1679                 return NULL;
1680         if (err)
1681                 return ERR_PTR(err);
1682
1683         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1684
1685         ext4_fname_free_filename(&fname);
1686         return bh;
1687 }
1688
1689 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1690                                              struct dentry *dentry,
1691                                              struct ext4_dir_entry_2 **res_dir)
1692 {
1693         int err;
1694         struct ext4_filename fname;
1695         struct buffer_head *bh;
1696
1697         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1698         if (err == -ENOENT)
1699                 return NULL;
1700         if (err)
1701                 return ERR_PTR(err);
1702
1703         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1704
1705         ext4_fname_free_filename(&fname);
1706         return bh;
1707 }
1708
1709 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1710                         struct ext4_filename *fname,
1711                         struct ext4_dir_entry_2 **res_dir)
1712 {
1713         struct super_block * sb = dir->i_sb;
1714         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1715         struct buffer_head *bh;
1716         ext4_lblk_t block;
1717         int retval;
1718
1719 #ifdef CONFIG_FS_ENCRYPTION
1720         *res_dir = NULL;
1721 #endif
1722         frame = dx_probe(fname, dir, NULL, frames);
1723         if (IS_ERR(frame))
1724                 return ERR_CAST(frame);
1725         do {
1726                 block = dx_get_block(frame->at);
1727                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1728                 if (IS_ERR(bh))
1729                         goto errout;
1730
1731                 retval = search_dirblock(bh, dir, fname,
1732                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1733                                          res_dir);
1734                 if (retval == 1)
1735                         goto success;
1736                 brelse(bh);
1737                 if (retval < 0) {
1738                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1739                         goto errout;
1740                 }
1741
1742                 /* Check to see if we should continue to search */
1743                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1744                                                frames, NULL);
1745                 if (retval < 0) {
1746                         ext4_warning_inode(dir,
1747                                 "error %d reading directory index block",
1748                                 retval);
1749                         bh = ERR_PTR(retval);
1750                         goto errout;
1751                 }
1752         } while (retval == 1);
1753
1754         bh = NULL;
1755 errout:
1756         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1757 success:
1758         dx_release(frames);
1759         return bh;
1760 }
1761
1762 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1763 {
1764         struct inode *inode;
1765         struct ext4_dir_entry_2 *de;
1766         struct buffer_head *bh;
1767
1768         if (dentry->d_name.len > EXT4_NAME_LEN)
1769                 return ERR_PTR(-ENAMETOOLONG);
1770
1771         bh = ext4_lookup_entry(dir, dentry, &de);
1772         if (IS_ERR(bh))
1773                 return ERR_CAST(bh);
1774         inode = NULL;
1775         if (bh) {
1776                 __u32 ino = le32_to_cpu(de->inode);
1777                 brelse(bh);
1778                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1779                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1780                         return ERR_PTR(-EFSCORRUPTED);
1781                 }
1782                 if (unlikely(ino == dir->i_ino)) {
1783                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1784                                          dentry);
1785                         return ERR_PTR(-EFSCORRUPTED);
1786                 }
1787                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1788                 if (inode == ERR_PTR(-ESTALE)) {
1789                         EXT4_ERROR_INODE(dir,
1790                                          "deleted inode referenced: %u",
1791                                          ino);
1792                         return ERR_PTR(-EFSCORRUPTED);
1793                 }
1794                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1795                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1796                     !fscrypt_has_permitted_context(dir, inode)) {
1797                         ext4_warning(inode->i_sb,
1798                                      "Inconsistent encryption contexts: %lu/%lu",
1799                                      dir->i_ino, inode->i_ino);
1800                         iput(inode);
1801                         return ERR_PTR(-EPERM);
1802                 }
1803         }
1804
1805         if (IS_ENABLED(CONFIG_UNICODE) && !inode && IS_CASEFOLDED(dir)) {
1806                 /* Eventually we want to call d_add_ci(dentry, NULL)
1807                  * for negative dentries in the encoding case as
1808                  * well.  For now, prevent the negative dentry
1809                  * from being cached.
1810                  */
1811                 return NULL;
1812         }
1813
1814         return d_splice_alias(inode, dentry);
1815 }
1816
1817
1818 struct dentry *ext4_get_parent(struct dentry *child)
1819 {
1820         __u32 ino;
1821         struct ext4_dir_entry_2 * de;
1822         struct buffer_head *bh;
1823
1824         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1825         if (IS_ERR(bh))
1826                 return ERR_CAST(bh);
1827         if (!bh)
1828                 return ERR_PTR(-ENOENT);
1829         ino = le32_to_cpu(de->inode);
1830         brelse(bh);
1831
1832         if (!ext4_valid_inum(child->d_sb, ino)) {
1833                 EXT4_ERROR_INODE(d_inode(child),
1834                                  "bad parent inode number: %u", ino);
1835                 return ERR_PTR(-EFSCORRUPTED);
1836         }
1837
1838         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1839 }
1840
1841 /*
1842  * Move count entries from end of map between two memory locations.
1843  * Returns pointer to last entry moved.
1844  */
1845 static struct ext4_dir_entry_2 *
1846 dx_move_dirents(struct inode *dir, char *from, char *to,
1847                 struct dx_map_entry *map, int count,
1848                 unsigned blocksize)
1849 {
1850         unsigned rec_len = 0;
1851
1852         while (count--) {
1853                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1854                                                 (from + (map->offs<<2));
1855                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1856
1857                 memcpy (to, de, rec_len);
1858                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1859                                 ext4_rec_len_to_disk(rec_len, blocksize);
1860
1861                 /* wipe dir_entry excluding the rec_len field */
1862                 de->inode = 0;
1863                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1864                                                                 blocksize) -
1865                                          offsetof(struct ext4_dir_entry_2,
1866                                                                 name_len));
1867
1868                 map++;
1869                 to += rec_len;
1870         }
1871         return (struct ext4_dir_entry_2 *) (to - rec_len);
1872 }
1873
1874 /*
1875  * Compact each dir entry in the range to the minimal rec_len.
1876  * Returns pointer to last entry in range.
1877  */
1878 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1879                                                         unsigned int blocksize)
1880 {
1881         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1882         unsigned rec_len = 0;
1883
1884         prev = to = de;
1885         while ((char*)de < base + blocksize) {
1886                 next = ext4_next_entry(de, blocksize);
1887                 if (de->inode && de->name_len) {
1888                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1889                         if (de > to)
1890                                 memmove(to, de, rec_len);
1891                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1892                         prev = to;
1893                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1894                 }
1895                 de = next;
1896         }
1897         return prev;
1898 }
1899
1900 /*
1901  * Split a full leaf block to make room for a new dir entry.
1902  * Allocate a new block, and move entries so that they are approx. equally full.
1903  * Returns pointer to de in block into which the new entry will be inserted.
1904  */
1905 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1906                         struct buffer_head **bh,struct dx_frame *frame,
1907                         struct dx_hash_info *hinfo)
1908 {
1909         unsigned blocksize = dir->i_sb->s_blocksize;
1910         unsigned continued;
1911         int count;
1912         struct buffer_head *bh2;
1913         ext4_lblk_t newblock;
1914         u32 hash2;
1915         struct dx_map_entry *map;
1916         char *data1 = (*bh)->b_data, *data2;
1917         unsigned split, move, size;
1918         struct ext4_dir_entry_2 *de = NULL, *de2;
1919         int     csum_size = 0;
1920         int     err = 0, i;
1921
1922         if (ext4_has_feature_metadata_csum(dir->i_sb))
1923                 csum_size = sizeof(struct ext4_dir_entry_tail);
1924
1925         bh2 = ext4_append(handle, dir, &newblock);
1926         if (IS_ERR(bh2)) {
1927                 brelse(*bh);
1928                 *bh = NULL;
1929                 return ERR_CAST(bh2);
1930         }
1931
1932         BUFFER_TRACE(*bh, "get_write_access");
1933         err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1934                                             EXT4_JTR_NONE);
1935         if (err)
1936                 goto journal_error;
1937
1938         BUFFER_TRACE(frame->bh, "get_write_access");
1939         err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1940                                             EXT4_JTR_NONE);
1941         if (err)
1942                 goto journal_error;
1943
1944         data2 = bh2->b_data;
1945
1946         /* create map in the end of data2 block */
1947         map = (struct dx_map_entry *) (data2 + blocksize);
1948         count = dx_make_map(dir, *bh, hinfo, map);
1949         if (count < 0) {
1950                 err = count;
1951                 goto journal_error;
1952         }
1953         map -= count;
1954         dx_sort_map(map, count);
1955         /* Ensure that neither split block is over half full */
1956         size = 0;
1957         move = 0;
1958         for (i = count-1; i >= 0; i--) {
1959                 /* is more than half of this entry in 2nd half of the block? */
1960                 if (size + map[i].size/2 > blocksize/2)
1961                         break;
1962                 size += map[i].size;
1963                 move++;
1964         }
1965         /*
1966          * map index at which we will split
1967          *
1968          * If the sum of active entries didn't exceed half the block size, just
1969          * split it in half by count; each resulting block will have at least
1970          * half the space free.
1971          */
1972         if (i >= 0)
1973                 split = count - move;
1974         else
1975                 split = count/2;
1976
1977         if (WARN_ON_ONCE(split == 0)) {
1978                 /* Should never happen, but avoid out-of-bounds access below */
1979                 ext4_error_inode_block(dir, (*bh)->b_blocknr, 0,
1980                         "bad indexed directory? hash=%08x:%08x count=%d move=%u",
1981                         hinfo->hash, hinfo->minor_hash, count, move);
1982                 err = -EFSCORRUPTED;
1983                 goto out;
1984         }
1985
1986         hash2 = map[split].hash;
1987         continued = hash2 == map[split - 1].hash;
1988         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1989                         (unsigned long)dx_get_block(frame->at),
1990                                         hash2, split, count-split));
1991
1992         /* Fancy dance to stay within two buffers */
1993         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
1994                               blocksize);
1995         de = dx_pack_dirents(dir, data1, blocksize);
1996         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1997                                            (char *) de,
1998                                            blocksize);
1999         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2000                                             (char *) de2,
2001                                             blocksize);
2002         if (csum_size) {
2003                 ext4_initialize_dirent_tail(*bh, blocksize);
2004                 ext4_initialize_dirent_tail(bh2, blocksize);
2005         }
2006
2007         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2008                         blocksize, 1));
2009         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2010                         blocksize, 1));
2011
2012         /* Which block gets the new entry? */
2013         if (hinfo->hash >= hash2) {
2014                 swap(*bh, bh2);
2015                 de = de2;
2016         }
2017         dx_insert_block(frame, hash2 + continued, newblock);
2018         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2019         if (err)
2020                 goto journal_error;
2021         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2022         if (err)
2023                 goto journal_error;
2024         brelse(bh2);
2025         dxtrace(dx_show_index("frame", frame->entries));
2026         return de;
2027
2028 journal_error:
2029         ext4_std_error(dir->i_sb, err);
2030 out:
2031         brelse(*bh);
2032         brelse(bh2);
2033         *bh = NULL;
2034         return ERR_PTR(err);
2035 }
2036
2037 int ext4_find_dest_de(struct inode *dir, struct buffer_head *bh,
2038                       void *buf, int buf_size,
2039                       struct ext4_filename *fname,
2040                       struct ext4_dir_entry_2 **dest_de)
2041 {
2042         struct ext4_dir_entry_2 *de;
2043         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2044         int nlen, rlen;
2045         unsigned int offset = 0;
2046         char *top;
2047
2048         de = buf;
2049         top = buf + buf_size - reclen;
2050         while ((char *) de <= top) {
2051                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2052                                          buf, buf_size, offset))
2053                         return -EFSCORRUPTED;
2054                 if (ext4_match(dir, fname, de))
2055                         return -EEXIST;
2056                 nlen = ext4_dir_rec_len(de->name_len, dir);
2057                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2058                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2059                         break;
2060                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2061                 offset += rlen;
2062         }
2063         if ((char *) de > top)
2064                 return -ENOSPC;
2065
2066         *dest_de = de;
2067         return 0;
2068 }
2069
2070 void ext4_insert_dentry(struct inode *dir,
2071                         struct inode *inode,
2072                         struct ext4_dir_entry_2 *de,
2073                         int buf_size,
2074                         struct ext4_filename *fname)
2075 {
2076
2077         int nlen, rlen;
2078
2079         nlen = ext4_dir_rec_len(de->name_len, dir);
2080         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2081         if (de->inode) {
2082                 struct ext4_dir_entry_2 *de1 =
2083                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2084                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2085                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2086                 de = de1;
2087         }
2088         de->file_type = EXT4_FT_UNKNOWN;
2089         de->inode = cpu_to_le32(inode->i_ino);
2090         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2091         de->name_len = fname_len(fname);
2092         memcpy(de->name, fname_name(fname), fname_len(fname));
2093         if (ext4_hash_in_dirent(dir)) {
2094                 struct dx_hash_info *hinfo = &fname->hinfo;
2095
2096                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2097                 EXT4_DIRENT_HASHES(de)->minor_hash =
2098                                                 cpu_to_le32(hinfo->minor_hash);
2099         }
2100 }
2101
2102 /*
2103  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2104  * it points to a directory entry which is guaranteed to be large
2105  * enough for new directory entry.  If de is NULL, then
2106  * add_dirent_to_buf will attempt search the directory block for
2107  * space.  It will return -ENOSPC if no space is available, and -EIO
2108  * and -EEXIST if directory entry already exists.
2109  */
2110 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2111                              struct inode *dir,
2112                              struct inode *inode, struct ext4_dir_entry_2 *de,
2113                              struct buffer_head *bh)
2114 {
2115         unsigned int    blocksize = dir->i_sb->s_blocksize;
2116         int             csum_size = 0;
2117         int             err, err2;
2118
2119         if (ext4_has_feature_metadata_csum(inode->i_sb))
2120                 csum_size = sizeof(struct ext4_dir_entry_tail);
2121
2122         if (!de) {
2123                 err = ext4_find_dest_de(dir, bh, bh->b_data,
2124                                         blocksize - csum_size, fname, &de);
2125                 if (err)
2126                         return err;
2127         }
2128         BUFFER_TRACE(bh, "get_write_access");
2129         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2130                                             EXT4_JTR_NONE);
2131         if (err) {
2132                 ext4_std_error(dir->i_sb, err);
2133                 return err;
2134         }
2135
2136         /* By now the buffer is marked for journaling */
2137         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2138
2139         /*
2140          * XXX shouldn't update any times until successful
2141          * completion of syscall, but too many callers depend
2142          * on this.
2143          *
2144          * XXX similarly, too many callers depend on
2145          * ext4_new_inode() setting the times, but error
2146          * recovery deletes the inode, so the worst that can
2147          * happen is that the times are slightly out of date
2148          * and/or different from the directory change time.
2149          */
2150         inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2151         ext4_update_dx_flag(dir);
2152         inode_inc_iversion(dir);
2153         err2 = ext4_mark_inode_dirty(handle, dir);
2154         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2155         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2156         if (err)
2157                 ext4_std_error(dir->i_sb, err);
2158         return err ? err : err2;
2159 }
2160
2161 static bool ext4_check_dx_root(struct inode *dir, struct dx_root *root)
2162 {
2163         struct fake_dirent *fde;
2164         const char *error_msg;
2165         unsigned int rlen;
2166         unsigned int blocksize = dir->i_sb->s_blocksize;
2167         char *blockend = (char *)root + dir->i_sb->s_blocksize;
2168
2169         fde = &root->dot;
2170         if (unlikely(fde->name_len != 1)) {
2171                 error_msg = "invalid name_len for '.'";
2172                 goto corrupted;
2173         }
2174         if (unlikely(strncmp(root->dot_name, ".", fde->name_len))) {
2175                 error_msg = "invalid name for '.'";
2176                 goto corrupted;
2177         }
2178         rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2179         if (unlikely((char *)fde + rlen >= blockend)) {
2180                 error_msg = "invalid rec_len for '.'";
2181                 goto corrupted;
2182         }
2183
2184         fde = &root->dotdot;
2185         if (unlikely(fde->name_len != 2)) {
2186                 error_msg = "invalid name_len for '..'";
2187                 goto corrupted;
2188         }
2189         if (unlikely(strncmp(root->dotdot_name, "..", fde->name_len))) {
2190                 error_msg = "invalid name for '..'";
2191                 goto corrupted;
2192         }
2193         rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2194         if (unlikely((char *)fde + rlen >= blockend)) {
2195                 error_msg = "invalid rec_len for '..'";
2196                 goto corrupted;
2197         }
2198
2199         return true;
2200
2201 corrupted:
2202         EXT4_ERROR_INODE(dir, "Corrupt dir, %s, running e2fsck is recommended",
2203                          error_msg);
2204         return false;
2205 }
2206
2207 /*
2208  * This converts a one block unindexed directory to a 3 block indexed
2209  * directory, and adds the dentry to the indexed directory.
2210  */
2211 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2212                             struct inode *dir,
2213                             struct inode *inode, struct buffer_head *bh)
2214 {
2215         struct buffer_head *bh2;
2216         struct dx_root  *root;
2217         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2218         struct dx_entry *entries;
2219         struct ext4_dir_entry_2 *de, *de2;
2220         char            *data2, *top;
2221         unsigned        len;
2222         int             retval;
2223         unsigned        blocksize;
2224         ext4_lblk_t  block;
2225         struct fake_dirent *fde;
2226         int csum_size = 0;
2227
2228         if (ext4_has_feature_metadata_csum(inode->i_sb))
2229                 csum_size = sizeof(struct ext4_dir_entry_tail);
2230
2231         blocksize =  dir->i_sb->s_blocksize;
2232         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2233         BUFFER_TRACE(bh, "get_write_access");
2234         retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2235                                                EXT4_JTR_NONE);
2236         if (retval) {
2237                 ext4_std_error(dir->i_sb, retval);
2238                 brelse(bh);
2239                 return retval;
2240         }
2241
2242         root = (struct dx_root *) bh->b_data;
2243         if (!ext4_check_dx_root(dir, root)) {
2244                 brelse(bh);
2245                 return -EFSCORRUPTED;
2246         }
2247
2248         /* The 0th block becomes the root, move the dirents out */
2249         fde = &root->dotdot;
2250         de = (struct ext4_dir_entry_2 *)((char *)fde +
2251                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2252         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2253
2254         /* Allocate new block for the 0th block's dirents */
2255         bh2 = ext4_append(handle, dir, &block);
2256         if (IS_ERR(bh2)) {
2257                 brelse(bh);
2258                 return PTR_ERR(bh2);
2259         }
2260         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2261         data2 = bh2->b_data;
2262
2263         memcpy(data2, de, len);
2264         memset(de, 0, len); /* wipe old data */
2265         de = (struct ext4_dir_entry_2 *) data2;
2266         top = data2 + len;
2267         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2268                 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2269                                         (char *)de - data2)) {
2270                         brelse(bh2);
2271                         brelse(bh);
2272                         return -EFSCORRUPTED;
2273                 }
2274                 de = de2;
2275         }
2276         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2277                                            (char *) de, blocksize);
2278
2279         if (csum_size)
2280                 ext4_initialize_dirent_tail(bh2, blocksize);
2281
2282         /* Initialize the root; the dot dirents already exist */
2283         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2284         de->rec_len = ext4_rec_len_to_disk(
2285                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2286         memset (&root->info, 0, sizeof(root->info));
2287         root->info.info_length = sizeof(root->info);
2288         if (ext4_hash_in_dirent(dir))
2289                 root->info.hash_version = DX_HASH_SIPHASH;
2290         else
2291                 root->info.hash_version =
2292                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2293
2294         entries = root->entries;
2295         dx_set_block(entries, 1);
2296         dx_set_count(entries, 1);
2297         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2298
2299         /* Initialize as for dx_probe */
2300         fname->hinfo.hash_version = root->info.hash_version;
2301         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2302                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2303         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2304
2305         /* casefolded encrypted hashes are computed on fname setup */
2306         if (!ext4_hash_in_dirent(dir)) {
2307                 int err = ext4fs_dirhash(dir, fname_name(fname),
2308                                          fname_len(fname), &fname->hinfo);
2309                 if (err < 0) {
2310                         brelse(bh2);
2311                         brelse(bh);
2312                         return err;
2313                 }
2314         }
2315         memset(frames, 0, sizeof(frames));
2316         frame = frames;
2317         frame->entries = entries;
2318         frame->at = entries;
2319         frame->bh = bh;
2320
2321         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2322         if (retval)
2323                 goto out_frames;
2324         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2325         if (retval)
2326                 goto out_frames;
2327
2328         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2329         if (IS_ERR(de)) {
2330                 retval = PTR_ERR(de);
2331                 goto out_frames;
2332         }
2333
2334         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2335 out_frames:
2336         /*
2337          * Even if the block split failed, we have to properly write
2338          * out all the changes we did so far. Otherwise we can end up
2339          * with corrupted filesystem.
2340          */
2341         if (retval)
2342                 ext4_mark_inode_dirty(handle, dir);
2343         dx_release(frames);
2344         brelse(bh2);
2345         return retval;
2346 }
2347
2348 /*
2349  *      ext4_add_entry()
2350  *
2351  * adds a file entry to the specified directory, using the same
2352  * semantics as ext4_find_entry(). It returns NULL if it failed.
2353  *
2354  * NOTE!! The inode part of 'de' is left at 0 - which means you
2355  * may not sleep between calling this and putting something into
2356  * the entry, as someone else might have used it while you slept.
2357  */
2358 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2359                           struct inode *inode)
2360 {
2361         struct inode *dir = d_inode(dentry->d_parent);
2362         struct buffer_head *bh = NULL;
2363         struct ext4_dir_entry_2 *de;
2364         struct super_block *sb;
2365         struct ext4_filename fname;
2366         int     retval;
2367         int     dx_fallback=0;
2368         unsigned blocksize;
2369         ext4_lblk_t block, blocks;
2370         int     csum_size = 0;
2371
2372         if (ext4_has_feature_metadata_csum(inode->i_sb))
2373                 csum_size = sizeof(struct ext4_dir_entry_tail);
2374
2375         sb = dir->i_sb;
2376         blocksize = sb->s_blocksize;
2377
2378         if (fscrypt_is_nokey_name(dentry))
2379                 return -ENOKEY;
2380
2381         if (!generic_ci_validate_strict_name(dir, &dentry->d_name))
2382                 return -EINVAL;
2383
2384         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2385         if (retval)
2386                 return retval;
2387
2388         if (ext4_has_inline_data(dir)) {
2389                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2390                 if (retval < 0)
2391                         goto out;
2392                 if (retval == 1) {
2393                         retval = 0;
2394                         goto out;
2395                 }
2396         }
2397
2398         if (is_dx(dir)) {
2399                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2400                 if (!retval || (retval != ERR_BAD_DX_DIR))
2401                         goto out;
2402                 /* Can we just ignore htree data? */
2403                 if (ext4_has_feature_metadata_csum(sb)) {
2404                         EXT4_ERROR_INODE(dir,
2405                                 "Directory has corrupted htree index.");
2406                         retval = -EFSCORRUPTED;
2407                         goto out;
2408                 }
2409                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2410                 dx_fallback++;
2411                 retval = ext4_mark_inode_dirty(handle, dir);
2412                 if (unlikely(retval))
2413                         goto out;
2414         }
2415         blocks = dir->i_size >> sb->s_blocksize_bits;
2416         for (block = 0; block < blocks; block++) {
2417                 bh = ext4_read_dirblock(dir, block, DIRENT);
2418                 if (bh == NULL) {
2419                         bh = ext4_bread(handle, dir, block,
2420                                         EXT4_GET_BLOCKS_CREATE);
2421                         goto add_to_new_block;
2422                 }
2423                 if (IS_ERR(bh)) {
2424                         retval = PTR_ERR(bh);
2425                         bh = NULL;
2426                         goto out;
2427                 }
2428                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2429                                            NULL, bh);
2430                 if (retval != -ENOSPC)
2431                         goto out;
2432
2433                 if (blocks == 1 && !dx_fallback &&
2434                     ext4_has_feature_dir_index(sb)) {
2435                         retval = make_indexed_dir(handle, &fname, dir,
2436                                                   inode, bh);
2437                         bh = NULL; /* make_indexed_dir releases bh */
2438                         goto out;
2439                 }
2440                 brelse(bh);
2441         }
2442         bh = ext4_append(handle, dir, &block);
2443 add_to_new_block:
2444         if (IS_ERR(bh)) {
2445                 retval = PTR_ERR(bh);
2446                 bh = NULL;
2447                 goto out;
2448         }
2449         de = (struct ext4_dir_entry_2 *) bh->b_data;
2450         de->inode = 0;
2451         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2452
2453         if (csum_size)
2454                 ext4_initialize_dirent_tail(bh, blocksize);
2455
2456         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2457 out:
2458         ext4_fname_free_filename(&fname);
2459         brelse(bh);
2460         if (retval == 0)
2461                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2462         return retval;
2463 }
2464
2465 /*
2466  * Returns 0 for success, or a negative error value
2467  */
2468 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2469                              struct inode *dir, struct inode *inode)
2470 {
2471         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2472         struct dx_entry *entries, *at;
2473         struct buffer_head *bh;
2474         struct super_block *sb = dir->i_sb;
2475         struct ext4_dir_entry_2 *de;
2476         int restart;
2477         int err;
2478
2479 again:
2480         restart = 0;
2481         frame = dx_probe(fname, dir, NULL, frames);
2482         if (IS_ERR(frame))
2483                 return PTR_ERR(frame);
2484         entries = frame->entries;
2485         at = frame->at;
2486         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2487         if (IS_ERR(bh)) {
2488                 err = PTR_ERR(bh);
2489                 bh = NULL;
2490                 goto cleanup;
2491         }
2492
2493         BUFFER_TRACE(bh, "get_write_access");
2494         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2495         if (err)
2496                 goto journal_error;
2497
2498         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2499         if (err != -ENOSPC)
2500                 goto cleanup;
2501
2502         err = 0;
2503         /* Block full, should compress but for now just split */
2504         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2505                        dx_get_count(entries), dx_get_limit(entries)));
2506         /* Need to split index? */
2507         if (dx_get_count(entries) == dx_get_limit(entries)) {
2508                 ext4_lblk_t newblock;
2509                 int levels = frame - frames + 1;
2510                 unsigned int icount;
2511                 int add_level = 1;
2512                 struct dx_entry *entries2;
2513                 struct dx_node *node2;
2514                 struct buffer_head *bh2;
2515
2516                 while (frame > frames) {
2517                         if (dx_get_count((frame - 1)->entries) <
2518                             dx_get_limit((frame - 1)->entries)) {
2519                                 add_level = 0;
2520                                 break;
2521                         }
2522                         frame--; /* split higher index block */
2523                         at = frame->at;
2524                         entries = frame->entries;
2525                         restart = 1;
2526                 }
2527                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2528                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2529                                          "reach max htree level :%d",
2530                                          dir->i_ino, levels);
2531                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2532                                 ext4_warning(sb, "Large directory feature is "
2533                                                  "not enabled on this "
2534                                                  "filesystem");
2535                         }
2536                         err = -ENOSPC;
2537                         goto cleanup;
2538                 }
2539                 icount = dx_get_count(entries);
2540                 bh2 = ext4_append(handle, dir, &newblock);
2541                 if (IS_ERR(bh2)) {
2542                         err = PTR_ERR(bh2);
2543                         goto cleanup;
2544                 }
2545                 node2 = (struct dx_node *)(bh2->b_data);
2546                 entries2 = node2->entries;
2547                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2548                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2549                                                            sb->s_blocksize);
2550                 BUFFER_TRACE(frame->bh, "get_write_access");
2551                 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2552                                                     EXT4_JTR_NONE);
2553                 if (err) {
2554                         brelse(bh2);
2555                         goto journal_error;
2556                 }
2557                 if (!add_level) {
2558                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2559                         unsigned hash2 = dx_get_hash(entries + icount1);
2560                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2561                                        icount1, icount2));
2562
2563                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2564                         err = ext4_journal_get_write_access(handle, sb,
2565                                                             (frame - 1)->bh,
2566                                                             EXT4_JTR_NONE);
2567                         if (err) {
2568                                 brelse(bh2);
2569                                 goto journal_error;
2570                         }
2571
2572                         memcpy((char *) entries2, (char *) (entries + icount1),
2573                                icount2 * sizeof(struct dx_entry));
2574                         dx_set_count(entries, icount1);
2575                         dx_set_count(entries2, icount2);
2576                         dx_set_limit(entries2, dx_node_limit(dir));
2577
2578                         /* Which index block gets the new entry? */
2579                         if (at - entries >= icount1) {
2580                                 frame->at = at - entries - icount1 + entries2;
2581                                 frame->entries = entries = entries2;
2582                                 swap(frame->bh, bh2);
2583                         }
2584                         dx_insert_block((frame - 1), hash2, newblock);
2585                         dxtrace(dx_show_index("node", frame->entries));
2586                         dxtrace(dx_show_index("node",
2587                                ((struct dx_node *) bh2->b_data)->entries));
2588                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2589                         if (err) {
2590                                 brelse(bh2);
2591                                 goto journal_error;
2592                         }
2593                         brelse (bh2);
2594                         err = ext4_handle_dirty_dx_node(handle, dir,
2595                                                    (frame - 1)->bh);
2596                         if (err)
2597                                 goto journal_error;
2598                         err = ext4_handle_dirty_dx_node(handle, dir,
2599                                                         frame->bh);
2600                         if (restart || err)
2601                                 goto journal_error;
2602                 } else {
2603                         struct dx_root *dxroot;
2604                         memcpy((char *) entries2, (char *) entries,
2605                                icount * sizeof(struct dx_entry));
2606                         dx_set_limit(entries2, dx_node_limit(dir));
2607
2608                         /* Set up root */
2609                         dx_set_count(entries, 1);
2610                         dx_set_block(entries + 0, newblock);
2611                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2612                         dxroot->info.indirect_levels += 1;
2613                         dxtrace(printk(KERN_DEBUG
2614                                        "Creating %d level index...\n",
2615                                        dxroot->info.indirect_levels));
2616                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2617                         if (err) {
2618                                 brelse(bh2);
2619                                 goto journal_error;
2620                         }
2621                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2622                         brelse(bh2);
2623                         restart = 1;
2624                         goto journal_error;
2625                 }
2626         }
2627         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2628         if (IS_ERR(de)) {
2629                 err = PTR_ERR(de);
2630                 goto cleanup;
2631         }
2632         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2633         goto cleanup;
2634
2635 journal_error:
2636         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2637 cleanup:
2638         brelse(bh);
2639         dx_release(frames);
2640         /* @restart is true means htree-path has been changed, we need to
2641          * repeat dx_probe() to find out valid htree-path
2642          */
2643         if (restart && err == 0)
2644                 goto again;
2645         return err;
2646 }
2647
2648 /*
2649  * ext4_generic_delete_entry deletes a directory entry by merging it
2650  * with the previous entry
2651  */
2652 int ext4_generic_delete_entry(struct inode *dir,
2653                               struct ext4_dir_entry_2 *de_del,
2654                               struct buffer_head *bh,
2655                               void *entry_buf,
2656                               int buf_size,
2657                               int csum_size)
2658 {
2659         struct ext4_dir_entry_2 *de, *pde;
2660         unsigned int blocksize = dir->i_sb->s_blocksize;
2661         int i;
2662
2663         i = 0;
2664         pde = NULL;
2665         de = entry_buf;
2666         while (i < buf_size - csum_size) {
2667                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2668                                          entry_buf, buf_size, i))
2669                         return -EFSCORRUPTED;
2670                 if (de == de_del)  {
2671                         if (pde) {
2672                                 pde->rec_len = ext4_rec_len_to_disk(
2673                                         ext4_rec_len_from_disk(pde->rec_len,
2674                                                                blocksize) +
2675                                         ext4_rec_len_from_disk(de->rec_len,
2676                                                                blocksize),
2677                                         blocksize);
2678
2679                                 /* wipe entire dir_entry */
2680                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2681                                                                 blocksize));
2682                         } else {
2683                                 /* wipe dir_entry excluding the rec_len field */
2684                                 de->inode = 0;
2685                                 memset(&de->name_len, 0,
2686                                         ext4_rec_len_from_disk(de->rec_len,
2687                                                                 blocksize) -
2688                                         offsetof(struct ext4_dir_entry_2,
2689                                                                 name_len));
2690                         }
2691
2692                         inode_inc_iversion(dir);
2693                         return 0;
2694                 }
2695                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2696                 pde = de;
2697                 de = ext4_next_entry(de, blocksize);
2698         }
2699         return -ENOENT;
2700 }
2701
2702 static int ext4_delete_entry(handle_t *handle,
2703                              struct inode *dir,
2704                              struct ext4_dir_entry_2 *de_del,
2705                              struct buffer_head *bh)
2706 {
2707         int err, csum_size = 0;
2708
2709         if (ext4_has_inline_data(dir)) {
2710                 int has_inline_data = 1;
2711                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2712                                                &has_inline_data);
2713                 if (has_inline_data)
2714                         return err;
2715         }
2716
2717         if (ext4_has_feature_metadata_csum(dir->i_sb))
2718                 csum_size = sizeof(struct ext4_dir_entry_tail);
2719
2720         BUFFER_TRACE(bh, "get_write_access");
2721         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2722                                             EXT4_JTR_NONE);
2723         if (unlikely(err))
2724                 goto out;
2725
2726         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2727                                         dir->i_sb->s_blocksize, csum_size);
2728         if (err)
2729                 goto out;
2730
2731         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2732         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2733         if (unlikely(err))
2734                 goto out;
2735
2736         return 0;
2737 out:
2738         if (err != -ENOENT)
2739                 ext4_std_error(dir->i_sb, err);
2740         return err;
2741 }
2742
2743 /*
2744  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2745  * since this indicates that nlinks count was previously 1 to avoid overflowing
2746  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2747  * that subdirectory link counts are not being maintained accurately.
2748  *
2749  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2750  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2751  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2752  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2753  */
2754 static void ext4_inc_count(struct inode *inode)
2755 {
2756         inc_nlink(inode);
2757         if (is_dx(inode) &&
2758             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2759                 set_nlink(inode, 1);
2760 }
2761
2762 /*
2763  * If a directory had nlink == 1, then we should let it be 1. This indicates
2764  * directory has >EXT4_LINK_MAX subdirs.
2765  */
2766 static void ext4_dec_count(struct inode *inode)
2767 {
2768         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2769                 drop_nlink(inode);
2770 }
2771
2772
2773 /*
2774  * Add non-directory inode to a directory. On success, the inode reference is
2775  * consumed by dentry is instantiation. This is also indicated by clearing of
2776  * *inodep pointer. On failure, the caller is responsible for dropping the
2777  * inode reference in the safe context.
2778  */
2779 static int ext4_add_nondir(handle_t *handle,
2780                 struct dentry *dentry, struct inode **inodep)
2781 {
2782         struct inode *dir = d_inode(dentry->d_parent);
2783         struct inode *inode = *inodep;
2784         int err = ext4_add_entry(handle, dentry, inode);
2785         if (!err) {
2786                 err = ext4_mark_inode_dirty(handle, inode);
2787                 if (IS_DIRSYNC(dir))
2788                         ext4_handle_sync(handle);
2789                 d_instantiate_new(dentry, inode);
2790                 *inodep = NULL;
2791                 return err;
2792         }
2793         drop_nlink(inode);
2794         ext4_mark_inode_dirty(handle, inode);
2795         ext4_orphan_add(handle, inode);
2796         unlock_new_inode(inode);
2797         return err;
2798 }
2799
2800 /*
2801  * By the time this is called, we already have created
2802  * the directory cache entry for the new file, but it
2803  * is so far negative - it has no inode.
2804  *
2805  * If the create succeeds, we fill in the inode information
2806  * with d_instantiate().
2807  */
2808 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2809                        struct dentry *dentry, umode_t mode, bool excl)
2810 {
2811         handle_t *handle;
2812         struct inode *inode;
2813         int err, credits, retries = 0;
2814
2815         err = dquot_initialize(dir);
2816         if (err)
2817                 return err;
2818
2819         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2820                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2821 retry:
2822         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2823                                             0, NULL, EXT4_HT_DIR, credits);
2824         handle = ext4_journal_current_handle();
2825         err = PTR_ERR(inode);
2826         if (!IS_ERR(inode)) {
2827                 inode->i_op = &ext4_file_inode_operations;
2828                 inode->i_fop = &ext4_file_operations;
2829                 ext4_set_aops(inode);
2830                 err = ext4_add_nondir(handle, dentry, &inode);
2831                 if (!err)
2832                         ext4_fc_track_create(handle, dentry);
2833         }
2834         if (handle)
2835                 ext4_journal_stop(handle);
2836         if (!IS_ERR_OR_NULL(inode))
2837                 iput(inode);
2838         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2839                 goto retry;
2840         return err;
2841 }
2842
2843 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2844                       struct dentry *dentry, umode_t mode, dev_t rdev)
2845 {
2846         handle_t *handle;
2847         struct inode *inode;
2848         int err, credits, retries = 0;
2849
2850         err = dquot_initialize(dir);
2851         if (err)
2852                 return err;
2853
2854         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2855                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2856 retry:
2857         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2858                                             0, NULL, EXT4_HT_DIR, credits);
2859         handle = ext4_journal_current_handle();
2860         err = PTR_ERR(inode);
2861         if (!IS_ERR(inode)) {
2862                 init_special_inode(inode, inode->i_mode, rdev);
2863                 inode->i_op = &ext4_special_inode_operations;
2864                 err = ext4_add_nondir(handle, dentry, &inode);
2865                 if (!err)
2866                         ext4_fc_track_create(handle, dentry);
2867         }
2868         if (handle)
2869                 ext4_journal_stop(handle);
2870         if (!IS_ERR_OR_NULL(inode))
2871                 iput(inode);
2872         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2873                 goto retry;
2874         return err;
2875 }
2876
2877 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2878                         struct file *file, umode_t mode)
2879 {
2880         handle_t *handle;
2881         struct inode *inode;
2882         int err, retries = 0;
2883
2884         err = dquot_initialize(dir);
2885         if (err)
2886                 return err;
2887
2888 retry:
2889         inode = ext4_new_inode_start_handle(idmap, dir, mode,
2890                                             NULL, 0, NULL,
2891                                             EXT4_HT_DIR,
2892                         EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
2893                           4 + EXT4_XATTR_TRANS_BLOCKS);
2894         handle = ext4_journal_current_handle();
2895         err = PTR_ERR(inode);
2896         if (!IS_ERR(inode)) {
2897                 inode->i_op = &ext4_file_inode_operations;
2898                 inode->i_fop = &ext4_file_operations;
2899                 ext4_set_aops(inode);
2900                 d_tmpfile(file, inode);
2901                 err = ext4_orphan_add(handle, inode);
2902                 if (err)
2903                         goto err_unlock_inode;
2904                 mark_inode_dirty(inode);
2905                 unlock_new_inode(inode);
2906         }
2907         if (handle)
2908                 ext4_journal_stop(handle);
2909         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2910                 goto retry;
2911         return finish_open_simple(file, err);
2912 err_unlock_inode:
2913         ext4_journal_stop(handle);
2914         unlock_new_inode(inode);
2915         return err;
2916 }
2917
2918 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2919                           struct ext4_dir_entry_2 *de,
2920                           int blocksize, int csum_size,
2921                           unsigned int parent_ino, int dotdot_real_len)
2922 {
2923         de->inode = cpu_to_le32(inode->i_ino);
2924         de->name_len = 1;
2925         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2926                                            blocksize);
2927         strcpy(de->name, ".");
2928         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2929
2930         de = ext4_next_entry(de, blocksize);
2931         de->inode = cpu_to_le32(parent_ino);
2932         de->name_len = 2;
2933         if (!dotdot_real_len)
2934                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2935                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2936                                         blocksize);
2937         else
2938                 de->rec_len = ext4_rec_len_to_disk(
2939                                         ext4_dir_rec_len(de->name_len, NULL),
2940                                         blocksize);
2941         strcpy(de->name, "..");
2942         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2943
2944         return ext4_next_entry(de, blocksize);
2945 }
2946
2947 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2948                              struct inode *inode)
2949 {
2950         struct buffer_head *dir_block = NULL;
2951         struct ext4_dir_entry_2 *de;
2952         ext4_lblk_t block = 0;
2953         unsigned int blocksize = dir->i_sb->s_blocksize;
2954         int csum_size = 0;
2955         int err;
2956
2957         if (ext4_has_feature_metadata_csum(dir->i_sb))
2958                 csum_size = sizeof(struct ext4_dir_entry_tail);
2959
2960         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2961                 err = ext4_try_create_inline_dir(handle, dir, inode);
2962                 if (err < 0 && err != -ENOSPC)
2963                         goto out;
2964                 if (!err)
2965                         goto out;
2966         }
2967
2968         inode->i_size = 0;
2969         dir_block = ext4_append(handle, inode, &block);
2970         if (IS_ERR(dir_block))
2971                 return PTR_ERR(dir_block);
2972         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2973         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2974         set_nlink(inode, 2);
2975         if (csum_size)
2976                 ext4_initialize_dirent_tail(dir_block, blocksize);
2977
2978         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2979         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2980         if (err)
2981                 goto out;
2982         set_buffer_verified(dir_block);
2983 out:
2984         brelse(dir_block);
2985         return err;
2986 }
2987
2988 static struct dentry *ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2989                                  struct dentry *dentry, umode_t mode)
2990 {
2991         handle_t *handle;
2992         struct inode *inode;
2993         int err, err2 = 0, credits, retries = 0;
2994
2995         if (EXT4_DIR_LINK_MAX(dir))
2996                 return ERR_PTR(-EMLINK);
2997
2998         err = dquot_initialize(dir);
2999         if (err)
3000                 return ERR_PTR(err);
3001
3002         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3003                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3004 retry:
3005         inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3006                                             &dentry->d_name,
3007                                             0, NULL, EXT4_HT_DIR, credits);
3008         handle = ext4_journal_current_handle();
3009         err = PTR_ERR(inode);
3010         if (IS_ERR(inode))
3011                 goto out_stop;
3012
3013         inode->i_op = &ext4_dir_inode_operations;
3014         inode->i_fop = &ext4_dir_operations;
3015         err = ext4_init_new_dir(handle, dir, inode);
3016         if (err)
3017                 goto out_clear_inode;
3018         err = ext4_mark_inode_dirty(handle, inode);
3019         if (!err)
3020                 err = ext4_add_entry(handle, dentry, inode);
3021         if (err) {
3022 out_clear_inode:
3023                 clear_nlink(inode);
3024                 ext4_orphan_add(handle, inode);
3025                 unlock_new_inode(inode);
3026                 err2 = ext4_mark_inode_dirty(handle, inode);
3027                 if (unlikely(err2))
3028                         err = err2;
3029                 ext4_journal_stop(handle);
3030                 iput(inode);
3031                 goto out_retry;
3032         }
3033         ext4_inc_count(dir);
3034
3035         ext4_update_dx_flag(dir);
3036         err = ext4_mark_inode_dirty(handle, dir);
3037         if (err)
3038                 goto out_clear_inode;
3039         d_instantiate_new(dentry, inode);
3040         ext4_fc_track_create(handle, dentry);
3041         if (IS_DIRSYNC(dir))
3042                 ext4_handle_sync(handle);
3043
3044 out_stop:
3045         if (handle)
3046                 ext4_journal_stop(handle);
3047 out_retry:
3048         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3049                 goto retry;
3050         return ERR_PTR(err);
3051 }
3052
3053 /*
3054  * routine to check that the specified directory is empty (for rmdir)
3055  */
3056 bool ext4_empty_dir(struct inode *inode)
3057 {
3058         unsigned int offset;
3059         struct buffer_head *bh;
3060         struct ext4_dir_entry_2 *de;
3061         struct super_block *sb;
3062
3063         if (ext4_has_inline_data(inode)) {
3064                 int has_inline_data = 1;
3065                 int ret;
3066
3067                 ret = empty_inline_dir(inode, &has_inline_data);
3068                 if (has_inline_data)
3069                         return ret;
3070         }
3071
3072         sb = inode->i_sb;
3073         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3074                                         ext4_dir_rec_len(2, NULL)) {
3075                 EXT4_ERROR_INODE(inode, "invalid size");
3076                 return false;
3077         }
3078         bh = ext4_read_dirblock(inode, 0, EITHER);
3079         if (IS_ERR(bh))
3080                 return false;
3081
3082         de = (struct ext4_dir_entry_2 *) bh->b_data;
3083         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3084                                  0) ||
3085             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3086                 ext4_warning_inode(inode, "directory missing '.'");
3087                 brelse(bh);
3088                 return false;
3089         }
3090         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3091         de = ext4_next_entry(de, sb->s_blocksize);
3092         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3093                                  offset) ||
3094             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3095                 ext4_warning_inode(inode, "directory missing '..'");
3096                 brelse(bh);
3097                 return false;
3098         }
3099         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3100         while (offset < inode->i_size) {
3101                 if (!(offset & (sb->s_blocksize - 1))) {
3102                         unsigned int lblock;
3103                         brelse(bh);
3104                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3105                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3106                         if (bh == NULL) {
3107                                 offset += sb->s_blocksize;
3108                                 continue;
3109                         }
3110                         if (IS_ERR(bh))
3111                                 return false;
3112                 }
3113                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3114                                         (offset & (sb->s_blocksize - 1)));
3115                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3116                                          bh->b_data, bh->b_size, offset) ||
3117                     le32_to_cpu(de->inode)) {
3118                         brelse(bh);
3119                         return false;
3120                 }
3121                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3122         }
3123         brelse(bh);
3124         return true;
3125 }
3126
3127 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3128 {
3129         int retval;
3130         struct inode *inode;
3131         struct buffer_head *bh;
3132         struct ext4_dir_entry_2 *de;
3133         handle_t *handle = NULL;
3134
3135         retval = ext4_emergency_state(dir->i_sb);
3136         if (unlikely(retval))
3137                 return retval;
3138
3139         /* Initialize quotas before so that eventual writes go in
3140          * separate transaction */
3141         retval = dquot_initialize(dir);
3142         if (retval)
3143                 return retval;
3144         retval = dquot_initialize(d_inode(dentry));
3145         if (retval)
3146                 return retval;
3147
3148         retval = -ENOENT;
3149         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3150         if (IS_ERR(bh))
3151                 return PTR_ERR(bh);
3152         if (!bh)
3153                 goto end_rmdir;
3154
3155         inode = d_inode(dentry);
3156
3157         retval = -EFSCORRUPTED;
3158         if (le32_to_cpu(de->inode) != inode->i_ino)
3159                 goto end_rmdir;
3160
3161         retval = -ENOTEMPTY;
3162         if (!ext4_empty_dir(inode))
3163                 goto end_rmdir;
3164
3165         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3166                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3167         if (IS_ERR(handle)) {
3168                 retval = PTR_ERR(handle);
3169                 handle = NULL;
3170                 goto end_rmdir;
3171         }
3172
3173         if (IS_DIRSYNC(dir))
3174                 ext4_handle_sync(handle);
3175
3176         retval = ext4_delete_entry(handle, dir, de, bh);
3177         if (retval)
3178                 goto end_rmdir;
3179         if (!EXT4_DIR_LINK_EMPTY(inode))
3180                 ext4_warning_inode(inode,
3181                              "empty directory '%.*s' has too many links (%u)",
3182                              dentry->d_name.len, dentry->d_name.name,
3183                              inode->i_nlink);
3184         inode_inc_iversion(inode);
3185         clear_nlink(inode);
3186         /* There's no need to set i_disksize: the fact that i_nlink is
3187          * zero will ensure that the right thing happens during any
3188          * recovery. */
3189         inode->i_size = 0;
3190         ext4_orphan_add(handle, inode);
3191         inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3192         inode_set_ctime_current(inode);
3193         retval = ext4_mark_inode_dirty(handle, inode);
3194         if (retval)
3195                 goto end_rmdir;
3196         ext4_dec_count(dir);
3197         ext4_update_dx_flag(dir);
3198         ext4_fc_track_unlink(handle, dentry);
3199         retval = ext4_mark_inode_dirty(handle, dir);
3200
3201         /* VFS negative dentries are incompatible with Encoding and
3202          * Case-insensitiveness. Eventually we'll want avoid
3203          * invalidating the dentries here, alongside with returning the
3204          * negative dentries at ext4_lookup(), when it is better
3205          * supported by the VFS for the CI case.
3206          */
3207         if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3208                 d_invalidate(dentry);
3209
3210 end_rmdir:
3211         brelse(bh);
3212         if (handle)
3213                 ext4_journal_stop(handle);
3214         return retval;
3215 }
3216
3217 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3218                   struct inode *inode,
3219                   struct dentry *dentry /* NULL during fast_commit recovery */)
3220 {
3221         int retval = -ENOENT;
3222         struct buffer_head *bh;
3223         struct ext4_dir_entry_2 *de;
3224         handle_t *handle;
3225         int skip_remove_dentry = 0;
3226
3227         /*
3228          * Keep this outside the transaction; it may have to set up the
3229          * directory's encryption key, which isn't GFP_NOFS-safe.
3230          */
3231         bh = ext4_find_entry(dir, d_name, &de, NULL);
3232         if (IS_ERR(bh))
3233                 return PTR_ERR(bh);
3234
3235         if (!bh)
3236                 return -ENOENT;
3237
3238         if (le32_to_cpu(de->inode) != inode->i_ino) {
3239                 /*
3240                  * It's okay if we find dont find dentry which matches
3241                  * the inode. That's because it might have gotten
3242                  * renamed to a different inode number
3243                  */
3244                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3245                         skip_remove_dentry = 1;
3246                 else
3247                         goto out_bh;
3248         }
3249
3250         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3251                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3252         if (IS_ERR(handle)) {
3253                 retval = PTR_ERR(handle);
3254                 goto out_bh;
3255         }
3256
3257         if (IS_DIRSYNC(dir))
3258                 ext4_handle_sync(handle);
3259
3260         if (!skip_remove_dentry) {
3261                 retval = ext4_delete_entry(handle, dir, de, bh);
3262                 if (retval)
3263                         goto out_handle;
3264                 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3265                 ext4_update_dx_flag(dir);
3266                 retval = ext4_mark_inode_dirty(handle, dir);
3267                 if (retval)
3268                         goto out_handle;
3269         } else {
3270                 retval = 0;
3271         }
3272         if (inode->i_nlink == 0)
3273                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3274                                    d_name->len, d_name->name);
3275         else
3276                 drop_nlink(inode);
3277         if (!inode->i_nlink)
3278                 ext4_orphan_add(handle, inode);
3279         inode_set_ctime_current(inode);
3280         retval = ext4_mark_inode_dirty(handle, inode);
3281         if (dentry && !retval)
3282                 ext4_fc_track_unlink(handle, dentry);
3283 out_handle:
3284         ext4_journal_stop(handle);
3285 out_bh:
3286         brelse(bh);
3287         return retval;
3288 }
3289
3290 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3291 {
3292         int retval;
3293
3294         retval = ext4_emergency_state(dir->i_sb);
3295         if (unlikely(retval))
3296                 return retval;
3297
3298         trace_ext4_unlink_enter(dir, dentry);
3299         /*
3300          * Initialize quotas before so that eventual writes go
3301          * in separate transaction
3302          */
3303         retval = dquot_initialize(dir);
3304         if (retval)
3305                 goto out_trace;
3306         retval = dquot_initialize(d_inode(dentry));
3307         if (retval)
3308                 goto out_trace;
3309
3310         retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3311
3312         /* VFS negative dentries are incompatible with Encoding and
3313          * Case-insensitiveness. Eventually we'll want avoid
3314          * invalidating the dentries here, alongside with returning the
3315          * negative dentries at ext4_lookup(), when it is  better
3316          * supported by the VFS for the CI case.
3317          */
3318         if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3319                 d_invalidate(dentry);
3320
3321 out_trace:
3322         trace_ext4_unlink_exit(dentry, retval);
3323         return retval;
3324 }
3325
3326 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3327                                    struct fscrypt_str *disk_link)
3328 {
3329         struct buffer_head *bh;
3330         char *kaddr;
3331         int err = 0;
3332
3333         bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3334         if (IS_ERR(bh))
3335                 return PTR_ERR(bh);
3336
3337         BUFFER_TRACE(bh, "get_write_access");
3338         err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3339         if (err)
3340                 goto out;
3341
3342         kaddr = (char *)bh->b_data;
3343         memcpy(kaddr, disk_link->name, disk_link->len);
3344         inode->i_size = disk_link->len - 1;
3345         EXT4_I(inode)->i_disksize = inode->i_size;
3346         err = ext4_handle_dirty_metadata(handle, inode, bh);
3347 out:
3348         brelse(bh);
3349         return err;
3350 }
3351
3352 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3353                         struct dentry *dentry, const char *symname)
3354 {
3355         handle_t *handle;
3356         struct inode *inode;
3357         int err, len = strlen(symname);
3358         int credits;
3359         struct fscrypt_str disk_link;
3360         int retries = 0;
3361
3362         err = ext4_emergency_state(dir->i_sb);
3363         if (unlikely(err))
3364                 return err;
3365
3366         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3367                                       &disk_link);
3368         if (err)
3369                 return err;
3370
3371         err = dquot_initialize(dir);
3372         if (err)
3373                 return err;
3374
3375         /*
3376          * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3377          * directory. +3 for inode, inode bitmap, group descriptor allocation.
3378          * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3379          * modification.
3380          */
3381         credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3382                   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3383 retry:
3384         inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3385                                             &dentry->d_name, 0, NULL,
3386                                             EXT4_HT_DIR, credits);
3387         handle = ext4_journal_current_handle();
3388         if (IS_ERR(inode)) {
3389                 if (handle)
3390                         ext4_journal_stop(handle);
3391                 err = PTR_ERR(inode);
3392                 goto out_retry;
3393         }
3394
3395         if (IS_ENCRYPTED(inode)) {
3396                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3397                 if (err)
3398                         goto err_drop_inode;
3399                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3400         } else {
3401                 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3402                         inode->i_op = &ext4_symlink_inode_operations;
3403                 } else {
3404                         inode->i_op = &ext4_fast_symlink_inode_operations;
3405                 }
3406         }
3407
3408         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3409                 /* alloc symlink block and fill it */
3410                 err = ext4_init_symlink_block(handle, inode, &disk_link);
3411                 if (err)
3412                         goto err_drop_inode;
3413         } else {
3414                 /* clear the extent format for fast symlink */
3415                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3416                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3417                        disk_link.len);
3418                 inode->i_size = disk_link.len - 1;
3419                 EXT4_I(inode)->i_disksize = inode->i_size;
3420                 if (!IS_ENCRYPTED(inode))
3421                         inode_set_cached_link(inode, (char *)&EXT4_I(inode)->i_data,
3422                                               inode->i_size);
3423         }
3424         err = ext4_add_nondir(handle, dentry, &inode);
3425         if (handle)
3426                 ext4_journal_stop(handle);
3427         iput(inode);
3428         goto out_retry;
3429
3430 err_drop_inode:
3431         clear_nlink(inode);
3432         ext4_mark_inode_dirty(handle, inode);
3433         ext4_orphan_add(handle, inode);
3434         unlock_new_inode(inode);
3435         if (handle)
3436                 ext4_journal_stop(handle);
3437         iput(inode);
3438 out_retry:
3439         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3440                 goto retry;
3441         if (disk_link.name != (unsigned char *)symname)
3442                 kfree(disk_link.name);
3443         return err;
3444 }
3445
3446 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3447 {
3448         handle_t *handle;
3449         int err, retries = 0;
3450 retry:
3451         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3452                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3453                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3454         if (IS_ERR(handle))
3455                 return PTR_ERR(handle);
3456
3457         if (IS_DIRSYNC(dir))
3458                 ext4_handle_sync(handle);
3459
3460         inode_set_ctime_current(inode);
3461         ext4_inc_count(inode);
3462         ihold(inode);
3463
3464         err = ext4_add_entry(handle, dentry, inode);
3465         if (!err) {
3466                 err = ext4_mark_inode_dirty(handle, inode);
3467                 /* this can happen only for tmpfile being
3468                  * linked the first time
3469                  */
3470                 if (inode->i_nlink == 1)
3471                         ext4_orphan_del(handle, inode);
3472                 d_instantiate(dentry, inode);
3473                 ext4_fc_track_link(handle, dentry);
3474         } else {
3475                 drop_nlink(inode);
3476                 iput(inode);
3477         }
3478         ext4_journal_stop(handle);
3479         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3480                 goto retry;
3481         return err;
3482 }
3483
3484 static int ext4_link(struct dentry *old_dentry,
3485                      struct inode *dir, struct dentry *dentry)
3486 {
3487         struct inode *inode = d_inode(old_dentry);
3488         int err;
3489
3490         if (inode->i_nlink >= EXT4_LINK_MAX)
3491                 return -EMLINK;
3492
3493         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3494         if (err)
3495                 return err;
3496
3497         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3498             (!projid_eq(EXT4_I(dir)->i_projid,
3499                         EXT4_I(old_dentry->d_inode)->i_projid)))
3500                 return -EXDEV;
3501
3502         err = dquot_initialize(dir);
3503         if (err)
3504                 return err;
3505         return __ext4_link(dir, inode, dentry);
3506 }
3507
3508 /*
3509  * Try to find buffer head where contains the parent block.
3510  * It should be the inode block if it is inlined or the 1st block
3511  * if it is a normal dir.
3512  */
3513 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3514                                         struct inode *inode,
3515                                         int *retval,
3516                                         struct ext4_dir_entry_2 **parent_de,
3517                                         int *inlined)
3518 {
3519         struct buffer_head *bh;
3520
3521         if (!ext4_has_inline_data(inode)) {
3522                 struct ext4_dir_entry_2 *de;
3523                 unsigned int offset;
3524
3525                 bh = ext4_read_dirblock(inode, 0, EITHER);
3526                 if (IS_ERR(bh)) {
3527                         *retval = PTR_ERR(bh);
3528                         return NULL;
3529                 }
3530
3531                 de = (struct ext4_dir_entry_2 *) bh->b_data;
3532                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3533                                          bh->b_size, 0) ||
3534                     le32_to_cpu(de->inode) != inode->i_ino ||
3535                     strcmp(".", de->name)) {
3536                         EXT4_ERROR_INODE(inode, "directory missing '.'");
3537                         brelse(bh);
3538                         *retval = -EFSCORRUPTED;
3539                         return NULL;
3540                 }
3541                 offset = ext4_rec_len_from_disk(de->rec_len,
3542                                                 inode->i_sb->s_blocksize);
3543                 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3544                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3545                                          bh->b_size, offset) ||
3546                     le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3547                         EXT4_ERROR_INODE(inode, "directory missing '..'");
3548                         brelse(bh);
3549                         *retval = -EFSCORRUPTED;
3550                         return NULL;
3551                 }
3552                 *parent_de = de;
3553
3554                 return bh;
3555         }
3556
3557         *inlined = 1;
3558         return ext4_get_first_inline_block(inode, parent_de, retval);
3559 }
3560
3561 struct ext4_renament {
3562         struct inode *dir;
3563         struct dentry *dentry;
3564         struct inode *inode;
3565         bool is_dir;
3566         int dir_nlink_delta;
3567
3568         /* entry for "dentry" */
3569         struct buffer_head *bh;
3570         struct ext4_dir_entry_2 *de;
3571         int inlined;
3572
3573         /* entry for ".." in inode if it's a directory */
3574         struct buffer_head *dir_bh;
3575         struct ext4_dir_entry_2 *parent_de;
3576         int dir_inlined;
3577 };
3578
3579 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3580 {
3581         int retval;
3582
3583         ent->is_dir = true;
3584         if (!is_cross)
3585                 return 0;
3586
3587         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3588                                               &retval, &ent->parent_de,
3589                                               &ent->dir_inlined);
3590         if (!ent->dir_bh)
3591                 return retval;
3592         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3593                 return -EFSCORRUPTED;
3594         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3595         return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3596                                              ent->dir_bh, EXT4_JTR_NONE);
3597 }
3598
3599 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3600                                   unsigned dir_ino)
3601 {
3602         int retval;
3603
3604         if (!ent->dir_bh)
3605                 return 0;
3606
3607         ent->parent_de->inode = cpu_to_le32(dir_ino);
3608         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3609         if (!ent->dir_inlined) {
3610                 if (is_dx(ent->inode)) {
3611                         retval = ext4_handle_dirty_dx_node(handle,
3612                                                            ent->inode,
3613                                                            ent->dir_bh);
3614                 } else {
3615                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3616                                                             ent->dir_bh);
3617                 }
3618         } else {
3619                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3620         }
3621         if (retval) {
3622                 ext4_std_error(ent->dir->i_sb, retval);
3623                 return retval;
3624         }
3625         return 0;
3626 }
3627
3628 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3629                        unsigned ino, unsigned file_type)
3630 {
3631         int retval, retval2;
3632
3633         BUFFER_TRACE(ent->bh, "get write access");
3634         retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3635                                                EXT4_JTR_NONE);
3636         if (retval)
3637                 return retval;
3638         ent->de->inode = cpu_to_le32(ino);
3639         if (ext4_has_feature_filetype(ent->dir->i_sb))
3640                 ent->de->file_type = file_type;
3641         inode_inc_iversion(ent->dir);
3642         inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3643         retval = ext4_mark_inode_dirty(handle, ent->dir);
3644         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3645         if (!ent->inlined) {
3646                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3647                 if (unlikely(retval2)) {
3648                         ext4_std_error(ent->dir->i_sb, retval2);
3649                         return retval2;
3650                 }
3651         }
3652         return retval;
3653 }
3654
3655 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3656                           unsigned ino, unsigned file_type)
3657 {
3658         struct ext4_renament old = *ent;
3659         int retval = 0;
3660
3661         /*
3662          * old->de could have moved from under us during make indexed dir,
3663          * so the old->de may no longer valid and need to find it again
3664          * before reset old inode info.
3665          */
3666         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3667                                  &old.inlined);
3668         if (IS_ERR(old.bh))
3669                 retval = PTR_ERR(old.bh);
3670         if (!old.bh)
3671                 retval = -ENOENT;
3672         if (retval) {
3673                 ext4_std_error(old.dir->i_sb, retval);
3674                 return;
3675         }
3676
3677         ext4_setent(handle, &old, ino, file_type);
3678         brelse(old.bh);
3679 }
3680
3681 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3682                                   const struct qstr *d_name)
3683 {
3684         int retval = -ENOENT;
3685         struct buffer_head *bh;
3686         struct ext4_dir_entry_2 *de;
3687
3688         bh = ext4_find_entry(dir, d_name, &de, NULL);
3689         if (IS_ERR(bh))
3690                 return PTR_ERR(bh);
3691         if (bh) {
3692                 retval = ext4_delete_entry(handle, dir, de, bh);
3693                 brelse(bh);
3694         }
3695         return retval;
3696 }
3697
3698 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3699                                int force_reread)
3700 {
3701         int retval;
3702         /*
3703          * ent->de could have moved from under us during htree split, so make
3704          * sure that we are deleting the right entry.  We might also be pointing
3705          * to a stale entry in the unused part of ent->bh so just checking inum
3706          * and the name isn't enough.
3707          */
3708         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3709             ent->de->name_len != ent->dentry->d_name.len ||
3710             strncmp(ent->de->name, ent->dentry->d_name.name,
3711                     ent->de->name_len) ||
3712             force_reread) {
3713                 retval = ext4_find_delete_entry(handle, ent->dir,
3714                                                 &ent->dentry->d_name);
3715         } else {
3716                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3717                 if (retval == -ENOENT) {
3718                         retval = ext4_find_delete_entry(handle, ent->dir,
3719                                                         &ent->dentry->d_name);
3720                 }
3721         }
3722
3723         if (retval) {
3724                 ext4_warning_inode(ent->dir,
3725                                    "Deleting old file: nlink %d, error=%d",
3726                                    ent->dir->i_nlink, retval);
3727         }
3728 }
3729
3730 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3731 {
3732         if (ent->dir_nlink_delta) {
3733                 if (ent->dir_nlink_delta == -1)
3734                         ext4_dec_count(ent->dir);
3735                 else
3736                         ext4_inc_count(ent->dir);
3737                 ext4_mark_inode_dirty(handle, ent->dir);
3738         }
3739 }
3740
3741 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3742                                               struct ext4_renament *ent,
3743                                               int credits, handle_t **h)
3744 {
3745         struct inode *wh;
3746         handle_t *handle;
3747         int retries = 0;
3748
3749         /*
3750          * for inode block, sb block, group summaries,
3751          * and inode bitmap
3752          */
3753         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3754                     EXT4_XATTR_TRANS_BLOCKS + 4);
3755 retry:
3756         wh = ext4_new_inode_start_handle(idmap, ent->dir,
3757                                          S_IFCHR | WHITEOUT_MODE,
3758                                          &ent->dentry->d_name, 0, NULL,
3759                                          EXT4_HT_DIR, credits);
3760
3761         handle = ext4_journal_current_handle();
3762         if (IS_ERR(wh)) {
3763                 if (handle)
3764                         ext4_journal_stop(handle);
3765                 if (PTR_ERR(wh) == -ENOSPC &&
3766                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3767                         goto retry;
3768         } else {
3769                 *h = handle;
3770                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3771                 wh->i_op = &ext4_special_inode_operations;
3772         }
3773         return wh;
3774 }
3775
3776 /*
3777  * Anybody can rename anything with this: the permission checks are left to the
3778  * higher-level routines.
3779  *
3780  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3781  * while new_{dentry,inode) refers to the destination dentry/inode
3782  * This comes from rename(const char *oldpath, const char *newpath)
3783  */
3784 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3785                        struct dentry *old_dentry, struct inode *new_dir,
3786                        struct dentry *new_dentry, unsigned int flags)
3787 {
3788         handle_t *handle = NULL;
3789         struct ext4_renament old = {
3790                 .dir = old_dir,
3791                 .dentry = old_dentry,
3792                 .inode = d_inode(old_dentry),
3793         };
3794         struct ext4_renament new = {
3795                 .dir = new_dir,
3796                 .dentry = new_dentry,
3797                 .inode = d_inode(new_dentry),
3798         };
3799         int force_reread;
3800         int retval;
3801         struct inode *whiteout = NULL;
3802         int credits;
3803         u8 old_file_type;
3804
3805         if (new.inode && new.inode->i_nlink == 0) {
3806                 EXT4_ERROR_INODE(new.inode,
3807                                  "target of rename is already freed");
3808                 return -EFSCORRUPTED;
3809         }
3810
3811         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3812             (!projid_eq(EXT4_I(new_dir)->i_projid,
3813                         EXT4_I(old_dentry->d_inode)->i_projid)))
3814                 return -EXDEV;
3815
3816         retval = dquot_initialize(old.dir);
3817         if (retval)
3818                 return retval;
3819         retval = dquot_initialize(old.inode);
3820         if (retval)
3821                 return retval;
3822         retval = dquot_initialize(new.dir);
3823         if (retval)
3824                 return retval;
3825
3826         /* Initialize quotas before so that eventual writes go
3827          * in separate transaction */
3828         if (new.inode) {
3829                 retval = dquot_initialize(new.inode);
3830                 if (retval)
3831                         return retval;
3832         }
3833
3834         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3835                                  &old.inlined);
3836         if (IS_ERR(old.bh))
3837                 return PTR_ERR(old.bh);
3838
3839         /*
3840          *  Check for inode number is _not_ due to possible IO errors.
3841          *  We might rmdir the source, keep it as pwd of some process
3842          *  and merrily kill the link to whatever was created under the
3843          *  same name. Goodbye sticky bit ;-<
3844          */
3845         retval = -ENOENT;
3846         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3847                 goto release_bh;
3848
3849         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3850                                  &new.de, &new.inlined);
3851         if (IS_ERR(new.bh)) {
3852                 retval = PTR_ERR(new.bh);
3853                 new.bh = NULL;
3854                 goto release_bh;
3855         }
3856         if (new.bh) {
3857                 if (!new.inode) {
3858                         brelse(new.bh);
3859                         new.bh = NULL;
3860                 }
3861         }
3862         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3863                 ext4_alloc_da_blocks(old.inode);
3864
3865         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3866                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3867         if (!(flags & RENAME_WHITEOUT)) {
3868                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3869                 if (IS_ERR(handle)) {
3870                         retval = PTR_ERR(handle);
3871                         goto release_bh;
3872                 }
3873         } else {
3874                 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3875                 if (IS_ERR(whiteout)) {
3876                         retval = PTR_ERR(whiteout);
3877                         goto release_bh;
3878                 }
3879         }
3880
3881         old_file_type = old.de->file_type;
3882         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3883                 ext4_handle_sync(handle);
3884
3885         if (S_ISDIR(old.inode->i_mode)) {
3886                 if (new.inode) {
3887                         retval = -ENOTEMPTY;
3888                         if (!ext4_empty_dir(new.inode))
3889                                 goto end_rename;
3890                 } else {
3891                         retval = -EMLINK;
3892                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3893                                 goto end_rename;
3894                 }
3895                 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3896                 if (retval)
3897                         goto end_rename;
3898         }
3899         /*
3900          * If we're renaming a file within an inline_data dir and adding or
3901          * setting the new dirent causes a conversion from inline_data to
3902          * extents/blockmap, we need to force the dirent delete code to
3903          * re-read the directory, or else we end up trying to delete a dirent
3904          * from what is now the extent tree root (or a block map).
3905          */
3906         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3907                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3908
3909         if (whiteout) {
3910                 /*
3911                  * Do this before adding a new entry, so the old entry is sure
3912                  * to be still pointing to the valid old entry.
3913                  */
3914                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3915                                      EXT4_FT_CHRDEV);
3916                 if (retval)
3917                         goto end_rename;
3918                 retval = ext4_mark_inode_dirty(handle, whiteout);
3919                 if (unlikely(retval))
3920                         goto end_rename;
3921
3922         }
3923         if (!new.bh) {
3924                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3925                 if (retval)
3926                         goto end_rename;
3927         } else {
3928                 retval = ext4_setent(handle, &new,
3929                                      old.inode->i_ino, old_file_type);
3930                 if (retval)
3931                         goto end_rename;
3932         }
3933         if (force_reread)
3934                 force_reread = !ext4_test_inode_flag(new.dir,
3935                                                      EXT4_INODE_INLINE_DATA);
3936
3937         /*
3938          * Like most other Unix systems, set the ctime for inodes on a
3939          * rename.
3940          */
3941         inode_set_ctime_current(old.inode);
3942         retval = ext4_mark_inode_dirty(handle, old.inode);
3943         if (unlikely(retval))
3944                 goto end_rename;
3945
3946         if (!whiteout) {
3947                 /*
3948                  * ok, that's it
3949                  */
3950                 ext4_rename_delete(handle, &old, force_reread);
3951         }
3952
3953         if (new.inode) {
3954                 ext4_dec_count(new.inode);
3955                 inode_set_ctime_current(new.inode);
3956         }
3957         inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3958         ext4_update_dx_flag(old.dir);
3959         if (old.is_dir) {
3960                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3961                 if (retval)
3962                         goto end_rename;
3963
3964                 ext4_dec_count(old.dir);
3965                 if (new.inode) {
3966                         /* checked ext4_empty_dir above, can't have another
3967                          * parent, ext4_dec_count() won't work for many-linked
3968                          * dirs */
3969                         clear_nlink(new.inode);
3970                 } else {
3971                         ext4_inc_count(new.dir);
3972                         ext4_update_dx_flag(new.dir);
3973                         retval = ext4_mark_inode_dirty(handle, new.dir);
3974                         if (unlikely(retval))
3975                                 goto end_rename;
3976                 }
3977         }
3978         retval = ext4_mark_inode_dirty(handle, old.dir);
3979         if (unlikely(retval))
3980                 goto end_rename;
3981
3982         if (old.is_dir) {
3983                 /*
3984                  * We disable fast commits here that's because the
3985                  * replay code is not yet capable of changing dot dot
3986                  * dirents in directories.
3987                  */
3988                 ext4_fc_mark_ineligible(old.inode->i_sb,
3989                         EXT4_FC_REASON_RENAME_DIR, handle);
3990         } else {
3991                 struct super_block *sb = old.inode->i_sb;
3992
3993                 if (new.inode)
3994                         ext4_fc_track_unlink(handle, new.dentry);
3995                 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
3996                     !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
3997                     !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
3998                         __ext4_fc_track_link(handle, old.inode, new.dentry);
3999                         __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4000                         if (whiteout)
4001                                 __ext4_fc_track_create(handle, whiteout,
4002                                                        old.dentry);
4003                 }
4004         }
4005
4006         if (new.inode) {
4007                 retval = ext4_mark_inode_dirty(handle, new.inode);
4008                 if (unlikely(retval))
4009                         goto end_rename;
4010                 if (!new.inode->i_nlink)
4011                         ext4_orphan_add(handle, new.inode);
4012         }
4013         retval = 0;
4014
4015 end_rename:
4016         if (whiteout) {
4017                 if (retval) {
4018                         ext4_resetent(handle, &old,
4019                                       old.inode->i_ino, old_file_type);
4020                         drop_nlink(whiteout);
4021                         ext4_mark_inode_dirty(handle, whiteout);
4022                         ext4_orphan_add(handle, whiteout);
4023                 }
4024                 unlock_new_inode(whiteout);
4025                 ext4_journal_stop(handle);
4026                 iput(whiteout);
4027         } else {
4028                 ext4_journal_stop(handle);
4029         }
4030 release_bh:
4031         brelse(old.dir_bh);
4032         brelse(old.bh);
4033         brelse(new.bh);
4034
4035         return retval;
4036 }
4037
4038 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4039                              struct inode *new_dir, struct dentry *new_dentry)
4040 {
4041         handle_t *handle = NULL;
4042         struct ext4_renament old = {
4043                 .dir = old_dir,
4044                 .dentry = old_dentry,
4045                 .inode = d_inode(old_dentry),
4046         };
4047         struct ext4_renament new = {
4048                 .dir = new_dir,
4049                 .dentry = new_dentry,
4050                 .inode = d_inode(new_dentry),
4051         };
4052         u8 new_file_type;
4053         int retval;
4054
4055         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4056              !projid_eq(EXT4_I(new_dir)->i_projid,
4057                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
4058             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4059              !projid_eq(EXT4_I(old_dir)->i_projid,
4060                         EXT4_I(new_dentry->d_inode)->i_projid)))
4061                 return -EXDEV;
4062
4063         retval = dquot_initialize(old.dir);
4064         if (retval)
4065                 return retval;
4066         retval = dquot_initialize(new.dir);
4067         if (retval)
4068                 return retval;
4069
4070         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4071                                  &old.de, &old.inlined);
4072         if (IS_ERR(old.bh))
4073                 return PTR_ERR(old.bh);
4074         /*
4075          *  Check for inode number is _not_ due to possible IO errors.
4076          *  We might rmdir the source, keep it as pwd of some process
4077          *  and merrily kill the link to whatever was created under the
4078          *  same name. Goodbye sticky bit ;-<
4079          */
4080         retval = -ENOENT;
4081         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4082                 goto end_rename;
4083
4084         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4085                                  &new.de, &new.inlined);
4086         if (IS_ERR(new.bh)) {
4087                 retval = PTR_ERR(new.bh);
4088                 new.bh = NULL;
4089                 goto end_rename;
4090         }
4091
4092         /* RENAME_EXCHANGE case: old *and* new must both exist */
4093         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4094                 goto end_rename;
4095
4096         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4097                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4098                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4099         if (IS_ERR(handle)) {
4100                 retval = PTR_ERR(handle);
4101                 handle = NULL;
4102                 goto end_rename;
4103         }
4104
4105         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4106                 ext4_handle_sync(handle);
4107
4108         if (S_ISDIR(old.inode->i_mode)) {
4109                 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4110                 if (retval)
4111                         goto end_rename;
4112         }
4113         if (S_ISDIR(new.inode->i_mode)) {
4114                 retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4115                 if (retval)
4116                         goto end_rename;
4117         }
4118
4119         /*
4120          * Other than the special case of overwriting a directory, parents'
4121          * nlink only needs to be modified if this is a cross directory rename.
4122          */
4123         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4124                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4125                 new.dir_nlink_delta = -old.dir_nlink_delta;
4126                 retval = -EMLINK;
4127                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4128                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4129                         goto end_rename;
4130         }
4131
4132         new_file_type = new.de->file_type;
4133         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4134         if (retval)
4135                 goto end_rename;
4136
4137         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4138         if (retval)
4139                 goto end_rename;
4140
4141         /*
4142          * Like most other Unix systems, set the ctime for inodes on a
4143          * rename.
4144          */
4145         inode_set_ctime_current(old.inode);
4146         inode_set_ctime_current(new.inode);
4147         retval = ext4_mark_inode_dirty(handle, old.inode);
4148         if (unlikely(retval))
4149                 goto end_rename;
4150         retval = ext4_mark_inode_dirty(handle, new.inode);
4151         if (unlikely(retval))
4152                 goto end_rename;
4153         ext4_fc_mark_ineligible(new.inode->i_sb,
4154                                 EXT4_FC_REASON_CROSS_RENAME, handle);
4155         if (old.dir_bh) {
4156                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4157                 if (retval)
4158                         goto end_rename;
4159         }
4160         if (new.dir_bh) {
4161                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4162                 if (retval)
4163                         goto end_rename;
4164         }
4165         ext4_update_dir_count(handle, &old);
4166         ext4_update_dir_count(handle, &new);
4167         retval = 0;
4168
4169 end_rename:
4170         brelse(old.dir_bh);
4171         brelse(new.dir_bh);
4172         brelse(old.bh);
4173         brelse(new.bh);
4174         if (handle)
4175                 ext4_journal_stop(handle);
4176         return retval;
4177 }
4178
4179 static int ext4_rename2(struct mnt_idmap *idmap,
4180                         struct inode *old_dir, struct dentry *old_dentry,
4181                         struct inode *new_dir, struct dentry *new_dentry,
4182                         unsigned int flags)
4183 {
4184         int err;
4185
4186         err = ext4_emergency_state(old_dir->i_sb);
4187         if (unlikely(err))
4188                 return err;
4189
4190         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4191                 return -EINVAL;
4192
4193         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4194                                      flags);
4195         if (err)
4196                 return err;
4197
4198         if (flags & RENAME_EXCHANGE) {
4199                 return ext4_cross_rename(old_dir, old_dentry,
4200                                          new_dir, new_dentry);
4201         }
4202
4203         return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4204 }
4205
4206 /*
4207  * directories can handle most operations...
4208  */
4209 const struct inode_operations ext4_dir_inode_operations = {
4210         .create         = ext4_create,
4211         .lookup         = ext4_lookup,
4212         .link           = ext4_link,
4213         .unlink         = ext4_unlink,
4214         .symlink        = ext4_symlink,
4215         .mkdir          = ext4_mkdir,
4216         .rmdir          = ext4_rmdir,
4217         .mknod          = ext4_mknod,
4218         .tmpfile        = ext4_tmpfile,
4219         .rename         = ext4_rename2,
4220         .setattr        = ext4_setattr,
4221         .getattr        = ext4_getattr,
4222         .listxattr      = ext4_listxattr,
4223         .get_inode_acl  = ext4_get_acl,
4224         .set_acl        = ext4_set_acl,
4225         .fiemap         = ext4_fiemap,
4226         .fileattr_get   = ext4_fileattr_get,
4227         .fileattr_set   = ext4_fileattr_set,
4228 };
4229
4230 const struct inode_operations ext4_special_inode_operations = {
4231         .setattr        = ext4_setattr,
4232         .getattr        = ext4_getattr,
4233         .listxattr      = ext4_listxattr,
4234         .get_inode_acl  = ext4_get_acl,
4235         .set_acl        = ext4_set_acl,
4236 };