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         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 155                 ext4_error_inode(inode, func, line, block,
 156                                  "Directory hole found for htree %s block",
 157                                  (type == INDEX) ? "index" : "leaf");
 158                 return ERR_PTR(-EFSCORRUPTED);
 159         }
 160         if (!bh)
 161                 return NULL;
 162         dirent = (struct ext4_dir_entry *) bh->b_data;
 163         /* Determine whether or not we have an index block */
 164         if (is_dx(inode)) {
 165                 if (block == 0)
 166                         is_dx_block = 1;
 167                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 168                                                 inode->i_sb->s_blocksize) ==
 169                          inode->i_sb->s_blocksize)
 170                         is_dx_block = 1;
 171         }
 172         if (!is_dx_block && type == INDEX) {
 173                 ext4_error_inode(inode, func, line, block,
 174                        "directory leaf block found instead of index block");
 175                 brelse(bh);
 176                 return ERR_PTR(-EFSCORRUPTED);
 177         }
 178         if (!ext4_has_metadata_csum(inode->i_sb) ||
 179             buffer_verified(bh))
 180                 return bh;
 181
 182         /*
 183          * An empty leaf block can get mistaken for a index block; for
 184          * this reason, we can only check the index checksum when the
 185          * caller is sure it should be an index block.
 186          */
 187         if (is_dx_block && type == INDEX) {
 188                 if (ext4_dx_csum_verify(inode, dirent) &&
 189                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 190                         set_buffer_verified(bh);
 191                 else {
 192                         ext4_error_inode_err(inode, func, line, block,
 193                                              EFSBADCRC,
 194                                              "Directory index failed checksum");
 195                         brelse(bh);
 196                         return ERR_PTR(-EFSBADCRC);
 197                 }
 198         }
 199         if (!is_dx_block) {
 200                 if (ext4_dirblock_csum_verify(inode, bh) &&
 201                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 202                         set_buffer_verified(bh);
 203                 else {
 204                         ext4_error_inode_err(inode, func, line, block,
 205                                              EFSBADCRC,
 206                                              "Directory block failed checksum");
 207                         brelse(bh);
 208                         return ERR_PTR(-EFSBADCRC);
 209                 }
 210         }
 211         return bh;
 212 }
 213
 214 #ifdef DX_DEBUG
 215 #define dxtrace(command) command
 216 #else
 217 #define dxtrace(command)
 218 #endif
 219
 220 struct fake_dirent
 221 {
 222         __le32 inode;
 223         __le16 rec_len;
 224         u8 name_len;
 225         u8 file_type;
 226 };
 227
 228 struct dx_countlimit
 229 {
 230         __le16 limit;
 231         __le16 count;
 232 };
 233
 234 struct dx_entry
 235 {
 236         __le32 hash;
 237         __le32 block;
 238 };
 239
 240 /*
 241  * dx_root_info is laid out so that if it should somehow get overlaid by a
 242  * dirent the two low bits of the hash version will be zero.  Therefore, the
 243  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 244  */
 245
 246 struct dx_root
 247 {
 248         struct fake_dirent dot;
 249         char dot_name[4];
 250         struct fake_dirent dotdot;
 251         char dotdot_name[4];
 252         struct dx_root_info
 253         {
 254                 __le32 reserved_zero;
 255                 u8 hash_version;
 256                 u8 info_length; /* 8 */
 257                 u8 indirect_levels;
 258                 u8 unused_flags;
 259         }
 260         info;
 261         struct dx_entry entries[];
 262 };
 263
 264 struct dx_node
 265 {
 266         struct fake_dirent fake;
 267         struct dx_entry entries[];
 268 };
 269
 270
 271 struct dx_frame
 272 {
 273         struct buffer_head *bh;
 274         struct dx_entry *entries;
 275         struct dx_entry *at;
 276 };
 277
 278 struct dx_map_entry
 279 {
 280         u32 hash;
 281         u16 offs;
 282         u16 size;
 283 };
 284
 285 /*
 286  * This goes at the end of each htree block.
 287  */
 288 struct dx_tail {
 289         u32 dt_reserved;
 290         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 291 };
 292
 293 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 294 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 295 static inline unsigned dx_get_hash(struct dx_entry *entry);
 296 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 297 static unsigned dx_get_count(struct dx_entry *entries);
 298 static unsigned dx_get_limit(struct dx_entry *entries);
 299 static void dx_set_count(struct dx_entry *entries, unsigned value);
 300 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 301 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 302 static unsigned dx_node_limit(struct inode *dir);
 303 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 304                                  struct inode *dir,
 305                                  struct dx_hash_info *hinfo,
 306                                  struct dx_frame *frame);
 307 static void dx_release(struct dx_frame *frames);
 308 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
 309                        struct dx_hash_info *hinfo,
 310                        struct dx_map_entry *map_tail);
 311 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 312 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
 313                                         char *to, struct dx_map_entry *offsets,
 314                                         int count, unsigned int blocksize);
 315 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 316                                                 unsigned int blocksize);
 317 static void dx_insert_block(struct dx_frame *frame,
 318                                         u32 hash, ext4_lblk_t block);
 319 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 320                                  struct dx_frame *frame,
 321                                  struct dx_frame *frames,
 322                                  __u32 *start_hash);
 323 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 324                 struct ext4_filename *fname,
 325                 struct ext4_dir_entry_2 **res_dir);
 326 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 327                              struct inode *dir, struct inode *inode);
 328
 329 /* checksumming functions */
 330 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 331                                  unsigned int blocksize)
 332 {
 333         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 334
 335         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 336         t->det_rec_len = ext4_rec_len_to_disk(
 337                         sizeof(struct ext4_dir_entry_tail), blocksize);
 338         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 339 }
 340
 341 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 342 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 343                                                    struct buffer_head *bh)
 344 {
 345         struct ext4_dir_entry_tail *t;
 346         int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
 347
 348 #ifdef PARANOID
 349         struct ext4_dir_entry *d, *top;
 350
 351         d = (struct ext4_dir_entry *)bh->b_data;
 352         top = (struct ext4_dir_entry *)(bh->b_data +
 353                 (blocksize - sizeof(struct ext4_dir_entry_tail)));
 354         while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
 355                 d = (struct ext4_dir_entry *)(((void *)d) +
 356                     ext4_rec_len_from_disk(d->rec_len, blocksize));
 357
 358         if (d != top)
 359                 return NULL;
 360
 361         t = (struct ext4_dir_entry_tail *)d;
 362 #else
 363         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 364 #endif
 365
 366         if (t->det_reserved_zero1 ||
 367             (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
 368              sizeof(struct ext4_dir_entry_tail)) ||
 369             t->det_reserved_zero2 ||
 370             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 371                 return NULL;
 372
 373         return t;
 374 }
 375
 376 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 377 {
 378         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 379         struct ext4_inode_info *ei = EXT4_I(inode);
 380         __u32 csum;
 381
 382         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 383         return cpu_to_le32(csum);
 384 }
 385
 386 #define warn_no_space_for_csum(inode)                                   \
 387         __warn_no_space_for_csum((inode), __func__, __LINE__)
 388
 389 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 390                                      unsigned int line)
 391 {
 392         __ext4_warning_inode(inode, func, line,
 393                 "No space for directory leaf checksum. Please run e2fsck -D.");
 394 }
 395
 396 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 397 {
 398         struct ext4_dir_entry_tail *t;
 399
 400         if (!ext4_has_metadata_csum(inode->i_sb))
 401                 return 1;
 402
 403         t = get_dirent_tail(inode, bh);
 404         if (!t) {
 405                 warn_no_space_for_csum(inode);
 406                 return 0;
 407         }
 408
 409         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 410                                                   (char *)t - bh->b_data))
 411                 return 0;
 412
 413         return 1;
 414 }
 415
 416 static void ext4_dirblock_csum_set(struct inode *inode,
 417                                  struct buffer_head *bh)
 418 {
 419         struct ext4_dir_entry_tail *t;
 420
 421         if (!ext4_has_metadata_csum(inode->i_sb))
 422                 return;
 423
 424         t = get_dirent_tail(inode, bh);
 425         if (!t) {
 426                 warn_no_space_for_csum(inode);
 427                 return;
 428         }
 429
 430         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 431                                              (char *)t - bh->b_data);
 432 }
 433
 434 int ext4_handle_dirty_dirblock(handle_t *handle,
 435                                struct inode *inode,
 436                                struct buffer_head *bh)
 437 {
 438         ext4_dirblock_csum_set(inode, bh);
 439         return ext4_handle_dirty_metadata(handle, inode, bh);
 440 }
 441
 442 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 443                                                struct ext4_dir_entry *dirent,
 444                                                int *offset)
 445 {
 446         struct ext4_dir_entry *dp;
 447         struct dx_root_info *root;
 448         int count_offset;
 449         int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
 450         unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
 451
 452         if (rlen == blocksize)
 453                 count_offset = 8;
 454         else if (rlen == 12) {
 455                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 456                 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
 457                         return NULL;
 458                 root = (struct dx_root_info *)(((void *)dp + 12));
 459                 if (root->reserved_zero ||
 460                     root->info_length != sizeof(struct dx_root_info))
 461                         return NULL;
 462                 count_offset = 32;
 463         } else
 464                 return NULL;
 465
 466         if (offset)
 467                 *offset = count_offset;
 468         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 469 }
 470
 471 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 472                            int count_offset, int count, struct dx_tail *t)
 473 {
 474         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 475         struct ext4_inode_info *ei = EXT4_I(inode);
 476         __u32 csum;
 477         int size;
 478         __u32 dummy_csum = 0;
 479         int offset = offsetof(struct dx_tail, dt_checksum);
 480
 481         size = count_offset + (count * sizeof(struct dx_entry));
 482         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 483         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 484         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 485
 486         return cpu_to_le32(csum);
 487 }
 488
 489 static int ext4_dx_csum_verify(struct inode *inode,
 490                                struct ext4_dir_entry *dirent)
 491 {
 492         struct dx_countlimit *c;
 493         struct dx_tail *t;
 494         int count_offset, limit, count;
 495
 496         if (!ext4_has_metadata_csum(inode->i_sb))
 497                 return 1;
 498
 499         c = get_dx_countlimit(inode, dirent, &count_offset);
 500         if (!c) {
 501                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 502                 return 0;
 503         }
 504         limit = le16_to_cpu(c->limit);
 505         count = le16_to_cpu(c->count);
 506         if (count_offset + (limit * sizeof(struct dx_entry)) >
 507             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 508                 warn_no_space_for_csum(inode);
 509                 return 0;
 510         }
 511         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 512
 513         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 514                                             count, t))
 515                 return 0;
 516         return 1;
 517 }
 518
 519 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 520 {
 521         struct dx_countlimit *c;
 522         struct dx_tail *t;
 523         int count_offset, limit, count;
 524
 525         if (!ext4_has_metadata_csum(inode->i_sb))
 526                 return;
 527
 528         c = get_dx_countlimit(inode, dirent, &count_offset);
 529         if (!c) {
 530                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 531                 return;
 532         }
 533         limit = le16_to_cpu(c->limit);
 534         count = le16_to_cpu(c->count);
 535         if (count_offset + (limit * sizeof(struct dx_entry)) >
 536             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 537                 warn_no_space_for_csum(inode);
 538                 return;
 539         }
 540         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 541
 542         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 543 }
 544
 545 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 546                                             struct inode *inode,
 547                                             struct buffer_head *bh)
 548 {
 549         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 550         return ext4_handle_dirty_metadata(handle, inode, bh);
 551 }
 552
 553 /*
 554  * p is at least 6 bytes before the end of page
 555  */
 556 static inline struct ext4_dir_entry_2 *
 557 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 558 {
 559         return (struct ext4_dir_entry_2 *)((char *)p +
 560                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 561 }
 562
 563 /*
 564  * Future: use high four bits of block for coalesce-on-delete flags
 565  * Mask them off for now.
 566  */
 567
 568 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 569 {
 570         return le32_to_cpu(entry->block) & 0x0fffffff;
 571 }
 572
 573 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 574 {
 575         entry->block = cpu_to_le32(value);
 576 }
 577
 578 static inline unsigned dx_get_hash(struct dx_entry *entry)
 579 {
 580         return le32_to_cpu(entry->hash);
 581 }
 582
 583 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 584 {
 585         entry->hash = cpu_to_le32(value);
 586 }
 587
 588 static inline unsigned dx_get_count(struct dx_entry *entries)
 589 {
 590         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 591 }
 592
 593 static inline unsigned dx_get_limit(struct dx_entry *entries)
 594 {
 595         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 596 }
 597
 598 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 599 {
 600         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 601 }
 602
 603 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 604 {
 605         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 606 }
 607
 608 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 609 {
 610         unsigned int entry_space = dir->i_sb->s_blocksize -
 611                         ext4_dir_rec_len(1, NULL) -
 612                         ext4_dir_rec_len(2, NULL) - infosize;
 613
 614         if (ext4_has_metadata_csum(dir->i_sb))
 615                 entry_space -= sizeof(struct dx_tail);
 616         return entry_space / sizeof(struct dx_entry);
 617 }
 618
 619 static inline unsigned dx_node_limit(struct inode *dir)
 620 {
 621         unsigned int entry_space = dir->i_sb->s_blocksize -
 622                         ext4_dir_rec_len(0, dir);
 623
 624         if (ext4_has_metadata_csum(dir->i_sb))
 625                 entry_space -= sizeof(struct dx_tail);
 626         return entry_space / sizeof(struct dx_entry);
 627 }
 628
 629 /*
 630  * Debug
 631  */
 632 #ifdef DX_DEBUG
 633 static void dx_show_index(char * label, struct dx_entry *entries)
 634 {
 635         int i, n = dx_get_count (entries);
 636         printk(KERN_DEBUG "%s index", label);
 637         for (i = 0; i < n; i++) {
 638                 printk(KERN_CONT " %x->%lu",
 639                        i ? dx_get_hash(entries + i) : 0,
 640                        (unsigned long)dx_get_block(entries + i));
 641         }
 642         printk(KERN_CONT "\n");
 643 }
 644
 645 struct stats
 646 {
 647         unsigned names;
 648         unsigned space;
 649         unsigned bcount;
 650 };
 651
 652 static struct stats dx_show_leaf(struct inode *dir,
 653                                 struct dx_hash_info *hinfo,
 654                                 struct ext4_dir_entry_2 *de,
 655                                 int size, int show_names)
 656 {
 657         unsigned names = 0, space = 0;
 658         char *base = (char *) de;
 659         struct dx_hash_info h = *hinfo;
 660
 661         printk("names: ");
 662         while ((char *) de < base + size)
 663         {
 664                 if (de->inode)
 665                 {
 666                         if (show_names)
 667                         {
 668 #ifdef CONFIG_FS_ENCRYPTION
 669                                 int len;
 670                                 char *name;
 671                                 struct fscrypt_str fname_crypto_str =
 672                                         FSTR_INIT(NULL, 0);
 673                                 int res = 0;
 674
 675                                 name  = de->name;
 676                                 len = de->name_len;
 677                                 if (!IS_ENCRYPTED(dir)) {
 678                                         /* Directory is not encrypted */
 679                                         (void) ext4fs_dirhash(dir, de->name,
 680                                                 de->name_len, &h);
 681                                         printk("%*.s:(U)%x.%u ", len,
 682                                                name, h.hash,
 683                                                (unsigned) ((char *) de
 684                                                            - base));
 685                                 } else {
 686                                         struct fscrypt_str de_name =
 687                                                 FSTR_INIT(name, len);
 688
 689                                         /* Directory is encrypted */
 690                                         res = fscrypt_fname_alloc_buffer(
 691                                                 len, &fname_crypto_str);
 692                                         if (res)
 693                                                 printk(KERN_WARNING "Error "
 694                                                         "allocating crypto "
 695                                                         "buffer--skipping "
 696                                                         "crypto\n");
 697                                         res = fscrypt_fname_disk_to_usr(dir,
 698                                                 0, 0, &de_name,
 699                                                 &fname_crypto_str);
 700                                         if (res) {
 701                                                 printk(KERN_WARNING "Error "
 702                                                         "converting filename "
 703                                                         "from disk to usr"
 704                                                         "\n");
 705                                                 name = "??";
 706                                                 len = 2;
 707                                         } else {
 708                                                 name = fname_crypto_str.name;
 709                                                 len = fname_crypto_str.len;
 710                                         }
 711                                         if (IS_CASEFOLDED(dir))
 712                                                 h.hash = EXT4_DIRENT_HASH(de);
 713                                         else
 714                                                 (void) ext4fs_dirhash(dir,
 715                                                         de->name,
 716                                                         de->name_len, &h);
 717                                         printk("%*.s:(E)%x.%u ", len, name,
 718                                                h.hash, (unsigned) ((char *) de
 719                                                                    - base));
 720                                         fscrypt_fname_free_buffer(
 721                                                         &fname_crypto_str);
 722                                 }
 723 #else
 724                                 int len = de->name_len;
 725                                 char *name = de->name;
 726                                 (void) ext4fs_dirhash(dir, de->name,
 727                                                       de->name_len, &h);
 728                                 printk("%*.s:%x.%u ", len, name, h.hash,
 729                                        (unsigned) ((char *) de - base));
 730 #endif
 731                         }
 732                         space += ext4_dir_rec_len(de->name_len, dir);
 733                         names++;
 734                 }
 735                 de = ext4_next_entry(de, size);
 736         }
 737         printk(KERN_CONT "(%i)\n", names);
 738         return (struct stats) { names, space, 1 };
 739 }
 740
 741 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 742                              struct dx_entry *entries, int levels)
 743 {
 744         unsigned blocksize = dir->i_sb->s_blocksize;
 745         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 746         unsigned bcount = 0;
 747         struct buffer_head *bh;
 748         printk("%i indexed blocks...\n", count);
 749         for (i = 0; i < count; i++, entries++)
 750         {
 751                 ext4_lblk_t block = dx_get_block(entries);
 752                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 753                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 754                 struct stats stats;
 755                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 756                 bh = ext4_bread(NULL,dir, block, 0);
 757                 if (!bh || IS_ERR(bh))
 758                         continue;
 759                 stats = levels?
 760                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 761                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 762                         bh->b_data, blocksize, 0);
 763                 names += stats.names;
 764                 space += stats.space;
 765                 bcount += stats.bcount;
 766                 brelse(bh);
 767         }
 768         if (bcount)
 769                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 770                        levels ? "" : "   ", names, space/bcount,
 771                        (space/bcount)*100/blocksize);
 772         return (struct stats) { names, space, bcount};
 773 }
 774
 775 /*
 776  * Linear search cross check
 777  */
 778 static inline void htree_rep_invariant_check(struct dx_entry *at,
 779                                              struct dx_entry *target,
 780                                              u32 hash, unsigned int n)
 781 {
 782         while (n--) {
 783                 dxtrace(printk(KERN_CONT ","));
 784                 if (dx_get_hash(++at) > hash) {
 785                         at--;
 786                         break;
 787                 }
 788         }
 789         ASSERT(at == target - 1);
 790 }
 791 #else /* DX_DEBUG */
 792 static inline void htree_rep_invariant_check(struct dx_entry *at,
 793                                              struct dx_entry *target,
 794                                              u32 hash, unsigned int n)
 795 {
 796 }
 797 #endif /* DX_DEBUG */
 798
 799 /*
 800  * Probe for a directory leaf block to search.
 801  *
 802  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 803  * error in the directory index, and the caller should fall back to
 804  * searching the directory normally.  The callers of dx_probe **MUST**
 805  * check for this error code, and make sure it never gets reflected
 806  * back to userspace.
 807  */
 808 static struct dx_frame *
 809 dx_probe(struct ext4_filename *fname, struct inode *dir,
 810          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 811 {
 812         unsigned count, indirect, level, i;
 813         struct dx_entry *at, *entries, *p, *q, *m;
 814         struct dx_root *root;
 815         struct dx_frame *frame = frame_in;
 816         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 817         u32 hash;
 818         ext4_lblk_t block;
 819         ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
 820
 821         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 822         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 823         if (IS_ERR(frame->bh))
 824                 return (struct dx_frame *) frame->bh;
 825
 826         root = (struct dx_root *) frame->bh->b_data;
 827         if (root->info.hash_version != DX_HASH_TEA &&
 828             root->info.hash_version != DX_HASH_HALF_MD4 &&
 829             root->info.hash_version != DX_HASH_LEGACY &&
 830             root->info.hash_version != DX_HASH_SIPHASH) {
 831                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 832                                    root->info.hash_version);
 833                 goto fail;
 834         }
 835         if (ext4_hash_in_dirent(dir)) {
 836                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 837                         ext4_warning_inode(dir,
 838                                 "Hash in dirent, but hash is not SIPHASH");
 839                         goto fail;
 840                 }
 841         } else {
 842                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 843                         ext4_warning_inode(dir,
 844                                 "Hash code is SIPHASH, but hash not in dirent");
 845                         goto fail;
 846                 }
 847         }
 848         if (fname)
 849                 hinfo = &fname->hinfo;
 850         hinfo->hash_version = root->info.hash_version;
 851         if (hinfo->hash_version <= DX_HASH_TEA)
 852                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 853         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 854         /* hash is already computed for encrypted casefolded directory */
 855         if (fname && fname_name(fname) &&
 856             !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
 857                 int ret = ext4fs_dirhash(dir, fname_name(fname),
 858                                          fname_len(fname), hinfo);
 859                 if (ret < 0) {
 860                         ret_err = ERR_PTR(ret);
 861                         goto fail;
 862                 }
 863         }
 864         hash = hinfo->hash;
 865
 866         if (root->info.unused_flags & 1) {
 867                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 868                                    root->info.unused_flags);
 869                 goto fail;
 870         }
 871
 872         indirect = root->info.indirect_levels;
 873         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 874                 ext4_warning(dir->i_sb,
 875                              "Directory (ino: %lu) htree depth %#06x exceed"
 876                              "supported value", dir->i_ino,
 877                              ext4_dir_htree_level(dir->i_sb));
 878                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 879                         ext4_warning(dir->i_sb, "Enable large directory "
 880                                                 "feature to access it");
 881                 }
 882                 goto fail;
 883         }
 884
 885         entries = (struct dx_entry *)(((char *)&root->info) +
 886                                       root->info.info_length);
 887
 888         if (dx_get_limit(entries) != dx_root_limit(dir,
 889                                                    root->info.info_length)) {
 890                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 891                                    dx_get_limit(entries),
 892                                    dx_root_limit(dir, root->info.info_length));
 893                 goto fail;
 894         }
 895
 896         dxtrace(printk("Look up %x", hash));
 897         level = 0;
 898         blocks[0] = 0;
 899         while (1) {
 900                 count = dx_get_count(entries);
 901                 if (!count || count > dx_get_limit(entries)) {
 902                         ext4_warning_inode(dir,
 903                                            "dx entry: count %u beyond limit %u",
 904                                            count, dx_get_limit(entries));
 905                         goto fail;
 906                 }
 907
 908                 p = entries + 1;
 909                 q = entries + count - 1;
 910                 while (p <= q) {
 911                         m = p + (q - p) / 2;
 912                         dxtrace(printk(KERN_CONT "."));
 913                         if (dx_get_hash(m) > hash)
 914                                 q = m - 1;
 915                         else
 916                                 p = m + 1;
 917                 }
 918
 919                 htree_rep_invariant_check(entries, p, hash, count - 1);
 920
 921                 at = p - 1;
 922                 dxtrace(printk(KERN_CONT " %x->%u\n",
 923                                at == entries ? 0 : dx_get_hash(at),
 924                                dx_get_block(at)));
 925                 frame->entries = entries;
 926                 frame->at = at;
 927
 928                 block = dx_get_block(at);
 929                 for (i = 0; i <= level; i++) {
 930                         if (blocks[i] == block) {
 931                                 ext4_warning_inode(dir,
 932                                         "dx entry: tree cycle block %u points back to block %u",
 933                                         blocks[level], block);
 934                                 goto fail;
 935                         }
 936                 }
 937                 if (++level > indirect)
 938                         return frame;
 939                 blocks[level] = block;
 940                 frame++;
 941                 frame->bh = ext4_read_dirblock(dir, block, INDEX);
 942                 if (IS_ERR(frame->bh)) {
 943                         ret_err = (struct dx_frame *) frame->bh;
 944                         frame->bh = NULL;
 945                         goto fail;
 946                 }
 947
 948                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 949
 950                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 951                         ext4_warning_inode(dir,
 952                                 "dx entry: limit %u != node limit %u",
 953                                 dx_get_limit(entries), dx_node_limit(dir));
 954                         goto fail;
 955                 }
 956         }
 957 fail:
 958         while (frame >= frame_in) {
 959                 brelse(frame->bh);
 960                 frame--;
 961         }
 962
 963         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 964                 ext4_warning_inode(dir,
 965                         "Corrupt directory, running e2fsck is recommended");
 966         return ret_err;
 967 }
 968
 969 static void dx_release(struct dx_frame *frames)
 970 {
 971         struct dx_root_info *info;
 972         int i;
 973         unsigned int indirect_levels;
 974
 975         if (frames[0].bh == NULL)
 976                 return;
 977
 978         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 979         /* save local copy, "info" may be freed after brelse() */
 980         indirect_levels = info->indirect_levels;
 981         for (i = 0; i <= indirect_levels; i++) {
 982                 if (frames[i].bh == NULL)
 983                         break;
 984                 brelse(frames[i].bh);
 985                 frames[i].bh = NULL;
 986         }
 987 }
 988
 989 /*
 990  * This function increments the frame pointer to search the next leaf
 991  * block, and reads in the necessary intervening nodes if the search
 992  * should be necessary.  Whether or not the search is necessary is
 993  * controlled by the hash parameter.  If the hash value is even, then
 994  * the search is only continued if the next block starts with that
 995  * hash value.  This is used if we are searching for a specific file.
 996  *
 997  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 998  *
 999  * This function returns 1 if the caller should continue to search,
1000  * or 0 if it should not.  If there is an error reading one of the
1001  * index blocks, it will a negative error code.
1002  *
1003  * If start_hash is non-null, it will be filled in with the starting
1004  * hash of the next page.
1005  */
1006 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1007                                  struct dx_frame *frame,
1008                                  struct dx_frame *frames,
1009                                  __u32 *start_hash)
1010 {
1011         struct dx_frame *p;
1012         struct buffer_head *bh;
1013         int num_frames = 0;
1014         __u32 bhash;
1015
1016         p = frame;
1017         /*
1018          * Find the next leaf page by incrementing the frame pointer.
1019          * If we run out of entries in the interior node, loop around and
1020          * increment pointer in the parent node.  When we break out of
1021          * this loop, num_frames indicates the number of interior
1022          * nodes need to be read.
1023          */
1024         while (1) {
1025                 if (++(p->at) < p->entries + dx_get_count(p->entries))
1026                         break;
1027                 if (p == frames)
1028                         return 0;
1029                 num_frames++;
1030                 p--;
1031         }
1032
1033         /*
1034          * If the hash is 1, then continue only if the next page has a
1035          * continuation hash of any value.  This is used for readdir
1036          * handling.  Otherwise, check to see if the hash matches the
1037          * desired continuation hash.  If it doesn't, return since
1038          * there's no point to read in the successive index pages.
1039          */
1040         bhash = dx_get_hash(p->at);
1041         if (start_hash)
1042                 *start_hash = bhash;
1043         if ((hash & 1) == 0) {
1044                 if ((bhash & ~1) != hash)
1045                         return 0;
1046         }
1047         /*
1048          * If the hash is HASH_NB_ALWAYS, we always go to the next
1049          * block so no check is necessary
1050          */
1051         while (num_frames--) {
1052                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1053                 if (IS_ERR(bh))
1054                         return PTR_ERR(bh);
1055                 p++;
1056                 brelse(p->bh);
1057                 p->bh = bh;
1058                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1059         }
1060         return 1;
1061 }
1062
1063
1064 /*
1065  * This function fills a red-black tree with information from a
1066  * directory block.  It returns the number directory entries loaded
1067  * into the tree.  If there is an error it is returned in err.
1068  */
1069 static int htree_dirblock_to_tree(struct file *dir_file,
1070                                   struct inode *dir, ext4_lblk_t block,
1071                                   struct dx_hash_info *hinfo,
1072                                   __u32 start_hash, __u32 start_minor_hash)
1073 {
1074         struct buffer_head *bh;
1075         struct ext4_dir_entry_2 *de, *top;
1076         int err = 0, count = 0;
1077         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1078         int csum = ext4_has_metadata_csum(dir->i_sb);
1079
1080         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1081                                                         (unsigned long)block));
1082         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1083         if (IS_ERR(bh))
1084                 return PTR_ERR(bh);
1085
1086         de = (struct ext4_dir_entry_2 *) bh->b_data;
1087         /* csum entries are not larger in the casefolded encrypted case */
1088         top = (struct ext4_dir_entry_2 *) ((char *) de +
1089                                            dir->i_sb->s_blocksize -
1090                                            ext4_dir_rec_len(0,
1091                                                            csum ? NULL : dir));
1092         /* Check if the directory is encrypted */
1093         if (IS_ENCRYPTED(dir)) {
1094                 err = fscrypt_prepare_readdir(dir);
1095                 if (err < 0) {
1096                         brelse(bh);
1097                         return err;
1098                 }
1099                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1100                                                  &fname_crypto_str);
1101                 if (err < 0) {
1102                         brelse(bh);
1103                         return err;
1104                 }
1105         }
1106
1107         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1108                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1109                                 bh->b_data, bh->b_size,
1110                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1111                                          + ((char *)de - bh->b_data))) {
1112                         /* silently ignore the rest of the block */
1113                         break;
1114                 }
1115                 if (ext4_hash_in_dirent(dir)) {
1116                         if (de->name_len && de->inode) {
1117                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1118                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1119                         } else {
1120                                 hinfo->hash = 0;
1121                                 hinfo->minor_hash = 0;
1122                         }
1123                 } else {
1124                         err = ext4fs_dirhash(dir, de->name,
1125                                              de->name_len, hinfo);
1126                         if (err < 0) {
1127                                 count = err;
1128                                 goto errout;
1129                         }
1130                 }
1131                 if ((hinfo->hash < start_hash) ||
1132                     ((hinfo->hash == start_hash) &&
1133                      (hinfo->minor_hash < start_minor_hash)))
1134                         continue;
1135                 if (de->inode == 0)
1136                         continue;
1137                 if (!IS_ENCRYPTED(dir)) {
1138                         tmp_str.name = de->name;
1139                         tmp_str.len = de->name_len;
1140                         err = ext4_htree_store_dirent(dir_file,
1141                                    hinfo->hash, hinfo->minor_hash, de,
1142                                    &tmp_str);
1143                 } else {
1144                         int save_len = fname_crypto_str.len;
1145                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1146                                                                 de->name_len);
1147
1148                         /* Directory is encrypted */
1149                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1150                                         hinfo->minor_hash, &de_name,
1151                                         &fname_crypto_str);
1152                         if (err) {
1153                                 count = err;
1154                                 goto errout;
1155                         }
1156                         err = ext4_htree_store_dirent(dir_file,
1157                                    hinfo->hash, hinfo->minor_hash, de,
1158                                         &fname_crypto_str);
1159                         fname_crypto_str.len = save_len;
1160                 }
1161                 if (err != 0) {
1162                         count = err;
1163                         goto errout;
1164                 }
1165                 count++;
1166         }
1167 errout:
1168         brelse(bh);
1169         fscrypt_fname_free_buffer(&fname_crypto_str);
1170         return count;
1171 }
1172
1173
1174 /*
1175  * This function fills a red-black tree with information from a
1176  * directory.  We start scanning the directory in hash order, starting
1177  * at start_hash and start_minor_hash.
1178  *
1179  * This function returns the number of entries inserted into the tree,
1180  * or a negative error code.
1181  */
1182 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1183                          __u32 start_minor_hash, __u32 *next_hash)
1184 {
1185         struct dx_hash_info hinfo;
1186         struct ext4_dir_entry_2 *de;
1187         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1188         struct inode *dir;
1189         ext4_lblk_t block;
1190         int count = 0;
1191         int ret, err;
1192         __u32 hashval;
1193         struct fscrypt_str tmp_str;
1194
1195         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1196                        start_hash, start_minor_hash));
1197         dir = file_inode(dir_file);
1198         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1199                 if (ext4_hash_in_dirent(dir))
1200                         hinfo.hash_version = DX_HASH_SIPHASH;
1201                 else
1202                         hinfo.hash_version =
1203                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1204                 if (hinfo.hash_version <= DX_HASH_TEA)
1205                         hinfo.hash_version +=
1206                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1207                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1208                 if (ext4_has_inline_data(dir)) {
1209                         int has_inline_data = 1;
1210                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1211                                                        &hinfo, start_hash,
1212                                                        start_minor_hash,
1213                                                        &has_inline_data);
1214                         if (has_inline_data) {
1215                                 *next_hash = ~0;
1216                                 return count;
1217                         }
1218                 }
1219                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1220                                                start_hash, start_minor_hash);
1221                 *next_hash = ~0;
1222                 return count;
1223         }
1224         hinfo.hash = start_hash;
1225         hinfo.minor_hash = 0;
1226         frame = dx_probe(NULL, dir, &hinfo, frames);
1227         if (IS_ERR(frame))
1228                 return PTR_ERR(frame);
1229
1230         /* Add '.' and '..' from the htree header */
1231         if (!start_hash && !start_minor_hash) {
1232                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1233                 tmp_str.name = de->name;
1234                 tmp_str.len = de->name_len;
1235                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1236                                               de, &tmp_str);
1237                 if (err != 0)
1238                         goto errout;
1239                 count++;
1240         }
1241         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1242                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1243                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1244                 tmp_str.name = de->name;
1245                 tmp_str.len = de->name_len;
1246                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1247                                               de, &tmp_str);
1248                 if (err != 0)
1249                         goto errout;
1250                 count++;
1251         }
1252
1253         while (1) {
1254                 if (fatal_signal_pending(current)) {
1255                         err = -ERESTARTSYS;
1256                         goto errout;
1257                 }
1258                 cond_resched();
1259                 block = dx_get_block(frame->at);
1260                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1261                                              start_hash, start_minor_hash);
1262                 if (ret < 0) {
1263                         err = ret;
1264                         goto errout;
1265                 }
1266                 count += ret;
1267                 hashval = ~0;
1268                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1269                                             frame, frames, &hashval);
1270                 *next_hash = hashval;
1271                 if (ret < 0) {
1272                         err = ret;
1273                         goto errout;
1274                 }
1275                 /*
1276                  * Stop if:  (a) there are no more entries, or
1277                  * (b) we have inserted at least one entry and the
1278                  * next hash value is not a continuation
1279                  */
1280                 if ((ret == 0) ||
1281                     (count && ((hashval & 1) == 0)))
1282                         break;
1283         }
1284         dx_release(frames);
1285         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1286                        "next hash: %x\n", count, *next_hash));
1287         return count;
1288 errout:
1289         dx_release(frames);
1290         return (err);
1291 }
1292
1293 static inline int search_dirblock(struct buffer_head *bh,
1294                                   struct inode *dir,
1295                                   struct ext4_filename *fname,
1296                                   unsigned int offset,
1297                                   struct ext4_dir_entry_2 **res_dir)
1298 {
1299         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1300                                fname, offset, res_dir);
1301 }
1302
1303 /*
1304  * Directory block splitting, compacting
1305  */
1306
1307 /*
1308  * Create map of hash values, offsets, and sizes, stored at end of block.
1309  * Returns number of entries mapped.
1310  */
1311 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1312                        struct dx_hash_info *hinfo,
1313                        struct dx_map_entry *map_tail)
1314 {
1315         int count = 0;
1316         struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1317         unsigned int buflen = bh->b_size;
1318         char *base = bh->b_data;
1319         struct dx_hash_info h = *hinfo;
1320         int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1321
1322         if (ext4_has_metadata_csum(dir->i_sb))
1323                 buflen -= sizeof(struct ext4_dir_entry_tail);
1324
1325         while ((char *) de < base + buflen) {
1326                 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1327                                          ((char *)de) - base))
1328                         return -EFSCORRUPTED;
1329                 if (de->name_len && de->inode) {
1330                         if (ext4_hash_in_dirent(dir))
1331                                 h.hash = EXT4_DIRENT_HASH(de);
1332                         else {
1333                                 int err = ext4fs_dirhash(dir, de->name,
1334                                                      de->name_len, &h);
1335                                 if (err < 0)
1336                                         return err;
1337                         }
1338                         map_tail--;
1339                         map_tail->hash = h.hash;
1340                         map_tail->offs = ((char *) de - base)>>2;
1341                         map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1342                                                                 blocksize);
1343                         count++;
1344                         cond_resched();
1345                 }
1346                 de = ext4_next_entry(de, blocksize);
1347         }
1348         return count;
1349 }
1350
1351 /* Sort map by hash value */
1352 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1353 {
1354         struct dx_map_entry *p, *q, *top = map + count - 1;
1355         int more;
1356         /* Combsort until bubble sort doesn't suck */
1357         while (count > 2) {
1358                 count = count*10/13;
1359                 if (count - 9 < 2) /* 9, 10 -> 11 */
1360                         count = 11;
1361                 for (p = top, q = p - count; q >= map; p--, q--)
1362                         if (p->hash < q->hash)
1363                                 swap(*p, *q);
1364         }
1365         /* Garden variety bubble sort */
1366         do {
1367                 more = 0;
1368                 q = top;
1369                 while (q-- > map) {
1370                         if (q[1].hash >= q[0].hash)
1371                                 continue;
1372                         swap(*(q+1), *q);
1373                         more = 1;
1374                 }
1375         } while(more);
1376 }
1377
1378 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1379 {
1380         struct dx_entry *entries = frame->entries;
1381         struct dx_entry *old = frame->at, *new = old + 1;
1382         int count = dx_get_count(entries);
1383
1384         ASSERT(count < dx_get_limit(entries));
1385         ASSERT(old < entries + count);
1386         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1387         dx_set_hash(new, hash);
1388         dx_set_block(new, block);
1389         dx_set_count(entries, count + 1);
1390 }
1391
1392 #if IS_ENABLED(CONFIG_UNICODE)
1393 /*
1394  * Test whether a case-insensitive directory entry matches the filename
1395  * being searched for.  If quick is set, assume the name being looked up
1396  * is already in the casefolded form.
1397  *
1398  * Returns: 0 if the directory entry matches, more than 0 if it
1399  * doesn't match or less than zero on error.
1400  */
1401 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1402                            u8 *de_name, size_t de_name_len, bool quick)
1403 {
1404         const struct super_block *sb = parent->i_sb;
1405         const struct unicode_map *um = sb->s_encoding;
1406         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1407         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1408         int ret;
1409
1410         if (IS_ENCRYPTED(parent)) {
1411                 const struct fscrypt_str encrypted_name =
1412                                 FSTR_INIT(de_name, de_name_len);
1413
1414                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1415                 if (!decrypted_name.name)
1416                         return -ENOMEM;
1417                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1418                                                 &decrypted_name);
1419                 if (ret < 0)
1420                         goto out;
1421                 entry.name = decrypted_name.name;
1422                 entry.len = decrypted_name.len;
1423         }
1424
1425         if (quick)
1426                 ret = utf8_strncasecmp_folded(um, name, &entry);
1427         else
1428                 ret = utf8_strncasecmp(um, name, &entry);
1429         if (ret < 0) {
1430                 /* Handle invalid character sequence as either an error
1431                  * or as an opaque byte sequence.
1432                  */
1433                 if (sb_has_strict_encoding(sb))
1434                         ret = -EINVAL;
1435                 else if (name->len != entry.len)
1436                         ret = 1;
1437                 else
1438                         ret = !!memcmp(name->name, entry.name, entry.len);
1439         }
1440 out:
1441         kfree(decrypted_name.name);
1442         return ret;
1443 }
1444
1445 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1446                                   struct ext4_filename *name)
1447 {
1448         struct fscrypt_str *cf_name = &name->cf_name;
1449         struct dx_hash_info *hinfo = &name->hinfo;
1450         int len;
1451
1452         if (!IS_CASEFOLDED(dir) ||
1453             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1454                 cf_name->name = NULL;
1455                 return 0;
1456         }
1457
1458         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1459         if (!cf_name->name)
1460                 return -ENOMEM;
1461
1462         len = utf8_casefold(dir->i_sb->s_encoding,
1463                             iname, cf_name->name,
1464                             EXT4_NAME_LEN);
1465         if (len <= 0) {
1466                 kfree(cf_name->name);
1467                 cf_name->name = NULL;
1468         }
1469         cf_name->len = (unsigned) len;
1470         if (!IS_ENCRYPTED(dir))
1471                 return 0;
1472
1473         hinfo->hash_version = DX_HASH_SIPHASH;
1474         hinfo->seed = NULL;
1475         if (cf_name->name)
1476                 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1477         else
1478                 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1479 }
1480 #endif
1481
1482 /*
1483  * Test whether a directory entry matches the filename being searched for.
1484  *
1485  * Return: %true if the directory entry matches, otherwise %false.
1486  */
1487 static bool ext4_match(struct inode *parent,
1488                               const struct ext4_filename *fname,
1489                               struct ext4_dir_entry_2 *de)
1490 {
1491         struct fscrypt_name f;
1492
1493         if (!de->inode)
1494                 return false;
1495
1496         f.usr_fname = fname->usr_fname;
1497         f.disk_name = fname->disk_name;
1498 #ifdef CONFIG_FS_ENCRYPTION
1499         f.crypto_buf = fname->crypto_buf;
1500 #endif
1501
1502 #if IS_ENABLED(CONFIG_UNICODE)
1503         if (IS_CASEFOLDED(parent) &&
1504             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1505                 if (fname->cf_name.name) {
1506                         struct qstr cf = {.name = fname->cf_name.name,
1507                                           .len = fname->cf_name.len};
1508                         if (IS_ENCRYPTED(parent)) {
1509                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1510                                         fname->hinfo.minor_hash !=
1511                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1512
1513                                         return false;
1514                                 }
1515                         }
1516                         return !ext4_ci_compare(parent, &cf, de->name,
1517                                                         de->name_len, true);
1518                 }
1519                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1520                                                 de->name_len, false);
1521         }
1522 #endif
1523
1524         return fscrypt_match_name(&f, de->name, de->name_len);
1525 }
1526
1527 /*
1528  * Returns 0 if not found, -1 on failure, and 1 on success
1529  */
1530 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1531                     struct inode *dir, struct ext4_filename *fname,
1532                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1533 {
1534         struct ext4_dir_entry_2 * de;
1535         char * dlimit;
1536         int de_len;
1537
1538         de = (struct ext4_dir_entry_2 *)search_buf;
1539         dlimit = search_buf + buf_size;
1540         while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1541                 /* this code is executed quadratically often */
1542                 /* do minimal checking `by hand' */
1543                 if (de->name + de->name_len <= dlimit &&
1544                     ext4_match(dir, fname, de)) {
1545                         /* found a match - just to be sure, do
1546                          * a full check */
1547                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1548                                                  buf_size, offset))
1549                                 return -1;
1550                         *res_dir = de;
1551                         return 1;
1552                 }
1553                 /* prevent looping on a bad block */
1554                 de_len = ext4_rec_len_from_disk(de->rec_len,
1555                                                 dir->i_sb->s_blocksize);
1556                 if (de_len <= 0)
1557                         return -1;
1558                 offset += de_len;
1559                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1560         }
1561         return 0;
1562 }
1563
1564 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1565                                struct ext4_dir_entry *de)
1566 {
1567         struct super_block *sb = dir->i_sb;
1568
1569         if (!is_dx(dir))
1570                 return 0;
1571         if (block == 0)
1572                 return 1;
1573         if (de->inode == 0 &&
1574             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1575                         sb->s_blocksize)
1576                 return 1;
1577         return 0;
1578 }
1579
1580 /*
1581  *      __ext4_find_entry()
1582  *
1583  * finds an entry in the specified directory with the wanted name. It
1584  * returns the cache buffer in which the entry was found, and the entry
1585  * itself (as a parameter - res_dir). It does NOT read the inode of the
1586  * entry - you'll have to do that yourself if you want to.
1587  *
1588  * The returned buffer_head has ->b_count elevated.  The caller is expected
1589  * to brelse() it when appropriate.
1590  */
1591 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1592                                              struct ext4_filename *fname,
1593                                              struct ext4_dir_entry_2 **res_dir,
1594                                              int *inlined)
1595 {
1596         struct super_block *sb;
1597         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1598         struct buffer_head *bh, *ret = NULL;
1599         ext4_lblk_t start, block;
1600         const u8 *name = fname->usr_fname->name;
1601         size_t ra_max = 0;      /* Number of bh's in the readahead
1602                                    buffer, bh_use[] */
1603         size_t ra_ptr = 0;      /* Current index into readahead
1604                                    buffer */
1605         ext4_lblk_t  nblocks;
1606         int i, namelen, retval;
1607
1608         *res_dir = NULL;
1609         sb = dir->i_sb;
1610         namelen = fname->usr_fname->len;
1611         if (namelen > EXT4_NAME_LEN)
1612                 return NULL;
1613
1614         if (ext4_has_inline_data(dir)) {
1615                 int has_inline_data = 1;
1616                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1617                                              &has_inline_data);
1618                 if (inlined)
1619                         *inlined = has_inline_data;
1620                 if (has_inline_data)
1621                         goto cleanup_and_exit;
1622         }
1623
1624         if ((namelen <= 2) && (name[0] == '.') &&
1625             (name[1] == '.' || name[1] == '\0')) {
1626                 /*
1627                  * "." or ".." will only be in the first block
1628                  * NFS may look up ".."; "." should be handled by the VFS
1629                  */
1630                 block = start = 0;
1631                 nblocks = 1;
1632                 goto restart;
1633         }
1634         if (is_dx(dir)) {
1635                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1636                 /*
1637                  * On success, or if the error was file not found,
1638                  * return.  Otherwise, fall back to doing a search the
1639                  * old fashioned way.
1640                  */
1641                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1642                         goto cleanup_and_exit;
1643                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1644                                "falling back\n"));
1645                 ret = NULL;
1646         }
1647         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1648         if (!nblocks) {
1649                 ret = NULL;
1650                 goto cleanup_and_exit;
1651         }
1652         start = EXT4_I(dir)->i_dir_start_lookup;
1653         if (start >= nblocks)
1654                 start = 0;
1655         block = start;
1656 restart:
1657         do {
1658                 /*
1659                  * We deal with the read-ahead logic here.
1660                  */
1661                 cond_resched();
1662                 if (ra_ptr >= ra_max) {
1663                         /* Refill the readahead buffer */
1664                         ra_ptr = 0;
1665                         if (block < start)
1666                                 ra_max = start - block;
1667                         else
1668                                 ra_max = nblocks - block;
1669                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1670                         retval = ext4_bread_batch(dir, block, ra_max,
1671                                                   false /* wait */, bh_use);
1672                         if (retval) {
1673                                 ret = ERR_PTR(retval);
1674                                 ra_max = 0;
1675                                 goto cleanup_and_exit;
1676                         }
1677                 }
1678                 if ((bh = bh_use[ra_ptr++]) == NULL)
1679                         goto next;
1680                 wait_on_buffer(bh);
1681                 if (!buffer_uptodate(bh)) {
1682                         EXT4_ERROR_INODE_ERR(dir, EIO,
1683                                              "reading directory lblock %lu",
1684                                              (unsigned long) block);
1685                         brelse(bh);
1686                         ret = ERR_PTR(-EIO);
1687                         goto cleanup_and_exit;
1688                 }
1689                 if (!buffer_verified(bh) &&
1690                     !is_dx_internal_node(dir, block,
1691                                          (struct ext4_dir_entry *)bh->b_data) &&
1692                     !ext4_dirblock_csum_verify(dir, bh)) {
1693                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1694                                              "checksumming directory "
1695                                              "block %lu", (unsigned long)block);
1696                         brelse(bh);
1697                         ret = ERR_PTR(-EFSBADCRC);
1698                         goto cleanup_and_exit;
1699                 }
1700                 set_buffer_verified(bh);
1701                 i = search_dirblock(bh, dir, fname,
1702                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1703                 if (i == 1) {
1704                         EXT4_I(dir)->i_dir_start_lookup = block;
1705                         ret = bh;
1706                         goto cleanup_and_exit;
1707                 } else {
1708                         brelse(bh);
1709                         if (i < 0)
1710                                 goto cleanup_and_exit;
1711                 }
1712         next:
1713                 if (++block >= nblocks)
1714                         block = 0;
1715         } while (block != start);
1716
1717         /*
1718          * If the directory has grown while we were searching, then
1719          * search the last part of the directory before giving up.
1720          */
1721         block = nblocks;
1722         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1723         if (block < nblocks) {
1724                 start = 0;
1725                 goto restart;
1726         }
1727
1728 cleanup_and_exit:
1729         /* Clean up the read-ahead blocks */
1730         for (; ra_ptr < ra_max; ra_ptr++)
1731                 brelse(bh_use[ra_ptr]);
1732         return ret;
1733 }
1734
1735 static struct buffer_head *ext4_find_entry(struct inode *dir,
1736                                            const struct qstr *d_name,
1737                                            struct ext4_dir_entry_2 **res_dir,
1738                                            int *inlined)
1739 {
1740         int err;
1741         struct ext4_filename fname;
1742         struct buffer_head *bh;
1743
1744         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1745         if (err == -ENOENT)
1746                 return NULL;
1747         if (err)
1748                 return ERR_PTR(err);
1749
1750         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1751
1752         ext4_fname_free_filename(&fname);
1753         return bh;
1754 }
1755
1756 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1757                                              struct dentry *dentry,
1758                                              struct ext4_dir_entry_2 **res_dir)
1759 {
1760         int err;
1761         struct ext4_filename fname;
1762         struct buffer_head *bh;
1763
1764         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1765         generic_set_encrypted_ci_d_ops(dentry);
1766         if (err == -ENOENT)
1767                 return NULL;
1768         if (err)
1769                 return ERR_PTR(err);
1770
1771         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1772
1773         ext4_fname_free_filename(&fname);
1774         return bh;
1775 }
1776
1777 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1778                         struct ext4_filename *fname,
1779                         struct ext4_dir_entry_2 **res_dir)
1780 {
1781         struct super_block * sb = dir->i_sb;
1782         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1783         struct buffer_head *bh;
1784         ext4_lblk_t block;
1785         int retval;
1786
1787 #ifdef CONFIG_FS_ENCRYPTION
1788         *res_dir = NULL;
1789 #endif
1790         frame = dx_probe(fname, dir, NULL, frames);
1791         if (IS_ERR(frame))
1792                 return (struct buffer_head *) frame;
1793         do {
1794                 block = dx_get_block(frame->at);
1795                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1796                 if (IS_ERR(bh))
1797                         goto errout;
1798
1799                 retval = search_dirblock(bh, dir, fname,
1800                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1801                                          res_dir);
1802                 if (retval == 1)
1803                         goto success;
1804                 brelse(bh);
1805                 if (retval == -1) {
1806                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1807                         goto errout;
1808                 }
1809
1810                 /* Check to see if we should continue to search */
1811                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1812                                                frames, NULL);
1813                 if (retval < 0) {
1814                         ext4_warning_inode(dir,
1815                                 "error %d reading directory index block",
1816                                 retval);
1817                         bh = ERR_PTR(retval);
1818                         goto errout;
1819                 }
1820         } while (retval == 1);
1821
1822         bh = NULL;
1823 errout:
1824         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1825 success:
1826         dx_release(frames);
1827         return bh;
1828 }
1829
1830 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1831 {
1832         struct inode *inode;
1833         struct ext4_dir_entry_2 *de;
1834         struct buffer_head *bh;
1835
1836         if (dentry->d_name.len > EXT4_NAME_LEN)
1837                 return ERR_PTR(-ENAMETOOLONG);
1838
1839         bh = ext4_lookup_entry(dir, dentry, &de);
1840         if (IS_ERR(bh))
1841                 return ERR_CAST(bh);
1842         inode = NULL;
1843         if (bh) {
1844                 __u32 ino = le32_to_cpu(de->inode);
1845                 brelse(bh);
1846                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1847                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1848                         return ERR_PTR(-EFSCORRUPTED);
1849                 }
1850                 if (unlikely(ino == dir->i_ino)) {
1851                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1852                                          dentry);
1853                         return ERR_PTR(-EFSCORRUPTED);
1854                 }
1855                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1856                 if (inode == ERR_PTR(-ESTALE)) {
1857                         EXT4_ERROR_INODE(dir,
1858                                          "deleted inode referenced: %u",
1859                                          ino);
1860                         return ERR_PTR(-EFSCORRUPTED);
1861                 }
1862                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1863                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1864                     !fscrypt_has_permitted_context(dir, inode)) {
1865                         ext4_warning(inode->i_sb,
1866                                      "Inconsistent encryption contexts: %lu/%lu",
1867                                      dir->i_ino, inode->i_ino);
1868                         iput(inode);
1869                         return ERR_PTR(-EPERM);
1870                 }
1871         }
1872
1873 #if IS_ENABLED(CONFIG_UNICODE)
1874         if (!inode && IS_CASEFOLDED(dir)) {
1875                 /* Eventually we want to call d_add_ci(dentry, NULL)
1876                  * for negative dentries in the encoding case as
1877                  * well.  For now, prevent the negative dentry
1878                  * from being cached.
1879                  */
1880                 return NULL;
1881         }
1882 #endif
1883         return d_splice_alias(inode, dentry);
1884 }
1885
1886
1887 struct dentry *ext4_get_parent(struct dentry *child)
1888 {
1889         __u32 ino;
1890         struct ext4_dir_entry_2 * de;
1891         struct buffer_head *bh;
1892
1893         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1894         if (IS_ERR(bh))
1895                 return ERR_CAST(bh);
1896         if (!bh)
1897                 return ERR_PTR(-ENOENT);
1898         ino = le32_to_cpu(de->inode);
1899         brelse(bh);
1900
1901         if (!ext4_valid_inum(child->d_sb, ino)) {
1902                 EXT4_ERROR_INODE(d_inode(child),
1903                                  "bad parent inode number: %u", ino);
1904                 return ERR_PTR(-EFSCORRUPTED);
1905         }
1906
1907         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1908 }
1909
1910 /*
1911  * Move count entries from end of map between two memory locations.
1912  * Returns pointer to last entry moved.
1913  */
1914 static struct ext4_dir_entry_2 *
1915 dx_move_dirents(struct inode *dir, char *from, char *to,
1916                 struct dx_map_entry *map, int count,
1917                 unsigned blocksize)
1918 {
1919         unsigned rec_len = 0;
1920
1921         while (count--) {
1922                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1923                                                 (from + (map->offs<<2));
1924                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1925
1926                 memcpy (to, de, rec_len);
1927                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1928                                 ext4_rec_len_to_disk(rec_len, blocksize);
1929
1930                 /* wipe dir_entry excluding the rec_len field */
1931                 de->inode = 0;
1932                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1933                                                                 blocksize) -
1934                                          offsetof(struct ext4_dir_entry_2,
1935                                                                 name_len));
1936
1937                 map++;
1938                 to += rec_len;
1939         }
1940         return (struct ext4_dir_entry_2 *) (to - rec_len);
1941 }
1942
1943 /*
1944  * Compact each dir entry in the range to the minimal rec_len.
1945  * Returns pointer to last entry in range.
1946  */
1947 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1948                                                         unsigned int blocksize)
1949 {
1950         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1951         unsigned rec_len = 0;
1952
1953         prev = to = de;
1954         while ((char*)de < base + blocksize) {
1955                 next = ext4_next_entry(de, blocksize);
1956                 if (de->inode && de->name_len) {
1957                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1958                         if (de > to)
1959                                 memmove(to, de, rec_len);
1960                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1961                         prev = to;
1962                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1963                 }
1964                 de = next;
1965         }
1966         return prev;
1967 }
1968
1969 /*
1970  * Split a full leaf block to make room for a new dir entry.
1971  * Allocate a new block, and move entries so that they are approx. equally full.
1972  * Returns pointer to de in block into which the new entry will be inserted.
1973  */
1974 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1975                         struct buffer_head **bh,struct dx_frame *frame,
1976                         struct dx_hash_info *hinfo)
1977 {
1978         unsigned blocksize = dir->i_sb->s_blocksize;
1979         unsigned continued;
1980         int count;
1981         struct buffer_head *bh2;
1982         ext4_lblk_t newblock;
1983         u32 hash2;
1984         struct dx_map_entry *map;
1985         char *data1 = (*bh)->b_data, *data2;
1986         unsigned split, move, size;
1987         struct ext4_dir_entry_2 *de = NULL, *de2;
1988         int     csum_size = 0;
1989         int     err = 0, i;
1990
1991         if (ext4_has_metadata_csum(dir->i_sb))
1992                 csum_size = sizeof(struct ext4_dir_entry_tail);
1993
1994         bh2 = ext4_append(handle, dir, &newblock);
1995         if (IS_ERR(bh2)) {
1996                 brelse(*bh);
1997                 *bh = NULL;
1998                 return (struct ext4_dir_entry_2 *) bh2;
1999         }
2000
2001         BUFFER_TRACE(*bh, "get_write_access");
2002         err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
2003                                             EXT4_JTR_NONE);
2004         if (err)
2005                 goto journal_error;
2006
2007         BUFFER_TRACE(frame->bh, "get_write_access");
2008         err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2009                                             EXT4_JTR_NONE);
2010         if (err)
2011                 goto journal_error;
2012
2013         data2 = bh2->b_data;
2014
2015         /* create map in the end of data2 block */
2016         map = (struct dx_map_entry *) (data2 + blocksize);
2017         count = dx_make_map(dir, *bh, hinfo, map);
2018         if (count < 0) {
2019                 err = count;
2020                 goto journal_error;
2021         }
2022         map -= count;
2023         dx_sort_map(map, count);
2024         /* Ensure that neither split block is over half full */
2025         size = 0;
2026         move = 0;
2027         for (i = count-1; i >= 0; i--) {
2028                 /* is more than half of this entry in 2nd half of the block? */
2029                 if (size + map[i].size/2 > blocksize/2)
2030                         break;
2031                 size += map[i].size;
2032                 move++;
2033         }
2034         /*
2035          * map index at which we will split
2036          *
2037          * If the sum of active entries didn't exceed half the block size, just
2038          * split it in half by count; each resulting block will have at least
2039          * half the space free.
2040          */
2041         if (i > 0)
2042                 split = count - move;
2043         else
2044                 split = count/2;
2045
2046         hash2 = map[split].hash;
2047         continued = hash2 == map[split - 1].hash;
2048         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2049                         (unsigned long)dx_get_block(frame->at),
2050                                         hash2, split, count-split));
2051
2052         /* Fancy dance to stay within two buffers */
2053         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2054                               blocksize);
2055         de = dx_pack_dirents(dir, data1, blocksize);
2056         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2057                                            (char *) de,
2058                                            blocksize);
2059         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2060                                             (char *) de2,
2061                                             blocksize);
2062         if (csum_size) {
2063                 ext4_initialize_dirent_tail(*bh, blocksize);
2064                 ext4_initialize_dirent_tail(bh2, blocksize);
2065         }
2066
2067         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2068                         blocksize, 1));
2069         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2070                         blocksize, 1));
2071
2072         /* Which block gets the new entry? */
2073         if (hinfo->hash >= hash2) {
2074                 swap(*bh, bh2);
2075                 de = de2;
2076         }
2077         dx_insert_block(frame, hash2 + continued, newblock);
2078         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2079         if (err)
2080                 goto journal_error;
2081         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2082         if (err)
2083                 goto journal_error;
2084         brelse(bh2);
2085         dxtrace(dx_show_index("frame", frame->entries));
2086         return de;
2087
2088 journal_error:
2089         brelse(*bh);
2090         brelse(bh2);
2091         *bh = NULL;
2092         ext4_std_error(dir->i_sb, err);
2093         return ERR_PTR(err);
2094 }
2095
2096 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2097                       struct buffer_head *bh,
2098                       void *buf, int buf_size,
2099                       struct ext4_filename *fname,
2100                       struct ext4_dir_entry_2 **dest_de)
2101 {
2102         struct ext4_dir_entry_2 *de;
2103         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2104         int nlen, rlen;
2105         unsigned int offset = 0;
2106         char *top;
2107
2108         de = buf;
2109         top = buf + buf_size - reclen;
2110         while ((char *) de <= top) {
2111                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2112                                          buf, buf_size, offset))
2113                         return -EFSCORRUPTED;
2114                 if (ext4_match(dir, fname, de))
2115                         return -EEXIST;
2116                 nlen = ext4_dir_rec_len(de->name_len, dir);
2117                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2118                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2119                         break;
2120                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2121                 offset += rlen;
2122         }
2123         if ((char *) de > top)
2124                 return -ENOSPC;
2125
2126         *dest_de = de;
2127         return 0;
2128 }
2129
2130 void ext4_insert_dentry(struct inode *dir,
2131                         struct inode *inode,
2132                         struct ext4_dir_entry_2 *de,
2133                         int buf_size,
2134                         struct ext4_filename *fname)
2135 {
2136
2137         int nlen, rlen;
2138
2139         nlen = ext4_dir_rec_len(de->name_len, dir);
2140         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2141         if (de->inode) {
2142                 struct ext4_dir_entry_2 *de1 =
2143                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2144                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2145                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2146                 de = de1;
2147         }
2148         de->file_type = EXT4_FT_UNKNOWN;
2149         de->inode = cpu_to_le32(inode->i_ino);
2150         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2151         de->name_len = fname_len(fname);
2152         memcpy(de->name, fname_name(fname), fname_len(fname));
2153         if (ext4_hash_in_dirent(dir)) {
2154                 struct dx_hash_info *hinfo = &fname->hinfo;
2155
2156                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2157                 EXT4_DIRENT_HASHES(de)->minor_hash =
2158                                                 cpu_to_le32(hinfo->minor_hash);
2159         }
2160 }
2161
2162 /*
2163  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2164  * it points to a directory entry which is guaranteed to be large
2165  * enough for new directory entry.  If de is NULL, then
2166  * add_dirent_to_buf will attempt search the directory block for
2167  * space.  It will return -ENOSPC if no space is available, and -EIO
2168  * and -EEXIST if directory entry already exists.
2169  */
2170 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2171                              struct inode *dir,
2172                              struct inode *inode, struct ext4_dir_entry_2 *de,
2173                              struct buffer_head *bh)
2174 {
2175         unsigned int    blocksize = dir->i_sb->s_blocksize;
2176         int             csum_size = 0;
2177         int             err, err2;
2178
2179         if (ext4_has_metadata_csum(inode->i_sb))
2180                 csum_size = sizeof(struct ext4_dir_entry_tail);
2181
2182         if (!de) {
2183                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2184                                         blocksize - csum_size, fname, &de);
2185                 if (err)
2186                         return err;
2187         }
2188         BUFFER_TRACE(bh, "get_write_access");
2189         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2190                                             EXT4_JTR_NONE);
2191         if (err) {
2192                 ext4_std_error(dir->i_sb, err);
2193                 return err;
2194         }
2195
2196         /* By now the buffer is marked for journaling */
2197         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2198
2199         /*
2200          * XXX shouldn't update any times until successful
2201          * completion of syscall, but too many callers depend
2202          * on this.
2203          *
2204          * XXX similarly, too many callers depend on
2205          * ext4_new_inode() setting the times, but error
2206          * recovery deletes the inode, so the worst that can
2207          * happen is that the times are slightly out of date
2208          * and/or different from the directory change time.
2209          */
2210         inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2211         ext4_update_dx_flag(dir);
2212         inode_inc_iversion(dir);
2213         err2 = ext4_mark_inode_dirty(handle, dir);
2214         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2215         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2216         if (err)
2217                 ext4_std_error(dir->i_sb, err);
2218         return err ? err : err2;
2219 }
2220
2221 /*
2222  * This converts a one block unindexed directory to a 3 block indexed
2223  * directory, and adds the dentry to the indexed directory.
2224  */
2225 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2226                             struct inode *dir,
2227                             struct inode *inode, struct buffer_head *bh)
2228 {
2229         struct buffer_head *bh2;
2230         struct dx_root  *root;
2231         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2232         struct dx_entry *entries;
2233         struct ext4_dir_entry_2 *de, *de2;
2234         char            *data2, *top;
2235         unsigned        len;
2236         int             retval;
2237         unsigned        blocksize;
2238         ext4_lblk_t  block;
2239         struct fake_dirent *fde;
2240         int csum_size = 0;
2241
2242         if (ext4_has_metadata_csum(inode->i_sb))
2243                 csum_size = sizeof(struct ext4_dir_entry_tail);
2244
2245         blocksize =  dir->i_sb->s_blocksize;
2246         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2247         BUFFER_TRACE(bh, "get_write_access");
2248         retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2249                                                EXT4_JTR_NONE);
2250         if (retval) {
2251                 ext4_std_error(dir->i_sb, retval);
2252                 brelse(bh);
2253                 return retval;
2254         }
2255         root = (struct dx_root *) bh->b_data;
2256
2257         /* The 0th block becomes the root, move the dirents out */
2258         fde = &root->dotdot;
2259         de = (struct ext4_dir_entry_2 *)((char *)fde +
2260                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2261         if ((char *) de >= (((char *) root) + blocksize)) {
2262                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2263                 brelse(bh);
2264                 return -EFSCORRUPTED;
2265         }
2266         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2267
2268         /* Allocate new block for the 0th block's dirents */
2269         bh2 = ext4_append(handle, dir, &block);
2270         if (IS_ERR(bh2)) {
2271                 brelse(bh);
2272                 return PTR_ERR(bh2);
2273         }
2274         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2275         data2 = bh2->b_data;
2276
2277         memcpy(data2, de, len);
2278         memset(de, 0, len); /* wipe old data */
2279         de = (struct ext4_dir_entry_2 *) data2;
2280         top = data2 + len;
2281         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2282                 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2283                                         (char *)de - data2)) {
2284                         brelse(bh2);
2285                         brelse(bh);
2286                         return -EFSCORRUPTED;
2287                 }
2288                 de = de2;
2289         }
2290         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2291                                            (char *) de, blocksize);
2292
2293         if (csum_size)
2294                 ext4_initialize_dirent_tail(bh2, blocksize);
2295
2296         /* Initialize the root; the dot dirents already exist */
2297         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2298         de->rec_len = ext4_rec_len_to_disk(
2299                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2300         memset (&root->info, 0, sizeof(root->info));
2301         root->info.info_length = sizeof(root->info);
2302         if (ext4_hash_in_dirent(dir))
2303                 root->info.hash_version = DX_HASH_SIPHASH;
2304         else
2305                 root->info.hash_version =
2306                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2307
2308         entries = root->entries;
2309         dx_set_block(entries, 1);
2310         dx_set_count(entries, 1);
2311         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2312
2313         /* Initialize as for dx_probe */
2314         fname->hinfo.hash_version = root->info.hash_version;
2315         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2316                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2317         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2318
2319         /* casefolded encrypted hashes are computed on fname setup */
2320         if (!ext4_hash_in_dirent(dir)) {
2321                 int err = ext4fs_dirhash(dir, fname_name(fname),
2322                                          fname_len(fname), &fname->hinfo);
2323                 if (err < 0) {
2324                         brelse(bh2);
2325                         brelse(bh);
2326                         return err;
2327                 }
2328         }
2329         memset(frames, 0, sizeof(frames));
2330         frame = frames;
2331         frame->entries = entries;
2332         frame->at = entries;
2333         frame->bh = bh;
2334
2335         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2336         if (retval)
2337                 goto out_frames;
2338         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2339         if (retval)
2340                 goto out_frames;
2341
2342         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2343         if (IS_ERR(de)) {
2344                 retval = PTR_ERR(de);
2345                 goto out_frames;
2346         }
2347
2348         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2349 out_frames:
2350         /*
2351          * Even if the block split failed, we have to properly write
2352          * out all the changes we did so far. Otherwise we can end up
2353          * with corrupted filesystem.
2354          */
2355         if (retval)
2356                 ext4_mark_inode_dirty(handle, dir);
2357         dx_release(frames);
2358         brelse(bh2);
2359         return retval;
2360 }
2361
2362 /*
2363  *      ext4_add_entry()
2364  *
2365  * adds a file entry to the specified directory, using the same
2366  * semantics as ext4_find_entry(). It returns NULL if it failed.
2367  *
2368  * NOTE!! The inode part of 'de' is left at 0 - which means you
2369  * may not sleep between calling this and putting something into
2370  * the entry, as someone else might have used it while you slept.
2371  */
2372 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2373                           struct inode *inode)
2374 {
2375         struct inode *dir = d_inode(dentry->d_parent);
2376         struct buffer_head *bh = NULL;
2377         struct ext4_dir_entry_2 *de;
2378         struct super_block *sb;
2379         struct ext4_filename fname;
2380         int     retval;
2381         int     dx_fallback=0;
2382         unsigned blocksize;
2383         ext4_lblk_t block, blocks;
2384         int     csum_size = 0;
2385
2386         if (ext4_has_metadata_csum(inode->i_sb))
2387                 csum_size = sizeof(struct ext4_dir_entry_tail);
2388
2389         sb = dir->i_sb;
2390         blocksize = sb->s_blocksize;
2391         if (!dentry->d_name.len)
2392                 return -EINVAL;
2393
2394         if (fscrypt_is_nokey_name(dentry))
2395                 return -ENOKEY;
2396
2397 #if IS_ENABLED(CONFIG_UNICODE)
2398         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2399             utf8_validate(sb->s_encoding, &dentry->d_name))
2400                 return -EINVAL;
2401 #endif
2402
2403         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2404         if (retval)
2405                 return retval;
2406
2407         if (ext4_has_inline_data(dir)) {
2408                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2409                 if (retval < 0)
2410                         goto out;
2411                 if (retval == 1) {
2412                         retval = 0;
2413                         goto out;
2414                 }
2415         }
2416
2417         if (is_dx(dir)) {
2418                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2419                 if (!retval || (retval != ERR_BAD_DX_DIR))
2420                         goto out;
2421                 /* Can we just ignore htree data? */
2422                 if (ext4_has_metadata_csum(sb)) {
2423                         EXT4_ERROR_INODE(dir,
2424                                 "Directory has corrupted htree index.");
2425                         retval = -EFSCORRUPTED;
2426                         goto out;
2427                 }
2428                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2429                 dx_fallback++;
2430                 retval = ext4_mark_inode_dirty(handle, dir);
2431                 if (unlikely(retval))
2432                         goto out;
2433         }
2434         blocks = dir->i_size >> sb->s_blocksize_bits;
2435         for (block = 0; block < blocks; block++) {
2436                 bh = ext4_read_dirblock(dir, block, DIRENT);
2437                 if (bh == NULL) {
2438                         bh = ext4_bread(handle, dir, block,
2439                                         EXT4_GET_BLOCKS_CREATE);
2440                         goto add_to_new_block;
2441                 }
2442                 if (IS_ERR(bh)) {
2443                         retval = PTR_ERR(bh);
2444                         bh = NULL;
2445                         goto out;
2446                 }
2447                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2448                                            NULL, bh);
2449                 if (retval != -ENOSPC)
2450                         goto out;
2451
2452                 if (blocks == 1 && !dx_fallback &&
2453                     ext4_has_feature_dir_index(sb)) {
2454                         retval = make_indexed_dir(handle, &fname, dir,
2455                                                   inode, bh);
2456                         bh = NULL; /* make_indexed_dir releases bh */
2457                         goto out;
2458                 }
2459                 brelse(bh);
2460         }
2461         bh = ext4_append(handle, dir, &block);
2462 add_to_new_block:
2463         if (IS_ERR(bh)) {
2464                 retval = PTR_ERR(bh);
2465                 bh = NULL;
2466                 goto out;
2467         }
2468         de = (struct ext4_dir_entry_2 *) bh->b_data;
2469         de->inode = 0;
2470         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2471
2472         if (csum_size)
2473                 ext4_initialize_dirent_tail(bh, blocksize);
2474
2475         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2476 out:
2477         ext4_fname_free_filename(&fname);
2478         brelse(bh);
2479         if (retval == 0)
2480                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2481         return retval;
2482 }
2483
2484 /*
2485  * Returns 0 for success, or a negative error value
2486  */
2487 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2488                              struct inode *dir, struct inode *inode)
2489 {
2490         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2491         struct dx_entry *entries, *at;
2492         struct buffer_head *bh;
2493         struct super_block *sb = dir->i_sb;
2494         struct ext4_dir_entry_2 *de;
2495         int restart;
2496         int err;
2497
2498 again:
2499         restart = 0;
2500         frame = dx_probe(fname, dir, NULL, frames);
2501         if (IS_ERR(frame))
2502                 return PTR_ERR(frame);
2503         entries = frame->entries;
2504         at = frame->at;
2505         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2506         if (IS_ERR(bh)) {
2507                 err = PTR_ERR(bh);
2508                 bh = NULL;
2509                 goto cleanup;
2510         }
2511
2512         BUFFER_TRACE(bh, "get_write_access");
2513         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2514         if (err)
2515                 goto journal_error;
2516
2517         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2518         if (err != -ENOSPC)
2519                 goto cleanup;
2520
2521         err = 0;
2522         /* Block full, should compress but for now just split */
2523         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2524                        dx_get_count(entries), dx_get_limit(entries)));
2525         /* Need to split index? */
2526         if (dx_get_count(entries) == dx_get_limit(entries)) {
2527                 ext4_lblk_t newblock;
2528                 int levels = frame - frames + 1;
2529                 unsigned int icount;
2530                 int add_level = 1;
2531                 struct dx_entry *entries2;
2532                 struct dx_node *node2;
2533                 struct buffer_head *bh2;
2534
2535                 while (frame > frames) {
2536                         if (dx_get_count((frame - 1)->entries) <
2537                             dx_get_limit((frame - 1)->entries)) {
2538                                 add_level = 0;
2539                                 break;
2540                         }
2541                         frame--; /* split higher index block */
2542                         at = frame->at;
2543                         entries = frame->entries;
2544                         restart = 1;
2545                 }
2546                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2547                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2548                                          "reach max htree level :%d",
2549                                          dir->i_ino, levels);
2550                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2551                                 ext4_warning(sb, "Large directory feature is "
2552                                                  "not enabled on this "
2553                                                  "filesystem");
2554                         }
2555                         err = -ENOSPC;
2556                         goto cleanup;
2557                 }
2558                 icount = dx_get_count(entries);
2559                 bh2 = ext4_append(handle, dir, &newblock);
2560                 if (IS_ERR(bh2)) {
2561                         err = PTR_ERR(bh2);
2562                         goto cleanup;
2563                 }
2564                 node2 = (struct dx_node *)(bh2->b_data);
2565                 entries2 = node2->entries;
2566                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2567                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2568                                                            sb->s_blocksize);
2569                 BUFFER_TRACE(frame->bh, "get_write_access");
2570                 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2571                                                     EXT4_JTR_NONE);
2572                 if (err)
2573                         goto journal_error;
2574                 if (!add_level) {
2575                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2576                         unsigned hash2 = dx_get_hash(entries + icount1);
2577                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2578                                        icount1, icount2));
2579
2580                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2581                         err = ext4_journal_get_write_access(handle, sb,
2582                                                             (frame - 1)->bh,
2583                                                             EXT4_JTR_NONE);
2584                         if (err)
2585                                 goto journal_error;
2586
2587                         memcpy((char *) entries2, (char *) (entries + icount1),
2588                                icount2 * sizeof(struct dx_entry));
2589                         dx_set_count(entries, icount1);
2590                         dx_set_count(entries2, icount2);
2591                         dx_set_limit(entries2, dx_node_limit(dir));
2592
2593                         /* Which index block gets the new entry? */
2594                         if (at - entries >= icount1) {
2595                                 frame->at = at - entries - icount1 + entries2;
2596                                 frame->entries = entries = entries2;
2597                                 swap(frame->bh, bh2);
2598                         }
2599                         dx_insert_block((frame - 1), hash2, newblock);
2600                         dxtrace(dx_show_index("node", frame->entries));
2601                         dxtrace(dx_show_index("node",
2602                                ((struct dx_node *) bh2->b_data)->entries));
2603                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2604                         if (err)
2605                                 goto journal_error;
2606                         brelse (bh2);
2607                         err = ext4_handle_dirty_dx_node(handle, dir,
2608                                                    (frame - 1)->bh);
2609                         if (err)
2610                                 goto journal_error;
2611                         err = ext4_handle_dirty_dx_node(handle, dir,
2612                                                         frame->bh);
2613                         if (restart || err)
2614                                 goto journal_error;
2615                 } else {
2616                         struct dx_root *dxroot;
2617                         memcpy((char *) entries2, (char *) entries,
2618                                icount * sizeof(struct dx_entry));
2619                         dx_set_limit(entries2, dx_node_limit(dir));
2620
2621                         /* Set up root */
2622                         dx_set_count(entries, 1);
2623                         dx_set_block(entries + 0, newblock);
2624                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2625                         dxroot->info.indirect_levels += 1;
2626                         dxtrace(printk(KERN_DEBUG
2627                                        "Creating %d level index...\n",
2628                                        dxroot->info.indirect_levels));
2629                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2630                         if (err)
2631                                 goto journal_error;
2632                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2633                         brelse(bh2);
2634                         restart = 1;
2635                         goto journal_error;
2636                 }
2637         }
2638         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2639         if (IS_ERR(de)) {
2640                 err = PTR_ERR(de);
2641                 goto cleanup;
2642         }
2643         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2644         goto cleanup;
2645
2646 journal_error:
2647         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2648 cleanup:
2649         brelse(bh);
2650         dx_release(frames);
2651         /* @restart is true means htree-path has been changed, we need to
2652          * repeat dx_probe() to find out valid htree-path
2653          */
2654         if (restart && err == 0)
2655                 goto again;
2656         return err;
2657 }
2658
2659 /*
2660  * ext4_generic_delete_entry deletes a directory entry by merging it
2661  * with the previous entry
2662  */
2663 int ext4_generic_delete_entry(struct inode *dir,
2664                               struct ext4_dir_entry_2 *de_del,
2665                               struct buffer_head *bh,
2666                               void *entry_buf,
2667                               int buf_size,
2668                               int csum_size)
2669 {
2670         struct ext4_dir_entry_2 *de, *pde;
2671         unsigned int blocksize = dir->i_sb->s_blocksize;
2672         int i;
2673
2674         i = 0;
2675         pde = NULL;
2676         de = entry_buf;
2677         while (i < buf_size - csum_size) {
2678                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2679                                          entry_buf, buf_size, i))
2680                         return -EFSCORRUPTED;
2681                 if (de == de_del)  {
2682                         if (pde) {
2683                                 pde->rec_len = ext4_rec_len_to_disk(
2684                                         ext4_rec_len_from_disk(pde->rec_len,
2685                                                                blocksize) +
2686                                         ext4_rec_len_from_disk(de->rec_len,
2687                                                                blocksize),
2688                                         blocksize);
2689
2690                                 /* wipe entire dir_entry */
2691                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2692                                                                 blocksize));
2693                         } else {
2694                                 /* wipe dir_entry excluding the rec_len field */
2695                                 de->inode = 0;
2696                                 memset(&de->name_len, 0,
2697                                         ext4_rec_len_from_disk(de->rec_len,
2698                                                                 blocksize) -
2699                                         offsetof(struct ext4_dir_entry_2,
2700                                                                 name_len));
2701                         }
2702
2703                         inode_inc_iversion(dir);
2704                         return 0;
2705                 }
2706                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2707                 pde = de;
2708                 de = ext4_next_entry(de, blocksize);
2709         }
2710         return -ENOENT;
2711 }
2712
2713 static int ext4_delete_entry(handle_t *handle,
2714                              struct inode *dir,
2715                              struct ext4_dir_entry_2 *de_del,
2716                              struct buffer_head *bh)
2717 {
2718         int err, csum_size = 0;
2719
2720         if (ext4_has_inline_data(dir)) {
2721                 int has_inline_data = 1;
2722                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2723                                                &has_inline_data);
2724                 if (has_inline_data)
2725                         return err;
2726         }
2727
2728         if (ext4_has_metadata_csum(dir->i_sb))
2729                 csum_size = sizeof(struct ext4_dir_entry_tail);
2730
2731         BUFFER_TRACE(bh, "get_write_access");
2732         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2733                                             EXT4_JTR_NONE);
2734         if (unlikely(err))
2735                 goto out;
2736
2737         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2738                                         dir->i_sb->s_blocksize, csum_size);
2739         if (err)
2740                 goto out;
2741
2742         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2743         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2744         if (unlikely(err))
2745                 goto out;
2746
2747         return 0;
2748 out:
2749         if (err != -ENOENT)
2750                 ext4_std_error(dir->i_sb, err);
2751         return err;
2752 }
2753
2754 /*
2755  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2756  * since this indicates that nlinks count was previously 1 to avoid overflowing
2757  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2758  * that subdirectory link counts are not being maintained accurately.
2759  *
2760  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2761  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2762  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2763  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2764  */
2765 static void ext4_inc_count(struct inode *inode)
2766 {
2767         inc_nlink(inode);
2768         if (is_dx(inode) &&
2769             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2770                 set_nlink(inode, 1);
2771 }
2772
2773 /*
2774  * If a directory had nlink == 1, then we should let it be 1. This indicates
2775  * directory has >EXT4_LINK_MAX subdirs.
2776  */
2777 static void ext4_dec_count(struct inode *inode)
2778 {
2779         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2780                 drop_nlink(inode);
2781 }
2782
2783
2784 /*
2785  * Add non-directory inode to a directory. On success, the inode reference is
2786  * consumed by dentry is instantiation. This is also indicated by clearing of
2787  * *inodep pointer. On failure, the caller is responsible for dropping the
2788  * inode reference in the safe context.
2789  */
2790 static int ext4_add_nondir(handle_t *handle,
2791                 struct dentry *dentry, struct inode **inodep)
2792 {
2793         struct inode *dir = d_inode(dentry->d_parent);
2794         struct inode *inode = *inodep;
2795         int err = ext4_add_entry(handle, dentry, inode);
2796         if (!err) {
2797                 err = ext4_mark_inode_dirty(handle, inode);
2798                 if (IS_DIRSYNC(dir))
2799                         ext4_handle_sync(handle);
2800                 d_instantiate_new(dentry, inode);
2801                 *inodep = NULL;
2802                 return err;
2803         }
2804         drop_nlink(inode);
2805         ext4_mark_inode_dirty(handle, inode);
2806         ext4_orphan_add(handle, inode);
2807         unlock_new_inode(inode);
2808         return err;
2809 }
2810
2811 /*
2812  * By the time this is called, we already have created
2813  * the directory cache entry for the new file, but it
2814  * is so far negative - it has no inode.
2815  *
2816  * If the create succeeds, we fill in the inode information
2817  * with d_instantiate().
2818  */
2819 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2820                        struct dentry *dentry, umode_t mode, bool excl)
2821 {
2822         handle_t *handle;
2823         struct inode *inode;
2824         int err, credits, retries = 0;
2825
2826         err = dquot_initialize(dir);
2827         if (err)
2828                 return err;
2829
2830         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2831                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2832 retry:
2833         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2834                                             0, NULL, EXT4_HT_DIR, credits);
2835         handle = ext4_journal_current_handle();
2836         err = PTR_ERR(inode);
2837         if (!IS_ERR(inode)) {
2838                 inode->i_op = &ext4_file_inode_operations;
2839                 inode->i_fop = &ext4_file_operations;
2840                 ext4_set_aops(inode);
2841                 err = ext4_add_nondir(handle, dentry, &inode);
2842                 if (!err)
2843                         ext4_fc_track_create(handle, dentry);
2844         }
2845         if (handle)
2846                 ext4_journal_stop(handle);
2847         if (!IS_ERR_OR_NULL(inode))
2848                 iput(inode);
2849         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2850                 goto retry;
2851         return err;
2852 }
2853
2854 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2855                       struct dentry *dentry, umode_t mode, dev_t rdev)
2856 {
2857         handle_t *handle;
2858         struct inode *inode;
2859         int err, credits, retries = 0;
2860
2861         err = dquot_initialize(dir);
2862         if (err)
2863                 return err;
2864
2865         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2866                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2867 retry:
2868         inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2869                                             0, NULL, EXT4_HT_DIR, credits);
2870         handle = ext4_journal_current_handle();
2871         err = PTR_ERR(inode);
2872         if (!IS_ERR(inode)) {
2873                 init_special_inode(inode, inode->i_mode, rdev);
2874                 inode->i_op = &ext4_special_inode_operations;
2875                 err = ext4_add_nondir(handle, dentry, &inode);
2876                 if (!err)
2877                         ext4_fc_track_create(handle, dentry);
2878         }
2879         if (handle)
2880                 ext4_journal_stop(handle);
2881         if (!IS_ERR_OR_NULL(inode))
2882                 iput(inode);
2883         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2884                 goto retry;
2885         return err;
2886 }
2887
2888 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2889                         struct file *file, umode_t mode)
2890 {
2891         handle_t *handle;
2892         struct inode *inode;
2893         int err, retries = 0;
2894
2895         err = dquot_initialize(dir);
2896         if (err)
2897                 return err;
2898
2899 retry:
2900         inode = ext4_new_inode_start_handle(idmap, dir, mode,
2901                                             NULL, 0, NULL,
2902                                             EXT4_HT_DIR,
2903                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2904                           4 + EXT4_XATTR_TRANS_BLOCKS);
2905         handle = ext4_journal_current_handle();
2906         err = PTR_ERR(inode);
2907         if (!IS_ERR(inode)) {
2908                 inode->i_op = &ext4_file_inode_operations;
2909                 inode->i_fop = &ext4_file_operations;
2910                 ext4_set_aops(inode);
2911                 d_tmpfile(file, inode);
2912                 err = ext4_orphan_add(handle, inode);
2913                 if (err)
2914                         goto err_unlock_inode;
2915                 mark_inode_dirty(inode);
2916                 unlock_new_inode(inode);
2917         }
2918         if (handle)
2919                 ext4_journal_stop(handle);
2920         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2921                 goto retry;
2922         return finish_open_simple(file, err);
2923 err_unlock_inode:
2924         ext4_journal_stop(handle);
2925         unlock_new_inode(inode);
2926         return err;
2927 }
2928
2929 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2930                           struct ext4_dir_entry_2 *de,
2931                           int blocksize, int csum_size,
2932                           unsigned int parent_ino, int dotdot_real_len)
2933 {
2934         de->inode = cpu_to_le32(inode->i_ino);
2935         de->name_len = 1;
2936         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2937                                            blocksize);
2938         strcpy(de->name, ".");
2939         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2940
2941         de = ext4_next_entry(de, blocksize);
2942         de->inode = cpu_to_le32(parent_ino);
2943         de->name_len = 2;
2944         if (!dotdot_real_len)
2945                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2946                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2947                                         blocksize);
2948         else
2949                 de->rec_len = ext4_rec_len_to_disk(
2950                                         ext4_dir_rec_len(de->name_len, NULL),
2951                                         blocksize);
2952         strcpy(de->name, "..");
2953         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2954
2955         return ext4_next_entry(de, blocksize);
2956 }
2957
2958 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2959                              struct inode *inode)
2960 {
2961         struct buffer_head *dir_block = NULL;
2962         struct ext4_dir_entry_2 *de;
2963         ext4_lblk_t block = 0;
2964         unsigned int blocksize = dir->i_sb->s_blocksize;
2965         int csum_size = 0;
2966         int err;
2967
2968         if (ext4_has_metadata_csum(dir->i_sb))
2969                 csum_size = sizeof(struct ext4_dir_entry_tail);
2970
2971         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2972                 err = ext4_try_create_inline_dir(handle, dir, inode);
2973                 if (err < 0 && err != -ENOSPC)
2974                         goto out;
2975                 if (!err)
2976                         goto out;
2977         }
2978
2979         inode->i_size = 0;
2980         dir_block = ext4_append(handle, inode, &block);
2981         if (IS_ERR(dir_block))
2982                 return PTR_ERR(dir_block);
2983         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2984         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2985         set_nlink(inode, 2);
2986         if (csum_size)
2987                 ext4_initialize_dirent_tail(dir_block, blocksize);
2988
2989         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2990         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2991         if (err)
2992                 goto out;
2993         set_buffer_verified(dir_block);
2994 out:
2995         brelse(dir_block);
2996         return err;
2997 }
2998
2999 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3000                       struct dentry *dentry, umode_t mode)
3001 {
3002         handle_t *handle;
3003         struct inode *inode;
3004         int err, err2 = 0, credits, retries = 0;
3005
3006         if (EXT4_DIR_LINK_MAX(dir))
3007                 return -EMLINK;
3008
3009         err = dquot_initialize(dir);
3010         if (err)
3011                 return err;
3012
3013         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3014                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3015 retry:
3016         inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3017                                             &dentry->d_name,
3018                                             0, NULL, EXT4_HT_DIR, credits);
3019         handle = ext4_journal_current_handle();
3020         err = PTR_ERR(inode);
3021         if (IS_ERR(inode))
3022                 goto out_stop;
3023
3024         inode->i_op = &ext4_dir_inode_operations;
3025         inode->i_fop = &ext4_dir_operations;
3026         err = ext4_init_new_dir(handle, dir, inode);
3027         if (err)
3028                 goto out_clear_inode;
3029         err = ext4_mark_inode_dirty(handle, inode);
3030         if (!err)
3031                 err = ext4_add_entry(handle, dentry, inode);
3032         if (err) {
3033 out_clear_inode:
3034                 clear_nlink(inode);
3035                 ext4_orphan_add(handle, inode);
3036                 unlock_new_inode(inode);
3037                 err2 = ext4_mark_inode_dirty(handle, inode);
3038                 if (unlikely(err2))
3039                         err = err2;
3040                 ext4_journal_stop(handle);
3041                 iput(inode);
3042                 goto out_retry;
3043         }
3044         ext4_inc_count(dir);
3045
3046         ext4_update_dx_flag(dir);
3047         err = ext4_mark_inode_dirty(handle, dir);
3048         if (err)
3049                 goto out_clear_inode;
3050         d_instantiate_new(dentry, inode);
3051         ext4_fc_track_create(handle, dentry);
3052         if (IS_DIRSYNC(dir))
3053                 ext4_handle_sync(handle);
3054
3055 out_stop:
3056         if (handle)
3057                 ext4_journal_stop(handle);
3058 out_retry:
3059         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3060                 goto retry;
3061         return err;
3062 }
3063
3064 /*
3065  * routine to check that the specified directory is empty (for rmdir)
3066  */
3067 bool ext4_empty_dir(struct inode *inode)
3068 {
3069         unsigned int offset;
3070         struct buffer_head *bh;
3071         struct ext4_dir_entry_2 *de;
3072         struct super_block *sb;
3073
3074         if (ext4_has_inline_data(inode)) {
3075                 int has_inline_data = 1;
3076                 int ret;
3077
3078                 ret = empty_inline_dir(inode, &has_inline_data);
3079                 if (has_inline_data)
3080                         return ret;
3081         }
3082
3083         sb = inode->i_sb;
3084         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3085                                         ext4_dir_rec_len(2, NULL)) {
3086                 EXT4_ERROR_INODE(inode, "invalid size");
3087                 return false;
3088         }
3089         /* The first directory block must not be a hole,
3090          * so treat it as DIRENT_HTREE
3091          */
3092         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3093         if (IS_ERR(bh))
3094                 return false;
3095
3096         de = (struct ext4_dir_entry_2 *) bh->b_data;
3097         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3098                                  0) ||
3099             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3100                 ext4_warning_inode(inode, "directory missing '.'");
3101                 brelse(bh);
3102                 return false;
3103         }
3104         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3105         de = ext4_next_entry(de, sb->s_blocksize);
3106         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3107                                  offset) ||
3108             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3109                 ext4_warning_inode(inode, "directory missing '..'");
3110                 brelse(bh);
3111                 return false;
3112         }
3113         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3114         while (offset < inode->i_size) {
3115                 if (!(offset & (sb->s_blocksize - 1))) {
3116                         unsigned int lblock;
3117                         brelse(bh);
3118                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3119                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3120                         if (bh == NULL) {
3121                                 offset += sb->s_blocksize;
3122                                 continue;
3123                         }
3124                         if (IS_ERR(bh))
3125                                 return false;
3126                 }
3127                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3128                                         (offset & (sb->s_blocksize - 1)));
3129                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3130                                          bh->b_data, bh->b_size, offset) ||
3131                     le32_to_cpu(de->inode)) {
3132                         brelse(bh);
3133                         return false;
3134                 }
3135                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3136         }
3137         brelse(bh);
3138         return true;
3139 }
3140
3141 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3142 {
3143         int retval;
3144         struct inode *inode;
3145         struct buffer_head *bh;
3146         struct ext4_dir_entry_2 *de;
3147         handle_t *handle = NULL;
3148
3149         if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3150                 return -EIO;
3151
3152         /* Initialize quotas before so that eventual writes go in
3153          * separate transaction */
3154         retval = dquot_initialize(dir);
3155         if (retval)
3156                 return retval;
3157         retval = dquot_initialize(d_inode(dentry));
3158         if (retval)
3159                 return retval;
3160
3161         retval = -ENOENT;
3162         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3163         if (IS_ERR(bh))
3164                 return PTR_ERR(bh);
3165         if (!bh)
3166                 goto end_rmdir;
3167
3168         inode = d_inode(dentry);
3169
3170         retval = -EFSCORRUPTED;
3171         if (le32_to_cpu(de->inode) != inode->i_ino)
3172                 goto end_rmdir;
3173
3174         retval = -ENOTEMPTY;
3175         if (!ext4_empty_dir(inode))
3176                 goto end_rmdir;
3177
3178         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3179                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3180         if (IS_ERR(handle)) {
3181                 retval = PTR_ERR(handle);
3182                 handle = NULL;
3183                 goto end_rmdir;
3184         }
3185
3186         if (IS_DIRSYNC(dir))
3187                 ext4_handle_sync(handle);
3188
3189         retval = ext4_delete_entry(handle, dir, de, bh);
3190         if (retval)
3191                 goto end_rmdir;
3192         if (!EXT4_DIR_LINK_EMPTY(inode))
3193                 ext4_warning_inode(inode,
3194                              "empty directory '%.*s' has too many links (%u)",
3195                              dentry->d_name.len, dentry->d_name.name,
3196                              inode->i_nlink);
3197         inode_inc_iversion(inode);
3198         clear_nlink(inode);
3199         /* There's no need to set i_disksize: the fact that i_nlink is
3200          * zero will ensure that the right thing happens during any
3201          * recovery. */
3202         inode->i_size = 0;
3203         ext4_orphan_add(handle, inode);
3204         inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3205         inode_set_ctime_current(inode);
3206         retval = ext4_mark_inode_dirty(handle, inode);
3207         if (retval)
3208                 goto end_rmdir;
3209         ext4_dec_count(dir);
3210         ext4_update_dx_flag(dir);
3211         ext4_fc_track_unlink(handle, dentry);
3212         retval = ext4_mark_inode_dirty(handle, dir);
3213
3214 #if IS_ENABLED(CONFIG_UNICODE)
3215         /* VFS negative dentries are incompatible with Encoding and
3216          * Case-insensitiveness. Eventually we'll want avoid
3217          * invalidating the dentries here, alongside with returning the
3218          * negative dentries at ext4_lookup(), when it is better
3219          * supported by the VFS for the CI case.
3220          */
3221         if (IS_CASEFOLDED(dir))
3222                 d_invalidate(dentry);
3223 #endif
3224
3225 end_rmdir:
3226         brelse(bh);
3227         if (handle)
3228                 ext4_journal_stop(handle);
3229         return retval;
3230 }
3231
3232 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3233                   struct inode *inode,
3234                   struct dentry *dentry /* NULL during fast_commit recovery */)
3235 {
3236         int retval = -ENOENT;
3237         struct buffer_head *bh;
3238         struct ext4_dir_entry_2 *de;
3239         handle_t *handle;
3240         int skip_remove_dentry = 0;
3241
3242         /*
3243          * Keep this outside the transaction; it may have to set up the
3244          * directory's encryption key, which isn't GFP_NOFS-safe.
3245          */
3246         bh = ext4_find_entry(dir, d_name, &de, NULL);
3247         if (IS_ERR(bh))
3248                 return PTR_ERR(bh);
3249
3250         if (!bh)
3251                 return -ENOENT;
3252
3253         if (le32_to_cpu(de->inode) != inode->i_ino) {
3254                 /*
3255                  * It's okay if we find dont find dentry which matches
3256                  * the inode. That's because it might have gotten
3257                  * renamed to a different inode number
3258                  */
3259                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3260                         skip_remove_dentry = 1;
3261                 else
3262                         goto out_bh;
3263         }
3264
3265         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3266                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3267         if (IS_ERR(handle)) {
3268                 retval = PTR_ERR(handle);
3269                 goto out_bh;
3270         }
3271
3272         if (IS_DIRSYNC(dir))
3273                 ext4_handle_sync(handle);
3274
3275         if (!skip_remove_dentry) {
3276                 retval = ext4_delete_entry(handle, dir, de, bh);
3277                 if (retval)
3278                         goto out_handle;
3279                 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3280                 ext4_update_dx_flag(dir);
3281                 retval = ext4_mark_inode_dirty(handle, dir);
3282                 if (retval)
3283                         goto out_handle;
3284         } else {
3285                 retval = 0;
3286         }
3287         if (inode->i_nlink == 0)
3288                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3289                                    d_name->len, d_name->name);
3290         else
3291                 drop_nlink(inode);
3292         if (!inode->i_nlink)
3293                 ext4_orphan_add(handle, inode);
3294         inode_set_ctime_current(inode);
3295         retval = ext4_mark_inode_dirty(handle, inode);
3296         if (dentry && !retval)
3297                 ext4_fc_track_unlink(handle, dentry);
3298 out_handle:
3299         ext4_journal_stop(handle);
3300 out_bh:
3301         brelse(bh);
3302         return retval;
3303 }
3304
3305 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3306 {
3307         int retval;
3308
3309         if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3310                 return -EIO;
3311
3312         trace_ext4_unlink_enter(dir, dentry);
3313         /*
3314          * Initialize quotas before so that eventual writes go
3315          * in separate transaction
3316          */
3317         retval = dquot_initialize(dir);
3318         if (retval)
3319                 goto out_trace;
3320         retval = dquot_initialize(d_inode(dentry));
3321         if (retval)
3322                 goto out_trace;
3323
3324         retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3325 #if IS_ENABLED(CONFIG_UNICODE)
3326         /* VFS negative dentries are incompatible with Encoding and
3327          * Case-insensitiveness. Eventually we'll want avoid
3328          * invalidating the dentries here, alongside with returning the
3329          * negative dentries at ext4_lookup(), when it is  better
3330          * supported by the VFS for the CI case.
3331          */
3332         if (IS_CASEFOLDED(dir))
3333                 d_invalidate(dentry);
3334 #endif
3335
3336 out_trace:
3337         trace_ext4_unlink_exit(dentry, retval);
3338         return retval;
3339 }
3340
3341 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3342                                    struct fscrypt_str *disk_link)
3343 {
3344         struct buffer_head *bh;
3345         char *kaddr;
3346         int err = 0;
3347
3348         bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3349         if (IS_ERR(bh))
3350                 return PTR_ERR(bh);
3351
3352         BUFFER_TRACE(bh, "get_write_access");
3353         err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3354         if (err)
3355                 goto out;
3356
3357         kaddr = (char *)bh->b_data;
3358         memcpy(kaddr, disk_link->name, disk_link->len);
3359         inode->i_size = disk_link->len - 1;
3360         EXT4_I(inode)->i_disksize = inode->i_size;
3361         err = ext4_handle_dirty_metadata(handle, inode, bh);
3362 out:
3363         brelse(bh);
3364         return err;
3365 }
3366
3367 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3368                         struct dentry *dentry, const char *symname)
3369 {
3370         handle_t *handle;
3371         struct inode *inode;
3372         int err, len = strlen(symname);
3373         int credits;
3374         struct fscrypt_str disk_link;
3375         int retries = 0;
3376
3377         if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3378                 return -EIO;
3379
3380         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3381                                       &disk_link);
3382         if (err)
3383                 return err;
3384
3385         err = dquot_initialize(dir);
3386         if (err)
3387                 return err;
3388
3389         /*
3390          * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3391          * directory. +3 for inode, inode bitmap, group descriptor allocation.
3392          * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3393          * modification.
3394          */
3395         credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3396                   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3397 retry:
3398         inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3399                                             &dentry->d_name, 0, NULL,
3400                                             EXT4_HT_DIR, credits);
3401         handle = ext4_journal_current_handle();
3402         if (IS_ERR(inode)) {
3403                 if (handle)
3404                         ext4_journal_stop(handle);
3405                 err = PTR_ERR(inode);
3406                 goto out_retry;
3407         }
3408
3409         if (IS_ENCRYPTED(inode)) {
3410                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3411                 if (err)
3412                         goto err_drop_inode;
3413                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3414         } else {
3415                 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3416                         inode->i_op = &ext4_symlink_inode_operations;
3417                 } else {
3418                         inode->i_op = &ext4_fast_symlink_inode_operations;
3419                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3420                 }
3421         }
3422
3423         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3424                 /* alloc symlink block and fill it */
3425                 err = ext4_init_symlink_block(handle, inode, &disk_link);
3426                 if (err)
3427                         goto err_drop_inode;
3428         } else {
3429                 /* clear the extent format for fast symlink */
3430                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3431                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3432                        disk_link.len);
3433                 inode->i_size = disk_link.len - 1;
3434                 EXT4_I(inode)->i_disksize = inode->i_size;
3435         }
3436         err = ext4_add_nondir(handle, dentry, &inode);
3437         if (handle)
3438                 ext4_journal_stop(handle);
3439         iput(inode);
3440         goto out_retry;
3441
3442 err_drop_inode:
3443         clear_nlink(inode);
3444         ext4_mark_inode_dirty(handle, inode);
3445         ext4_orphan_add(handle, inode);
3446         unlock_new_inode(inode);
3447         if (handle)
3448                 ext4_journal_stop(handle);
3449         iput(inode);
3450 out_retry:
3451         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3452                 goto retry;
3453         if (disk_link.name != (unsigned char *)symname)
3454                 kfree(disk_link.name);
3455         return err;
3456 }
3457
3458 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3459 {
3460         handle_t *handle;
3461         int err, retries = 0;
3462 retry:
3463         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3464                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3465                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3466         if (IS_ERR(handle))
3467                 return PTR_ERR(handle);
3468
3469         if (IS_DIRSYNC(dir))
3470                 ext4_handle_sync(handle);
3471
3472         inode_set_ctime_current(inode);
3473         ext4_inc_count(inode);
3474         ihold(inode);
3475
3476         err = ext4_add_entry(handle, dentry, inode);
3477         if (!err) {
3478                 err = ext4_mark_inode_dirty(handle, inode);
3479                 /* this can happen only for tmpfile being
3480                  * linked the first time
3481                  */
3482                 if (inode->i_nlink == 1)
3483                         ext4_orphan_del(handle, inode);
3484                 d_instantiate(dentry, inode);
3485                 ext4_fc_track_link(handle, dentry);
3486         } else {
3487                 drop_nlink(inode);
3488                 iput(inode);
3489         }
3490         ext4_journal_stop(handle);
3491         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3492                 goto retry;
3493         return err;
3494 }
3495
3496 static int ext4_link(struct dentry *old_dentry,
3497                      struct inode *dir, struct dentry *dentry)
3498 {
3499         struct inode *inode = d_inode(old_dentry);
3500         int err;
3501
3502         if (inode->i_nlink >= EXT4_LINK_MAX)
3503                 return -EMLINK;
3504
3505         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3506         if (err)
3507                 return err;
3508
3509         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3510             (!projid_eq(EXT4_I(dir)->i_projid,
3511                         EXT4_I(old_dentry->d_inode)->i_projid)))
3512                 return -EXDEV;
3513
3514         err = dquot_initialize(dir);
3515         if (err)
3516                 return err;
3517         return __ext4_link(dir, inode, dentry);
3518 }
3519
3520 /*
3521  * Try to find buffer head where contains the parent block.
3522  * It should be the inode block if it is inlined or the 1st block
3523  * if it is a normal dir.
3524  */
3525 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3526                                         struct inode *inode,
3527                                         int *retval,
3528                                         struct ext4_dir_entry_2 **parent_de,
3529                                         int *inlined)
3530 {
3531         struct buffer_head *bh;
3532
3533         if (!ext4_has_inline_data(inode)) {
3534                 struct ext4_dir_entry_2 *de;
3535                 unsigned int offset;
3536
3537                 /* The first directory block must not be a hole, so
3538                  * treat it as DIRENT_HTREE
3539                  */
3540                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3541                 if (IS_ERR(bh)) {
3542                         *retval = PTR_ERR(bh);
3543                         return NULL;
3544                 }
3545
3546                 de = (struct ext4_dir_entry_2 *) bh->b_data;
3547                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3548                                          bh->b_size, 0) ||
3549                     le32_to_cpu(de->inode) != inode->i_ino ||
3550                     strcmp(".", de->name)) {
3551                         EXT4_ERROR_INODE(inode, "directory missing '.'");
3552                         brelse(bh);
3553                         *retval = -EFSCORRUPTED;
3554                         return NULL;
3555                 }
3556                 offset = ext4_rec_len_from_disk(de->rec_len,
3557                                                 inode->i_sb->s_blocksize);
3558                 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3559                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3560                                          bh->b_size, offset) ||
3561                     le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3562                         EXT4_ERROR_INODE(inode, "directory missing '..'");
3563                         brelse(bh);
3564                         *retval = -EFSCORRUPTED;
3565                         return NULL;
3566                 }
3567                 *parent_de = de;
3568
3569                 return bh;
3570         }
3571
3572         *inlined = 1;
3573         return ext4_get_first_inline_block(inode, parent_de, retval);
3574 }
3575
3576 struct ext4_renament {
3577         struct inode *dir;
3578         struct dentry *dentry;
3579         struct inode *inode;
3580         bool is_dir;
3581         int dir_nlink_delta;
3582
3583         /* entry for "dentry" */
3584         struct buffer_head *bh;
3585         struct ext4_dir_entry_2 *de;
3586         int inlined;
3587
3588         /* entry for ".." in inode if it's a directory */
3589         struct buffer_head *dir_bh;
3590         struct ext4_dir_entry_2 *parent_de;
3591         int dir_inlined;
3592 };
3593
3594 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3595 {
3596         int retval;
3597
3598         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3599                                               &retval, &ent->parent_de,
3600                                               &ent->dir_inlined);
3601         if (!ent->dir_bh)
3602                 return retval;
3603         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3604                 return -EFSCORRUPTED;
3605         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3606         return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3607                                              ent->dir_bh, EXT4_JTR_NONE);
3608 }
3609
3610 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3611                                   unsigned dir_ino)
3612 {
3613         int retval;
3614
3615         ent->parent_de->inode = cpu_to_le32(dir_ino);
3616         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3617         if (!ent->dir_inlined) {
3618                 if (is_dx(ent->inode)) {
3619                         retval = ext4_handle_dirty_dx_node(handle,
3620                                                            ent->inode,
3621                                                            ent->dir_bh);
3622                 } else {
3623                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3624                                                             ent->dir_bh);
3625                 }
3626         } else {
3627                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3628         }
3629         if (retval) {
3630                 ext4_std_error(ent->dir->i_sb, retval);
3631                 return retval;
3632         }
3633         return 0;
3634 }
3635
3636 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3637                        unsigned ino, unsigned file_type)
3638 {
3639         int retval, retval2;
3640
3641         BUFFER_TRACE(ent->bh, "get write access");
3642         retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3643                                                EXT4_JTR_NONE);
3644         if (retval)
3645                 return retval;
3646         ent->de->inode = cpu_to_le32(ino);
3647         if (ext4_has_feature_filetype(ent->dir->i_sb))
3648                 ent->de->file_type = file_type;
3649         inode_inc_iversion(ent->dir);
3650         inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3651         retval = ext4_mark_inode_dirty(handle, ent->dir);
3652         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3653         if (!ent->inlined) {
3654                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3655                 if (unlikely(retval2)) {
3656                         ext4_std_error(ent->dir->i_sb, retval2);
3657                         return retval2;
3658                 }
3659         }
3660         return retval;
3661 }
3662
3663 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3664                           unsigned ino, unsigned file_type)
3665 {
3666         struct ext4_renament old = *ent;
3667         int retval = 0;
3668
3669         /*
3670          * old->de could have moved from under us during make indexed dir,
3671          * so the old->de may no longer valid and need to find it again
3672          * before reset old inode info.
3673          */
3674         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3675                                  &old.inlined);
3676         if (IS_ERR(old.bh))
3677                 retval = PTR_ERR(old.bh);
3678         if (!old.bh)
3679                 retval = -ENOENT;
3680         if (retval) {
3681                 ext4_std_error(old.dir->i_sb, retval);
3682                 return;
3683         }
3684
3685         ext4_setent(handle, &old, ino, file_type);
3686         brelse(old.bh);
3687 }
3688
3689 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3690                                   const struct qstr *d_name)
3691 {
3692         int retval = -ENOENT;
3693         struct buffer_head *bh;
3694         struct ext4_dir_entry_2 *de;
3695
3696         bh = ext4_find_entry(dir, d_name, &de, NULL);
3697         if (IS_ERR(bh))
3698                 return PTR_ERR(bh);
3699         if (bh) {
3700                 retval = ext4_delete_entry(handle, dir, de, bh);
3701                 brelse(bh);
3702         }
3703         return retval;
3704 }
3705
3706 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3707                                int force_reread)
3708 {
3709         int retval;
3710         /*
3711          * ent->de could have moved from under us during htree split, so make
3712          * sure that we are deleting the right entry.  We might also be pointing
3713          * to a stale entry in the unused part of ent->bh so just checking inum
3714          * and the name isn't enough.
3715          */
3716         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3717             ent->de->name_len != ent->dentry->d_name.len ||
3718             strncmp(ent->de->name, ent->dentry->d_name.name,
3719                     ent->de->name_len) ||
3720             force_reread) {
3721                 retval = ext4_find_delete_entry(handle, ent->dir,
3722                                                 &ent->dentry->d_name);
3723         } else {
3724                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3725                 if (retval == -ENOENT) {
3726                         retval = ext4_find_delete_entry(handle, ent->dir,
3727                                                         &ent->dentry->d_name);
3728                 }
3729         }
3730
3731         if (retval) {
3732                 ext4_warning_inode(ent->dir,
3733                                    "Deleting old file: nlink %d, error=%d",
3734                                    ent->dir->i_nlink, retval);
3735         }
3736 }
3737
3738 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3739 {
3740         if (ent->dir_nlink_delta) {
3741                 if (ent->dir_nlink_delta == -1)
3742                         ext4_dec_count(ent->dir);
3743                 else
3744                         ext4_inc_count(ent->dir);
3745                 ext4_mark_inode_dirty(handle, ent->dir);
3746         }
3747 }
3748
3749 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3750                                               struct ext4_renament *ent,
3751                                               int credits, handle_t **h)
3752 {
3753         struct inode *wh;
3754         handle_t *handle;
3755         int retries = 0;
3756
3757         /*
3758          * for inode block, sb block, group summaries,
3759          * and inode bitmap
3760          */
3761         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3762                     EXT4_XATTR_TRANS_BLOCKS + 4);
3763 retry:
3764         wh = ext4_new_inode_start_handle(idmap, ent->dir,
3765                                          S_IFCHR | WHITEOUT_MODE,
3766                                          &ent->dentry->d_name, 0, NULL,
3767                                          EXT4_HT_DIR, credits);
3768
3769         handle = ext4_journal_current_handle();
3770         if (IS_ERR(wh)) {
3771                 if (handle)
3772                         ext4_journal_stop(handle);
3773                 if (PTR_ERR(wh) == -ENOSPC &&
3774                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3775                         goto retry;
3776         } else {
3777                 *h = handle;
3778                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3779                 wh->i_op = &ext4_special_inode_operations;
3780         }
3781         return wh;
3782 }
3783
3784 /*
3785  * Anybody can rename anything with this: the permission checks are left to the
3786  * higher-level routines.
3787  *
3788  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3789  * while new_{dentry,inode) refers to the destination dentry/inode
3790  * This comes from rename(const char *oldpath, const char *newpath)
3791  */
3792 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3793                        struct dentry *old_dentry, struct inode *new_dir,
3794                        struct dentry *new_dentry, unsigned int flags)
3795 {
3796         handle_t *handle = NULL;
3797         struct ext4_renament old = {
3798                 .dir = old_dir,
3799                 .dentry = old_dentry,
3800                 .inode = d_inode(old_dentry),
3801         };
3802         struct ext4_renament new = {
3803                 .dir = new_dir,
3804                 .dentry = new_dentry,
3805                 .inode = d_inode(new_dentry),
3806         };
3807         int force_reread;
3808         int retval;
3809         struct inode *whiteout = NULL;
3810         int credits;
3811         u8 old_file_type;
3812
3813         if (new.inode && new.inode->i_nlink == 0) {
3814                 EXT4_ERROR_INODE(new.inode,
3815                                  "target of rename is already freed");
3816                 return -EFSCORRUPTED;
3817         }
3818
3819         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3820             (!projid_eq(EXT4_I(new_dir)->i_projid,
3821                         EXT4_I(old_dentry->d_inode)->i_projid)))
3822                 return -EXDEV;
3823
3824         retval = dquot_initialize(old.dir);
3825         if (retval)
3826                 return retval;
3827         retval = dquot_initialize(old.inode);
3828         if (retval)
3829                 return retval;
3830         retval = dquot_initialize(new.dir);
3831         if (retval)
3832                 return retval;
3833
3834         /* Initialize quotas before so that eventual writes go
3835          * in separate transaction */
3836         if (new.inode) {
3837                 retval = dquot_initialize(new.inode);
3838                 if (retval)
3839                         return retval;
3840         }
3841
3842         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3843                                  &old.inlined);
3844         if (IS_ERR(old.bh))
3845                 return PTR_ERR(old.bh);
3846
3847         /*
3848          *  Check for inode number is _not_ due to possible IO errors.
3849          *  We might rmdir the source, keep it as pwd of some process
3850          *  and merrily kill the link to whatever was created under the
3851          *  same name. Goodbye sticky bit ;-<
3852          */
3853         retval = -ENOENT;
3854         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3855                 goto release_bh;
3856
3857         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3858                                  &new.de, &new.inlined);
3859         if (IS_ERR(new.bh)) {
3860                 retval = PTR_ERR(new.bh);
3861                 new.bh = NULL;
3862                 goto release_bh;
3863         }
3864         if (new.bh) {
3865                 if (!new.inode) {
3866                         brelse(new.bh);
3867                         new.bh = NULL;
3868                 }
3869         }
3870         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3871                 ext4_alloc_da_blocks(old.inode);
3872
3873         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3874                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3875         if (!(flags & RENAME_WHITEOUT)) {
3876                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3877                 if (IS_ERR(handle)) {
3878                         retval = PTR_ERR(handle);
3879                         goto release_bh;
3880                 }
3881         } else {
3882                 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3883                 if (IS_ERR(whiteout)) {
3884                         retval = PTR_ERR(whiteout);
3885                         goto release_bh;
3886                 }
3887         }
3888
3889         old_file_type = old.de->file_type;
3890         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3891                 ext4_handle_sync(handle);
3892
3893         if (S_ISDIR(old.inode->i_mode)) {
3894                 if (new.inode) {
3895                         retval = -ENOTEMPTY;
3896                         if (!ext4_empty_dir(new.inode))
3897                                 goto end_rename;
3898                 } else {
3899                         retval = -EMLINK;
3900                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3901                                 goto end_rename;
3902                 }
3903                 retval = ext4_rename_dir_prepare(handle, &old);
3904                 if (retval)
3905                         goto end_rename;
3906         }
3907         /*
3908          * If we're renaming a file within an inline_data dir and adding or
3909          * setting the new dirent causes a conversion from inline_data to
3910          * extents/blockmap, we need to force the dirent delete code to
3911          * re-read the directory, or else we end up trying to delete a dirent
3912          * from what is now the extent tree root (or a block map).
3913          */
3914         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3915                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3916
3917         if (whiteout) {
3918                 /*
3919                  * Do this before adding a new entry, so the old entry is sure
3920                  * to be still pointing to the valid old entry.
3921                  */
3922                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3923                                      EXT4_FT_CHRDEV);
3924                 if (retval)
3925                         goto end_rename;
3926                 retval = ext4_mark_inode_dirty(handle, whiteout);
3927                 if (unlikely(retval))
3928                         goto end_rename;
3929
3930         }
3931         if (!new.bh) {
3932                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3933                 if (retval)
3934                         goto end_rename;
3935         } else {
3936                 retval = ext4_setent(handle, &new,
3937                                      old.inode->i_ino, old_file_type);
3938                 if (retval)
3939                         goto end_rename;
3940         }
3941         if (force_reread)
3942                 force_reread = !ext4_test_inode_flag(new.dir,
3943                                                      EXT4_INODE_INLINE_DATA);
3944
3945         /*
3946          * Like most other Unix systems, set the ctime for inodes on a
3947          * rename.
3948          */
3949         inode_set_ctime_current(old.inode);
3950         retval = ext4_mark_inode_dirty(handle, old.inode);
3951         if (unlikely(retval))
3952                 goto end_rename;
3953
3954         if (!whiteout) {
3955                 /*
3956                  * ok, that's it
3957                  */
3958                 ext4_rename_delete(handle, &old, force_reread);
3959         }
3960
3961         if (new.inode) {
3962                 ext4_dec_count(new.inode);
3963                 inode_set_ctime_current(new.inode);
3964         }
3965         inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3966         ext4_update_dx_flag(old.dir);
3967         if (old.dir_bh) {
3968                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3969                 if (retval)
3970                         goto end_rename;
3971
3972                 ext4_dec_count(old.dir);
3973                 if (new.inode) {
3974                         /* checked ext4_empty_dir above, can't have another
3975                          * parent, ext4_dec_count() won't work for many-linked
3976                          * dirs */
3977                         clear_nlink(new.inode);
3978                 } else {
3979                         ext4_inc_count(new.dir);
3980                         ext4_update_dx_flag(new.dir);
3981                         retval = ext4_mark_inode_dirty(handle, new.dir);
3982                         if (unlikely(retval))
3983                                 goto end_rename;
3984                 }
3985         }
3986         retval = ext4_mark_inode_dirty(handle, old.dir);
3987         if (unlikely(retval))
3988                 goto end_rename;
3989
3990         if (S_ISDIR(old.inode->i_mode)) {
3991                 /*
3992                  * We disable fast commits here that's because the
3993                  * replay code is not yet capable of changing dot dot
3994                  * dirents in directories.
3995                  */
3996                 ext4_fc_mark_ineligible(old.inode->i_sb,
3997                         EXT4_FC_REASON_RENAME_DIR, handle);
3998         } else {
3999                 struct super_block *sb = old.inode->i_sb;
4000
4001                 if (new.inode)
4002                         ext4_fc_track_unlink(handle, new.dentry);
4003                 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4004                     !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4005                     !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4006                         __ext4_fc_track_link(handle, old.inode, new.dentry);
4007                         __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4008                         if (whiteout)
4009                                 __ext4_fc_track_create(handle, whiteout,
4010                                                        old.dentry);
4011                 }
4012         }
4013
4014         if (new.inode) {
4015                 retval = ext4_mark_inode_dirty(handle, new.inode);
4016                 if (unlikely(retval))
4017                         goto end_rename;
4018                 if (!new.inode->i_nlink)
4019                         ext4_orphan_add(handle, new.inode);
4020         }
4021         retval = 0;
4022
4023 end_rename:
4024         if (whiteout) {
4025                 if (retval) {
4026                         ext4_resetent(handle, &old,
4027                                       old.inode->i_ino, old_file_type);
4028                         drop_nlink(whiteout);
4029                         ext4_mark_inode_dirty(handle, whiteout);
4030                         ext4_orphan_add(handle, whiteout);
4031                 }
4032                 unlock_new_inode(whiteout);
4033                 ext4_journal_stop(handle);
4034                 iput(whiteout);
4035         } else {
4036                 ext4_journal_stop(handle);
4037         }
4038 release_bh:
4039         brelse(old.dir_bh);
4040         brelse(old.bh);
4041         brelse(new.bh);
4042
4043         return retval;
4044 }
4045
4046 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4047                              struct inode *new_dir, struct dentry *new_dentry)
4048 {
4049         handle_t *handle = NULL;
4050         struct ext4_renament old = {
4051                 .dir = old_dir,
4052                 .dentry = old_dentry,
4053                 .inode = d_inode(old_dentry),
4054         };
4055         struct ext4_renament new = {
4056                 .dir = new_dir,
4057                 .dentry = new_dentry,
4058                 .inode = d_inode(new_dentry),
4059         };
4060         u8 new_file_type;
4061         int retval;
4062
4063         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4064              !projid_eq(EXT4_I(new_dir)->i_projid,
4065                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
4066             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4067              !projid_eq(EXT4_I(old_dir)->i_projid,
4068                         EXT4_I(new_dentry->d_inode)->i_projid)))
4069                 return -EXDEV;
4070
4071         retval = dquot_initialize(old.dir);
4072         if (retval)
4073                 return retval;
4074         retval = dquot_initialize(new.dir);
4075         if (retval)
4076                 return retval;
4077
4078         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4079                                  &old.de, &old.inlined);
4080         if (IS_ERR(old.bh))
4081                 return PTR_ERR(old.bh);
4082         /*
4083          *  Check for inode number is _not_ due to possible IO errors.
4084          *  We might rmdir the source, keep it as pwd of some process
4085          *  and merrily kill the link to whatever was created under the
4086          *  same name. Goodbye sticky bit ;-<
4087          */
4088         retval = -ENOENT;
4089         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4090                 goto end_rename;
4091
4092         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4093                                  &new.de, &new.inlined);
4094         if (IS_ERR(new.bh)) {
4095                 retval = PTR_ERR(new.bh);
4096                 new.bh = NULL;
4097                 goto end_rename;
4098         }
4099
4100         /* RENAME_EXCHANGE case: old *and* new must both exist */
4101         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4102                 goto end_rename;
4103
4104         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4105                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4106                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4107         if (IS_ERR(handle)) {
4108                 retval = PTR_ERR(handle);
4109                 handle = NULL;
4110                 goto end_rename;
4111         }
4112
4113         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4114                 ext4_handle_sync(handle);
4115
4116         if (S_ISDIR(old.inode->i_mode)) {
4117                 old.is_dir = true;
4118                 retval = ext4_rename_dir_prepare(handle, &old);
4119                 if (retval)
4120                         goto end_rename;
4121         }
4122         if (S_ISDIR(new.inode->i_mode)) {
4123                 new.is_dir = true;
4124                 retval = ext4_rename_dir_prepare(handle, &new);
4125                 if (retval)
4126                         goto end_rename;
4127         }
4128
4129         /*
4130          * Other than the special case of overwriting a directory, parents'
4131          * nlink only needs to be modified if this is a cross directory rename.
4132          */
4133         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4134                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4135                 new.dir_nlink_delta = -old.dir_nlink_delta;
4136                 retval = -EMLINK;
4137                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4138                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4139                         goto end_rename;
4140         }
4141
4142         new_file_type = new.de->file_type;
4143         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4144         if (retval)
4145                 goto end_rename;
4146
4147         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4148         if (retval)
4149                 goto end_rename;
4150
4151         /*
4152          * Like most other Unix systems, set the ctime for inodes on a
4153          * rename.
4154          */
4155         inode_set_ctime_current(old.inode);
4156         inode_set_ctime_current(new.inode);
4157         retval = ext4_mark_inode_dirty(handle, old.inode);
4158         if (unlikely(retval))
4159                 goto end_rename;
4160         retval = ext4_mark_inode_dirty(handle, new.inode);
4161         if (unlikely(retval))
4162                 goto end_rename;
4163         ext4_fc_mark_ineligible(new.inode->i_sb,
4164                                 EXT4_FC_REASON_CROSS_RENAME, handle);
4165         if (old.dir_bh) {
4166                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4167                 if (retval)
4168                         goto end_rename;
4169         }
4170         if (new.dir_bh) {
4171                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4172                 if (retval)
4173                         goto end_rename;
4174         }
4175         ext4_update_dir_count(handle, &old);
4176         ext4_update_dir_count(handle, &new);
4177         retval = 0;
4178
4179 end_rename:
4180         brelse(old.dir_bh);
4181         brelse(new.dir_bh);
4182         brelse(old.bh);
4183         brelse(new.bh);
4184         if (handle)
4185                 ext4_journal_stop(handle);
4186         return retval;
4187 }
4188
4189 static int ext4_rename2(struct mnt_idmap *idmap,
4190                         struct inode *old_dir, struct dentry *old_dentry,
4191                         struct inode *new_dir, struct dentry *new_dentry,
4192                         unsigned int flags)
4193 {
4194         int err;
4195
4196         if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4197                 return -EIO;
4198
4199         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4200                 return -EINVAL;
4201
4202         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4203                                      flags);
4204         if (err)
4205                 return err;
4206
4207         if (flags & RENAME_EXCHANGE) {
4208                 return ext4_cross_rename(old_dir, old_dentry,
4209                                          new_dir, new_dentry);
4210         }
4211
4212         return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4213 }
4214
4215 /*
4216  * directories can handle most operations...
4217  */
4218 const struct inode_operations ext4_dir_inode_operations = {
4219         .create         = ext4_create,
4220         .lookup         = ext4_lookup,
4221         .link           = ext4_link,
4222         .unlink         = ext4_unlink,
4223         .symlink        = ext4_symlink,
4224         .mkdir          = ext4_mkdir,
4225         .rmdir          = ext4_rmdir,
4226         .mknod          = ext4_mknod,
4227         .tmpfile        = ext4_tmpfile,
4228         .rename         = ext4_rename2,
4229         .setattr        = ext4_setattr,
4230         .getattr        = ext4_getattr,
4231         .listxattr      = ext4_listxattr,
4232         .get_inode_acl  = ext4_get_acl,
4233         .set_acl        = ext4_set_acl,
4234         .fiemap         = ext4_fiemap,
4235         .fileattr_get   = ext4_fileattr_get,
4236         .fileattr_set   = ext4_fileattr_set,
4237 };
4238
4239 const struct inode_operations ext4_special_inode_operations = {
4240         .setattr        = ext4_setattr,
4241         .getattr        = ext4_getattr,
4242         .listxattr      = ext4_listxattr,
4243         .get_inode_acl  = ext4_get_acl,
4244         .set_acl        = ext4_set_acl,
4245 };