fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 113                 bh = ERR_PTR(-EIO);
 114         else
 115                 bh = ext4_bread(NULL, inode, block, 0);
 116         if (IS_ERR(bh)) {
 117                 __ext4_warning(inode->i_sb, func, line,
 118                                "inode #%lu: lblock %lu: comm %s: "
 119                                "error %ld reading directory block",
 120                                inode->i_ino, (unsigned long)block,
 121                                current->comm, PTR_ERR(bh));
 122
 123                 return bh;
 124         }
 125         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 126                 ext4_error_inode(inode, func, line, block,
 127                                  "Directory hole found for htree %s block",
 128                                  (type == INDEX) ? "index" : "leaf");
 129                 return ERR_PTR(-EFSCORRUPTED);
 130         }
 131         if (!bh)
 132                 return NULL;
 133         dirent = (struct ext4_dir_entry *) bh->b_data;
 134         /* Determine whether or not we have an index block */
 135         if (is_dx(inode)) {
 136                 if (block == 0)
 137                         is_dx_block = 1;
 138                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 139                                                 inode->i_sb->s_blocksize) ==
 140                          inode->i_sb->s_blocksize)
 141                         is_dx_block = 1;
 142         }
 143         if (!is_dx_block && type == INDEX) {
 144                 ext4_error_inode(inode, func, line, block,
 145                        "directory leaf block found instead of index block");
 146                 brelse(bh);
 147                 return ERR_PTR(-EFSCORRUPTED);
 148         }
 149         if (!ext4_has_metadata_csum(inode->i_sb) ||
 150             buffer_verified(bh))
 151                 return bh;
 152
 153         /*
 154          * An empty leaf block can get mistaken for a index block; for
 155          * this reason, we can only check the index checksum when the
 156          * caller is sure it should be an index block.
 157          */
 158         if (is_dx_block && type == INDEX) {
 159                 if (ext4_dx_csum_verify(inode, dirent) &&
 160                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 161                         set_buffer_verified(bh);
 162                 else {
 163                         ext4_error_inode_err(inode, func, line, block,
 164                                              EFSBADCRC,
 165                                              "Directory index failed checksum");
 166                         brelse(bh);
 167                         return ERR_PTR(-EFSBADCRC);
 168                 }
 169         }
 170         if (!is_dx_block) {
 171                 if (ext4_dirblock_csum_verify(inode, bh) &&
 172                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 173                         set_buffer_verified(bh);
 174                 else {
 175                         ext4_error_inode_err(inode, func, line, block,
 176                                              EFSBADCRC,
 177                                              "Directory block failed checksum");
 178                         brelse(bh);
 179                         return ERR_PTR(-EFSBADCRC);
 180                 }
 181         }
 182         return bh;
 183 }
 184
 185 #ifdef DX_DEBUG
 186 #define dxtrace(command) command
 187 #else
 188 #define dxtrace(command)
 189 #endif
 190
 191 struct fake_dirent
 192 {
 193         __le32 inode;
 194         __le16 rec_len;
 195         u8 name_len;
 196         u8 file_type;
 197 };
 198
 199 struct dx_countlimit
 200 {
 201         __le16 limit;
 202         __le16 count;
 203 };
 204
 205 struct dx_entry
 206 {
 207         __le32 hash;
 208         __le32 block;
 209 };
 210
 211 /*
 212  * dx_root_info is laid out so that if it should somehow get overlaid by a
 213  * dirent the two low bits of the hash version will be zero.  Therefore, the
 214  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 215  */
 216
 217 struct dx_root
 218 {
 219         struct fake_dirent dot;
 220         char dot_name[4];
 221         struct fake_dirent dotdot;
 222         char dotdot_name[4];
 223         struct dx_root_info
 224         {
 225                 __le32 reserved_zero;
 226                 u8 hash_version;
 227                 u8 info_length; /* 8 */
 228                 u8 indirect_levels;
 229                 u8 unused_flags;
 230         }
 231         info;
 232         struct dx_entry entries[];
 233 };
 234
 235 struct dx_node
 236 {
 237         struct fake_dirent fake;
 238         struct dx_entry entries[];
 239 };
 240
 241
 242 struct dx_frame
 243 {
 244         struct buffer_head *bh;
 245         struct dx_entry *entries;
 246         struct dx_entry *at;
 247 };
 248
 249 struct dx_map_entry
 250 {
 251         u32 hash;
 252         u16 offs;
 253         u16 size;
 254 };
 255
 256 /*
 257  * This goes at the end of each htree block.
 258  */
 259 struct dx_tail {
 260         u32 dt_reserved;
 261         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 262 };
 263
 264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 266 static inline unsigned dx_get_hash(struct dx_entry *entry);
 267 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 268 static unsigned dx_get_count(struct dx_entry *entries);
 269 static unsigned dx_get_limit(struct dx_entry *entries);
 270 static void dx_set_count(struct dx_entry *entries, unsigned value);
 271 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 273 static unsigned dx_node_limit(struct inode *dir);
 274 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 275                                  struct inode *dir,
 276                                  struct dx_hash_info *hinfo,
 277                                  struct dx_frame *frame);
 278 static void dx_release(struct dx_frame *frames);
 279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 280                        unsigned blocksize, struct dx_hash_info *hinfo,
 281                        struct dx_map_entry map[]);
 282 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 283 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
 284                                         char *to, struct dx_map_entry *offsets,
 285                                         int count, unsigned int blocksize);
 286 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 287                                                 unsigned int blocksize);
 288 static void dx_insert_block(struct dx_frame *frame,
 289                                         u32 hash, ext4_lblk_t block);
 290 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 291                                  struct dx_frame *frame,
 292                                  struct dx_frame *frames,
 293                                  __u32 *start_hash);
 294 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 295                 struct ext4_filename *fname,
 296                 struct ext4_dir_entry_2 **res_dir);
 297 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 298                              struct inode *dir, struct inode *inode);
 299
 300 /* checksumming functions */
 301 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 302                                  unsigned int blocksize)
 303 {
 304         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 305
 306         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 307         t->det_rec_len = ext4_rec_len_to_disk(
 308                         sizeof(struct ext4_dir_entry_tail), blocksize);
 309         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 310 }
 311
 312 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 313 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 314                                                    struct buffer_head *bh)
 315 {
 316         struct ext4_dir_entry_tail *t;
 317
 318 #ifdef PARANOID
 319         struct ext4_dir_entry *d, *top;
 320
 321         d = (struct ext4_dir_entry *)bh->b_data;
 322         top = (struct ext4_dir_entry *)(bh->b_data +
 323                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 324                  sizeof(struct ext4_dir_entry_tail)));
 325         while (d < top && d->rec_len)
 326                 d = (struct ext4_dir_entry *)(((void *)d) +
 327                     le16_to_cpu(d->rec_len));
 328
 329         if (d != top)
 330                 return NULL;
 331
 332         t = (struct ext4_dir_entry_tail *)d;
 333 #else
 334         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 335 #endif
 336
 337         if (t->det_reserved_zero1 ||
 338             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 339             t->det_reserved_zero2 ||
 340             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 341                 return NULL;
 342
 343         return t;
 344 }
 345
 346 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 347 {
 348         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 349         struct ext4_inode_info *ei = EXT4_I(inode);
 350         __u32 csum;
 351
 352         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 353         return cpu_to_le32(csum);
 354 }
 355
 356 #define warn_no_space_for_csum(inode)                                   \
 357         __warn_no_space_for_csum((inode), __func__, __LINE__)
 358
 359 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 360                                      unsigned int line)
 361 {
 362         __ext4_warning_inode(inode, func, line,
 363                 "No space for directory leaf checksum. Please run e2fsck -D.");
 364 }
 365
 366 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 367 {
 368         struct ext4_dir_entry_tail *t;
 369
 370         if (!ext4_has_metadata_csum(inode->i_sb))
 371                 return 1;
 372
 373         t = get_dirent_tail(inode, bh);
 374         if (!t) {
 375                 warn_no_space_for_csum(inode);
 376                 return 0;
 377         }
 378
 379         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 380                                                   (char *)t - bh->b_data))
 381                 return 0;
 382
 383         return 1;
 384 }
 385
 386 static void ext4_dirblock_csum_set(struct inode *inode,
 387                                  struct buffer_head *bh)
 388 {
 389         struct ext4_dir_entry_tail *t;
 390
 391         if (!ext4_has_metadata_csum(inode->i_sb))
 392                 return;
 393
 394         t = get_dirent_tail(inode, bh);
 395         if (!t) {
 396                 warn_no_space_for_csum(inode);
 397                 return;
 398         }
 399
 400         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 401                                              (char *)t - bh->b_data);
 402 }
 403
 404 int ext4_handle_dirty_dirblock(handle_t *handle,
 405                                struct inode *inode,
 406                                struct buffer_head *bh)
 407 {
 408         ext4_dirblock_csum_set(inode, bh);
 409         return ext4_handle_dirty_metadata(handle, inode, bh);
 410 }
 411
 412 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 413                                                struct ext4_dir_entry *dirent,
 414                                                int *offset)
 415 {
 416         struct ext4_dir_entry *dp;
 417         struct dx_root_info *root;
 418         int count_offset;
 419
 420         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 421                 count_offset = 8;
 422         else if (le16_to_cpu(dirent->rec_len) == 12) {
 423                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 424                 if (le16_to_cpu(dp->rec_len) !=
 425                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 426                         return NULL;
 427                 root = (struct dx_root_info *)(((void *)dp + 12));
 428                 if (root->reserved_zero ||
 429                     root->info_length != sizeof(struct dx_root_info))
 430                         return NULL;
 431                 count_offset = 32;
 432         } else
 433                 return NULL;
 434
 435         if (offset)
 436                 *offset = count_offset;
 437         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 438 }
 439
 440 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 441                            int count_offset, int count, struct dx_tail *t)
 442 {
 443         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 444         struct ext4_inode_info *ei = EXT4_I(inode);
 445         __u32 csum;
 446         int size;
 447         __u32 dummy_csum = 0;
 448         int offset = offsetof(struct dx_tail, dt_checksum);
 449
 450         size = count_offset + (count * sizeof(struct dx_entry));
 451         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 452         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 453         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 454
 455         return cpu_to_le32(csum);
 456 }
 457
 458 static int ext4_dx_csum_verify(struct inode *inode,
 459                                struct ext4_dir_entry *dirent)
 460 {
 461         struct dx_countlimit *c;
 462         struct dx_tail *t;
 463         int count_offset, limit, count;
 464
 465         if (!ext4_has_metadata_csum(inode->i_sb))
 466                 return 1;
 467
 468         c = get_dx_countlimit(inode, dirent, &count_offset);
 469         if (!c) {
 470                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 471                 return 0;
 472         }
 473         limit = le16_to_cpu(c->limit);
 474         count = le16_to_cpu(c->count);
 475         if (count_offset + (limit * sizeof(struct dx_entry)) >
 476             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 477                 warn_no_space_for_csum(inode);
 478                 return 0;
 479         }
 480         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 481
 482         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 483                                             count, t))
 484                 return 0;
 485         return 1;
 486 }
 487
 488 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 489 {
 490         struct dx_countlimit *c;
 491         struct dx_tail *t;
 492         int count_offset, limit, count;
 493
 494         if (!ext4_has_metadata_csum(inode->i_sb))
 495                 return;
 496
 497         c = get_dx_countlimit(inode, dirent, &count_offset);
 498         if (!c) {
 499                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 500                 return;
 501         }
 502         limit = le16_to_cpu(c->limit);
 503         count = le16_to_cpu(c->count);
 504         if (count_offset + (limit * sizeof(struct dx_entry)) >
 505             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 506                 warn_no_space_for_csum(inode);
 507                 return;
 508         }
 509         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 510
 511         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 512 }
 513
 514 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 515                                             struct inode *inode,
 516                                             struct buffer_head *bh)
 517 {
 518         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 519         return ext4_handle_dirty_metadata(handle, inode, bh);
 520 }
 521
 522 /*
 523  * p is at least 6 bytes before the end of page
 524  */
 525 static inline struct ext4_dir_entry_2 *
 526 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 527 {
 528         return (struct ext4_dir_entry_2 *)((char *)p +
 529                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 530 }
 531
 532 /*
 533  * Future: use high four bits of block for coalesce-on-delete flags
 534  * Mask them off for now.
 535  */
 536
 537 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 538 {
 539         return le32_to_cpu(entry->block) & 0x0fffffff;
 540 }
 541
 542 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 543 {
 544         entry->block = cpu_to_le32(value);
 545 }
 546
 547 static inline unsigned dx_get_hash(struct dx_entry *entry)
 548 {
 549         return le32_to_cpu(entry->hash);
 550 }
 551
 552 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 553 {
 554         entry->hash = cpu_to_le32(value);
 555 }
 556
 557 static inline unsigned dx_get_count(struct dx_entry *entries)
 558 {
 559         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 560 }
 561
 562 static inline unsigned dx_get_limit(struct dx_entry *entries)
 563 {
 564         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 565 }
 566
 567 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 568 {
 569         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 570 }
 571
 572 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 573 {
 574         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 575 }
 576
 577 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 578 {
 579         unsigned int entry_space = dir->i_sb->s_blocksize -
 580                         ext4_dir_rec_len(1, NULL) -
 581                         ext4_dir_rec_len(2, NULL) - infosize;
 582
 583         if (ext4_has_metadata_csum(dir->i_sb))
 584                 entry_space -= sizeof(struct dx_tail);
 585         return entry_space / sizeof(struct dx_entry);
 586 }
 587
 588 static inline unsigned dx_node_limit(struct inode *dir)
 589 {
 590         unsigned int entry_space = dir->i_sb->s_blocksize -
 591                         ext4_dir_rec_len(0, dir);
 592
 593         if (ext4_has_metadata_csum(dir->i_sb))
 594                 entry_space -= sizeof(struct dx_tail);
 595         return entry_space / sizeof(struct dx_entry);
 596 }
 597
 598 /*
 599  * Debug
 600  */
 601 #ifdef DX_DEBUG
 602 static void dx_show_index(char * label, struct dx_entry *entries)
 603 {
 604         int i, n = dx_get_count (entries);
 605         printk(KERN_DEBUG "%s index", label);
 606         for (i = 0; i < n; i++) {
 607                 printk(KERN_CONT " %x->%lu",
 608                        i ? dx_get_hash(entries + i) : 0,
 609                        (unsigned long)dx_get_block(entries + i));
 610         }
 611         printk(KERN_CONT "\n");
 612 }
 613
 614 struct stats
 615 {
 616         unsigned names;
 617         unsigned space;
 618         unsigned bcount;
 619 };
 620
 621 static struct stats dx_show_leaf(struct inode *dir,
 622                                 struct dx_hash_info *hinfo,
 623                                 struct ext4_dir_entry_2 *de,
 624                                 int size, int show_names)
 625 {
 626         unsigned names = 0, space = 0;
 627         char *base = (char *) de;
 628         struct dx_hash_info h = *hinfo;
 629
 630         printk("names: ");
 631         while ((char *) de < base + size)
 632         {
 633                 if (de->inode)
 634                 {
 635                         if (show_names)
 636                         {
 637 #ifdef CONFIG_FS_ENCRYPTION
 638                                 int len;
 639                                 char *name;
 640                                 struct fscrypt_str fname_crypto_str =
 641                                         FSTR_INIT(NULL, 0);
 642                                 int res = 0;
 643
 644                                 name  = de->name;
 645                                 len = de->name_len;
 646                                 if (!IS_ENCRYPTED(dir)) {
 647                                         /* Directory is not encrypted */
 648                                         ext4fs_dirhash(dir, de->name,
 649                                                 de->name_len, &h);
 650                                         printk("%*.s:(U)%x.%u ", len,
 651                                                name, h.hash,
 652                                                (unsigned) ((char *) de
 653                                                            - base));
 654                                 } else {
 655                                         struct fscrypt_str de_name =
 656                                                 FSTR_INIT(name, len);
 657
 658                                         /* Directory is encrypted */
 659                                         res = fscrypt_fname_alloc_buffer(
 660                                                 len, &fname_crypto_str);
 661                                         if (res)
 662                                                 printk(KERN_WARNING "Error "
 663                                                         "allocating crypto "
 664                                                         "buffer--skipping "
 665                                                         "crypto\n");
 666                                         res = fscrypt_fname_disk_to_usr(dir,
 667                                                 0, 0, &de_name,
 668                                                 &fname_crypto_str);
 669                                         if (res) {
 670                                                 printk(KERN_WARNING "Error "
 671                                                         "converting filename "
 672                                                         "from disk to usr"
 673                                                         "\n");
 674                                                 name = "??";
 675                                                 len = 2;
 676                                         } else {
 677                                                 name = fname_crypto_str.name;
 678                                                 len = fname_crypto_str.len;
 679                                         }
 680                                         if (IS_CASEFOLDED(dir))
 681                                                 h.hash = EXT4_DIRENT_HASH(de);
 682                                         else
 683                                                 ext4fs_dirhash(dir, de->name,
 684                                                        de->name_len, &h);
 685                                         printk("%*.s:(E)%x.%u ", len, name,
 686                                                h.hash, (unsigned) ((char *) de
 687                                                                    - base));
 688                                         fscrypt_fname_free_buffer(
 689                                                         &fname_crypto_str);
 690                                 }
 691 #else
 692                                 int len = de->name_len;
 693                                 char *name = de->name;
 694                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 695                                 printk("%*.s:%x.%u ", len, name, h.hash,
 696                                        (unsigned) ((char *) de - base));
 697 #endif
 698                         }
 699                         space += ext4_dir_rec_len(de->name_len, dir);
 700                         names++;
 701                 }
 702                 de = ext4_next_entry(de, size);
 703         }
 704         printk(KERN_CONT "(%i)\n", names);
 705         return (struct stats) { names, space, 1 };
 706 }
 707
 708 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 709                              struct dx_entry *entries, int levels)
 710 {
 711         unsigned blocksize = dir->i_sb->s_blocksize;
 712         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 713         unsigned bcount = 0;
 714         struct buffer_head *bh;
 715         printk("%i indexed blocks...\n", count);
 716         for (i = 0; i < count; i++, entries++)
 717         {
 718                 ext4_lblk_t block = dx_get_block(entries);
 719                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 720                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 721                 struct stats stats;
 722                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 723                 bh = ext4_bread(NULL,dir, block, 0);
 724                 if (!bh || IS_ERR(bh))
 725                         continue;
 726                 stats = levels?
 727                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 728                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 729                         bh->b_data, blocksize, 0);
 730                 names += stats.names;
 731                 space += stats.space;
 732                 bcount += stats.bcount;
 733                 brelse(bh);
 734         }
 735         if (bcount)
 736                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 737                        levels ? "" : "   ", names, space/bcount,
 738                        (space/bcount)*100/blocksize);
 739         return (struct stats) { names, space, bcount};
 740 }
 741
 742 /*
 743  * Linear search cross check
 744  */
 745 static inline void htree_rep_invariant_check(struct dx_entry *at,
 746                                              struct dx_entry *target,
 747                                              u32 hash, unsigned int n)
 748 {
 749         while (n--) {
 750                 dxtrace(printk(KERN_CONT ","));
 751                 if (dx_get_hash(++at) > hash) {
 752                         at--;
 753                         break;
 754                 }
 755         }
 756         ASSERT(at == target - 1);
 757 }
 758 #else /* DX_DEBUG */
 759 static inline void htree_rep_invariant_check(struct dx_entry *at,
 760                                              struct dx_entry *target,
 761                                              u32 hash, unsigned int n)
 762 {
 763 }
 764 #endif /* DX_DEBUG */
 765
 766 /*
 767  * Probe for a directory leaf block to search.
 768  *
 769  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 770  * error in the directory index, and the caller should fall back to
 771  * searching the directory normally.  The callers of dx_probe **MUST**
 772  * check for this error code, and make sure it never gets reflected
 773  * back to userspace.
 774  */
 775 static struct dx_frame *
 776 dx_probe(struct ext4_filename *fname, struct inode *dir,
 777          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 778 {
 779         unsigned count, indirect;
 780         struct dx_entry *at, *entries, *p, *q, *m;
 781         struct dx_root *root;
 782         struct dx_frame *frame = frame_in;
 783         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 784         u32 hash;
 785
 786         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 787         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 788         if (IS_ERR(frame->bh))
 789                 return (struct dx_frame *) frame->bh;
 790
 791         root = (struct dx_root *) frame->bh->b_data;
 792         if (root->info.hash_version != DX_HASH_TEA &&
 793             root->info.hash_version != DX_HASH_HALF_MD4 &&
 794             root->info.hash_version != DX_HASH_LEGACY &&
 795             root->info.hash_version != DX_HASH_SIPHASH) {
 796                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 797                                    root->info.hash_version);
 798                 goto fail;
 799         }
 800         if (ext4_hash_in_dirent(dir)) {
 801                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 802                         ext4_warning_inode(dir,
 803                                 "Hash in dirent, but hash is not SIPHASH");
 804                         goto fail;
 805                 }
 806         } else {
 807                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 808                         ext4_warning_inode(dir,
 809                                 "Hash code is SIPHASH, but hash not in dirent");
 810                         goto fail;
 811                 }
 812         }
 813         if (fname)
 814                 hinfo = &fname->hinfo;
 815         hinfo->hash_version = root->info.hash_version;
 816         if (hinfo->hash_version <= DX_HASH_TEA)
 817                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 818         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 819         /* hash is already computed for encrypted casefolded directory */
 820         if (fname && fname_name(fname) &&
 821                                 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
 822                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 823         hash = hinfo->hash;
 824
 825         if (root->info.unused_flags & 1) {
 826                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 827                                    root->info.unused_flags);
 828                 goto fail;
 829         }
 830
 831         indirect = root->info.indirect_levels;
 832         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 833                 ext4_warning(dir->i_sb,
 834                              "Directory (ino: %lu) htree depth %#06x exceed"
 835                              "supported value", dir->i_ino,
 836                              ext4_dir_htree_level(dir->i_sb));
 837                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 838                         ext4_warning(dir->i_sb, "Enable large directory "
 839                                                 "feature to access it");
 840                 }
 841                 goto fail;
 842         }
 843
 844         entries = (struct dx_entry *)(((char *)&root->info) +
 845                                       root->info.info_length);
 846
 847         if (dx_get_limit(entries) != dx_root_limit(dir,
 848                                                    root->info.info_length)) {
 849                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 850                                    dx_get_limit(entries),
 851                                    dx_root_limit(dir, root->info.info_length));
 852                 goto fail;
 853         }
 854
 855         dxtrace(printk("Look up %x", hash));
 856         while (1) {
 857                 count = dx_get_count(entries);
 858                 if (!count || count > dx_get_limit(entries)) {
 859                         ext4_warning_inode(dir,
 860                                            "dx entry: count %u beyond limit %u",
 861                                            count, dx_get_limit(entries));
 862                         goto fail;
 863                 }
 864
 865                 p = entries + 1;
 866                 q = entries + count - 1;
 867                 while (p <= q) {
 868                         m = p + (q - p) / 2;
 869                         dxtrace(printk(KERN_CONT "."));
 870                         if (dx_get_hash(m) > hash)
 871                                 q = m - 1;
 872                         else
 873                                 p = m + 1;
 874                 }
 875
 876                 htree_rep_invariant_check(entries, p, hash, count - 1);
 877
 878                 at = p - 1;
 879                 dxtrace(printk(KERN_CONT " %x->%u\n",
 880                                at == entries ? 0 : dx_get_hash(at),
 881                                dx_get_block(at)));
 882                 frame->entries = entries;
 883                 frame->at = at;
 884                 if (!indirect--)
 885                         return frame;
 886                 frame++;
 887                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 888                 if (IS_ERR(frame->bh)) {
 889                         ret_err = (struct dx_frame *) frame->bh;
 890                         frame->bh = NULL;
 891                         goto fail;
 892                 }
 893                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 894
 895                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 896                         ext4_warning_inode(dir,
 897                                 "dx entry: limit %u != node limit %u",
 898                                 dx_get_limit(entries), dx_node_limit(dir));
 899                         goto fail;
 900                 }
 901         }
 902 fail:
 903         while (frame >= frame_in) {
 904                 brelse(frame->bh);
 905                 frame--;
 906         }
 907
 908         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 909                 ext4_warning_inode(dir,
 910                         "Corrupt directory, running e2fsck is recommended");
 911         return ret_err;
 912 }
 913
 914 static void dx_release(struct dx_frame *frames)
 915 {
 916         struct dx_root_info *info;
 917         int i;
 918         unsigned int indirect_levels;
 919
 920         if (frames[0].bh == NULL)
 921                 return;
 922
 923         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 924         /* save local copy, "info" may be freed after brelse() */
 925         indirect_levels = info->indirect_levels;
 926         for (i = 0; i <= indirect_levels; i++) {
 927                 if (frames[i].bh == NULL)
 928                         break;
 929                 brelse(frames[i].bh);
 930                 frames[i].bh = NULL;
 931         }
 932 }
 933
 934 /*
 935  * This function increments the frame pointer to search the next leaf
 936  * block, and reads in the necessary intervening nodes if the search
 937  * should be necessary.  Whether or not the search is necessary is
 938  * controlled by the hash parameter.  If the hash value is even, then
 939  * the search is only continued if the next block starts with that
 940  * hash value.  This is used if we are searching for a specific file.
 941  *
 942  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 943  *
 944  * This function returns 1 if the caller should continue to search,
 945  * or 0 if it should not.  If there is an error reading one of the
 946  * index blocks, it will a negative error code.
 947  *
 948  * If start_hash is non-null, it will be filled in with the starting
 949  * hash of the next page.
 950  */
 951 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 952                                  struct dx_frame *frame,
 953                                  struct dx_frame *frames,
 954                                  __u32 *start_hash)
 955 {
 956         struct dx_frame *p;
 957         struct buffer_head *bh;
 958         int num_frames = 0;
 959         __u32 bhash;
 960
 961         p = frame;
 962         /*
 963          * Find the next leaf page by incrementing the frame pointer.
 964          * If we run out of entries in the interior node, loop around and
 965          * increment pointer in the parent node.  When we break out of
 966          * this loop, num_frames indicates the number of interior
 967          * nodes need to be read.
 968          */
 969         while (1) {
 970                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 971                         break;
 972                 if (p == frames)
 973                         return 0;
 974                 num_frames++;
 975                 p--;
 976         }
 977
 978         /*
 979          * If the hash is 1, then continue only if the next page has a
 980          * continuation hash of any value.  This is used for readdir
 981          * handling.  Otherwise, check to see if the hash matches the
 982          * desired continuation hash.  If it doesn't, return since
 983          * there's no point to read in the successive index pages.
 984          */
 985         bhash = dx_get_hash(p->at);
 986         if (start_hash)
 987                 *start_hash = bhash;
 988         if ((hash & 1) == 0) {
 989                 if ((bhash & ~1) != hash)
 990                         return 0;
 991         }
 992         /*
 993          * If the hash is HASH_NB_ALWAYS, we always go to the next
 994          * block so no check is necessary
 995          */
 996         while (num_frames--) {
 997                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 998                 if (IS_ERR(bh))
 999                         return PTR_ERR(bh);
1000                 p++;
1001                 brelse(p->bh);
1002                 p->bh = bh;
1003                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1004         }
1005         return 1;
1006 }
1007
1008
1009 /*
1010  * This function fills a red-black tree with information from a
1011  * directory block.  It returns the number directory entries loaded
1012  * into the tree.  If there is an error it is returned in err.
1013  */
1014 static int htree_dirblock_to_tree(struct file *dir_file,
1015                                   struct inode *dir, ext4_lblk_t block,
1016                                   struct dx_hash_info *hinfo,
1017                                   __u32 start_hash, __u32 start_minor_hash)
1018 {
1019         struct buffer_head *bh;
1020         struct ext4_dir_entry_2 *de, *top;
1021         int err = 0, count = 0;
1022         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1023         int csum = ext4_has_metadata_csum(dir->i_sb);
1024
1025         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1026                                                         (unsigned long)block));
1027         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1028         if (IS_ERR(bh))
1029                 return PTR_ERR(bh);
1030
1031         de = (struct ext4_dir_entry_2 *) bh->b_data;
1032         /* csum entries are not larger in the casefolded encrypted case */
1033         top = (struct ext4_dir_entry_2 *) ((char *) de +
1034                                            dir->i_sb->s_blocksize -
1035                                            ext4_dir_rec_len(0,
1036                                                            csum ? NULL : dir));
1037         /* Check if the directory is encrypted */
1038         if (IS_ENCRYPTED(dir)) {
1039                 err = fscrypt_prepare_readdir(dir);
1040                 if (err < 0) {
1041                         brelse(bh);
1042                         return err;
1043                 }
1044                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1045                                                  &fname_crypto_str);
1046                 if (err < 0) {
1047                         brelse(bh);
1048                         return err;
1049                 }
1050         }
1051
1052         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1053                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1054                                 bh->b_data, bh->b_size,
1055                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1056                                          + ((char *)de - bh->b_data))) {
1057                         /* silently ignore the rest of the block */
1058                         break;
1059                 }
1060                 if (ext4_hash_in_dirent(dir)) {
1061                         if (de->name_len && de->inode) {
1062                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1063                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1064                         } else {
1065                                 hinfo->hash = 0;
1066                                 hinfo->minor_hash = 0;
1067                         }
1068                 } else {
1069                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1070                 }
1071                 if ((hinfo->hash < start_hash) ||
1072                     ((hinfo->hash == start_hash) &&
1073                      (hinfo->minor_hash < start_minor_hash)))
1074                         continue;
1075                 if (de->inode == 0)
1076                         continue;
1077                 if (!IS_ENCRYPTED(dir)) {
1078                         tmp_str.name = de->name;
1079                         tmp_str.len = de->name_len;
1080                         err = ext4_htree_store_dirent(dir_file,
1081                                    hinfo->hash, hinfo->minor_hash, de,
1082                                    &tmp_str);
1083                 } else {
1084                         int save_len = fname_crypto_str.len;
1085                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1086                                                                 de->name_len);
1087
1088                         /* Directory is encrypted */
1089                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1090                                         hinfo->minor_hash, &de_name,
1091                                         &fname_crypto_str);
1092                         if (err) {
1093                                 count = err;
1094                                 goto errout;
1095                         }
1096                         err = ext4_htree_store_dirent(dir_file,
1097                                    hinfo->hash, hinfo->minor_hash, de,
1098                                         &fname_crypto_str);
1099                         fname_crypto_str.len = save_len;
1100                 }
1101                 if (err != 0) {
1102                         count = err;
1103                         goto errout;
1104                 }
1105                 count++;
1106         }
1107 errout:
1108         brelse(bh);
1109         fscrypt_fname_free_buffer(&fname_crypto_str);
1110         return count;
1111 }
1112
1113
1114 /*
1115  * This function fills a red-black tree with information from a
1116  * directory.  We start scanning the directory in hash order, starting
1117  * at start_hash and start_minor_hash.
1118  *
1119  * This function returns the number of entries inserted into the tree,
1120  * or a negative error code.
1121  */
1122 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1123                          __u32 start_minor_hash, __u32 *next_hash)
1124 {
1125         struct dx_hash_info hinfo;
1126         struct ext4_dir_entry_2 *de;
1127         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1128         struct inode *dir;
1129         ext4_lblk_t block;
1130         int count = 0;
1131         int ret, err;
1132         __u32 hashval;
1133         struct fscrypt_str tmp_str;
1134
1135         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1136                        start_hash, start_minor_hash));
1137         dir = file_inode(dir_file);
1138         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1139                 if (ext4_hash_in_dirent(dir))
1140                         hinfo.hash_version = DX_HASH_SIPHASH;
1141                 else
1142                         hinfo.hash_version =
1143                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1144                 if (hinfo.hash_version <= DX_HASH_TEA)
1145                         hinfo.hash_version +=
1146                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1147                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1148                 if (ext4_has_inline_data(dir)) {
1149                         int has_inline_data = 1;
1150                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1151                                                        &hinfo, start_hash,
1152                                                        start_minor_hash,
1153                                                        &has_inline_data);
1154                         if (has_inline_data) {
1155                                 *next_hash = ~0;
1156                                 return count;
1157                         }
1158                 }
1159                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1160                                                start_hash, start_minor_hash);
1161                 *next_hash = ~0;
1162                 return count;
1163         }
1164         hinfo.hash = start_hash;
1165         hinfo.minor_hash = 0;
1166         frame = dx_probe(NULL, dir, &hinfo, frames);
1167         if (IS_ERR(frame))
1168                 return PTR_ERR(frame);
1169
1170         /* Add '.' and '..' from the htree header */
1171         if (!start_hash && !start_minor_hash) {
1172                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1173                 tmp_str.name = de->name;
1174                 tmp_str.len = de->name_len;
1175                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1176                                               de, &tmp_str);
1177                 if (err != 0)
1178                         goto errout;
1179                 count++;
1180         }
1181         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1182                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1183                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1184                 tmp_str.name = de->name;
1185                 tmp_str.len = de->name_len;
1186                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1187                                               de, &tmp_str);
1188                 if (err != 0)
1189                         goto errout;
1190                 count++;
1191         }
1192
1193         while (1) {
1194                 if (fatal_signal_pending(current)) {
1195                         err = -ERESTARTSYS;
1196                         goto errout;
1197                 }
1198                 cond_resched();
1199                 block = dx_get_block(frame->at);
1200                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1201                                              start_hash, start_minor_hash);
1202                 if (ret < 0) {
1203                         err = ret;
1204                         goto errout;
1205                 }
1206                 count += ret;
1207                 hashval = ~0;
1208                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1209                                             frame, frames, &hashval);
1210                 *next_hash = hashval;
1211                 if (ret < 0) {
1212                         err = ret;
1213                         goto errout;
1214                 }
1215                 /*
1216                  * Stop if:  (a) there are no more entries, or
1217                  * (b) we have inserted at least one entry and the
1218                  * next hash value is not a continuation
1219                  */
1220                 if ((ret == 0) ||
1221                     (count && ((hashval & 1) == 0)))
1222                         break;
1223         }
1224         dx_release(frames);
1225         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1226                        "next hash: %x\n", count, *next_hash));
1227         return count;
1228 errout:
1229         dx_release(frames);
1230         return (err);
1231 }
1232
1233 static inline int search_dirblock(struct buffer_head *bh,
1234                                   struct inode *dir,
1235                                   struct ext4_filename *fname,
1236                                   unsigned int offset,
1237                                   struct ext4_dir_entry_2 **res_dir)
1238 {
1239         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1240                                fname, offset, res_dir);
1241 }
1242
1243 /*
1244  * Directory block splitting, compacting
1245  */
1246
1247 /*
1248  * Create map of hash values, offsets, and sizes, stored at end of block.
1249  * Returns number of entries mapped.
1250  */
1251 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1252                        unsigned blocksize, struct dx_hash_info *hinfo,
1253                        struct dx_map_entry *map_tail)
1254 {
1255         int count = 0;
1256         char *base = (char *) de;
1257         struct dx_hash_info h = *hinfo;
1258
1259         while ((char *) de < base + blocksize) {
1260                 if (de->name_len && de->inode) {
1261                         if (ext4_hash_in_dirent(dir))
1262                                 h.hash = EXT4_DIRENT_HASH(de);
1263                         else
1264                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1265                         map_tail--;
1266                         map_tail->hash = h.hash;
1267                         map_tail->offs = ((char *) de - base)>>2;
1268                         map_tail->size = le16_to_cpu(de->rec_len);
1269                         count++;
1270                         cond_resched();
1271                 }
1272                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1273                 de = ext4_next_entry(de, blocksize);
1274         }
1275         return count;
1276 }
1277
1278 /* Sort map by hash value */
1279 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1280 {
1281         struct dx_map_entry *p, *q, *top = map + count - 1;
1282         int more;
1283         /* Combsort until bubble sort doesn't suck */
1284         while (count > 2) {
1285                 count = count*10/13;
1286                 if (count - 9 < 2) /* 9, 10 -> 11 */
1287                         count = 11;
1288                 for (p = top, q = p - count; q >= map; p--, q--)
1289                         if (p->hash < q->hash)
1290                                 swap(*p, *q);
1291         }
1292         /* Garden variety bubble sort */
1293         do {
1294                 more = 0;
1295                 q = top;
1296                 while (q-- > map) {
1297                         if (q[1].hash >= q[0].hash)
1298                                 continue;
1299                         swap(*(q+1), *q);
1300                         more = 1;
1301                 }
1302         } while(more);
1303 }
1304
1305 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1306 {
1307         struct dx_entry *entries = frame->entries;
1308         struct dx_entry *old = frame->at, *new = old + 1;
1309         int count = dx_get_count(entries);
1310
1311         ASSERT(count < dx_get_limit(entries));
1312         ASSERT(old < entries + count);
1313         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1314         dx_set_hash(new, hash);
1315         dx_set_block(new, block);
1316         dx_set_count(entries, count + 1);
1317 }
1318
1319 #ifdef CONFIG_UNICODE
1320 /*
1321  * Test whether a case-insensitive directory entry matches the filename
1322  * being searched for.  If quick is set, assume the name being looked up
1323  * is already in the casefolded form.
1324  *
1325  * Returns: 0 if the directory entry matches, more than 0 if it
1326  * doesn't match or less than zero on error.
1327  */
1328 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1329                            u8 *de_name, size_t de_name_len, bool quick)
1330 {
1331         const struct super_block *sb = parent->i_sb;
1332         const struct unicode_map *um = sb->s_encoding;
1333         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1334         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1335         int ret;
1336
1337         if (IS_ENCRYPTED(parent)) {
1338                 const struct fscrypt_str encrypted_name =
1339                                 FSTR_INIT(de_name, de_name_len);
1340
1341                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1342                 if (!decrypted_name.name)
1343                         return -ENOMEM;
1344                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1345                                                 &decrypted_name);
1346                 if (ret < 0)
1347                         goto out;
1348                 entry.name = decrypted_name.name;
1349                 entry.len = decrypted_name.len;
1350         }
1351
1352         if (quick)
1353                 ret = utf8_strncasecmp_folded(um, name, &entry);
1354         else
1355                 ret = utf8_strncasecmp(um, name, &entry);
1356         if (ret < 0) {
1357                 /* Handle invalid character sequence as either an error
1358                  * or as an opaque byte sequence.
1359                  */
1360                 if (sb_has_strict_encoding(sb))
1361                         ret = -EINVAL;
1362                 else if (name->len != entry.len)
1363                         ret = 1;
1364                 else
1365                         ret = !!memcmp(name->name, entry.name, entry.len);
1366         }
1367 out:
1368         kfree(decrypted_name.name);
1369         return ret;
1370 }
1371
1372 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1373                                   struct ext4_filename *name)
1374 {
1375         struct fscrypt_str *cf_name = &name->cf_name;
1376         struct dx_hash_info *hinfo = &name->hinfo;
1377         int len;
1378
1379         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1380                 cf_name->name = NULL;
1381                 return 0;
1382         }
1383
1384         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1385         if (!cf_name->name)
1386                 return -ENOMEM;
1387
1388         len = utf8_casefold(dir->i_sb->s_encoding,
1389                             iname, cf_name->name,
1390                             EXT4_NAME_LEN);
1391         if (len <= 0) {
1392                 kfree(cf_name->name);
1393                 cf_name->name = NULL;
1394         }
1395         cf_name->len = (unsigned) len;
1396         if (!IS_ENCRYPTED(dir))
1397                 return 0;
1398
1399         hinfo->hash_version = DX_HASH_SIPHASH;
1400         hinfo->seed = NULL;
1401         if (cf_name->name)
1402                 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1403         else
1404                 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1405         return 0;
1406 }
1407 #endif
1408
1409 /*
1410  * Test whether a directory entry matches the filename being searched for.
1411  *
1412  * Return: %true if the directory entry matches, otherwise %false.
1413  */
1414 static bool ext4_match(struct inode *parent,
1415                               const struct ext4_filename *fname,
1416                               struct ext4_dir_entry_2 *de)
1417 {
1418         struct fscrypt_name f;
1419
1420         if (!de->inode)
1421                 return false;
1422
1423         f.usr_fname = fname->usr_fname;
1424         f.disk_name = fname->disk_name;
1425 #ifdef CONFIG_FS_ENCRYPTION
1426         f.crypto_buf = fname->crypto_buf;
1427 #endif
1428
1429 #ifdef CONFIG_UNICODE
1430         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1431                 if (fname->cf_name.name) {
1432                         struct qstr cf = {.name = fname->cf_name.name,
1433                                           .len = fname->cf_name.len};
1434                         if (IS_ENCRYPTED(parent)) {
1435                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1436                                         fname->hinfo.minor_hash !=
1437                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1438
1439                                         return 0;
1440                                 }
1441                         }
1442                         return !ext4_ci_compare(parent, &cf, de->name,
1443                                                         de->name_len, true);
1444                 }
1445                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1446                                                 de->name_len, false);
1447         }
1448 #endif
1449
1450         return fscrypt_match_name(&f, de->name, de->name_len);
1451 }
1452
1453 /*
1454  * Returns 0 if not found, -1 on failure, and 1 on success
1455  */
1456 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1457                     struct inode *dir, struct ext4_filename *fname,
1458                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1459 {
1460         struct ext4_dir_entry_2 * de;
1461         char * dlimit;
1462         int de_len;
1463
1464         de = (struct ext4_dir_entry_2 *)search_buf;
1465         dlimit = search_buf + buf_size;
1466         while ((char *) de < dlimit) {
1467                 /* this code is executed quadratically often */
1468                 /* do minimal checking `by hand' */
1469                 if ((char *) de + de->name_len <= dlimit &&
1470                     ext4_match(dir, fname, de)) {
1471                         /* found a match - just to be sure, do
1472                          * a full check */
1473                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1474                                                  buf_size, offset))
1475                                 return -1;
1476                         *res_dir = de;
1477                         return 1;
1478                 }
1479                 /* prevent looping on a bad block */
1480                 de_len = ext4_rec_len_from_disk(de->rec_len,
1481                                                 dir->i_sb->s_blocksize);
1482                 if (de_len <= 0)
1483                         return -1;
1484                 offset += de_len;
1485                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1486         }
1487         return 0;
1488 }
1489
1490 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1491                                struct ext4_dir_entry *de)
1492 {
1493         struct super_block *sb = dir->i_sb;
1494
1495         if (!is_dx(dir))
1496                 return 0;
1497         if (block == 0)
1498                 return 1;
1499         if (de->inode == 0 &&
1500             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1501                         sb->s_blocksize)
1502                 return 1;
1503         return 0;
1504 }
1505
1506 /*
1507  *      __ext4_find_entry()
1508  *
1509  * finds an entry in the specified directory with the wanted name. It
1510  * returns the cache buffer in which the entry was found, and the entry
1511  * itself (as a parameter - res_dir). It does NOT read the inode of the
1512  * entry - you'll have to do that yourself if you want to.
1513  *
1514  * The returned buffer_head has ->b_count elevated.  The caller is expected
1515  * to brelse() it when appropriate.
1516  */
1517 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1518                                              struct ext4_filename *fname,
1519                                              struct ext4_dir_entry_2 **res_dir,
1520                                              int *inlined)
1521 {
1522         struct super_block *sb;
1523         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1524         struct buffer_head *bh, *ret = NULL;
1525         ext4_lblk_t start, block;
1526         const u8 *name = fname->usr_fname->name;
1527         size_t ra_max = 0;      /* Number of bh's in the readahead
1528                                    buffer, bh_use[] */
1529         size_t ra_ptr = 0;      /* Current index into readahead
1530                                    buffer */
1531         ext4_lblk_t  nblocks;
1532         int i, namelen, retval;
1533
1534         *res_dir = NULL;
1535         sb = dir->i_sb;
1536         namelen = fname->usr_fname->len;
1537         if (namelen > EXT4_NAME_LEN)
1538                 return NULL;
1539
1540         if (ext4_has_inline_data(dir)) {
1541                 int has_inline_data = 1;
1542                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1543                                              &has_inline_data);
1544                 if (has_inline_data) {
1545                         if (inlined)
1546                                 *inlined = 1;
1547                         goto cleanup_and_exit;
1548                 }
1549         }
1550
1551         if ((namelen <= 2) && (name[0] == '.') &&
1552             (name[1] == '.' || name[1] == '\0')) {
1553                 /*
1554                  * "." or ".." will only be in the first block
1555                  * NFS may look up ".."; "." should be handled by the VFS
1556                  */
1557                 block = start = 0;
1558                 nblocks = 1;
1559                 goto restart;
1560         }
1561         if (is_dx(dir)) {
1562                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1563                 /*
1564                  * On success, or if the error was file not found,
1565                  * return.  Otherwise, fall back to doing a search the
1566                  * old fashioned way.
1567                  */
1568                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1569                         goto cleanup_and_exit;
1570                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1571                                "falling back\n"));
1572                 ret = NULL;
1573         }
1574         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1575         if (!nblocks) {
1576                 ret = NULL;
1577                 goto cleanup_and_exit;
1578         }
1579         start = EXT4_I(dir)->i_dir_start_lookup;
1580         if (start >= nblocks)
1581                 start = 0;
1582         block = start;
1583 restart:
1584         do {
1585                 /*
1586                  * We deal with the read-ahead logic here.
1587                  */
1588                 cond_resched();
1589                 if (ra_ptr >= ra_max) {
1590                         /* Refill the readahead buffer */
1591                         ra_ptr = 0;
1592                         if (block < start)
1593                                 ra_max = start - block;
1594                         else
1595                                 ra_max = nblocks - block;
1596                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1597                         retval = ext4_bread_batch(dir, block, ra_max,
1598                                                   false /* wait */, bh_use);
1599                         if (retval) {
1600                                 ret = ERR_PTR(retval);
1601                                 ra_max = 0;
1602                                 goto cleanup_and_exit;
1603                         }
1604                 }
1605                 if ((bh = bh_use[ra_ptr++]) == NULL)
1606                         goto next;
1607                 wait_on_buffer(bh);
1608                 if (!buffer_uptodate(bh)) {
1609                         EXT4_ERROR_INODE_ERR(dir, EIO,
1610                                              "reading directory lblock %lu",
1611                                              (unsigned long) block);
1612                         brelse(bh);
1613                         ret = ERR_PTR(-EIO);
1614                         goto cleanup_and_exit;
1615                 }
1616                 if (!buffer_verified(bh) &&
1617                     !is_dx_internal_node(dir, block,
1618                                          (struct ext4_dir_entry *)bh->b_data) &&
1619                     !ext4_dirblock_csum_verify(dir, bh)) {
1620                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1621                                              "checksumming directory "
1622                                              "block %lu", (unsigned long)block);
1623                         brelse(bh);
1624                         ret = ERR_PTR(-EFSBADCRC);
1625                         goto cleanup_and_exit;
1626                 }
1627                 set_buffer_verified(bh);
1628                 i = search_dirblock(bh, dir, fname,
1629                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1630                 if (i == 1) {
1631                         EXT4_I(dir)->i_dir_start_lookup = block;
1632                         ret = bh;
1633                         goto cleanup_and_exit;
1634                 } else {
1635                         brelse(bh);
1636                         if (i < 0)
1637                                 goto cleanup_and_exit;
1638                 }
1639         next:
1640                 if (++block >= nblocks)
1641                         block = 0;
1642         } while (block != start);
1643
1644         /*
1645          * If the directory has grown while we were searching, then
1646          * search the last part of the directory before giving up.
1647          */
1648         block = nblocks;
1649         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1650         if (block < nblocks) {
1651                 start = 0;
1652                 goto restart;
1653         }
1654
1655 cleanup_and_exit:
1656         /* Clean up the read-ahead blocks */
1657         for (; ra_ptr < ra_max; ra_ptr++)
1658                 brelse(bh_use[ra_ptr]);
1659         return ret;
1660 }
1661
1662 static struct buffer_head *ext4_find_entry(struct inode *dir,
1663                                            const struct qstr *d_name,
1664                                            struct ext4_dir_entry_2 **res_dir,
1665                                            int *inlined)
1666 {
1667         int err;
1668         struct ext4_filename fname;
1669         struct buffer_head *bh;
1670
1671         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1672         if (err == -ENOENT)
1673                 return NULL;
1674         if (err)
1675                 return ERR_PTR(err);
1676
1677         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1678
1679         ext4_fname_free_filename(&fname);
1680         return bh;
1681 }
1682
1683 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1684                                              struct dentry *dentry,
1685                                              struct ext4_dir_entry_2 **res_dir)
1686 {
1687         int err;
1688         struct ext4_filename fname;
1689         struct buffer_head *bh;
1690
1691         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1692         generic_set_encrypted_ci_d_ops(dentry);
1693         if (err == -ENOENT)
1694                 return NULL;
1695         if (err)
1696                 return ERR_PTR(err);
1697
1698         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1699
1700         ext4_fname_free_filename(&fname);
1701         return bh;
1702 }
1703
1704 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1705                         struct ext4_filename *fname,
1706                         struct ext4_dir_entry_2 **res_dir)
1707 {
1708         struct super_block * sb = dir->i_sb;
1709         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1710         struct buffer_head *bh;
1711         ext4_lblk_t block;
1712         int retval;
1713
1714 #ifdef CONFIG_FS_ENCRYPTION
1715         *res_dir = NULL;
1716 #endif
1717         frame = dx_probe(fname, dir, NULL, frames);
1718         if (IS_ERR(frame))
1719                 return (struct buffer_head *) frame;
1720         do {
1721                 block = dx_get_block(frame->at);
1722                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1723                 if (IS_ERR(bh))
1724                         goto errout;
1725
1726                 retval = search_dirblock(bh, dir, fname,
1727                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1728                                          res_dir);
1729                 if (retval == 1)
1730                         goto success;
1731                 brelse(bh);
1732                 if (retval == -1) {
1733                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1734                         goto errout;
1735                 }
1736
1737                 /* Check to see if we should continue to search */
1738                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1739                                                frames, NULL);
1740                 if (retval < 0) {
1741                         ext4_warning_inode(dir,
1742                                 "error %d reading directory index block",
1743                                 retval);
1744                         bh = ERR_PTR(retval);
1745                         goto errout;
1746                 }
1747         } while (retval == 1);
1748
1749         bh = NULL;
1750 errout:
1751         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1752 success:
1753         dx_release(frames);
1754         return bh;
1755 }
1756
1757 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1758 {
1759         struct inode *inode;
1760         struct ext4_dir_entry_2 *de;
1761         struct buffer_head *bh;
1762
1763         if (dentry->d_name.len > EXT4_NAME_LEN)
1764                 return ERR_PTR(-ENAMETOOLONG);
1765
1766         bh = ext4_lookup_entry(dir, dentry, &de);
1767         if (IS_ERR(bh))
1768                 return ERR_CAST(bh);
1769         inode = NULL;
1770         if (bh) {
1771                 __u32 ino = le32_to_cpu(de->inode);
1772                 brelse(bh);
1773                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1774                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1775                         return ERR_PTR(-EFSCORRUPTED);
1776                 }
1777                 if (unlikely(ino == dir->i_ino)) {
1778                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1779                                          dentry);
1780                         return ERR_PTR(-EFSCORRUPTED);
1781                 }
1782                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1783                 if (inode == ERR_PTR(-ESTALE)) {
1784                         EXT4_ERROR_INODE(dir,
1785                                          "deleted inode referenced: %u",
1786                                          ino);
1787                         return ERR_PTR(-EFSCORRUPTED);
1788                 }
1789                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1790                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1791                     !fscrypt_has_permitted_context(dir, inode)) {
1792                         ext4_warning(inode->i_sb,
1793                                      "Inconsistent encryption contexts: %lu/%lu",
1794                                      dir->i_ino, inode->i_ino);
1795                         iput(inode);
1796                         return ERR_PTR(-EPERM);
1797                 }
1798         }
1799
1800 #ifdef CONFIG_UNICODE
1801         if (!inode && IS_CASEFOLDED(dir)) {
1802                 /* Eventually we want to call d_add_ci(dentry, NULL)
1803                  * for negative dentries in the encoding case as
1804                  * well.  For now, prevent the negative dentry
1805                  * from being cached.
1806                  */
1807                 return NULL;
1808         }
1809 #endif
1810         return d_splice_alias(inode, dentry);
1811 }
1812
1813
1814 struct dentry *ext4_get_parent(struct dentry *child)
1815 {
1816         __u32 ino;
1817         struct ext4_dir_entry_2 * de;
1818         struct buffer_head *bh;
1819
1820         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1821         if (IS_ERR(bh))
1822                 return ERR_CAST(bh);
1823         if (!bh)
1824                 return ERR_PTR(-ENOENT);
1825         ino = le32_to_cpu(de->inode);
1826         brelse(bh);
1827
1828         if (!ext4_valid_inum(child->d_sb, ino)) {
1829                 EXT4_ERROR_INODE(d_inode(child),
1830                                  "bad parent inode number: %u", ino);
1831                 return ERR_PTR(-EFSCORRUPTED);
1832         }
1833
1834         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1835 }
1836
1837 /*
1838  * Move count entries from end of map between two memory locations.
1839  * Returns pointer to last entry moved.
1840  */
1841 static struct ext4_dir_entry_2 *
1842 dx_move_dirents(struct inode *dir, char *from, char *to,
1843                 struct dx_map_entry *map, int count,
1844                 unsigned blocksize)
1845 {
1846         unsigned rec_len = 0;
1847
1848         while (count--) {
1849                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1850                                                 (from + (map->offs<<2));
1851                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1852
1853                 memcpy (to, de, rec_len);
1854                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1855                                 ext4_rec_len_to_disk(rec_len, blocksize);
1856
1857                 /* wipe dir_entry excluding the rec_len field */
1858                 de->inode = 0;
1859                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1860                                                                 blocksize) -
1861                                          offsetof(struct ext4_dir_entry_2,
1862                                                                 name_len));
1863
1864                 map++;
1865                 to += rec_len;
1866         }
1867         return (struct ext4_dir_entry_2 *) (to - rec_len);
1868 }
1869
1870 /*
1871  * Compact each dir entry in the range to the minimal rec_len.
1872  * Returns pointer to last entry in range.
1873  */
1874 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1875                                                         unsigned int blocksize)
1876 {
1877         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1878         unsigned rec_len = 0;
1879
1880         prev = to = de;
1881         while ((char*)de < base + blocksize) {
1882                 next = ext4_next_entry(de, blocksize);
1883                 if (de->inode && de->name_len) {
1884                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1885                         if (de > to)
1886                                 memmove(to, de, rec_len);
1887                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1888                         prev = to;
1889                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1890                 }
1891                 de = next;
1892         }
1893         return prev;
1894 }
1895
1896 /*
1897  * Split a full leaf block to make room for a new dir entry.
1898  * Allocate a new block, and move entries so that they are approx. equally full.
1899  * Returns pointer to de in block into which the new entry will be inserted.
1900  */
1901 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1902                         struct buffer_head **bh,struct dx_frame *frame,
1903                         struct dx_hash_info *hinfo)
1904 {
1905         unsigned blocksize = dir->i_sb->s_blocksize;
1906         unsigned count, continued;
1907         struct buffer_head *bh2;
1908         ext4_lblk_t newblock;
1909         u32 hash2;
1910         struct dx_map_entry *map;
1911         char *data1 = (*bh)->b_data, *data2;
1912         unsigned split, move, size;
1913         struct ext4_dir_entry_2 *de = NULL, *de2;
1914         int     csum_size = 0;
1915         int     err = 0, i;
1916
1917         if (ext4_has_metadata_csum(dir->i_sb))
1918                 csum_size = sizeof(struct ext4_dir_entry_tail);
1919
1920         bh2 = ext4_append(handle, dir, &newblock);
1921         if (IS_ERR(bh2)) {
1922                 brelse(*bh);
1923                 *bh = NULL;
1924                 return (struct ext4_dir_entry_2 *) bh2;
1925         }
1926
1927         BUFFER_TRACE(*bh, "get_write_access");
1928         err = ext4_journal_get_write_access(handle, *bh);
1929         if (err)
1930                 goto journal_error;
1931
1932         BUFFER_TRACE(frame->bh, "get_write_access");
1933         err = ext4_journal_get_write_access(handle, frame->bh);
1934         if (err)
1935                 goto journal_error;
1936
1937         data2 = bh2->b_data;
1938
1939         /* create map in the end of data2 block */
1940         map = (struct dx_map_entry *) (data2 + blocksize);
1941         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1942                              blocksize, hinfo, map);
1943         map -= count;
1944         dx_sort_map(map, count);
1945         /* Ensure that neither split block is over half full */
1946         size = 0;
1947         move = 0;
1948         for (i = count-1; i >= 0; i--) {
1949                 /* is more than half of this entry in 2nd half of the block? */
1950                 if (size + map[i].size/2 > blocksize/2)
1951                         break;
1952                 size += map[i].size;
1953                 move++;
1954         }
1955         /*
1956          * map index at which we will split
1957          *
1958          * If the sum of active entries didn't exceed half the block size, just
1959          * split it in half by count; each resulting block will have at least
1960          * half the space free.
1961          */
1962         if (i > 0)
1963                 split = count - move;
1964         else
1965                 split = count/2;
1966
1967         hash2 = map[split].hash;
1968         continued = hash2 == map[split - 1].hash;
1969         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1970                         (unsigned long)dx_get_block(frame->at),
1971                                         hash2, split, count-split));
1972
1973         /* Fancy dance to stay within two buffers */
1974         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
1975                               blocksize);
1976         de = dx_pack_dirents(dir, data1, blocksize);
1977         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1978                                            (char *) de,
1979                                            blocksize);
1980         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1981                                             (char *) de2,
1982                                             blocksize);
1983         if (csum_size) {
1984                 ext4_initialize_dirent_tail(*bh, blocksize);
1985                 ext4_initialize_dirent_tail(bh2, blocksize);
1986         }
1987
1988         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1989                         blocksize, 1));
1990         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1991                         blocksize, 1));
1992
1993         /* Which block gets the new entry? */
1994         if (hinfo->hash >= hash2) {
1995                 swap(*bh, bh2);
1996                 de = de2;
1997         }
1998         dx_insert_block(frame, hash2 + continued, newblock);
1999         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2000         if (err)
2001                 goto journal_error;
2002         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2003         if (err)
2004                 goto journal_error;
2005         brelse(bh2);
2006         dxtrace(dx_show_index("frame", frame->entries));
2007         return de;
2008
2009 journal_error:
2010         brelse(*bh);
2011         brelse(bh2);
2012         *bh = NULL;
2013         ext4_std_error(dir->i_sb, err);
2014         return ERR_PTR(err);
2015 }
2016
2017 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2018                       struct buffer_head *bh,
2019                       void *buf, int buf_size,
2020                       struct ext4_filename *fname,
2021                       struct ext4_dir_entry_2 **dest_de)
2022 {
2023         struct ext4_dir_entry_2 *de;
2024         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2025         int nlen, rlen;
2026         unsigned int offset = 0;
2027         char *top;
2028
2029         de = (struct ext4_dir_entry_2 *)buf;
2030         top = buf + buf_size - reclen;
2031         while ((char *) de <= top) {
2032                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2033                                          buf, buf_size, offset))
2034                         return -EFSCORRUPTED;
2035                 if (ext4_match(dir, fname, de))
2036                         return -EEXIST;
2037                 nlen = ext4_dir_rec_len(de->name_len, dir);
2038                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2039                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2040                         break;
2041                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2042                 offset += rlen;
2043         }
2044         if ((char *) de > top)
2045                 return -ENOSPC;
2046
2047         *dest_de = de;
2048         return 0;
2049 }
2050
2051 void ext4_insert_dentry(struct inode *dir,
2052                         struct inode *inode,
2053                         struct ext4_dir_entry_2 *de,
2054                         int buf_size,
2055                         struct ext4_filename *fname)
2056 {
2057
2058         int nlen, rlen;
2059
2060         nlen = ext4_dir_rec_len(de->name_len, dir);
2061         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2062         if (de->inode) {
2063                 struct ext4_dir_entry_2 *de1 =
2064                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2065                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2066                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2067                 de = de1;
2068         }
2069         de->file_type = EXT4_FT_UNKNOWN;
2070         de->inode = cpu_to_le32(inode->i_ino);
2071         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2072         de->name_len = fname_len(fname);
2073         memcpy(de->name, fname_name(fname), fname_len(fname));
2074         if (ext4_hash_in_dirent(dir)) {
2075                 struct dx_hash_info *hinfo = &fname->hinfo;
2076
2077                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2078                 EXT4_DIRENT_HASHES(de)->minor_hash =
2079                                                 cpu_to_le32(hinfo->minor_hash);
2080         }
2081 }
2082
2083 /*
2084  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2085  * it points to a directory entry which is guaranteed to be large
2086  * enough for new directory entry.  If de is NULL, then
2087  * add_dirent_to_buf will attempt search the directory block for
2088  * space.  It will return -ENOSPC if no space is available, and -EIO
2089  * and -EEXIST if directory entry already exists.
2090  */
2091 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2092                              struct inode *dir,
2093                              struct inode *inode, struct ext4_dir_entry_2 *de,
2094                              struct buffer_head *bh)
2095 {
2096         unsigned int    blocksize = dir->i_sb->s_blocksize;
2097         int             csum_size = 0;
2098         int             err, err2;
2099
2100         if (ext4_has_metadata_csum(inode->i_sb))
2101                 csum_size = sizeof(struct ext4_dir_entry_tail);
2102
2103         if (!de) {
2104                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2105                                         blocksize - csum_size, fname, &de);
2106                 if (err)
2107                         return err;
2108         }
2109         BUFFER_TRACE(bh, "get_write_access");
2110         err = ext4_journal_get_write_access(handle, bh);
2111         if (err) {
2112                 ext4_std_error(dir->i_sb, err);
2113                 return err;
2114         }
2115
2116         /* By now the buffer is marked for journaling */
2117         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2118
2119         /*
2120          * XXX shouldn't update any times until successful
2121          * completion of syscall, but too many callers depend
2122          * on this.
2123          *
2124          * XXX similarly, too many callers depend on
2125          * ext4_new_inode() setting the times, but error
2126          * recovery deletes the inode, so the worst that can
2127          * happen is that the times are slightly out of date
2128          * and/or different from the directory change time.
2129          */
2130         dir->i_mtime = dir->i_ctime = current_time(dir);
2131         ext4_update_dx_flag(dir);
2132         inode_inc_iversion(dir);
2133         err2 = ext4_mark_inode_dirty(handle, dir);
2134         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2135         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2136         if (err)
2137                 ext4_std_error(dir->i_sb, err);
2138         return err ? err : err2;
2139 }
2140
2141 /*
2142  * This converts a one block unindexed directory to a 3 block indexed
2143  * directory, and adds the dentry to the indexed directory.
2144  */
2145 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2146                             struct inode *dir,
2147                             struct inode *inode, struct buffer_head *bh)
2148 {
2149         struct buffer_head *bh2;
2150         struct dx_root  *root;
2151         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2152         struct dx_entry *entries;
2153         struct ext4_dir_entry_2 *de, *de2;
2154         char            *data2, *top;
2155         unsigned        len;
2156         int             retval;
2157         unsigned        blocksize;
2158         ext4_lblk_t  block;
2159         struct fake_dirent *fde;
2160         int csum_size = 0;
2161
2162         if (ext4_has_metadata_csum(inode->i_sb))
2163                 csum_size = sizeof(struct ext4_dir_entry_tail);
2164
2165         blocksize =  dir->i_sb->s_blocksize;
2166         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2167         BUFFER_TRACE(bh, "get_write_access");
2168         retval = ext4_journal_get_write_access(handle, bh);
2169         if (retval) {
2170                 ext4_std_error(dir->i_sb, retval);
2171                 brelse(bh);
2172                 return retval;
2173         }
2174         root = (struct dx_root *) bh->b_data;
2175
2176         /* The 0th block becomes the root, move the dirents out */
2177         fde = &root->dotdot;
2178         de = (struct ext4_dir_entry_2 *)((char *)fde +
2179                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2180         if ((char *) de >= (((char *) root) + blocksize)) {
2181                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2182                 brelse(bh);
2183                 return -EFSCORRUPTED;
2184         }
2185         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2186
2187         /* Allocate new block for the 0th block's dirents */
2188         bh2 = ext4_append(handle, dir, &block);
2189         if (IS_ERR(bh2)) {
2190                 brelse(bh);
2191                 return PTR_ERR(bh2);
2192         }
2193         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2194         data2 = bh2->b_data;
2195
2196         memcpy(data2, de, len);
2197         memset(de, 0, len); /* wipe old data */
2198         de = (struct ext4_dir_entry_2 *) data2;
2199         top = data2 + len;
2200         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2201                 de = de2;
2202         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2203                                            (char *) de, blocksize);
2204
2205         if (csum_size)
2206                 ext4_initialize_dirent_tail(bh2, blocksize);
2207
2208         /* Initialize the root; the dot dirents already exist */
2209         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2210         de->rec_len = ext4_rec_len_to_disk(
2211                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2212         memset (&root->info, 0, sizeof(root->info));
2213         root->info.info_length = sizeof(root->info);
2214         if (ext4_hash_in_dirent(dir))
2215                 root->info.hash_version = DX_HASH_SIPHASH;
2216         else
2217                 root->info.hash_version =
2218                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2219
2220         entries = root->entries;
2221         dx_set_block(entries, 1);
2222         dx_set_count(entries, 1);
2223         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2224
2225         /* Initialize as for dx_probe */
2226         fname->hinfo.hash_version = root->info.hash_version;
2227         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2228                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2229         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2230
2231         /* casefolded encrypted hashes are computed on fname setup */
2232         if (!ext4_hash_in_dirent(dir))
2233                 ext4fs_dirhash(dir, fname_name(fname),
2234                                 fname_len(fname), &fname->hinfo);
2235
2236         memset(frames, 0, sizeof(frames));
2237         frame = frames;
2238         frame->entries = entries;
2239         frame->at = entries;
2240         frame->bh = bh;
2241
2242         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2243         if (retval)
2244                 goto out_frames;
2245         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2246         if (retval)
2247                 goto out_frames;
2248
2249         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2250         if (IS_ERR(de)) {
2251                 retval = PTR_ERR(de);
2252                 goto out_frames;
2253         }
2254
2255         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2256 out_frames:
2257         /*
2258          * Even if the block split failed, we have to properly write
2259          * out all the changes we did so far. Otherwise we can end up
2260          * with corrupted filesystem.
2261          */
2262         if (retval)
2263                 ext4_mark_inode_dirty(handle, dir);
2264         dx_release(frames);
2265         brelse(bh2);
2266         return retval;
2267 }
2268
2269 /*
2270  *      ext4_add_entry()
2271  *
2272  * adds a file entry to the specified directory, using the same
2273  * semantics as ext4_find_entry(). It returns NULL if it failed.
2274  *
2275  * NOTE!! The inode part of 'de' is left at 0 - which means you
2276  * may not sleep between calling this and putting something into
2277  * the entry, as someone else might have used it while you slept.
2278  */
2279 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2280                           struct inode *inode)
2281 {
2282         struct inode *dir = d_inode(dentry->d_parent);
2283         struct buffer_head *bh = NULL;
2284         struct ext4_dir_entry_2 *de;
2285         struct super_block *sb;
2286         struct ext4_filename fname;
2287         int     retval;
2288         int     dx_fallback=0;
2289         unsigned blocksize;
2290         ext4_lblk_t block, blocks;
2291         int     csum_size = 0;
2292
2293         if (ext4_has_metadata_csum(inode->i_sb))
2294                 csum_size = sizeof(struct ext4_dir_entry_tail);
2295
2296         sb = dir->i_sb;
2297         blocksize = sb->s_blocksize;
2298         if (!dentry->d_name.len)
2299                 return -EINVAL;
2300
2301         if (fscrypt_is_nokey_name(dentry))
2302                 return -ENOKEY;
2303
2304 #ifdef CONFIG_UNICODE
2305         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2306             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2307                 return -EINVAL;
2308 #endif
2309
2310         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2311         if (retval)
2312                 return retval;
2313
2314         if (ext4_has_inline_data(dir)) {
2315                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2316                 if (retval < 0)
2317                         goto out;
2318                 if (retval == 1) {
2319                         retval = 0;
2320                         goto out;
2321                 }
2322         }
2323
2324         if (is_dx(dir)) {
2325                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2326                 if (!retval || (retval != ERR_BAD_DX_DIR))
2327                         goto out;
2328                 /* Can we just ignore htree data? */
2329                 if (ext4_has_metadata_csum(sb)) {
2330                         EXT4_ERROR_INODE(dir,
2331                                 "Directory has corrupted htree index.");
2332                         retval = -EFSCORRUPTED;
2333                         goto out;
2334                 }
2335                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2336                 dx_fallback++;
2337                 retval = ext4_mark_inode_dirty(handle, dir);
2338                 if (unlikely(retval))
2339                         goto out;
2340         }
2341         blocks = dir->i_size >> sb->s_blocksize_bits;
2342         for (block = 0; block < blocks; block++) {
2343                 bh = ext4_read_dirblock(dir, block, DIRENT);
2344                 if (bh == NULL) {
2345                         bh = ext4_bread(handle, dir, block,
2346                                         EXT4_GET_BLOCKS_CREATE);
2347                         goto add_to_new_block;
2348                 }
2349                 if (IS_ERR(bh)) {
2350                         retval = PTR_ERR(bh);
2351                         bh = NULL;
2352                         goto out;
2353                 }
2354                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2355                                            NULL, bh);
2356                 if (retval != -ENOSPC)
2357                         goto out;
2358
2359                 if (blocks == 1 && !dx_fallback &&
2360                     ext4_has_feature_dir_index(sb)) {
2361                         retval = make_indexed_dir(handle, &fname, dir,
2362                                                   inode, bh);
2363                         bh = NULL; /* make_indexed_dir releases bh */
2364                         goto out;
2365                 }
2366                 brelse(bh);
2367         }
2368         bh = ext4_append(handle, dir, &block);
2369 add_to_new_block:
2370         if (IS_ERR(bh)) {
2371                 retval = PTR_ERR(bh);
2372                 bh = NULL;
2373                 goto out;
2374         }
2375         de = (struct ext4_dir_entry_2 *) bh->b_data;
2376         de->inode = 0;
2377         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2378
2379         if (csum_size)
2380                 ext4_initialize_dirent_tail(bh, blocksize);
2381
2382         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2383 out:
2384         ext4_fname_free_filename(&fname);
2385         brelse(bh);
2386         if (retval == 0)
2387                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2388         return retval;
2389 }
2390
2391 /*
2392  * Returns 0 for success, or a negative error value
2393  */
2394 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2395                              struct inode *dir, struct inode *inode)
2396 {
2397         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2398         struct dx_entry *entries, *at;
2399         struct buffer_head *bh;
2400         struct super_block *sb = dir->i_sb;
2401         struct ext4_dir_entry_2 *de;
2402         int restart;
2403         int err;
2404
2405 again:
2406         restart = 0;
2407         frame = dx_probe(fname, dir, NULL, frames);
2408         if (IS_ERR(frame))
2409                 return PTR_ERR(frame);
2410         entries = frame->entries;
2411         at = frame->at;
2412         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2413         if (IS_ERR(bh)) {
2414                 err = PTR_ERR(bh);
2415                 bh = NULL;
2416                 goto cleanup;
2417         }
2418
2419         BUFFER_TRACE(bh, "get_write_access");
2420         err = ext4_journal_get_write_access(handle, bh);
2421         if (err)
2422                 goto journal_error;
2423
2424         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2425         if (err != -ENOSPC)
2426                 goto cleanup;
2427
2428         err = 0;
2429         /* Block full, should compress but for now just split */
2430         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2431                        dx_get_count(entries), dx_get_limit(entries)));
2432         /* Need to split index? */
2433         if (dx_get_count(entries) == dx_get_limit(entries)) {
2434                 ext4_lblk_t newblock;
2435                 int levels = frame - frames + 1;
2436                 unsigned int icount;
2437                 int add_level = 1;
2438                 struct dx_entry *entries2;
2439                 struct dx_node *node2;
2440                 struct buffer_head *bh2;
2441
2442                 while (frame > frames) {
2443                         if (dx_get_count((frame - 1)->entries) <
2444                             dx_get_limit((frame - 1)->entries)) {
2445                                 add_level = 0;
2446                                 break;
2447                         }
2448                         frame--; /* split higher index block */
2449                         at = frame->at;
2450                         entries = frame->entries;
2451                         restart = 1;
2452                 }
2453                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2454                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2455                                          "reach max htree level :%d",
2456                                          dir->i_ino, levels);
2457                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2458                                 ext4_warning(sb, "Large directory feature is "
2459                                                  "not enabled on this "
2460                                                  "filesystem");
2461                         }
2462                         err = -ENOSPC;
2463                         goto cleanup;
2464                 }
2465                 icount = dx_get_count(entries);
2466                 bh2 = ext4_append(handle, dir, &newblock);
2467                 if (IS_ERR(bh2)) {
2468                         err = PTR_ERR(bh2);
2469                         goto cleanup;
2470                 }
2471                 node2 = (struct dx_node *)(bh2->b_data);
2472                 entries2 = node2->entries;
2473                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2474                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2475                                                            sb->s_blocksize);
2476                 BUFFER_TRACE(frame->bh, "get_write_access");
2477                 err = ext4_journal_get_write_access(handle, frame->bh);
2478                 if (err)
2479                         goto journal_error;
2480                 if (!add_level) {
2481                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2482                         unsigned hash2 = dx_get_hash(entries + icount1);
2483                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2484                                        icount1, icount2));
2485
2486                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2487                         err = ext4_journal_get_write_access(handle,
2488                                                              (frame - 1)->bh);
2489                         if (err)
2490                                 goto journal_error;
2491
2492                         memcpy((char *) entries2, (char *) (entries + icount1),
2493                                icount2 * sizeof(struct dx_entry));
2494                         dx_set_count(entries, icount1);
2495                         dx_set_count(entries2, icount2);
2496                         dx_set_limit(entries2, dx_node_limit(dir));
2497
2498                         /* Which index block gets the new entry? */
2499                         if (at - entries >= icount1) {
2500                                 frame->at = at = at - entries - icount1 + entries2;
2501                                 frame->entries = entries = entries2;
2502                                 swap(frame->bh, bh2);
2503                         }
2504                         dx_insert_block((frame - 1), hash2, newblock);
2505                         dxtrace(dx_show_index("node", frame->entries));
2506                         dxtrace(dx_show_index("node",
2507                                ((struct dx_node *) bh2->b_data)->entries));
2508                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2509                         if (err)
2510                                 goto journal_error;
2511                         brelse (bh2);
2512                         err = ext4_handle_dirty_dx_node(handle, dir,
2513                                                    (frame - 1)->bh);
2514                         if (err)
2515                                 goto journal_error;
2516                         err = ext4_handle_dirty_dx_node(handle, dir,
2517                                                         frame->bh);
2518                         if (err)
2519                                 goto journal_error;
2520                 } else {
2521                         struct dx_root *dxroot;
2522                         memcpy((char *) entries2, (char *) entries,
2523                                icount * sizeof(struct dx_entry));
2524                         dx_set_limit(entries2, dx_node_limit(dir));
2525
2526                         /* Set up root */
2527                         dx_set_count(entries, 1);
2528                         dx_set_block(entries + 0, newblock);
2529                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2530                         dxroot->info.indirect_levels += 1;
2531                         dxtrace(printk(KERN_DEBUG
2532                                        "Creating %d level index...\n",
2533                                        dxroot->info.indirect_levels));
2534                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2535                         if (err)
2536                                 goto journal_error;
2537                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2538                         brelse(bh2);
2539                         restart = 1;
2540                         goto journal_error;
2541                 }
2542         }
2543         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2544         if (IS_ERR(de)) {
2545                 err = PTR_ERR(de);
2546                 goto cleanup;
2547         }
2548         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2549         goto cleanup;
2550
2551 journal_error:
2552         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2553 cleanup:
2554         brelse(bh);
2555         dx_release(frames);
2556         /* @restart is true means htree-path has been changed, we need to
2557          * repeat dx_probe() to find out valid htree-path
2558          */
2559         if (restart && err == 0)
2560                 goto again;
2561         return err;
2562 }
2563
2564 /*
2565  * ext4_generic_delete_entry deletes a directory entry by merging it
2566  * with the previous entry
2567  */
2568 int ext4_generic_delete_entry(struct inode *dir,
2569                               struct ext4_dir_entry_2 *de_del,
2570                               struct buffer_head *bh,
2571                               void *entry_buf,
2572                               int buf_size,
2573                               int csum_size)
2574 {
2575         struct ext4_dir_entry_2 *de, *pde;
2576         unsigned int blocksize = dir->i_sb->s_blocksize;
2577         int i;
2578
2579         i = 0;
2580         pde = NULL;
2581         de = (struct ext4_dir_entry_2 *)entry_buf;
2582         while (i < buf_size - csum_size) {
2583                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2584                                          entry_buf, buf_size, i))
2585                         return -EFSCORRUPTED;
2586                 if (de == de_del)  {
2587                         if (pde) {
2588                                 pde->rec_len = ext4_rec_len_to_disk(
2589                                         ext4_rec_len_from_disk(pde->rec_len,
2590                                                                blocksize) +
2591                                         ext4_rec_len_from_disk(de->rec_len,
2592                                                                blocksize),
2593                                         blocksize);
2594
2595                                 /* wipe entire dir_entry */
2596                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2597                                                                 blocksize));
2598                         } else {
2599                                 /* wipe dir_entry excluding the rec_len field */
2600                                 de->inode = 0;
2601                                 memset(&de->name_len, 0,
2602                                         ext4_rec_len_from_disk(de->rec_len,
2603                                                                 blocksize) -
2604                                         offsetof(struct ext4_dir_entry_2,
2605                                                                 name_len));
2606                         }
2607
2608                         inode_inc_iversion(dir);
2609                         return 0;
2610                 }
2611                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2612                 pde = de;
2613                 de = ext4_next_entry(de, blocksize);
2614         }
2615         return -ENOENT;
2616 }
2617
2618 static int ext4_delete_entry(handle_t *handle,
2619                              struct inode *dir,
2620                              struct ext4_dir_entry_2 *de_del,
2621                              struct buffer_head *bh)
2622 {
2623         int err, csum_size = 0;
2624
2625         if (ext4_has_inline_data(dir)) {
2626                 int has_inline_data = 1;
2627                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2628                                                &has_inline_data);
2629                 if (has_inline_data)
2630                         return err;
2631         }
2632
2633         if (ext4_has_metadata_csum(dir->i_sb))
2634                 csum_size = sizeof(struct ext4_dir_entry_tail);
2635
2636         BUFFER_TRACE(bh, "get_write_access");
2637         err = ext4_journal_get_write_access(handle, bh);
2638         if (unlikely(err))
2639                 goto out;
2640
2641         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2642                                         dir->i_sb->s_blocksize, csum_size);
2643         if (err)
2644                 goto out;
2645
2646         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2647         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2648         if (unlikely(err))
2649                 goto out;
2650
2651         return 0;
2652 out:
2653         if (err != -ENOENT)
2654                 ext4_std_error(dir->i_sb, err);
2655         return err;
2656 }
2657
2658 /*
2659  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2660  * since this indicates that nlinks count was previously 1 to avoid overflowing
2661  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2662  * that subdirectory link counts are not being maintained accurately.
2663  *
2664  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2665  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2666  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2667  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2668  */
2669 static void ext4_inc_count(struct inode *inode)
2670 {
2671         inc_nlink(inode);
2672         if (is_dx(inode) &&
2673             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2674                 set_nlink(inode, 1);
2675 }
2676
2677 /*
2678  * If a directory had nlink == 1, then we should let it be 1. This indicates
2679  * directory has >EXT4_LINK_MAX subdirs.
2680  */
2681 static void ext4_dec_count(struct inode *inode)
2682 {
2683         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2684                 drop_nlink(inode);
2685 }
2686
2687
2688 /*
2689  * Add non-directory inode to a directory. On success, the inode reference is
2690  * consumed by dentry is instantiation. This is also indicated by clearing of
2691  * *inodep pointer. On failure, the caller is responsible for dropping the
2692  * inode reference in the safe context.
2693  */
2694 static int ext4_add_nondir(handle_t *handle,
2695                 struct dentry *dentry, struct inode **inodep)
2696 {
2697         struct inode *dir = d_inode(dentry->d_parent);
2698         struct inode *inode = *inodep;
2699         int err = ext4_add_entry(handle, dentry, inode);
2700         if (!err) {
2701                 err = ext4_mark_inode_dirty(handle, inode);
2702                 if (IS_DIRSYNC(dir))
2703                         ext4_handle_sync(handle);
2704                 d_instantiate_new(dentry, inode);
2705                 *inodep = NULL;
2706                 return err;
2707         }
2708         drop_nlink(inode);
2709         ext4_orphan_add(handle, inode);
2710         unlock_new_inode(inode);
2711         return err;
2712 }
2713
2714 /*
2715  * By the time this is called, we already have created
2716  * the directory cache entry for the new file, but it
2717  * is so far negative - it has no inode.
2718  *
2719  * If the create succeeds, we fill in the inode information
2720  * with d_instantiate().
2721  */
2722 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2723                        struct dentry *dentry, umode_t mode, bool excl)
2724 {
2725         handle_t *handle;
2726         struct inode *inode;
2727         int err, credits, retries = 0;
2728
2729         err = dquot_initialize(dir);
2730         if (err)
2731                 return err;
2732
2733         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2734                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2735 retry:
2736         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2737                                             0, NULL, EXT4_HT_DIR, credits);
2738         handle = ext4_journal_current_handle();
2739         err = PTR_ERR(inode);
2740         if (!IS_ERR(inode)) {
2741                 inode->i_op = &ext4_file_inode_operations;
2742                 inode->i_fop = &ext4_file_operations;
2743                 ext4_set_aops(inode);
2744                 err = ext4_add_nondir(handle, dentry, &inode);
2745                 if (!err)
2746                         ext4_fc_track_create(handle, dentry);
2747         }
2748         if (handle)
2749                 ext4_journal_stop(handle);
2750         if (!IS_ERR_OR_NULL(inode))
2751                 iput(inode);
2752         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2753                 goto retry;
2754         return err;
2755 }
2756
2757 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2758                       struct dentry *dentry, umode_t mode, dev_t rdev)
2759 {
2760         handle_t *handle;
2761         struct inode *inode;
2762         int err, credits, retries = 0;
2763
2764         err = dquot_initialize(dir);
2765         if (err)
2766                 return err;
2767
2768         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2769                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2770 retry:
2771         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2772                                             0, NULL, EXT4_HT_DIR, credits);
2773         handle = ext4_journal_current_handle();
2774         err = PTR_ERR(inode);
2775         if (!IS_ERR(inode)) {
2776                 init_special_inode(inode, inode->i_mode, rdev);
2777                 inode->i_op = &ext4_special_inode_operations;
2778                 err = ext4_add_nondir(handle, dentry, &inode);
2779                 if (!err)
2780                         ext4_fc_track_create(handle, dentry);
2781         }
2782         if (handle)
2783                 ext4_journal_stop(handle);
2784         if (!IS_ERR_OR_NULL(inode))
2785                 iput(inode);
2786         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2787                 goto retry;
2788         return err;
2789 }
2790
2791 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2792                         struct dentry *dentry, umode_t mode)
2793 {
2794         handle_t *handle;
2795         struct inode *inode;
2796         int err, retries = 0;
2797
2798         err = dquot_initialize(dir);
2799         if (err)
2800                 return err;
2801
2802 retry:
2803         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2804                                             NULL, 0, NULL,
2805                                             EXT4_HT_DIR,
2806                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2807                           4 + EXT4_XATTR_TRANS_BLOCKS);
2808         handle = ext4_journal_current_handle();
2809         err = PTR_ERR(inode);
2810         if (!IS_ERR(inode)) {
2811                 inode->i_op = &ext4_file_inode_operations;
2812                 inode->i_fop = &ext4_file_operations;
2813                 ext4_set_aops(inode);
2814                 d_tmpfile(dentry, inode);
2815                 err = ext4_orphan_add(handle, inode);
2816                 if (err)
2817                         goto err_unlock_inode;
2818                 mark_inode_dirty(inode);
2819                 unlock_new_inode(inode);
2820         }
2821         if (handle)
2822                 ext4_journal_stop(handle);
2823         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2824                 goto retry;
2825         return err;
2826 err_unlock_inode:
2827         ext4_journal_stop(handle);
2828         unlock_new_inode(inode);
2829         return err;
2830 }
2831
2832 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2833                           struct ext4_dir_entry_2 *de,
2834                           int blocksize, int csum_size,
2835                           unsigned int parent_ino, int dotdot_real_len)
2836 {
2837         de->inode = cpu_to_le32(inode->i_ino);
2838         de->name_len = 1;
2839         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2840                                            blocksize);
2841         strcpy(de->name, ".");
2842         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2843
2844         de = ext4_next_entry(de, blocksize);
2845         de->inode = cpu_to_le32(parent_ino);
2846         de->name_len = 2;
2847         if (!dotdot_real_len)
2848                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2849                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2850                                         blocksize);
2851         else
2852                 de->rec_len = ext4_rec_len_to_disk(
2853                                         ext4_dir_rec_len(de->name_len, NULL),
2854                                         blocksize);
2855         strcpy(de->name, "..");
2856         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2857
2858         return ext4_next_entry(de, blocksize);
2859 }
2860
2861 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2862                              struct inode *inode)
2863 {
2864         struct buffer_head *dir_block = NULL;
2865         struct ext4_dir_entry_2 *de;
2866         ext4_lblk_t block = 0;
2867         unsigned int blocksize = dir->i_sb->s_blocksize;
2868         int csum_size = 0;
2869         int err;
2870
2871         if (ext4_has_metadata_csum(dir->i_sb))
2872                 csum_size = sizeof(struct ext4_dir_entry_tail);
2873
2874         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2875                 err = ext4_try_create_inline_dir(handle, dir, inode);
2876                 if (err < 0 && err != -ENOSPC)
2877                         goto out;
2878                 if (!err)
2879                         goto out;
2880         }
2881
2882         inode->i_size = 0;
2883         dir_block = ext4_append(handle, inode, &block);
2884         if (IS_ERR(dir_block))
2885                 return PTR_ERR(dir_block);
2886         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2887         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2888         set_nlink(inode, 2);
2889         if (csum_size)
2890                 ext4_initialize_dirent_tail(dir_block, blocksize);
2891
2892         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2893         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2894         if (err)
2895                 goto out;
2896         set_buffer_verified(dir_block);
2897 out:
2898         brelse(dir_block);
2899         return err;
2900 }
2901
2902 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2903                       struct dentry *dentry, umode_t mode)
2904 {
2905         handle_t *handle;
2906         struct inode *inode;
2907         int err, err2 = 0, credits, retries = 0;
2908
2909         if (EXT4_DIR_LINK_MAX(dir))
2910                 return -EMLINK;
2911
2912         err = dquot_initialize(dir);
2913         if (err)
2914                 return err;
2915
2916         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2917                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2918 retry:
2919         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2920                                             &dentry->d_name,
2921                                             0, NULL, EXT4_HT_DIR, credits);
2922         handle = ext4_journal_current_handle();
2923         err = PTR_ERR(inode);
2924         if (IS_ERR(inode))
2925                 goto out_stop;
2926
2927         inode->i_op = &ext4_dir_inode_operations;
2928         inode->i_fop = &ext4_dir_operations;
2929         err = ext4_init_new_dir(handle, dir, inode);
2930         if (err)
2931                 goto out_clear_inode;
2932         err = ext4_mark_inode_dirty(handle, inode);
2933         if (!err)
2934                 err = ext4_add_entry(handle, dentry, inode);
2935         if (err) {
2936 out_clear_inode:
2937                 clear_nlink(inode);
2938                 ext4_orphan_add(handle, inode);
2939                 unlock_new_inode(inode);
2940                 err2 = ext4_mark_inode_dirty(handle, inode);
2941                 if (unlikely(err2))
2942                         err = err2;
2943                 ext4_journal_stop(handle);
2944                 iput(inode);
2945                 goto out_retry;
2946         }
2947         ext4_inc_count(dir);
2948
2949         ext4_update_dx_flag(dir);
2950         err = ext4_mark_inode_dirty(handle, dir);
2951         if (err)
2952                 goto out_clear_inode;
2953         d_instantiate_new(dentry, inode);
2954         ext4_fc_track_create(handle, dentry);
2955         if (IS_DIRSYNC(dir))
2956                 ext4_handle_sync(handle);
2957
2958 out_stop:
2959         if (handle)
2960                 ext4_journal_stop(handle);
2961 out_retry:
2962         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2963                 goto retry;
2964         return err;
2965 }
2966
2967 /*
2968  * routine to check that the specified directory is empty (for rmdir)
2969  */
2970 bool ext4_empty_dir(struct inode *inode)
2971 {
2972         unsigned int offset;
2973         struct buffer_head *bh;
2974         struct ext4_dir_entry_2 *de;
2975         struct super_block *sb;
2976
2977         if (ext4_has_inline_data(inode)) {
2978                 int has_inline_data = 1;
2979                 int ret;
2980
2981                 ret = empty_inline_dir(inode, &has_inline_data);
2982                 if (has_inline_data)
2983                         return ret;
2984         }
2985
2986         sb = inode->i_sb;
2987         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
2988                                         ext4_dir_rec_len(2, NULL)) {
2989                 EXT4_ERROR_INODE(inode, "invalid size");
2990                 return true;
2991         }
2992         /* The first directory block must not be a hole,
2993          * so treat it as DIRENT_HTREE
2994          */
2995         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2996         if (IS_ERR(bh))
2997                 return true;
2998
2999         de = (struct ext4_dir_entry_2 *) bh->b_data;
3000         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3001                                  0) ||
3002             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3003                 ext4_warning_inode(inode, "directory missing '.'");
3004                 brelse(bh);
3005                 return true;
3006         }
3007         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3008         de = ext4_next_entry(de, sb->s_blocksize);
3009         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3010                                  offset) ||
3011             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3012                 ext4_warning_inode(inode, "directory missing '..'");
3013                 brelse(bh);
3014                 return true;
3015         }
3016         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3017         while (offset < inode->i_size) {
3018                 if (!(offset & (sb->s_blocksize - 1))) {
3019                         unsigned int lblock;
3020                         brelse(bh);
3021                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3022                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3023                         if (bh == NULL) {
3024                                 offset += sb->s_blocksize;
3025                                 continue;
3026                         }
3027                         if (IS_ERR(bh))
3028                                 return true;
3029                 }
3030                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3031                                         (offset & (sb->s_blocksize - 1)));
3032                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3033                                          bh->b_data, bh->b_size, offset)) {
3034                         offset = (offset | (sb->s_blocksize - 1)) + 1;
3035                         continue;
3036                 }
3037                 if (le32_to_cpu(de->inode)) {
3038                         brelse(bh);
3039                         return false;
3040                 }
3041                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3042         }
3043         brelse(bh);
3044         return true;
3045 }
3046
3047 /*
3048  * ext4_orphan_add() links an unlinked or truncated inode into a list of
3049  * such inodes, starting at the superblock, in case we crash before the
3050  * file is closed/deleted, or in case the inode truncate spans multiple
3051  * transactions and the last transaction is not recovered after a crash.
3052  *
3053  * At filesystem recovery time, we walk this list deleting unlinked
3054  * inodes and truncating linked inodes in ext4_orphan_cleanup().
3055  *
3056  * Orphan list manipulation functions must be called under i_mutex unless
3057  * we are just creating the inode or deleting it.
3058  */
3059 int ext4_orphan_add(handle_t *handle, struct inode *inode)
3060 {
3061         struct super_block *sb = inode->i_sb;
3062         struct ext4_sb_info *sbi = EXT4_SB(sb);
3063         struct ext4_iloc iloc;
3064         int err = 0, rc;
3065         bool dirty = false;
3066
3067         if (!sbi->s_journal || is_bad_inode(inode))
3068                 return 0;
3069
3070         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3071                      !inode_is_locked(inode));
3072         /*
3073          * Exit early if inode already is on orphan list. This is a big speedup
3074          * since we don't have to contend on the global s_orphan_lock.
3075          */
3076         if (!list_empty(&EXT4_I(inode)->i_orphan))
3077                 return 0;
3078
3079         /*
3080          * Orphan handling is only valid for files with data blocks
3081          * being truncated, or files being unlinked. Note that we either
3082          * hold i_mutex, or the inode can not be referenced from outside,
3083          * so i_nlink should not be bumped due to race
3084          */
3085         ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3086                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
3087
3088         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3089         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3090         if (err)
3091                 goto out;
3092
3093         err = ext4_reserve_inode_write(handle, inode, &iloc);
3094         if (err)
3095                 goto out;
3096
3097         mutex_lock(&sbi->s_orphan_lock);
3098         /*
3099          * Due to previous errors inode may be already a part of on-disk
3100          * orphan list. If so skip on-disk list modification.
3101          */
3102         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
3103             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
3104                 /* Insert this inode at the head of the on-disk orphan list */
3105                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
3106                 lock_buffer(sbi->s_sbh);
3107                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
3108                 ext4_superblock_csum_set(sb);
3109                 unlock_buffer(sbi->s_sbh);
3110                 dirty = true;
3111         }
3112         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
3113         mutex_unlock(&sbi->s_orphan_lock);
3114
3115         if (dirty) {
3116                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3117                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3118                 if (!err)
3119                         err = rc;
3120                 if (err) {
3121                         /*
3122                          * We have to remove inode from in-memory list if
3123                          * addition to on disk orphan list failed. Stray orphan
3124                          * list entries can cause panics at unmount time.
3125                          */
3126                         mutex_lock(&sbi->s_orphan_lock);
3127                         list_del_init(&EXT4_I(inode)->i_orphan);
3128                         mutex_unlock(&sbi->s_orphan_lock);
3129                 }
3130         } else
3131                 brelse(iloc.bh);
3132
3133         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3134         jbd_debug(4, "orphan inode %lu will point to %d\n",
3135                         inode->i_ino, NEXT_ORPHAN(inode));
3136 out:
3137         ext4_std_error(sb, err);
3138         return err;
3139 }
3140
3141 /*
3142  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3143  * of such inodes stored on disk, because it is finally being cleaned up.
3144  */
3145 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3146 {
3147         struct list_head *prev;
3148         struct ext4_inode_info *ei = EXT4_I(inode);
3149         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3150         __u32 ino_next;
3151         struct ext4_iloc iloc;
3152         int err = 0;
3153
3154         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3155                 return 0;
3156
3157         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3158                      !inode_is_locked(inode));
3159         /* Do this quick check before taking global s_orphan_lock. */
3160         if (list_empty(&ei->i_orphan))
3161                 return 0;
3162
3163         if (handle) {
3164                 /* Grab inode buffer early before taking global s_orphan_lock */
3165                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3166         }
3167
3168         mutex_lock(&sbi->s_orphan_lock);
3169         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3170
3171         prev = ei->i_orphan.prev;
3172         list_del_init(&ei->i_orphan);
3173
3174         /* If we're on an error path, we may not have a valid
3175          * transaction handle with which to update the orphan list on
3176          * disk, but we still need to remove the inode from the linked
3177          * list in memory. */
3178         if (!handle || err) {
3179                 mutex_unlock(&sbi->s_orphan_lock);
3180                 goto out_err;
3181         }
3182
3183         ino_next = NEXT_ORPHAN(inode);
3184         if (prev == &sbi->s_orphan) {
3185                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3186                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3187                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3188                 if (err) {
3189                         mutex_unlock(&sbi->s_orphan_lock);
3190                         goto out_brelse;
3191                 }
3192                 lock_buffer(sbi->s_sbh);
3193                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3194                 ext4_superblock_csum_set(inode->i_sb);
3195                 unlock_buffer(sbi->s_sbh);
3196                 mutex_unlock(&sbi->s_orphan_lock);
3197                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3198         } else {
3199                 struct ext4_iloc iloc2;
3200                 struct inode *i_prev =
3201                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3202
3203                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3204                           i_prev->i_ino, ino_next);
3205                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3206                 if (err) {
3207                         mutex_unlock(&sbi->s_orphan_lock);
3208                         goto out_brelse;
3209                 }
3210                 NEXT_ORPHAN(i_prev) = ino_next;
3211                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3212                 mutex_unlock(&sbi->s_orphan_lock);
3213         }
3214         if (err)
3215                 goto out_brelse;
3216         NEXT_ORPHAN(inode) = 0;
3217         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3218 out_err:
3219         ext4_std_error(inode->i_sb, err);
3220         return err;
3221
3222 out_brelse:
3223         brelse(iloc.bh);
3224         goto out_err;
3225 }
3226
3227 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3228 {
3229         int retval;
3230         struct inode *inode;
3231         struct buffer_head *bh;
3232         struct ext4_dir_entry_2 *de;
3233         handle_t *handle = NULL;
3234
3235         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3236                 return -EIO;
3237
3238         /* Initialize quotas before so that eventual writes go in
3239          * separate transaction */
3240         retval = dquot_initialize(dir);
3241         if (retval)
3242                 return retval;
3243         retval = dquot_initialize(d_inode(dentry));
3244         if (retval)
3245                 return retval;
3246
3247         retval = -ENOENT;
3248         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3249         if (IS_ERR(bh))
3250                 return PTR_ERR(bh);
3251         if (!bh)
3252                 goto end_rmdir;
3253
3254         inode = d_inode(dentry);
3255
3256         retval = -EFSCORRUPTED;
3257         if (le32_to_cpu(de->inode) != inode->i_ino)
3258                 goto end_rmdir;
3259
3260         retval = -ENOTEMPTY;
3261         if (!ext4_empty_dir(inode))
3262                 goto end_rmdir;
3263
3264         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3265                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3266         if (IS_ERR(handle)) {
3267                 retval = PTR_ERR(handle);
3268                 handle = NULL;
3269                 goto end_rmdir;
3270         }
3271
3272         if (IS_DIRSYNC(dir))
3273                 ext4_handle_sync(handle);
3274
3275         retval = ext4_delete_entry(handle, dir, de, bh);
3276         if (retval)
3277                 goto end_rmdir;
3278         if (!EXT4_DIR_LINK_EMPTY(inode))
3279                 ext4_warning_inode(inode,
3280                              "empty directory '%.*s' has too many links (%u)",
3281                              dentry->d_name.len, dentry->d_name.name,
3282                              inode->i_nlink);
3283         inode_inc_iversion(inode);
3284         clear_nlink(inode);
3285         /* There's no need to set i_disksize: the fact that i_nlink is
3286          * zero will ensure that the right thing happens during any
3287          * recovery. */
3288         inode->i_size = 0;
3289         ext4_orphan_add(handle, inode);
3290         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3291         retval = ext4_mark_inode_dirty(handle, inode);
3292         if (retval)
3293                 goto end_rmdir;
3294         ext4_dec_count(dir);
3295         ext4_update_dx_flag(dir);
3296         ext4_fc_track_unlink(handle, dentry);
3297         retval = ext4_mark_inode_dirty(handle, dir);
3298
3299 #ifdef CONFIG_UNICODE
3300         /* VFS negative dentries are incompatible with Encoding and
3301          * Case-insensitiveness. Eventually we'll want avoid
3302          * invalidating the dentries here, alongside with returning the
3303          * negative dentries at ext4_lookup(), when it is better
3304          * supported by the VFS for the CI case.
3305          */
3306         if (IS_CASEFOLDED(dir))
3307                 d_invalidate(dentry);
3308 #endif
3309
3310 end_rmdir:
3311         brelse(bh);
3312         if (handle)
3313                 ext4_journal_stop(handle);
3314         return retval;
3315 }
3316
3317 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3318                   struct inode *inode)
3319 {
3320         int retval = -ENOENT;
3321         struct buffer_head *bh;
3322         struct ext4_dir_entry_2 *de;
3323         int skip_remove_dentry = 0;
3324
3325         bh = ext4_find_entry(dir, d_name, &de, NULL);
3326         if (IS_ERR(bh))
3327                 return PTR_ERR(bh);
3328
3329         if (!bh)
3330                 return -ENOENT;
3331
3332         if (le32_to_cpu(de->inode) != inode->i_ino) {
3333                 /*
3334                  * It's okay if we find dont find dentry which matches
3335                  * the inode. That's because it might have gotten
3336                  * renamed to a different inode number
3337                  */
3338                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3339                         skip_remove_dentry = 1;
3340                 else
3341                         goto out;
3342         }
3343
3344         if (IS_DIRSYNC(dir))
3345                 ext4_handle_sync(handle);
3346
3347         if (!skip_remove_dentry) {
3348                 retval = ext4_delete_entry(handle, dir, de, bh);
3349                 if (retval)
3350                         goto out;
3351                 dir->i_ctime = dir->i_mtime = current_time(dir);
3352                 ext4_update_dx_flag(dir);
3353                 retval = ext4_mark_inode_dirty(handle, dir);
3354                 if (retval)
3355                         goto out;
3356         } else {
3357                 retval = 0;
3358         }
3359         if (inode->i_nlink == 0)
3360                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3361                                    d_name->len, d_name->name);
3362         else
3363                 drop_nlink(inode);
3364         if (!inode->i_nlink)
3365                 ext4_orphan_add(handle, inode);
3366         inode->i_ctime = current_time(inode);
3367         retval = ext4_mark_inode_dirty(handle, inode);
3368
3369 out:
3370         brelse(bh);
3371         return retval;
3372 }
3373
3374 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3375 {
3376         handle_t *handle;
3377         int retval;
3378
3379         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3380                 return -EIO;
3381
3382         trace_ext4_unlink_enter(dir, dentry);
3383         /*
3384          * Initialize quotas before so that eventual writes go
3385          * in separate transaction
3386          */
3387         retval = dquot_initialize(dir);
3388         if (retval)
3389                 goto out_trace;
3390         retval = dquot_initialize(d_inode(dentry));
3391         if (retval)
3392                 goto out_trace;
3393
3394         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3395                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3396         if (IS_ERR(handle)) {
3397                 retval = PTR_ERR(handle);
3398                 goto out_trace;
3399         }
3400
3401         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3402         if (!retval)
3403                 ext4_fc_track_unlink(handle, dentry);
3404 #ifdef CONFIG_UNICODE
3405         /* VFS negative dentries are incompatible with Encoding and
3406          * Case-insensitiveness. Eventually we'll want avoid
3407          * invalidating the dentries here, alongside with returning the
3408          * negative dentries at ext4_lookup(), when it is  better
3409          * supported by the VFS for the CI case.
3410          */
3411         if (IS_CASEFOLDED(dir))
3412                 d_invalidate(dentry);
3413 #endif
3414         if (handle)
3415                 ext4_journal_stop(handle);
3416
3417 out_trace:
3418         trace_ext4_unlink_exit(dentry, retval);
3419         return retval;
3420 }
3421
3422 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3423                         struct dentry *dentry, const char *symname)
3424 {
3425         handle_t *handle;
3426         struct inode *inode;
3427         int err, len = strlen(symname);
3428         int credits;
3429         struct fscrypt_str disk_link;
3430
3431         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3432                 return -EIO;
3433
3434         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3435                                       &disk_link);
3436         if (err)
3437                 return err;
3438
3439         err = dquot_initialize(dir);
3440         if (err)
3441                 return err;
3442
3443         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3444                 /*
3445                  * For non-fast symlinks, we just allocate inode and put it on
3446                  * orphan list in the first transaction => we need bitmap,
3447                  * group descriptor, sb, inode block, quota blocks, and
3448                  * possibly selinux xattr blocks.
3449                  */
3450                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3451                           EXT4_XATTR_TRANS_BLOCKS;
3452         } else {
3453                 /*
3454                  * Fast symlink. We have to add entry to directory
3455                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3456                  * allocate new inode (bitmap, group descriptor, inode block,
3457                  * quota blocks, sb is already counted in previous macros).
3458                  */
3459                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3460                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3461         }
3462
3463         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3464                                             &dentry->d_name, 0, NULL,
3465                                             EXT4_HT_DIR, credits);
3466         handle = ext4_journal_current_handle();
3467         if (IS_ERR(inode)) {
3468                 if (handle)
3469                         ext4_journal_stop(handle);
3470                 return PTR_ERR(inode);
3471         }
3472
3473         if (IS_ENCRYPTED(inode)) {
3474                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3475                 if (err)
3476                         goto err_drop_inode;
3477                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3478         }
3479
3480         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3481                 if (!IS_ENCRYPTED(inode))
3482                         inode->i_op = &ext4_symlink_inode_operations;
3483                 inode_nohighmem(inode);
3484                 ext4_set_aops(inode);
3485                 /*
3486                  * We cannot call page_symlink() with transaction started
3487                  * because it calls into ext4_write_begin() which can wait
3488                  * for transaction commit if we are running out of space
3489                  * and thus we deadlock. So we have to stop transaction now
3490                  * and restart it when symlink contents is written.
3491                  *
3492                  * To keep fs consistent in case of crash, we have to put inode
3493                  * to orphan list in the mean time.
3494                  */
3495                 drop_nlink(inode);
3496                 err = ext4_orphan_add(handle, inode);
3497                 if (handle)
3498                         ext4_journal_stop(handle);
3499                 handle = NULL;
3500                 if (err)
3501                         goto err_drop_inode;
3502                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3503                 if (err)
3504                         goto err_drop_inode;
3505                 /*
3506                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3507                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3508                  */
3509                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3510                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3511                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3512                 if (IS_ERR(handle)) {
3513                         err = PTR_ERR(handle);
3514                         handle = NULL;
3515                         goto err_drop_inode;
3516                 }
3517                 set_nlink(inode, 1);
3518                 err = ext4_orphan_del(handle, inode);
3519                 if (err)
3520                         goto err_drop_inode;
3521         } else {
3522                 /* clear the extent format for fast symlink */
3523                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3524                 if (!IS_ENCRYPTED(inode)) {
3525                         inode->i_op = &ext4_fast_symlink_inode_operations;
3526                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3527                 }
3528                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3529                        disk_link.len);
3530                 inode->i_size = disk_link.len - 1;
3531         }
3532         EXT4_I(inode)->i_disksize = inode->i_size;
3533         err = ext4_add_nondir(handle, dentry, &inode);
3534         if (handle)
3535                 ext4_journal_stop(handle);
3536         if (inode)
3537                 iput(inode);
3538         goto out_free_encrypted_link;
3539
3540 err_drop_inode:
3541         if (handle)
3542                 ext4_journal_stop(handle);
3543         clear_nlink(inode);
3544         unlock_new_inode(inode);
3545         iput(inode);
3546 out_free_encrypted_link:
3547         if (disk_link.name != (unsigned char *)symname)
3548                 kfree(disk_link.name);
3549         return err;
3550 }
3551
3552 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3553 {
3554         handle_t *handle;
3555         int err, retries = 0;
3556 retry:
3557         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3558                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3559                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3560         if (IS_ERR(handle))
3561                 return PTR_ERR(handle);
3562
3563         if (IS_DIRSYNC(dir))
3564                 ext4_handle_sync(handle);
3565
3566         inode->i_ctime = current_time(inode);
3567         ext4_inc_count(inode);
3568         ihold(inode);
3569
3570         err = ext4_add_entry(handle, dentry, inode);
3571         if (!err) {
3572                 err = ext4_mark_inode_dirty(handle, inode);
3573                 /* this can happen only for tmpfile being
3574                  * linked the first time
3575                  */
3576                 if (inode->i_nlink == 1)
3577                         ext4_orphan_del(handle, inode);
3578                 d_instantiate(dentry, inode);
3579                 ext4_fc_track_link(handle, dentry);
3580         } else {
3581                 drop_nlink(inode);
3582                 iput(inode);
3583         }
3584         ext4_journal_stop(handle);
3585         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3586                 goto retry;
3587         return err;
3588 }
3589
3590 static int ext4_link(struct dentry *old_dentry,
3591                      struct inode *dir, struct dentry *dentry)
3592 {
3593         struct inode *inode = d_inode(old_dentry);
3594         int err;
3595
3596         if (inode->i_nlink >= EXT4_LINK_MAX)
3597                 return -EMLINK;
3598
3599         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3600         if (err)
3601                 return err;
3602
3603         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3604             (!projid_eq(EXT4_I(dir)->i_projid,
3605                         EXT4_I(old_dentry->d_inode)->i_projid)))
3606                 return -EXDEV;
3607
3608         err = dquot_initialize(dir);
3609         if (err)
3610                 return err;
3611         return __ext4_link(dir, inode, dentry);
3612 }
3613
3614 /*
3615  * Try to find buffer head where contains the parent block.
3616  * It should be the inode block if it is inlined or the 1st block
3617  * if it is a normal dir.
3618  */
3619 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3620                                         struct inode *inode,
3621                                         int *retval,
3622                                         struct ext4_dir_entry_2 **parent_de,
3623                                         int *inlined)
3624 {
3625         struct buffer_head *bh;
3626
3627         if (!ext4_has_inline_data(inode)) {
3628                 /* The first directory block must not be a hole, so
3629                  * treat it as DIRENT_HTREE
3630                  */
3631                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3632                 if (IS_ERR(bh)) {
3633                         *retval = PTR_ERR(bh);
3634                         return NULL;
3635                 }
3636                 *parent_de = ext4_next_entry(
3637                                         (struct ext4_dir_entry_2 *)bh->b_data,
3638                                         inode->i_sb->s_blocksize);
3639                 return bh;
3640         }
3641
3642         *inlined = 1;
3643         return ext4_get_first_inline_block(inode, parent_de, retval);
3644 }
3645
3646 struct ext4_renament {
3647         struct inode *dir;
3648         struct dentry *dentry;
3649         struct inode *inode;
3650         bool is_dir;
3651         int dir_nlink_delta;
3652
3653         /* entry for "dentry" */
3654         struct buffer_head *bh;
3655         struct ext4_dir_entry_2 *de;
3656         int inlined;
3657
3658         /* entry for ".." in inode if it's a directory */
3659         struct buffer_head *dir_bh;
3660         struct ext4_dir_entry_2 *parent_de;
3661         int dir_inlined;
3662 };
3663
3664 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3665 {
3666         int retval;
3667
3668         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3669                                               &retval, &ent->parent_de,
3670                                               &ent->dir_inlined);
3671         if (!ent->dir_bh)
3672                 return retval;
3673         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3674                 return -EFSCORRUPTED;
3675         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3676         return ext4_journal_get_write_access(handle, ent->dir_bh);
3677 }
3678
3679 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3680                                   unsigned dir_ino)
3681 {
3682         int retval;
3683
3684         ent->parent_de->inode = cpu_to_le32(dir_ino);
3685         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3686         if (!ent->dir_inlined) {
3687                 if (is_dx(ent->inode)) {
3688                         retval = ext4_handle_dirty_dx_node(handle,
3689                                                            ent->inode,
3690                                                            ent->dir_bh);
3691                 } else {
3692                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3693                                                             ent->dir_bh);
3694                 }
3695         } else {
3696                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3697         }
3698         if (retval) {
3699                 ext4_std_error(ent->dir->i_sb, retval);
3700                 return retval;
3701         }
3702         return 0;
3703 }
3704
3705 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3706                        unsigned ino, unsigned file_type)
3707 {
3708         int retval, retval2;
3709
3710         BUFFER_TRACE(ent->bh, "get write access");
3711         retval = ext4_journal_get_write_access(handle, ent->bh);
3712         if (retval)
3713                 return retval;
3714         ent->de->inode = cpu_to_le32(ino);
3715         if (ext4_has_feature_filetype(ent->dir->i_sb))
3716                 ent->de->file_type = file_type;
3717         inode_inc_iversion(ent->dir);
3718         ent->dir->i_ctime = ent->dir->i_mtime =
3719                 current_time(ent->dir);
3720         retval = ext4_mark_inode_dirty(handle, ent->dir);
3721         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3722         if (!ent->inlined) {
3723                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3724                 if (unlikely(retval2)) {
3725                         ext4_std_error(ent->dir->i_sb, retval2);
3726                         return retval2;
3727                 }
3728         }
3729         return retval;
3730 }
3731
3732 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3733                           unsigned ino, unsigned file_type)
3734 {
3735         struct ext4_renament old = *ent;
3736         int retval = 0;
3737
3738         /*
3739          * old->de could have moved from under us during make indexed dir,
3740          * so the old->de may no longer valid and need to find it again
3741          * before reset old inode info.
3742          */
3743         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3744         if (IS_ERR(old.bh))
3745                 retval = PTR_ERR(old.bh);
3746         if (!old.bh)
3747                 retval = -ENOENT;
3748         if (retval) {
3749                 ext4_std_error(old.dir->i_sb, retval);
3750                 return;
3751         }
3752
3753         ext4_setent(handle, &old, ino, file_type);
3754         brelse(old.bh);
3755 }
3756
3757 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3758                                   const struct qstr *d_name)
3759 {
3760         int retval = -ENOENT;
3761         struct buffer_head *bh;
3762         struct ext4_dir_entry_2 *de;
3763
3764         bh = ext4_find_entry(dir, d_name, &de, NULL);
3765         if (IS_ERR(bh))
3766                 return PTR_ERR(bh);
3767         if (bh) {
3768                 retval = ext4_delete_entry(handle, dir, de, bh);
3769                 brelse(bh);
3770         }
3771         return retval;
3772 }
3773
3774 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3775                                int force_reread)
3776 {
3777         int retval;
3778         /*
3779          * ent->de could have moved from under us during htree split, so make
3780          * sure that we are deleting the right entry.  We might also be pointing
3781          * to a stale entry in the unused part of ent->bh so just checking inum
3782          * and the name isn't enough.
3783          */
3784         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3785             ent->de->name_len != ent->dentry->d_name.len ||
3786             strncmp(ent->de->name, ent->dentry->d_name.name,
3787                     ent->de->name_len) ||
3788             force_reread) {
3789                 retval = ext4_find_delete_entry(handle, ent->dir,
3790                                                 &ent->dentry->d_name);
3791         } else {
3792                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3793                 if (retval == -ENOENT) {
3794                         retval = ext4_find_delete_entry(handle, ent->dir,
3795                                                         &ent->dentry->d_name);
3796                 }
3797         }
3798
3799         if (retval) {
3800                 ext4_warning_inode(ent->dir,
3801                                    "Deleting old file: nlink %d, error=%d",
3802                                    ent->dir->i_nlink, retval);
3803         }
3804 }
3805
3806 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3807 {
3808         if (ent->dir_nlink_delta) {
3809                 if (ent->dir_nlink_delta == -1)
3810                         ext4_dec_count(ent->dir);
3811                 else
3812                         ext4_inc_count(ent->dir);
3813                 ext4_mark_inode_dirty(handle, ent->dir);
3814         }
3815 }
3816
3817 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3818                                               struct ext4_renament *ent,
3819                                               int credits, handle_t **h)
3820 {
3821         struct inode *wh;
3822         handle_t *handle;
3823         int retries = 0;
3824
3825         /*
3826          * for inode block, sb block, group summaries,
3827          * and inode bitmap
3828          */
3829         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3830                     EXT4_XATTR_TRANS_BLOCKS + 4);
3831 retry:
3832         wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3833                                          S_IFCHR | WHITEOUT_MODE,
3834                                          &ent->dentry->d_name, 0, NULL,
3835                                          EXT4_HT_DIR, credits);
3836
3837         handle = ext4_journal_current_handle();
3838         if (IS_ERR(wh)) {
3839                 if (handle)
3840                         ext4_journal_stop(handle);
3841                 if (PTR_ERR(wh) == -ENOSPC &&
3842                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3843                         goto retry;
3844         } else {
3845                 *h = handle;
3846                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3847                 wh->i_op = &ext4_special_inode_operations;
3848         }
3849         return wh;
3850 }
3851
3852 /*
3853  * Anybody can rename anything with this: the permission checks are left to the
3854  * higher-level routines.
3855  *
3856  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3857  * while new_{dentry,inode) refers to the destination dentry/inode
3858  * This comes from rename(const char *oldpath, const char *newpath)
3859  */
3860 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3861                        struct dentry *old_dentry, struct inode *new_dir,
3862                        struct dentry *new_dentry, unsigned int flags)
3863 {
3864         handle_t *handle = NULL;
3865         struct ext4_renament old = {
3866                 .dir = old_dir,
3867                 .dentry = old_dentry,
3868                 .inode = d_inode(old_dentry),
3869         };
3870         struct ext4_renament new = {
3871                 .dir = new_dir,
3872                 .dentry = new_dentry,
3873                 .inode = d_inode(new_dentry),
3874         };
3875         int force_reread;
3876         int retval;
3877         struct inode *whiteout = NULL;
3878         int credits;
3879         u8 old_file_type;
3880
3881         if (new.inode && new.inode->i_nlink == 0) {
3882                 EXT4_ERROR_INODE(new.inode,
3883                                  "target of rename is already freed");
3884                 return -EFSCORRUPTED;
3885         }
3886
3887         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3888             (!projid_eq(EXT4_I(new_dir)->i_projid,
3889                         EXT4_I(old_dentry->d_inode)->i_projid)))
3890                 return -EXDEV;
3891
3892         retval = dquot_initialize(old.dir);
3893         if (retval)
3894                 return retval;
3895         retval = dquot_initialize(new.dir);
3896         if (retval)
3897                 return retval;
3898
3899         /* Initialize quotas before so that eventual writes go
3900          * in separate transaction */
3901         if (new.inode) {
3902                 retval = dquot_initialize(new.inode);
3903                 if (retval)
3904                         return retval;
3905         }
3906
3907         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3908         if (IS_ERR(old.bh))
3909                 return PTR_ERR(old.bh);
3910         /*
3911          *  Check for inode number is _not_ due to possible IO errors.
3912          *  We might rmdir the source, keep it as pwd of some process
3913          *  and merrily kill the link to whatever was created under the
3914          *  same name. Goodbye sticky bit ;-<
3915          */
3916         retval = -ENOENT;
3917         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3918                 goto release_bh;
3919
3920         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3921                                  &new.de, &new.inlined);
3922         if (IS_ERR(new.bh)) {
3923                 retval = PTR_ERR(new.bh);
3924                 new.bh = NULL;
3925                 goto release_bh;
3926         }
3927         if (new.bh) {
3928                 if (!new.inode) {
3929                         brelse(new.bh);
3930                         new.bh = NULL;
3931                 }
3932         }
3933         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3934                 ext4_alloc_da_blocks(old.inode);
3935
3936         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3937                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3938         if (!(flags & RENAME_WHITEOUT)) {
3939                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3940                 if (IS_ERR(handle)) {
3941                         retval = PTR_ERR(handle);
3942                         goto release_bh;
3943                 }
3944         } else {
3945                 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3946                 if (IS_ERR(whiteout)) {
3947                         retval = PTR_ERR(whiteout);
3948                         goto release_bh;
3949                 }
3950         }
3951
3952         old_file_type = old.de->file_type;
3953         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3954                 ext4_handle_sync(handle);
3955
3956         if (S_ISDIR(old.inode->i_mode)) {
3957                 if (new.inode) {
3958                         retval = -ENOTEMPTY;
3959                         if (!ext4_empty_dir(new.inode))
3960                                 goto end_rename;
3961                 } else {
3962                         retval = -EMLINK;
3963                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3964                                 goto end_rename;
3965                 }
3966                 retval = ext4_rename_dir_prepare(handle, &old);
3967                 if (retval)
3968                         goto end_rename;
3969         }
3970         /*
3971          * If we're renaming a file within an inline_data dir and adding or
3972          * setting the new dirent causes a conversion from inline_data to
3973          * extents/blockmap, we need to force the dirent delete code to
3974          * re-read the directory, or else we end up trying to delete a dirent
3975          * from what is now the extent tree root (or a block map).
3976          */
3977         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3978                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3979
3980         if (whiteout) {
3981                 /*
3982                  * Do this before adding a new entry, so the old entry is sure
3983                  * to be still pointing to the valid old entry.
3984                  */
3985                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3986                                      EXT4_FT_CHRDEV);
3987                 if (retval)
3988                         goto end_rename;
3989                 retval = ext4_mark_inode_dirty(handle, whiteout);
3990                 if (unlikely(retval))
3991                         goto end_rename;
3992
3993         }
3994         if (!new.bh) {
3995                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3996                 if (retval)
3997                         goto end_rename;
3998         } else {
3999                 retval = ext4_setent(handle, &new,
4000                                      old.inode->i_ino, old_file_type);
4001                 if (retval)
4002                         goto end_rename;
4003         }
4004         if (force_reread)
4005                 force_reread = !ext4_test_inode_flag(new.dir,
4006                                                      EXT4_INODE_INLINE_DATA);
4007
4008         /*
4009          * Like most other Unix systems, set the ctime for inodes on a
4010          * rename.
4011          */
4012         old.inode->i_ctime = current_time(old.inode);
4013         retval = ext4_mark_inode_dirty(handle, old.inode);
4014         if (unlikely(retval))
4015                 goto end_rename;
4016
4017         if (!whiteout) {
4018                 /*
4019                  * ok, that's it
4020                  */
4021                 ext4_rename_delete(handle, &old, force_reread);
4022         }
4023
4024         if (new.inode) {
4025                 ext4_dec_count(new.inode);
4026                 new.inode->i_ctime = current_time(new.inode);
4027         }
4028         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
4029         ext4_update_dx_flag(old.dir);
4030         if (old.dir_bh) {
4031                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4032                 if (retval)
4033                         goto end_rename;
4034
4035                 ext4_dec_count(old.dir);
4036                 if (new.inode) {
4037                         /* checked ext4_empty_dir above, can't have another
4038                          * parent, ext4_dec_count() won't work for many-linked
4039                          * dirs */
4040                         clear_nlink(new.inode);
4041                 } else {
4042                         ext4_inc_count(new.dir);
4043                         ext4_update_dx_flag(new.dir);
4044                         retval = ext4_mark_inode_dirty(handle, new.dir);
4045                         if (unlikely(retval))
4046                                 goto end_rename;
4047                 }
4048         }
4049         retval = ext4_mark_inode_dirty(handle, old.dir);
4050         if (unlikely(retval))
4051                 goto end_rename;
4052
4053         if (S_ISDIR(old.inode->i_mode)) {
4054                 /*
4055                  * We disable fast commits here that's because the
4056                  * replay code is not yet capable of changing dot dot
4057                  * dirents in directories.
4058                  */
4059                 ext4_fc_mark_ineligible(old.inode->i_sb,
4060                         EXT4_FC_REASON_RENAME_DIR);
4061         } else {
4062                 if (new.inode)
4063                         ext4_fc_track_unlink(handle, new.dentry);
4064                 __ext4_fc_track_link(handle, old.inode, new.dentry);
4065                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4066                 if (whiteout)
4067                         __ext4_fc_track_create(handle, whiteout, old.dentry);
4068         }
4069
4070         if (new.inode) {
4071                 retval = ext4_mark_inode_dirty(handle, new.inode);
4072                 if (unlikely(retval))
4073                         goto end_rename;
4074                 if (!new.inode->i_nlink)
4075                         ext4_orphan_add(handle, new.inode);
4076         }
4077         retval = 0;
4078
4079 end_rename:
4080         if (whiteout) {
4081                 if (retval) {
4082                         ext4_resetent(handle, &old,
4083                                       old.inode->i_ino, old_file_type);
4084                         drop_nlink(whiteout);
4085                         ext4_orphan_add(handle, whiteout);
4086                 }
4087                 unlock_new_inode(whiteout);
4088                 ext4_journal_stop(handle);
4089                 iput(whiteout);
4090         } else {
4091                 ext4_journal_stop(handle);
4092         }
4093 release_bh:
4094         brelse(old.dir_bh);
4095         brelse(old.bh);
4096         brelse(new.bh);
4097         return retval;
4098 }
4099
4100 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4101                              struct inode *new_dir, struct dentry *new_dentry)
4102 {
4103         handle_t *handle = NULL;
4104         struct ext4_renament old = {
4105                 .dir = old_dir,
4106                 .dentry = old_dentry,
4107                 .inode = d_inode(old_dentry),
4108         };
4109         struct ext4_renament new = {
4110                 .dir = new_dir,
4111                 .dentry = new_dentry,
4112                 .inode = d_inode(new_dentry),
4113         };
4114         u8 new_file_type;
4115         int retval;
4116         struct timespec64 ctime;
4117
4118         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4119              !projid_eq(EXT4_I(new_dir)->i_projid,
4120                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
4121             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4122              !projid_eq(EXT4_I(old_dir)->i_projid,
4123                         EXT4_I(new_dentry->d_inode)->i_projid)))
4124                 return -EXDEV;
4125
4126         retval = dquot_initialize(old.dir);
4127         if (retval)
4128                 return retval;
4129         retval = dquot_initialize(new.dir);
4130         if (retval)
4131                 return retval;
4132
4133         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4134                                  &old.de, &old.inlined);
4135         if (IS_ERR(old.bh))
4136                 return PTR_ERR(old.bh);
4137         /*
4138          *  Check for inode number is _not_ due to possible IO errors.
4139          *  We might rmdir the source, keep it as pwd of some process
4140          *  and merrily kill the link to whatever was created under the
4141          *  same name. Goodbye sticky bit ;-<
4142          */
4143         retval = -ENOENT;
4144         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4145                 goto end_rename;
4146
4147         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4148                                  &new.de, &new.inlined);
4149         if (IS_ERR(new.bh)) {
4150                 retval = PTR_ERR(new.bh);
4151                 new.bh = NULL;
4152                 goto end_rename;
4153         }
4154
4155         /* RENAME_EXCHANGE case: old *and* new must both exist */
4156         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4157                 goto end_rename;
4158
4159         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4160                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4161                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4162         if (IS_ERR(handle)) {
4163                 retval = PTR_ERR(handle);
4164                 handle = NULL;
4165                 goto end_rename;
4166         }
4167
4168         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4169                 ext4_handle_sync(handle);
4170
4171         if (S_ISDIR(old.inode->i_mode)) {
4172                 old.is_dir = true;
4173                 retval = ext4_rename_dir_prepare(handle, &old);
4174                 if (retval)
4175                         goto end_rename;
4176         }
4177         if (S_ISDIR(new.inode->i_mode)) {
4178                 new.is_dir = true;
4179                 retval = ext4_rename_dir_prepare(handle, &new);
4180                 if (retval)
4181                         goto end_rename;
4182         }
4183
4184         /*
4185          * Other than the special case of overwriting a directory, parents'
4186          * nlink only needs to be modified if this is a cross directory rename.
4187          */
4188         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4189                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4190                 new.dir_nlink_delta = -old.dir_nlink_delta;
4191                 retval = -EMLINK;
4192                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4193                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4194                         goto end_rename;
4195         }
4196
4197         new_file_type = new.de->file_type;
4198         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4199         if (retval)
4200                 goto end_rename;
4201
4202         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4203         if (retval)
4204                 goto end_rename;
4205
4206         /*
4207          * Like most other Unix systems, set the ctime for inodes on a
4208          * rename.
4209          */
4210         ctime = current_time(old.inode);
4211         old.inode->i_ctime = ctime;
4212         new.inode->i_ctime = ctime;
4213         retval = ext4_mark_inode_dirty(handle, old.inode);
4214         if (unlikely(retval))
4215                 goto end_rename;
4216         retval = ext4_mark_inode_dirty(handle, new.inode);
4217         if (unlikely(retval))
4218                 goto end_rename;
4219         ext4_fc_mark_ineligible(new.inode->i_sb,
4220                                 EXT4_FC_REASON_CROSS_RENAME);
4221         if (old.dir_bh) {
4222                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4223                 if (retval)
4224                         goto end_rename;
4225         }
4226         if (new.dir_bh) {
4227                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4228                 if (retval)
4229                         goto end_rename;
4230         }
4231         ext4_update_dir_count(handle, &old);
4232         ext4_update_dir_count(handle, &new);
4233         retval = 0;
4234
4235 end_rename:
4236         brelse(old.dir_bh);
4237         brelse(new.dir_bh);
4238         brelse(old.bh);
4239         brelse(new.bh);
4240         if (handle)
4241                 ext4_journal_stop(handle);
4242         return retval;
4243 }
4244
4245 static int ext4_rename2(struct user_namespace *mnt_userns,
4246                         struct inode *old_dir, struct dentry *old_dentry,
4247                         struct inode *new_dir, struct dentry *new_dentry,
4248                         unsigned int flags)
4249 {
4250         int err;
4251
4252         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4253                 return -EIO;
4254
4255         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4256                 return -EINVAL;
4257
4258         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4259                                      flags);
4260         if (err)
4261                 return err;
4262
4263         if (flags & RENAME_EXCHANGE) {
4264                 return ext4_cross_rename(old_dir, old_dentry,
4265                                          new_dir, new_dentry);
4266         }
4267
4268         return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4269 }
4270
4271 /*
4272  * directories can handle most operations...
4273  */
4274 const struct inode_operations ext4_dir_inode_operations = {
4275         .create         = ext4_create,
4276         .lookup         = ext4_lookup,
4277         .link           = ext4_link,
4278         .unlink         = ext4_unlink,
4279         .symlink        = ext4_symlink,
4280         .mkdir          = ext4_mkdir,
4281         .rmdir          = ext4_rmdir,
4282         .mknod          = ext4_mknod,
4283         .tmpfile        = ext4_tmpfile,
4284         .rename         = ext4_rename2,
4285         .setattr        = ext4_setattr,
4286         .getattr        = ext4_getattr,
4287         .listxattr      = ext4_listxattr,
4288         .get_acl        = ext4_get_acl,
4289         .set_acl        = ext4_set_acl,
4290         .fiemap         = ext4_fiemap,
4291         .fileattr_get   = ext4_fileattr_get,
4292         .fileattr_set   = ext4_fileattr_set,
4293 };
4294
4295 const struct inode_operations ext4_special_inode_operations = {
4296         .setattr        = ext4_setattr,
4297         .getattr        = ext4_getattr,
4298         .listxattr      = ext4_listxattr,
4299         .get_acl        = ext4_get_acl,
4300         .set_acl        = ext4_set_acl,
4301 };