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 ([email protected])
   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 ([email protected]), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o ([email protected]), 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(char *from, char *to,
 284                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 285 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 286 static void dx_insert_block(struct dx_frame *frame,
 287                                         u32 hash, ext4_lblk_t block);
 288 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 289                                  struct dx_frame *frame,
 290                                  struct dx_frame *frames,
 291                                  __u32 *start_hash);
 292 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 293                 struct ext4_filename *fname,
 294                 struct ext4_dir_entry_2 **res_dir);
 295 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 296                              struct inode *dir, struct inode *inode);
 297
 298 /* checksumming functions */
 299 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 300                                  unsigned int blocksize)
 301 {
 302         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 303
 304         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 305         t->det_rec_len = ext4_rec_len_to_disk(
 306                         sizeof(struct ext4_dir_entry_tail), blocksize);
 307         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 308 }
 309
 310 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 311 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 312                                                    struct buffer_head *bh)
 313 {
 314         struct ext4_dir_entry_tail *t;
 315
 316 #ifdef PARANOID
 317         struct ext4_dir_entry *d, *top;
 318
 319         d = (struct ext4_dir_entry *)bh->b_data;
 320         top = (struct ext4_dir_entry *)(bh->b_data +
 321                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 322                  sizeof(struct ext4_dir_entry_tail)));
 323         while (d < top && d->rec_len)
 324                 d = (struct ext4_dir_entry *)(((void *)d) +
 325                     le16_to_cpu(d->rec_len));
 326
 327         if (d != top)
 328                 return NULL;
 329
 330         t = (struct ext4_dir_entry_tail *)d;
 331 #else
 332         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 333 #endif
 334
 335         if (t->det_reserved_zero1 ||
 336             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 337             t->det_reserved_zero2 ||
 338             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 339                 return NULL;
 340
 341         return t;
 342 }
 343
 344 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 345 {
 346         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 347         struct ext4_inode_info *ei = EXT4_I(inode);
 348         __u32 csum;
 349
 350         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 351         return cpu_to_le32(csum);
 352 }
 353
 354 #define warn_no_space_for_csum(inode)                                   \
 355         __warn_no_space_for_csum((inode), __func__, __LINE__)
 356
 357 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 358                                      unsigned int line)
 359 {
 360         __ext4_warning_inode(inode, func, line,
 361                 "No space for directory leaf checksum. Please run e2fsck -D.");
 362 }
 363
 364 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 365 {
 366         struct ext4_dir_entry_tail *t;
 367
 368         if (!ext4_has_metadata_csum(inode->i_sb))
 369                 return 1;
 370
 371         t = get_dirent_tail(inode, bh);
 372         if (!t) {
 373                 warn_no_space_for_csum(inode);
 374                 return 0;
 375         }
 376
 377         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 378                                                   (char *)t - bh->b_data))
 379                 return 0;
 380
 381         return 1;
 382 }
 383
 384 static void ext4_dirblock_csum_set(struct inode *inode,
 385                                  struct buffer_head *bh)
 386 {
 387         struct ext4_dir_entry_tail *t;
 388
 389         if (!ext4_has_metadata_csum(inode->i_sb))
 390                 return;
 391
 392         t = get_dirent_tail(inode, bh);
 393         if (!t) {
 394                 warn_no_space_for_csum(inode);
 395                 return;
 396         }
 397
 398         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 399                                              (char *)t - bh->b_data);
 400 }
 401
 402 int ext4_handle_dirty_dirblock(handle_t *handle,
 403                                struct inode *inode,
 404                                struct buffer_head *bh)
 405 {
 406         ext4_dirblock_csum_set(inode, bh);
 407         return ext4_handle_dirty_metadata(handle, inode, bh);
 408 }
 409
 410 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 411                                                struct ext4_dir_entry *dirent,
 412                                                int *offset)
 413 {
 414         struct ext4_dir_entry *dp;
 415         struct dx_root_info *root;
 416         int count_offset;
 417
 418         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 419                 count_offset = 8;
 420         else if (le16_to_cpu(dirent->rec_len) == 12) {
 421                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 422                 if (le16_to_cpu(dp->rec_len) !=
 423                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 424                         return NULL;
 425                 root = (struct dx_root_info *)(((void *)dp + 12));
 426                 if (root->reserved_zero ||
 427                     root->info_length != sizeof(struct dx_root_info))
 428                         return NULL;
 429                 count_offset = 32;
 430         } else
 431                 return NULL;
 432
 433         if (offset)
 434                 *offset = count_offset;
 435         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 436 }
 437
 438 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 439                            int count_offset, int count, struct dx_tail *t)
 440 {
 441         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 442         struct ext4_inode_info *ei = EXT4_I(inode);
 443         __u32 csum;
 444         int size;
 445         __u32 dummy_csum = 0;
 446         int offset = offsetof(struct dx_tail, dt_checksum);
 447
 448         size = count_offset + (count * sizeof(struct dx_entry));
 449         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 450         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 451         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 452
 453         return cpu_to_le32(csum);
 454 }
 455
 456 static int ext4_dx_csum_verify(struct inode *inode,
 457                                struct ext4_dir_entry *dirent)
 458 {
 459         struct dx_countlimit *c;
 460         struct dx_tail *t;
 461         int count_offset, limit, count;
 462
 463         if (!ext4_has_metadata_csum(inode->i_sb))
 464                 return 1;
 465
 466         c = get_dx_countlimit(inode, dirent, &count_offset);
 467         if (!c) {
 468                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 469                 return 0;
 470         }
 471         limit = le16_to_cpu(c->limit);
 472         count = le16_to_cpu(c->count);
 473         if (count_offset + (limit * sizeof(struct dx_entry)) >
 474             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 475                 warn_no_space_for_csum(inode);
 476                 return 0;
 477         }
 478         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 479
 480         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 481                                             count, t))
 482                 return 0;
 483         return 1;
 484 }
 485
 486 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 487 {
 488         struct dx_countlimit *c;
 489         struct dx_tail *t;
 490         int count_offset, limit, count;
 491
 492         if (!ext4_has_metadata_csum(inode->i_sb))
 493                 return;
 494
 495         c = get_dx_countlimit(inode, dirent, &count_offset);
 496         if (!c) {
 497                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 498                 return;
 499         }
 500         limit = le16_to_cpu(c->limit);
 501         count = le16_to_cpu(c->count);
 502         if (count_offset + (limit * sizeof(struct dx_entry)) >
 503             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 504                 warn_no_space_for_csum(inode);
 505                 return;
 506         }
 507         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 508
 509         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 510 }
 511
 512 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 513                                             struct inode *inode,
 514                                             struct buffer_head *bh)
 515 {
 516         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 517         return ext4_handle_dirty_metadata(handle, inode, bh);
 518 }
 519
 520 /*
 521  * p is at least 6 bytes before the end of page
 522  */
 523 static inline struct ext4_dir_entry_2 *
 524 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 525 {
 526         return (struct ext4_dir_entry_2 *)((char *)p +
 527                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 528 }
 529
 530 /*
 531  * Future: use high four bits of block for coalesce-on-delete flags
 532  * Mask them off for now.
 533  */
 534
 535 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 536 {
 537         return le32_to_cpu(entry->block) & 0x0fffffff;
 538 }
 539
 540 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 541 {
 542         entry->block = cpu_to_le32(value);
 543 }
 544
 545 static inline unsigned dx_get_hash(struct dx_entry *entry)
 546 {
 547         return le32_to_cpu(entry->hash);
 548 }
 549
 550 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 551 {
 552         entry->hash = cpu_to_le32(value);
 553 }
 554
 555 static inline unsigned dx_get_count(struct dx_entry *entries)
 556 {
 557         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 558 }
 559
 560 static inline unsigned dx_get_limit(struct dx_entry *entries)
 561 {
 562         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 563 }
 564
 565 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 566 {
 567         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 568 }
 569
 570 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 571 {
 572         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 573 }
 574
 575 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 576 {
 577         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 578                 EXT4_DIR_REC_LEN(2) - infosize;
 579
 580         if (ext4_has_metadata_csum(dir->i_sb))
 581                 entry_space -= sizeof(struct dx_tail);
 582         return entry_space / sizeof(struct dx_entry);
 583 }
 584
 585 static inline unsigned dx_node_limit(struct inode *dir)
 586 {
 587         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 588
 589         if (ext4_has_metadata_csum(dir->i_sb))
 590                 entry_space -= sizeof(struct dx_tail);
 591         return entry_space / sizeof(struct dx_entry);
 592 }
 593
 594 /*
 595  * Debug
 596  */
 597 #ifdef DX_DEBUG
 598 static void dx_show_index(char * label, struct dx_entry *entries)
 599 {
 600         int i, n = dx_get_count (entries);
 601         printk(KERN_DEBUG "%s index", label);
 602         for (i = 0; i < n; i++) {
 603                 printk(KERN_CONT " %x->%lu",
 604                        i ? dx_get_hash(entries + i) : 0,
 605                        (unsigned long)dx_get_block(entries + i));
 606         }
 607         printk(KERN_CONT "\n");
 608 }
 609
 610 struct stats
 611 {
 612         unsigned names;
 613         unsigned space;
 614         unsigned bcount;
 615 };
 616
 617 static struct stats dx_show_leaf(struct inode *dir,
 618                                 struct dx_hash_info *hinfo,
 619                                 struct ext4_dir_entry_2 *de,
 620                                 int size, int show_names)
 621 {
 622         unsigned names = 0, space = 0;
 623         char *base = (char *) de;
 624         struct dx_hash_info h = *hinfo;
 625
 626         printk("names: ");
 627         while ((char *) de < base + size)
 628         {
 629                 if (de->inode)
 630                 {
 631                         if (show_names)
 632                         {
 633 #ifdef CONFIG_FS_ENCRYPTION
 634                                 int len;
 635                                 char *name;
 636                                 struct fscrypt_str fname_crypto_str =
 637                                         FSTR_INIT(NULL, 0);
 638                                 int res = 0;
 639
 640                                 name  = de->name;
 641                                 len = de->name_len;
 642                                 if (IS_ENCRYPTED(dir))
 643                                         res = fscrypt_get_encryption_info(dir);
 644                                 if (res) {
 645                                         printk(KERN_WARNING "Error setting up"
 646                                                " fname crypto: %d\n", res);
 647                                 }
 648                                 if (!fscrypt_has_encryption_key(dir)) {
 649                                         /* Directory is not encrypted */
 650                                         ext4fs_dirhash(dir, de->name,
 651                                                 de->name_len, &h);
 652                                         printk("%*.s:(U)%x.%u ", len,
 653                                                name, h.hash,
 654                                                (unsigned) ((char *) de
 655                                                            - base));
 656                                 } else {
 657                                         struct fscrypt_str de_name =
 658                                                 FSTR_INIT(name, len);
 659
 660                                         /* Directory is encrypted */
 661                                         res = fscrypt_fname_alloc_buffer(
 662                                                 len, &fname_crypto_str);
 663                                         if (res)
 664                                                 printk(KERN_WARNING "Error "
 665                                                         "allocating crypto "
 666                                                         "buffer--skipping "
 667                                                         "crypto\n");
 668                                         res = fscrypt_fname_disk_to_usr(dir,
 669                                                 0, 0, &de_name,
 670                                                 &fname_crypto_str);
 671                                         if (res) {
 672                                                 printk(KERN_WARNING "Error "
 673                                                         "converting filename "
 674                                                         "from disk to usr"
 675                                                         "\n");
 676                                                 name = "??";
 677                                                 len = 2;
 678                                         } else {
 679                                                 name = fname_crypto_str.name;
 680                                                 len = fname_crypto_str.len;
 681                                         }
 682                                         ext4fs_dirhash(dir, de->name,
 683                                                        de->name_len, &h);
 684                                         printk("%*.s:(E)%x.%u ", len, name,
 685                                                h.hash, (unsigned) ((char *) de
 686                                                                    - base));
 687                                         fscrypt_fname_free_buffer(
 688                                                         &fname_crypto_str);
 689                                 }
 690 #else
 691                                 int len = de->name_len;
 692                                 char *name = de->name;
 693                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 694                                 printk("%*.s:%x.%u ", len, name, h.hash,
 695                                        (unsigned) ((char *) de - base));
 696 #endif
 697                         }
 698                         space += EXT4_DIR_REC_LEN(de->name_len);
 699                         names++;
 700                 }
 701                 de = ext4_next_entry(de, size);
 702         }
 703         printk(KERN_CONT "(%i)\n", names);
 704         return (struct stats) { names, space, 1 };
 705 }
 706
 707 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 708                              struct dx_entry *entries, int levels)
 709 {
 710         unsigned blocksize = dir->i_sb->s_blocksize;
 711         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 712         unsigned bcount = 0;
 713         struct buffer_head *bh;
 714         printk("%i indexed blocks...\n", count);
 715         for (i = 0; i < count; i++, entries++)
 716         {
 717                 ext4_lblk_t block = dx_get_block(entries);
 718                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 719                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 720                 struct stats stats;
 721                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 722                 bh = ext4_bread(NULL,dir, block, 0);
 723                 if (!bh || IS_ERR(bh))
 724                         continue;
 725                 stats = levels?
 726                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 727                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 728                         bh->b_data, blocksize, 0);
 729                 names += stats.names;
 730                 space += stats.space;
 731                 bcount += stats.bcount;
 732                 brelse(bh);
 733         }
 734         if (bcount)
 735                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 736                        levels ? "" : "   ", names, space/bcount,
 737                        (space/bcount)*100/blocksize);
 738         return (struct stats) { names, space, bcount};
 739 }
 740 #endif /* DX_DEBUG */
 741
 742 /*
 743  * Probe for a directory leaf block to search.
 744  *
 745  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 746  * error in the directory index, and the caller should fall back to
 747  * searching the directory normally.  The callers of dx_probe **MUST**
 748  * check for this error code, and make sure it never gets reflected
 749  * back to userspace.
 750  */
 751 static struct dx_frame *
 752 dx_probe(struct ext4_filename *fname, struct inode *dir,
 753          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 754 {
 755         unsigned count, indirect;
 756         struct dx_entry *at, *entries, *p, *q, *m;
 757         struct dx_root *root;
 758         struct dx_frame *frame = frame_in;
 759         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 760         u32 hash;
 761
 762         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 763         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 764         if (IS_ERR(frame->bh))
 765                 return (struct dx_frame *) frame->bh;
 766
 767         root = (struct dx_root *) frame->bh->b_data;
 768         if (root->info.hash_version != DX_HASH_TEA &&
 769             root->info.hash_version != DX_HASH_HALF_MD4 &&
 770             root->info.hash_version != DX_HASH_LEGACY) {
 771                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 772                                    root->info.hash_version);
 773                 goto fail;
 774         }
 775         if (fname)
 776                 hinfo = &fname->hinfo;
 777         hinfo->hash_version = root->info.hash_version;
 778         if (hinfo->hash_version <= DX_HASH_TEA)
 779                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 780         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 781         if (fname && fname_name(fname))
 782                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 783         hash = hinfo->hash;
 784
 785         if (root->info.unused_flags & 1) {
 786                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 787                                    root->info.unused_flags);
 788                 goto fail;
 789         }
 790
 791         indirect = root->info.indirect_levels;
 792         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 793                 ext4_warning(dir->i_sb,
 794                              "Directory (ino: %lu) htree depth %#06x exceed"
 795                              "supported value", dir->i_ino,
 796                              ext4_dir_htree_level(dir->i_sb));
 797                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 798                         ext4_warning(dir->i_sb, "Enable large directory "
 799                                                 "feature to access it");
 800                 }
 801                 goto fail;
 802         }
 803
 804         entries = (struct dx_entry *)(((char *)&root->info) +
 805                                       root->info.info_length);
 806
 807         if (dx_get_limit(entries) != dx_root_limit(dir,
 808                                                    root->info.info_length)) {
 809                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 810                                    dx_get_limit(entries),
 811                                    dx_root_limit(dir, root->info.info_length));
 812                 goto fail;
 813         }
 814
 815         dxtrace(printk("Look up %x", hash));
 816         while (1) {
 817                 count = dx_get_count(entries);
 818                 if (!count || count > dx_get_limit(entries)) {
 819                         ext4_warning_inode(dir,
 820                                            "dx entry: count %u beyond limit %u",
 821                                            count, dx_get_limit(entries));
 822                         goto fail;
 823                 }
 824
 825                 p = entries + 1;
 826                 q = entries + count - 1;
 827                 while (p <= q) {
 828                         m = p + (q - p) / 2;
 829                         dxtrace(printk(KERN_CONT "."));
 830                         if (dx_get_hash(m) > hash)
 831                                 q = m - 1;
 832                         else
 833                                 p = m + 1;
 834                 }
 835
 836                 if (0) { // linear search cross check
 837                         unsigned n = count - 1;
 838                         at = entries;
 839                         while (n--)
 840                         {
 841                                 dxtrace(printk(KERN_CONT ","));
 842                                 if (dx_get_hash(++at) > hash)
 843                                 {
 844                                         at--;
 845                                         break;
 846                                 }
 847                         }
 848                         ASSERT(at == p - 1);
 849                 }
 850
 851                 at = p - 1;
 852                 dxtrace(printk(KERN_CONT " %x->%u\n",
 853                                at == entries ? 0 : dx_get_hash(at),
 854                                dx_get_block(at)));
 855                 frame->entries = entries;
 856                 frame->at = at;
 857                 if (!indirect--)
 858                         return frame;
 859                 frame++;
 860                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 861                 if (IS_ERR(frame->bh)) {
 862                         ret_err = (struct dx_frame *) frame->bh;
 863                         frame->bh = NULL;
 864                         goto fail;
 865                 }
 866                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 867
 868                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 869                         ext4_warning_inode(dir,
 870                                 "dx entry: limit %u != node limit %u",
 871                                 dx_get_limit(entries), dx_node_limit(dir));
 872                         goto fail;
 873                 }
 874         }
 875 fail:
 876         while (frame >= frame_in) {
 877                 brelse(frame->bh);
 878                 frame--;
 879         }
 880
 881         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 882                 ext4_warning_inode(dir,
 883                         "Corrupt directory, running e2fsck is recommended");
 884         return ret_err;
 885 }
 886
 887 static void dx_release(struct dx_frame *frames)
 888 {
 889         struct dx_root_info *info;
 890         int i;
 891         unsigned int indirect_levels;
 892
 893         if (frames[0].bh == NULL)
 894                 return;
 895
 896         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 897         /* save local copy, "info" may be freed after brelse() */
 898         indirect_levels = info->indirect_levels;
 899         for (i = 0; i <= indirect_levels; i++) {
 900                 if (frames[i].bh == NULL)
 901                         break;
 902                 brelse(frames[i].bh);
 903                 frames[i].bh = NULL;
 904         }
 905 }
 906
 907 /*
 908  * This function increments the frame pointer to search the next leaf
 909  * block, and reads in the necessary intervening nodes if the search
 910  * should be necessary.  Whether or not the search is necessary is
 911  * controlled by the hash parameter.  If the hash value is even, then
 912  * the search is only continued if the next block starts with that
 913  * hash value.  This is used if we are searching for a specific file.
 914  *
 915  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 916  *
 917  * This function returns 1 if the caller should continue to search,
 918  * or 0 if it should not.  If there is an error reading one of the
 919  * index blocks, it will a negative error code.
 920  *
 921  * If start_hash is non-null, it will be filled in with the starting
 922  * hash of the next page.
 923  */
 924 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 925                                  struct dx_frame *frame,
 926                                  struct dx_frame *frames,
 927                                  __u32 *start_hash)
 928 {
 929         struct dx_frame *p;
 930         struct buffer_head *bh;
 931         int num_frames = 0;
 932         __u32 bhash;
 933
 934         p = frame;
 935         /*
 936          * Find the next leaf page by incrementing the frame pointer.
 937          * If we run out of entries in the interior node, loop around and
 938          * increment pointer in the parent node.  When we break out of
 939          * this loop, num_frames indicates the number of interior
 940          * nodes need to be read.
 941          */
 942         while (1) {
 943                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 944                         break;
 945                 if (p == frames)
 946                         return 0;
 947                 num_frames++;
 948                 p--;
 949         }
 950
 951         /*
 952          * If the hash is 1, then continue only if the next page has a
 953          * continuation hash of any value.  This is used for readdir
 954          * handling.  Otherwise, check to see if the hash matches the
 955          * desired contiuation hash.  If it doesn't, return since
 956          * there's no point to read in the successive index pages.
 957          */
 958         bhash = dx_get_hash(p->at);
 959         if (start_hash)
 960                 *start_hash = bhash;
 961         if ((hash & 1) == 0) {
 962                 if ((bhash & ~1) != hash)
 963                         return 0;
 964         }
 965         /*
 966          * If the hash is HASH_NB_ALWAYS, we always go to the next
 967          * block so no check is necessary
 968          */
 969         while (num_frames--) {
 970                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 971                 if (IS_ERR(bh))
 972                         return PTR_ERR(bh);
 973                 p++;
 974                 brelse(p->bh);
 975                 p->bh = bh;
 976                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 977         }
 978         return 1;
 979 }
 980
 981
 982 /*
 983  * This function fills a red-black tree with information from a
 984  * directory block.  It returns the number directory entries loaded
 985  * into the tree.  If there is an error it is returned in err.
 986  */
 987 static int htree_dirblock_to_tree(struct file *dir_file,
 988                                   struct inode *dir, ext4_lblk_t block,
 989                                   struct dx_hash_info *hinfo,
 990                                   __u32 start_hash, __u32 start_minor_hash)
 991 {
 992         struct buffer_head *bh;
 993         struct ext4_dir_entry_2 *de, *top;
 994         int err = 0, count = 0;
 995         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 996
 997         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 998                                                         (unsigned long)block));
 999         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1000         if (IS_ERR(bh))
1001                 return PTR_ERR(bh);
1002
1003         de = (struct ext4_dir_entry_2 *) bh->b_data;
1004         top = (struct ext4_dir_entry_2 *) ((char *) de +
1005                                            dir->i_sb->s_blocksize -
1006                                            EXT4_DIR_REC_LEN(0));
1007         /* Check if the directory is encrypted */
1008         if (IS_ENCRYPTED(dir)) {
1009                 err = fscrypt_get_encryption_info(dir);
1010                 if (err < 0) {
1011                         brelse(bh);
1012                         return err;
1013                 }
1014                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1015                                                  &fname_crypto_str);
1016                 if (err < 0) {
1017                         brelse(bh);
1018                         return err;
1019                 }
1020         }
1021
1022         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1023                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1024                                 bh->b_data, bh->b_size,
1025                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1026                                          + ((char *)de - bh->b_data))) {
1027                         /* silently ignore the rest of the block */
1028                         break;
1029                 }
1030                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1031                 if ((hinfo->hash < start_hash) ||
1032                     ((hinfo->hash == start_hash) &&
1033                      (hinfo->minor_hash < start_minor_hash)))
1034                         continue;
1035                 if (de->inode == 0)
1036                         continue;
1037                 if (!IS_ENCRYPTED(dir)) {
1038                         tmp_str.name = de->name;
1039                         tmp_str.len = de->name_len;
1040                         err = ext4_htree_store_dirent(dir_file,
1041                                    hinfo->hash, hinfo->minor_hash, de,
1042                                    &tmp_str);
1043                 } else {
1044                         int save_len = fname_crypto_str.len;
1045                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1046                                                                 de->name_len);
1047
1048                         /* Directory is encrypted */
1049                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1050                                         hinfo->minor_hash, &de_name,
1051                                         &fname_crypto_str);
1052                         if (err) {
1053                                 count = err;
1054                                 goto errout;
1055                         }
1056                         err = ext4_htree_store_dirent(dir_file,
1057                                    hinfo->hash, hinfo->minor_hash, de,
1058                                         &fname_crypto_str);
1059                         fname_crypto_str.len = save_len;
1060                 }
1061                 if (err != 0) {
1062                         count = err;
1063                         goto errout;
1064                 }
1065                 count++;
1066         }
1067 errout:
1068         brelse(bh);
1069         fscrypt_fname_free_buffer(&fname_crypto_str);
1070         return count;
1071 }
1072
1073
1074 /*
1075  * This function fills a red-black tree with information from a
1076  * directory.  We start scanning the directory in hash order, starting
1077  * at start_hash and start_minor_hash.
1078  *
1079  * This function returns the number of entries inserted into the tree,
1080  * or a negative error code.
1081  */
1082 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1083                          __u32 start_minor_hash, __u32 *next_hash)
1084 {
1085         struct dx_hash_info hinfo;
1086         struct ext4_dir_entry_2 *de;
1087         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1088         struct inode *dir;
1089         ext4_lblk_t block;
1090         int count = 0;
1091         int ret, err;
1092         __u32 hashval;
1093         struct fscrypt_str tmp_str;
1094
1095         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1096                        start_hash, start_minor_hash));
1097         dir = file_inode(dir_file);
1098         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1099                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1100                 if (hinfo.hash_version <= DX_HASH_TEA)
1101                         hinfo.hash_version +=
1102                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1103                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1104                 if (ext4_has_inline_data(dir)) {
1105                         int has_inline_data = 1;
1106                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1107                                                        &hinfo, start_hash,
1108                                                        start_minor_hash,
1109                                                        &has_inline_data);
1110                         if (has_inline_data) {
1111                                 *next_hash = ~0;
1112                                 return count;
1113                         }
1114                 }
1115                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1116                                                start_hash, start_minor_hash);
1117                 *next_hash = ~0;
1118                 return count;
1119         }
1120         hinfo.hash = start_hash;
1121         hinfo.minor_hash = 0;
1122         frame = dx_probe(NULL, dir, &hinfo, frames);
1123         if (IS_ERR(frame))
1124                 return PTR_ERR(frame);
1125
1126         /* Add '.' and '..' from the htree header */
1127         if (!start_hash && !start_minor_hash) {
1128                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1129                 tmp_str.name = de->name;
1130                 tmp_str.len = de->name_len;
1131                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1132                                               de, &tmp_str);
1133                 if (err != 0)
1134                         goto errout;
1135                 count++;
1136         }
1137         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1138                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1139                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1140                 tmp_str.name = de->name;
1141                 tmp_str.len = de->name_len;
1142                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1143                                               de, &tmp_str);
1144                 if (err != 0)
1145                         goto errout;
1146                 count++;
1147         }
1148
1149         while (1) {
1150                 if (fatal_signal_pending(current)) {
1151                         err = -ERESTARTSYS;
1152                         goto errout;
1153                 }
1154                 cond_resched();
1155                 block = dx_get_block(frame->at);
1156                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1157                                              start_hash, start_minor_hash);
1158                 if (ret < 0) {
1159                         err = ret;
1160                         goto errout;
1161                 }
1162                 count += ret;
1163                 hashval = ~0;
1164                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1165                                             frame, frames, &hashval);
1166                 *next_hash = hashval;
1167                 if (ret < 0) {
1168                         err = ret;
1169                         goto errout;
1170                 }
1171                 /*
1172                  * Stop if:  (a) there are no more entries, or
1173                  * (b) we have inserted at least one entry and the
1174                  * next hash value is not a continuation
1175                  */
1176                 if ((ret == 0) ||
1177                     (count && ((hashval & 1) == 0)))
1178                         break;
1179         }
1180         dx_release(frames);
1181         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1182                        "next hash: %x\n", count, *next_hash));
1183         return count;
1184 errout:
1185         dx_release(frames);
1186         return (err);
1187 }
1188
1189 static inline int search_dirblock(struct buffer_head *bh,
1190                                   struct inode *dir,
1191                                   struct ext4_filename *fname,
1192                                   unsigned int offset,
1193                                   struct ext4_dir_entry_2 **res_dir)
1194 {
1195         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1196                                fname, offset, res_dir);
1197 }
1198
1199 /*
1200  * Directory block splitting, compacting
1201  */
1202
1203 /*
1204  * Create map of hash values, offsets, and sizes, stored at end of block.
1205  * Returns number of entries mapped.
1206  */
1207 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1208                        unsigned blocksize, struct dx_hash_info *hinfo,
1209                        struct dx_map_entry *map_tail)
1210 {
1211         int count = 0;
1212         char *base = (char *) de;
1213         struct dx_hash_info h = *hinfo;
1214
1215         while ((char *) de < base + blocksize) {
1216                 if (de->name_len && de->inode) {
1217                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1218                         map_tail--;
1219                         map_tail->hash = h.hash;
1220                         map_tail->offs = ((char *) de - base)>>2;
1221                         map_tail->size = le16_to_cpu(de->rec_len);
1222                         count++;
1223                         cond_resched();
1224                 }
1225                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1226                 de = ext4_next_entry(de, blocksize);
1227         }
1228         return count;
1229 }
1230
1231 /* Sort map by hash value */
1232 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1233 {
1234         struct dx_map_entry *p, *q, *top = map + count - 1;
1235         int more;
1236         /* Combsort until bubble sort doesn't suck */
1237         while (count > 2) {
1238                 count = count*10/13;
1239                 if (count - 9 < 2) /* 9, 10 -> 11 */
1240                         count = 11;
1241                 for (p = top, q = p - count; q >= map; p--, q--)
1242                         if (p->hash < q->hash)
1243                                 swap(*p, *q);
1244         }
1245         /* Garden variety bubble sort */
1246         do {
1247                 more = 0;
1248                 q = top;
1249                 while (q-- > map) {
1250                         if (q[1].hash >= q[0].hash)
1251                                 continue;
1252                         swap(*(q+1), *q);
1253                         more = 1;
1254                 }
1255         } while(more);
1256 }
1257
1258 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1259 {
1260         struct dx_entry *entries = frame->entries;
1261         struct dx_entry *old = frame->at, *new = old + 1;
1262         int count = dx_get_count(entries);
1263
1264         ASSERT(count < dx_get_limit(entries));
1265         ASSERT(old < entries + count);
1266         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1267         dx_set_hash(new, hash);
1268         dx_set_block(new, block);
1269         dx_set_count(entries, count + 1);
1270 }
1271
1272 #ifdef CONFIG_UNICODE
1273 /*
1274  * Test whether a case-insensitive directory entry matches the filename
1275  * being searched for.  If quick is set, assume the name being looked up
1276  * is already in the casefolded form.
1277  *
1278  * Returns: 0 if the directory entry matches, more than 0 if it
1279  * doesn't match or less than zero on error.
1280  */
1281 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1282                     const struct qstr *entry, bool quick)
1283 {
1284         const struct super_block *sb = parent->i_sb;
1285         const struct unicode_map *um = sb->s_encoding;
1286         int ret;
1287
1288         if (quick)
1289                 ret = utf8_strncasecmp_folded(um, name, entry);
1290         else
1291                 ret = utf8_strncasecmp(um, name, entry);
1292
1293         if (ret < 0) {
1294                 /* Handle invalid character sequence as either an error
1295                  * or as an opaque byte sequence.
1296                  */
1297                 if (sb_has_strict_encoding(sb))
1298                         return -EINVAL;
1299
1300                 if (name->len != entry->len)
1301                         return 1;
1302
1303                 return !!memcmp(name->name, entry->name, name->len);
1304         }
1305
1306         return ret;
1307 }
1308
1309 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1310                                   struct fscrypt_str *cf_name)
1311 {
1312         int len;
1313
1314         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1315                 cf_name->name = NULL;
1316                 return;
1317         }
1318
1319         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1320         if (!cf_name->name)
1321                 return;
1322
1323         len = utf8_casefold(dir->i_sb->s_encoding,
1324                             iname, cf_name->name,
1325                             EXT4_NAME_LEN);
1326         if (len <= 0) {
1327                 kfree(cf_name->name);
1328                 cf_name->name = NULL;
1329                 return;
1330         }
1331         cf_name->len = (unsigned) len;
1332
1333 }
1334 #endif
1335
1336 /*
1337  * Test whether a directory entry matches the filename being searched for.
1338  *
1339  * Return: %true if the directory entry matches, otherwise %false.
1340  */
1341 static inline bool ext4_match(const struct inode *parent,
1342                               const struct ext4_filename *fname,
1343                               const struct ext4_dir_entry_2 *de)
1344 {
1345         struct fscrypt_name f;
1346 #ifdef CONFIG_UNICODE
1347         const struct qstr entry = {.name = de->name, .len = de->name_len};
1348 #endif
1349
1350         if (!de->inode)
1351                 return false;
1352
1353         f.usr_fname = fname->usr_fname;
1354         f.disk_name = fname->disk_name;
1355 #ifdef CONFIG_FS_ENCRYPTION
1356         f.crypto_buf = fname->crypto_buf;
1357 #endif
1358
1359 #ifdef CONFIG_UNICODE
1360         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1361                 if (fname->cf_name.name) {
1362                         struct qstr cf = {.name = fname->cf_name.name,
1363                                           .len = fname->cf_name.len};
1364                         return !ext4_ci_compare(parent, &cf, &entry, true);
1365                 }
1366                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1367                                         false);
1368         }
1369 #endif
1370
1371         return fscrypt_match_name(&f, de->name, de->name_len);
1372 }
1373
1374 /*
1375  * Returns 0 if not found, -1 on failure, and 1 on success
1376  */
1377 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1378                     struct inode *dir, struct ext4_filename *fname,
1379                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1380 {
1381         struct ext4_dir_entry_2 * de;
1382         char * dlimit;
1383         int de_len;
1384
1385         de = (struct ext4_dir_entry_2 *)search_buf;
1386         dlimit = search_buf + buf_size;
1387         while ((char *) de < dlimit) {
1388                 /* this code is executed quadratically often */
1389                 /* do minimal checking `by hand' */
1390                 if ((char *) de + de->name_len <= dlimit &&
1391                     ext4_match(dir, fname, de)) {
1392                         /* found a match - just to be sure, do
1393                          * a full check */
1394                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1395                                                  buf_size, offset))
1396                                 return -1;
1397                         *res_dir = de;
1398                         return 1;
1399                 }
1400                 /* prevent looping on a bad block */
1401                 de_len = ext4_rec_len_from_disk(de->rec_len,
1402                                                 dir->i_sb->s_blocksize);
1403                 if (de_len <= 0)
1404                         return -1;
1405                 offset += de_len;
1406                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1407         }
1408         return 0;
1409 }
1410
1411 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1412                                struct ext4_dir_entry *de)
1413 {
1414         struct super_block *sb = dir->i_sb;
1415
1416         if (!is_dx(dir))
1417                 return 0;
1418         if (block == 0)
1419                 return 1;
1420         if (de->inode == 0 &&
1421             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1422                         sb->s_blocksize)
1423                 return 1;
1424         return 0;
1425 }
1426
1427 /*
1428  *      __ext4_find_entry()
1429  *
1430  * finds an entry in the specified directory with the wanted name. It
1431  * returns the cache buffer in which the entry was found, and the entry
1432  * itself (as a parameter - res_dir). It does NOT read the inode of the
1433  * entry - you'll have to do that yourself if you want to.
1434  *
1435  * The returned buffer_head has ->b_count elevated.  The caller is expected
1436  * to brelse() it when appropriate.
1437  */
1438 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1439                                              struct ext4_filename *fname,
1440                                              struct ext4_dir_entry_2 **res_dir,
1441                                              int *inlined)
1442 {
1443         struct super_block *sb;
1444         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1445         struct buffer_head *bh, *ret = NULL;
1446         ext4_lblk_t start, block;
1447         const u8 *name = fname->usr_fname->name;
1448         size_t ra_max = 0;      /* Number of bh's in the readahead
1449                                    buffer, bh_use[] */
1450         size_t ra_ptr = 0;      /* Current index into readahead
1451                                    buffer */
1452         ext4_lblk_t  nblocks;
1453         int i, namelen, retval;
1454
1455         *res_dir = NULL;
1456         sb = dir->i_sb;
1457         namelen = fname->usr_fname->len;
1458         if (namelen > EXT4_NAME_LEN)
1459                 return NULL;
1460
1461         if (ext4_has_inline_data(dir)) {
1462                 int has_inline_data = 1;
1463                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1464                                              &has_inline_data);
1465                 if (has_inline_data) {
1466                         if (inlined)
1467                                 *inlined = 1;
1468                         goto cleanup_and_exit;
1469                 }
1470         }
1471
1472         if ((namelen <= 2) && (name[0] == '.') &&
1473             (name[1] == '.' || name[1] == '\0')) {
1474                 /*
1475                  * "." or ".." will only be in the first block
1476                  * NFS may look up ".."; "." should be handled by the VFS
1477                  */
1478                 block = start = 0;
1479                 nblocks = 1;
1480                 goto restart;
1481         }
1482         if (is_dx(dir)) {
1483                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1484                 /*
1485                  * On success, or if the error was file not found,
1486                  * return.  Otherwise, fall back to doing a search the
1487                  * old fashioned way.
1488                  */
1489                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1490                         goto cleanup_and_exit;
1491                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1492                                "falling back\n"));
1493                 ret = NULL;
1494         }
1495         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1496         if (!nblocks) {
1497                 ret = NULL;
1498                 goto cleanup_and_exit;
1499         }
1500         start = EXT4_I(dir)->i_dir_start_lookup;
1501         if (start >= nblocks)
1502                 start = 0;
1503         block = start;
1504 restart:
1505         do {
1506                 /*
1507                  * We deal with the read-ahead logic here.
1508                  */
1509                 cond_resched();
1510                 if (ra_ptr >= ra_max) {
1511                         /* Refill the readahead buffer */
1512                         ra_ptr = 0;
1513                         if (block < start)
1514                                 ra_max = start - block;
1515                         else
1516                                 ra_max = nblocks - block;
1517                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1518                         retval = ext4_bread_batch(dir, block, ra_max,
1519                                                   false /* wait */, bh_use);
1520                         if (retval) {
1521                                 ret = ERR_PTR(retval);
1522                                 ra_max = 0;
1523                                 goto cleanup_and_exit;
1524                         }
1525                 }
1526                 if ((bh = bh_use[ra_ptr++]) == NULL)
1527                         goto next;
1528                 wait_on_buffer(bh);
1529                 if (!buffer_uptodate(bh)) {
1530                         EXT4_ERROR_INODE_ERR(dir, EIO,
1531                                              "reading directory lblock %lu",
1532                                              (unsigned long) block);
1533                         brelse(bh);
1534                         ret = ERR_PTR(-EIO);
1535                         goto cleanup_and_exit;
1536                 }
1537                 if (!buffer_verified(bh) &&
1538                     !is_dx_internal_node(dir, block,
1539                                          (struct ext4_dir_entry *)bh->b_data) &&
1540                     !ext4_dirblock_csum_verify(dir, bh)) {
1541                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1542                                              "checksumming directory "
1543                                              "block %lu", (unsigned long)block);
1544                         brelse(bh);
1545                         ret = ERR_PTR(-EFSBADCRC);
1546                         goto cleanup_and_exit;
1547                 }
1548                 set_buffer_verified(bh);
1549                 i = search_dirblock(bh, dir, fname,
1550                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1551                 if (i == 1) {
1552                         EXT4_I(dir)->i_dir_start_lookup = block;
1553                         ret = bh;
1554                         goto cleanup_and_exit;
1555                 } else {
1556                         brelse(bh);
1557                         if (i < 0)
1558                                 goto cleanup_and_exit;
1559                 }
1560         next:
1561                 if (++block >= nblocks)
1562                         block = 0;
1563         } while (block != start);
1564
1565         /*
1566          * If the directory has grown while we were searching, then
1567          * search the last part of the directory before giving up.
1568          */
1569         block = nblocks;
1570         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1571         if (block < nblocks) {
1572                 start = 0;
1573                 goto restart;
1574         }
1575
1576 cleanup_and_exit:
1577         /* Clean up the read-ahead blocks */
1578         for (; ra_ptr < ra_max; ra_ptr++)
1579                 brelse(bh_use[ra_ptr]);
1580         return ret;
1581 }
1582
1583 static struct buffer_head *ext4_find_entry(struct inode *dir,
1584                                            const struct qstr *d_name,
1585                                            struct ext4_dir_entry_2 **res_dir,
1586                                            int *inlined)
1587 {
1588         int err;
1589         struct ext4_filename fname;
1590         struct buffer_head *bh;
1591
1592         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1593         if (err == -ENOENT)
1594                 return NULL;
1595         if (err)
1596                 return ERR_PTR(err);
1597
1598         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1599
1600         ext4_fname_free_filename(&fname);
1601         return bh;
1602 }
1603
1604 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1605                                              struct dentry *dentry,
1606                                              struct ext4_dir_entry_2 **res_dir)
1607 {
1608         int err;
1609         struct ext4_filename fname;
1610         struct buffer_head *bh;
1611
1612         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1613         if (err == -ENOENT)
1614                 return NULL;
1615         if (err)
1616                 return ERR_PTR(err);
1617
1618         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1619
1620         ext4_fname_free_filename(&fname);
1621         return bh;
1622 }
1623
1624 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1625                         struct ext4_filename *fname,
1626                         struct ext4_dir_entry_2 **res_dir)
1627 {
1628         struct super_block * sb = dir->i_sb;
1629         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1630         struct buffer_head *bh;
1631         ext4_lblk_t block;
1632         int retval;
1633
1634 #ifdef CONFIG_FS_ENCRYPTION
1635         *res_dir = NULL;
1636 #endif
1637         frame = dx_probe(fname, dir, NULL, frames);
1638         if (IS_ERR(frame))
1639                 return (struct buffer_head *) frame;
1640         do {
1641                 block = dx_get_block(frame->at);
1642                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1643                 if (IS_ERR(bh))
1644                         goto errout;
1645
1646                 retval = search_dirblock(bh, dir, fname,
1647                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1648                                          res_dir);
1649                 if (retval == 1)
1650                         goto success;
1651                 brelse(bh);
1652                 if (retval == -1) {
1653                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1654                         goto errout;
1655                 }
1656
1657                 /* Check to see if we should continue to search */
1658                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1659                                                frames, NULL);
1660                 if (retval < 0) {
1661                         ext4_warning_inode(dir,
1662                                 "error %d reading directory index block",
1663                                 retval);
1664                         bh = ERR_PTR(retval);
1665                         goto errout;
1666                 }
1667         } while (retval == 1);
1668
1669         bh = NULL;
1670 errout:
1671         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1672 success:
1673         dx_release(frames);
1674         return bh;
1675 }
1676
1677 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1678 {
1679         struct inode *inode;
1680         struct ext4_dir_entry_2 *de;
1681         struct buffer_head *bh;
1682
1683         if (dentry->d_name.len > EXT4_NAME_LEN)
1684                 return ERR_PTR(-ENAMETOOLONG);
1685
1686         bh = ext4_lookup_entry(dir, dentry, &de);
1687         if (IS_ERR(bh))
1688                 return ERR_CAST(bh);
1689         inode = NULL;
1690         if (bh) {
1691                 __u32 ino = le32_to_cpu(de->inode);
1692                 brelse(bh);
1693                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1694                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1695                         return ERR_PTR(-EFSCORRUPTED);
1696                 }
1697                 if (unlikely(ino == dir->i_ino)) {
1698                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1699                                          dentry);
1700                         return ERR_PTR(-EFSCORRUPTED);
1701                 }
1702                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1703                 if (inode == ERR_PTR(-ESTALE)) {
1704                         EXT4_ERROR_INODE(dir,
1705                                          "deleted inode referenced: %u",
1706                                          ino);
1707                         return ERR_PTR(-EFSCORRUPTED);
1708                 }
1709                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1710                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1711                     !fscrypt_has_permitted_context(dir, inode)) {
1712                         ext4_warning(inode->i_sb,
1713                                      "Inconsistent encryption contexts: %lu/%lu",
1714                                      dir->i_ino, inode->i_ino);
1715                         iput(inode);
1716                         return ERR_PTR(-EPERM);
1717                 }
1718         }
1719
1720 #ifdef CONFIG_UNICODE
1721         if (!inode && IS_CASEFOLDED(dir)) {
1722                 /* Eventually we want to call d_add_ci(dentry, NULL)
1723                  * for negative dentries in the encoding case as
1724                  * well.  For now, prevent the negative dentry
1725                  * from being cached.
1726                  */
1727                 return NULL;
1728         }
1729 #endif
1730         return d_splice_alias(inode, dentry);
1731 }
1732
1733
1734 struct dentry *ext4_get_parent(struct dentry *child)
1735 {
1736         __u32 ino;
1737         static const struct qstr dotdot = QSTR_INIT("..", 2);
1738         struct ext4_dir_entry_2 * de;
1739         struct buffer_head *bh;
1740
1741         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1742         if (IS_ERR(bh))
1743                 return ERR_CAST(bh);
1744         if (!bh)
1745                 return ERR_PTR(-ENOENT);
1746         ino = le32_to_cpu(de->inode);
1747         brelse(bh);
1748
1749         if (!ext4_valid_inum(child->d_sb, ino)) {
1750                 EXT4_ERROR_INODE(d_inode(child),
1751                                  "bad parent inode number: %u", ino);
1752                 return ERR_PTR(-EFSCORRUPTED);
1753         }
1754
1755         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1756 }
1757
1758 /*
1759  * Move count entries from end of map between two memory locations.
1760  * Returns pointer to last entry moved.
1761  */
1762 static struct ext4_dir_entry_2 *
1763 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1764                 unsigned blocksize)
1765 {
1766         unsigned rec_len = 0;
1767
1768         while (count--) {
1769                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1770                                                 (from + (map->offs<<2));
1771                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1772                 memcpy (to, de, rec_len);
1773                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1774                                 ext4_rec_len_to_disk(rec_len, blocksize);
1775                 de->inode = 0;
1776                 map++;
1777                 to += rec_len;
1778         }
1779         return (struct ext4_dir_entry_2 *) (to - rec_len);
1780 }
1781
1782 /*
1783  * Compact each dir entry in the range to the minimal rec_len.
1784  * Returns pointer to last entry in range.
1785  */
1786 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1787 {
1788         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1789         unsigned rec_len = 0;
1790
1791         prev = to = de;
1792         while ((char*)de < base + blocksize) {
1793                 next = ext4_next_entry(de, blocksize);
1794                 if (de->inode && de->name_len) {
1795                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1796                         if (de > to)
1797                                 memmove(to, de, rec_len);
1798                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1799                         prev = to;
1800                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1801                 }
1802                 de = next;
1803         }
1804         return prev;
1805 }
1806
1807 /*
1808  * Split a full leaf block to make room for a new dir entry.
1809  * Allocate a new block, and move entries so that they are approx. equally full.
1810  * Returns pointer to de in block into which the new entry will be inserted.
1811  */
1812 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1813                         struct buffer_head **bh,struct dx_frame *frame,
1814                         struct dx_hash_info *hinfo)
1815 {
1816         unsigned blocksize = dir->i_sb->s_blocksize;
1817         unsigned count, continued;
1818         struct buffer_head *bh2;
1819         ext4_lblk_t newblock;
1820         u32 hash2;
1821         struct dx_map_entry *map;
1822         char *data1 = (*bh)->b_data, *data2;
1823         unsigned split, move, size;
1824         struct ext4_dir_entry_2 *de = NULL, *de2;
1825         int     csum_size = 0;
1826         int     err = 0, i;
1827
1828         if (ext4_has_metadata_csum(dir->i_sb))
1829                 csum_size = sizeof(struct ext4_dir_entry_tail);
1830
1831         bh2 = ext4_append(handle, dir, &newblock);
1832         if (IS_ERR(bh2)) {
1833                 brelse(*bh);
1834                 *bh = NULL;
1835                 return (struct ext4_dir_entry_2 *) bh2;
1836         }
1837
1838         BUFFER_TRACE(*bh, "get_write_access");
1839         err = ext4_journal_get_write_access(handle, *bh);
1840         if (err)
1841                 goto journal_error;
1842
1843         BUFFER_TRACE(frame->bh, "get_write_access");
1844         err = ext4_journal_get_write_access(handle, frame->bh);
1845         if (err)
1846                 goto journal_error;
1847
1848         data2 = bh2->b_data;
1849
1850         /* create map in the end of data2 block */
1851         map = (struct dx_map_entry *) (data2 + blocksize);
1852         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1853                              blocksize, hinfo, map);
1854         map -= count;
1855         dx_sort_map(map, count);
1856         /* Ensure that neither split block is over half full */
1857         size = 0;
1858         move = 0;
1859         for (i = count-1; i >= 0; i--) {
1860                 /* is more than half of this entry in 2nd half of the block? */
1861                 if (size + map[i].size/2 > blocksize/2)
1862                         break;
1863                 size += map[i].size;
1864                 move++;
1865         }
1866         /*
1867          * map index at which we will split
1868          *
1869          * If the sum of active entries didn't exceed half the block size, just
1870          * split it in half by count; each resulting block will have at least
1871          * half the space free.
1872          */
1873         if (i > 0)
1874                 split = count - move;
1875         else
1876                 split = count/2;
1877
1878         hash2 = map[split].hash;
1879         continued = hash2 == map[split - 1].hash;
1880         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1881                         (unsigned long)dx_get_block(frame->at),
1882                                         hash2, split, count-split));
1883
1884         /* Fancy dance to stay within two buffers */
1885         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1886                               blocksize);
1887         de = dx_pack_dirents(data1, blocksize);
1888         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1889                                            (char *) de,
1890                                            blocksize);
1891         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1892                                             (char *) de2,
1893                                             blocksize);
1894         if (csum_size) {
1895                 ext4_initialize_dirent_tail(*bh, blocksize);
1896                 ext4_initialize_dirent_tail(bh2, blocksize);
1897         }
1898
1899         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1900                         blocksize, 1));
1901         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1902                         blocksize, 1));
1903
1904         /* Which block gets the new entry? */
1905         if (hinfo->hash >= hash2) {
1906                 swap(*bh, bh2);
1907                 de = de2;
1908         }
1909         dx_insert_block(frame, hash2 + continued, newblock);
1910         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1911         if (err)
1912                 goto journal_error;
1913         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1914         if (err)
1915                 goto journal_error;
1916         brelse(bh2);
1917         dxtrace(dx_show_index("frame", frame->entries));
1918         return de;
1919
1920 journal_error:
1921         brelse(*bh);
1922         brelse(bh2);
1923         *bh = NULL;
1924         ext4_std_error(dir->i_sb, err);
1925         return ERR_PTR(err);
1926 }
1927
1928 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1929                       struct buffer_head *bh,
1930                       void *buf, int buf_size,
1931                       struct ext4_filename *fname,
1932                       struct ext4_dir_entry_2 **dest_de)
1933 {
1934         struct ext4_dir_entry_2 *de;
1935         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1936         int nlen, rlen;
1937         unsigned int offset = 0;
1938         char *top;
1939
1940         de = (struct ext4_dir_entry_2 *)buf;
1941         top = buf + buf_size - reclen;
1942         while ((char *) de <= top) {
1943                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1944                                          buf, buf_size, offset))
1945                         return -EFSCORRUPTED;
1946                 if (ext4_match(dir, fname, de))
1947                         return -EEXIST;
1948                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1949                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1950                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1951                         break;
1952                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1953                 offset += rlen;
1954         }
1955         if ((char *) de > top)
1956                 return -ENOSPC;
1957
1958         *dest_de = de;
1959         return 0;
1960 }
1961
1962 void ext4_insert_dentry(struct inode *inode,
1963                         struct ext4_dir_entry_2 *de,
1964                         int buf_size,
1965                         struct ext4_filename *fname)
1966 {
1967
1968         int nlen, rlen;
1969
1970         nlen = EXT4_DIR_REC_LEN(de->name_len);
1971         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1972         if (de->inode) {
1973                 struct ext4_dir_entry_2 *de1 =
1974                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1975                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1976                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1977                 de = de1;
1978         }
1979         de->file_type = EXT4_FT_UNKNOWN;
1980         de->inode = cpu_to_le32(inode->i_ino);
1981         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1982         de->name_len = fname_len(fname);
1983         memcpy(de->name, fname_name(fname), fname_len(fname));
1984 }
1985
1986 /*
1987  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1988  * it points to a directory entry which is guaranteed to be large
1989  * enough for new directory entry.  If de is NULL, then
1990  * add_dirent_to_buf will attempt search the directory block for
1991  * space.  It will return -ENOSPC if no space is available, and -EIO
1992  * and -EEXIST if directory entry already exists.
1993  */
1994 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1995                              struct inode *dir,
1996                              struct inode *inode, struct ext4_dir_entry_2 *de,
1997                              struct buffer_head *bh)
1998 {
1999         unsigned int    blocksize = dir->i_sb->s_blocksize;
2000         int             csum_size = 0;
2001         int             err, err2;
2002
2003         if (ext4_has_metadata_csum(inode->i_sb))
2004                 csum_size = sizeof(struct ext4_dir_entry_tail);
2005
2006         if (!de) {
2007                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2008                                         blocksize - csum_size, fname, &de);
2009                 if (err)
2010                         return err;
2011         }
2012         BUFFER_TRACE(bh, "get_write_access");
2013         err = ext4_journal_get_write_access(handle, bh);
2014         if (err) {
2015                 ext4_std_error(dir->i_sb, err);
2016                 return err;
2017         }
2018
2019         /* By now the buffer is marked for journaling */
2020         ext4_insert_dentry(inode, de, blocksize, fname);
2021
2022         /*
2023          * XXX shouldn't update any times until successful
2024          * completion of syscall, but too many callers depend
2025          * on this.
2026          *
2027          * XXX similarly, too many callers depend on
2028          * ext4_new_inode() setting the times, but error
2029          * recovery deletes the inode, so the worst that can
2030          * happen is that the times are slightly out of date
2031          * and/or different from the directory change time.
2032          */
2033         dir->i_mtime = dir->i_ctime = current_time(dir);
2034         ext4_update_dx_flag(dir);
2035         inode_inc_iversion(dir);
2036         err2 = ext4_mark_inode_dirty(handle, dir);
2037         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2038         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2039         if (err)
2040                 ext4_std_error(dir->i_sb, err);
2041         return err ? err : err2;
2042 }
2043
2044 /*
2045  * This converts a one block unindexed directory to a 3 block indexed
2046  * directory, and adds the dentry to the indexed directory.
2047  */
2048 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2049                             struct inode *dir,
2050                             struct inode *inode, struct buffer_head *bh)
2051 {
2052         struct buffer_head *bh2;
2053         struct dx_root  *root;
2054         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2055         struct dx_entry *entries;
2056         struct ext4_dir_entry_2 *de, *de2;
2057         char            *data2, *top;
2058         unsigned        len;
2059         int             retval;
2060         unsigned        blocksize;
2061         ext4_lblk_t  block;
2062         struct fake_dirent *fde;
2063         int csum_size = 0;
2064
2065         if (ext4_has_metadata_csum(inode->i_sb))
2066                 csum_size = sizeof(struct ext4_dir_entry_tail);
2067
2068         blocksize =  dir->i_sb->s_blocksize;
2069         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2070         BUFFER_TRACE(bh, "get_write_access");
2071         retval = ext4_journal_get_write_access(handle, bh);
2072         if (retval) {
2073                 ext4_std_error(dir->i_sb, retval);
2074                 brelse(bh);
2075                 return retval;
2076         }
2077         root = (struct dx_root *) bh->b_data;
2078
2079         /* The 0th block becomes the root, move the dirents out */
2080         fde = &root->dotdot;
2081         de = (struct ext4_dir_entry_2 *)((char *)fde +
2082                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2083         if ((char *) de >= (((char *) root) + blocksize)) {
2084                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2085                 brelse(bh);
2086                 return -EFSCORRUPTED;
2087         }
2088         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2089
2090         /* Allocate new block for the 0th block's dirents */
2091         bh2 = ext4_append(handle, dir, &block);
2092         if (IS_ERR(bh2)) {
2093                 brelse(bh);
2094                 return PTR_ERR(bh2);
2095         }
2096         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2097         data2 = bh2->b_data;
2098
2099         memcpy(data2, de, len);
2100         de = (struct ext4_dir_entry_2 *) data2;
2101         top = data2 + len;
2102         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2103                 de = de2;
2104         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2105                                            (char *) de, blocksize);
2106
2107         if (csum_size)
2108                 ext4_initialize_dirent_tail(bh2, blocksize);
2109
2110         /* Initialize the root; the dot dirents already exist */
2111         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2112         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2113                                            blocksize);
2114         memset (&root->info, 0, sizeof(root->info));
2115         root->info.info_length = sizeof(root->info);
2116         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2117         entries = root->entries;
2118         dx_set_block(entries, 1);
2119         dx_set_count(entries, 1);
2120         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2121
2122         /* Initialize as for dx_probe */
2123         fname->hinfo.hash_version = root->info.hash_version;
2124         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2125                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2126         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2127         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2128
2129         memset(frames, 0, sizeof(frames));
2130         frame = frames;
2131         frame->entries = entries;
2132         frame->at = entries;
2133         frame->bh = bh;
2134
2135         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2136         if (retval)
2137                 goto out_frames;
2138         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2139         if (retval)
2140                 goto out_frames;
2141
2142         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2143         if (IS_ERR(de)) {
2144                 retval = PTR_ERR(de);
2145                 goto out_frames;
2146         }
2147
2148         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2149 out_frames:
2150         /*
2151          * Even if the block split failed, we have to properly write
2152          * out all the changes we did so far. Otherwise we can end up
2153          * with corrupted filesystem.
2154          */
2155         if (retval)
2156                 ext4_mark_inode_dirty(handle, dir);
2157         dx_release(frames);
2158         brelse(bh2);
2159         return retval;
2160 }
2161
2162 /*
2163  *      ext4_add_entry()
2164  *
2165  * adds a file entry to the specified directory, using the same
2166  * semantics as ext4_find_entry(). It returns NULL if it failed.
2167  *
2168  * NOTE!! The inode part of 'de' is left at 0 - which means you
2169  * may not sleep between calling this and putting something into
2170  * the entry, as someone else might have used it while you slept.
2171  */
2172 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2173                           struct inode *inode)
2174 {
2175         struct inode *dir = d_inode(dentry->d_parent);
2176         struct buffer_head *bh = NULL;
2177         struct ext4_dir_entry_2 *de;
2178         struct super_block *sb;
2179         struct ext4_filename fname;
2180         int     retval;
2181         int     dx_fallback=0;
2182         unsigned blocksize;
2183         ext4_lblk_t block, blocks;
2184         int     csum_size = 0;
2185
2186         if (ext4_has_metadata_csum(inode->i_sb))
2187                 csum_size = sizeof(struct ext4_dir_entry_tail);
2188
2189         sb = dir->i_sb;
2190         blocksize = sb->s_blocksize;
2191         if (!dentry->d_name.len)
2192                 return -EINVAL;
2193
2194 #ifdef CONFIG_UNICODE
2195         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2196             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2197                 return -EINVAL;
2198 #endif
2199
2200         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2201         if (retval)
2202                 return retval;
2203
2204         if (ext4_has_inline_data(dir)) {
2205                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2206                 if (retval < 0)
2207                         goto out;
2208                 if (retval == 1) {
2209                         retval = 0;
2210                         goto out;
2211                 }
2212         }
2213
2214         if (is_dx(dir)) {
2215                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2216                 if (!retval || (retval != ERR_BAD_DX_DIR))
2217                         goto out;
2218                 /* Can we just ignore htree data? */
2219                 if (ext4_has_metadata_csum(sb)) {
2220                         EXT4_ERROR_INODE(dir,
2221                                 "Directory has corrupted htree index.");
2222                         retval = -EFSCORRUPTED;
2223                         goto out;
2224                 }
2225                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2226                 dx_fallback++;
2227                 retval = ext4_mark_inode_dirty(handle, dir);
2228                 if (unlikely(retval))
2229                         goto out;
2230         }
2231         blocks = dir->i_size >> sb->s_blocksize_bits;
2232         for (block = 0; block < blocks; block++) {
2233                 bh = ext4_read_dirblock(dir, block, DIRENT);
2234                 if (bh == NULL) {
2235                         bh = ext4_bread(handle, dir, block,
2236                                         EXT4_GET_BLOCKS_CREATE);
2237                         goto add_to_new_block;
2238                 }
2239                 if (IS_ERR(bh)) {
2240                         retval = PTR_ERR(bh);
2241                         bh = NULL;
2242                         goto out;
2243                 }
2244                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2245                                            NULL, bh);
2246                 if (retval != -ENOSPC)
2247                         goto out;
2248
2249                 if (blocks == 1 && !dx_fallback &&
2250                     ext4_has_feature_dir_index(sb)) {
2251                         retval = make_indexed_dir(handle, &fname, dir,
2252                                                   inode, bh);
2253                         bh = NULL; /* make_indexed_dir releases bh */
2254                         goto out;
2255                 }
2256                 brelse(bh);
2257         }
2258         bh = ext4_append(handle, dir, &block);
2259 add_to_new_block:
2260         if (IS_ERR(bh)) {
2261                 retval = PTR_ERR(bh);
2262                 bh = NULL;
2263                 goto out;
2264         }
2265         de = (struct ext4_dir_entry_2 *) bh->b_data;
2266         de->inode = 0;
2267         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2268
2269         if (csum_size)
2270                 ext4_initialize_dirent_tail(bh, blocksize);
2271
2272         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2273 out:
2274         ext4_fname_free_filename(&fname);
2275         brelse(bh);
2276         if (retval == 0)
2277                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2278         return retval;
2279 }
2280
2281 /*
2282  * Returns 0 for success, or a negative error value
2283  */
2284 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2285                              struct inode *dir, struct inode *inode)
2286 {
2287         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2288         struct dx_entry *entries, *at;
2289         struct buffer_head *bh;
2290         struct super_block *sb = dir->i_sb;
2291         struct ext4_dir_entry_2 *de;
2292         int restart;
2293         int err;
2294
2295 again:
2296         restart = 0;
2297         frame = dx_probe(fname, dir, NULL, frames);
2298         if (IS_ERR(frame))
2299                 return PTR_ERR(frame);
2300         entries = frame->entries;
2301         at = frame->at;
2302         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2303         if (IS_ERR(bh)) {
2304                 err = PTR_ERR(bh);
2305                 bh = NULL;
2306                 goto cleanup;
2307         }
2308
2309         BUFFER_TRACE(bh, "get_write_access");
2310         err = ext4_journal_get_write_access(handle, bh);
2311         if (err)
2312                 goto journal_error;
2313
2314         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2315         if (err != -ENOSPC)
2316                 goto cleanup;
2317
2318         err = 0;
2319         /* Block full, should compress but for now just split */
2320         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2321                        dx_get_count(entries), dx_get_limit(entries)));
2322         /* Need to split index? */
2323         if (dx_get_count(entries) == dx_get_limit(entries)) {
2324                 ext4_lblk_t newblock;
2325                 int levels = frame - frames + 1;
2326                 unsigned int icount;
2327                 int add_level = 1;
2328                 struct dx_entry *entries2;
2329                 struct dx_node *node2;
2330                 struct buffer_head *bh2;
2331
2332                 while (frame > frames) {
2333                         if (dx_get_count((frame - 1)->entries) <
2334                             dx_get_limit((frame - 1)->entries)) {
2335                                 add_level = 0;
2336                                 break;
2337                         }
2338                         frame--; /* split higher index block */
2339                         at = frame->at;
2340                         entries = frame->entries;
2341                         restart = 1;
2342                 }
2343                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2344                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2345                                          "reach max htree level :%d",
2346                                          dir->i_ino, levels);
2347                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2348                                 ext4_warning(sb, "Large directory feature is "
2349                                                  "not enabled on this "
2350                                                  "filesystem");
2351                         }
2352                         err = -ENOSPC;
2353                         goto cleanup;
2354                 }
2355                 icount = dx_get_count(entries);
2356                 bh2 = ext4_append(handle, dir, &newblock);
2357                 if (IS_ERR(bh2)) {
2358                         err = PTR_ERR(bh2);
2359                         goto cleanup;
2360                 }
2361                 node2 = (struct dx_node *)(bh2->b_data);
2362                 entries2 = node2->entries;
2363                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2364                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2365                                                            sb->s_blocksize);
2366                 BUFFER_TRACE(frame->bh, "get_write_access");
2367                 err = ext4_journal_get_write_access(handle, frame->bh);
2368                 if (err)
2369                         goto journal_error;
2370                 if (!add_level) {
2371                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2372                         unsigned hash2 = dx_get_hash(entries + icount1);
2373                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2374                                        icount1, icount2));
2375
2376                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2377                         err = ext4_journal_get_write_access(handle,
2378                                                              (frame - 1)->bh);
2379                         if (err)
2380                                 goto journal_error;
2381
2382                         memcpy((char *) entries2, (char *) (entries + icount1),
2383                                icount2 * sizeof(struct dx_entry));
2384                         dx_set_count(entries, icount1);
2385                         dx_set_count(entries2, icount2);
2386                         dx_set_limit(entries2, dx_node_limit(dir));
2387
2388                         /* Which index block gets the new entry? */
2389                         if (at - entries >= icount1) {
2390                                 frame->at = at = at - entries - icount1 + entries2;
2391                                 frame->entries = entries = entries2;
2392                                 swap(frame->bh, bh2);
2393                         }
2394                         dx_insert_block((frame - 1), hash2, newblock);
2395                         dxtrace(dx_show_index("node", frame->entries));
2396                         dxtrace(dx_show_index("node",
2397                                ((struct dx_node *) bh2->b_data)->entries));
2398                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2399                         if (err)
2400                                 goto journal_error;
2401                         brelse (bh2);
2402                         err = ext4_handle_dirty_dx_node(handle, dir,
2403                                                    (frame - 1)->bh);
2404                         if (err)
2405                                 goto journal_error;
2406                         if (restart) {
2407                                 err = ext4_handle_dirty_dx_node(handle, dir,
2408                                                            frame->bh);
2409                                 goto journal_error;
2410                         }
2411                 } else {
2412                         struct dx_root *dxroot;
2413                         memcpy((char *) entries2, (char *) entries,
2414                                icount * sizeof(struct dx_entry));
2415                         dx_set_limit(entries2, dx_node_limit(dir));
2416
2417                         /* Set up root */
2418                         dx_set_count(entries, 1);
2419                         dx_set_block(entries + 0, newblock);
2420                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2421                         dxroot->info.indirect_levels += 1;
2422                         dxtrace(printk(KERN_DEBUG
2423                                        "Creating %d level index...\n",
2424                                        dxroot->info.indirect_levels));
2425                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2426                         if (err)
2427                                 goto journal_error;
2428                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2429                         brelse(bh2);
2430                         restart = 1;
2431                         goto journal_error;
2432                 }
2433         }
2434         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2435         if (IS_ERR(de)) {
2436                 err = PTR_ERR(de);
2437                 goto cleanup;
2438         }
2439         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2440         goto cleanup;
2441
2442 journal_error:
2443         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2444 cleanup:
2445         brelse(bh);
2446         dx_release(frames);
2447         /* @restart is true means htree-path has been changed, we need to
2448          * repeat dx_probe() to find out valid htree-path
2449          */
2450         if (restart && err == 0)
2451                 goto again;
2452         return err;
2453 }
2454
2455 /*
2456  * ext4_generic_delete_entry deletes a directory entry by merging it
2457  * with the previous entry
2458  */
2459 int ext4_generic_delete_entry(struct inode *dir,
2460                               struct ext4_dir_entry_2 *de_del,
2461                               struct buffer_head *bh,
2462                               void *entry_buf,
2463                               int buf_size,
2464                               int csum_size)
2465 {
2466         struct ext4_dir_entry_2 *de, *pde;
2467         unsigned int blocksize = dir->i_sb->s_blocksize;
2468         int i;
2469
2470         i = 0;
2471         pde = NULL;
2472         de = (struct ext4_dir_entry_2 *)entry_buf;
2473         while (i < buf_size - csum_size) {
2474                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2475                                          entry_buf, buf_size, i))
2476                         return -EFSCORRUPTED;
2477                 if (de == de_del)  {
2478                         if (pde)
2479                                 pde->rec_len = ext4_rec_len_to_disk(
2480                                         ext4_rec_len_from_disk(pde->rec_len,
2481                                                                blocksize) +
2482                                         ext4_rec_len_from_disk(de->rec_len,
2483                                                                blocksize),
2484                                         blocksize);
2485                         else
2486                                 de->inode = 0;
2487                         inode_inc_iversion(dir);
2488                         return 0;
2489                 }
2490                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2491                 pde = de;
2492                 de = ext4_next_entry(de, blocksize);
2493         }
2494         return -ENOENT;
2495 }
2496
2497 static int ext4_delete_entry(handle_t *handle,
2498                              struct inode *dir,
2499                              struct ext4_dir_entry_2 *de_del,
2500                              struct buffer_head *bh)
2501 {
2502         int err, csum_size = 0;
2503
2504         if (ext4_has_inline_data(dir)) {
2505                 int has_inline_data = 1;
2506                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2507                                                &has_inline_data);
2508                 if (has_inline_data)
2509                         return err;
2510         }
2511
2512         if (ext4_has_metadata_csum(dir->i_sb))
2513                 csum_size = sizeof(struct ext4_dir_entry_tail);
2514
2515         BUFFER_TRACE(bh, "get_write_access");
2516         err = ext4_journal_get_write_access(handle, bh);
2517         if (unlikely(err))
2518                 goto out;
2519
2520         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2521                                         dir->i_sb->s_blocksize, csum_size);
2522         if (err)
2523                 goto out;
2524
2525         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2526         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2527         if (unlikely(err))
2528                 goto out;
2529
2530         return 0;
2531 out:
2532         if (err != -ENOENT)
2533                 ext4_std_error(dir->i_sb, err);
2534         return err;
2535 }
2536
2537 /*
2538  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2539  * since this indicates that nlinks count was previously 1 to avoid overflowing
2540  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2541  * that subdirectory link counts are not being maintained accurately.
2542  *
2543  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2544  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2545  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2546  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2547  */
2548 static void ext4_inc_count(struct inode *inode)
2549 {
2550         inc_nlink(inode);
2551         if (is_dx(inode) &&
2552             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2553                 set_nlink(inode, 1);
2554 }
2555
2556 /*
2557  * If a directory had nlink == 1, then we should let it be 1. This indicates
2558  * directory has >EXT4_LINK_MAX subdirs.
2559  */
2560 static void ext4_dec_count(struct inode *inode)
2561 {
2562         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2563                 drop_nlink(inode);
2564 }
2565
2566
2567 /*
2568  * Add non-directory inode to a directory. On success, the inode reference is
2569  * consumed by dentry is instantiation. This is also indicated by clearing of
2570  * *inodep pointer. On failure, the caller is responsible for dropping the
2571  * inode reference in the safe context.
2572  */
2573 static int ext4_add_nondir(handle_t *handle,
2574                 struct dentry *dentry, struct inode **inodep)
2575 {
2576         struct inode *dir = d_inode(dentry->d_parent);
2577         struct inode *inode = *inodep;
2578         int err = ext4_add_entry(handle, dentry, inode);
2579         if (!err) {
2580                 err = ext4_mark_inode_dirty(handle, inode);
2581                 if (IS_DIRSYNC(dir))
2582                         ext4_handle_sync(handle);
2583                 d_instantiate_new(dentry, inode);
2584                 *inodep = NULL;
2585                 return err;
2586         }
2587         drop_nlink(inode);
2588         ext4_orphan_add(handle, inode);
2589         unlock_new_inode(inode);
2590         return err;
2591 }
2592
2593 /*
2594  * By the time this is called, we already have created
2595  * the directory cache entry for the new file, but it
2596  * is so far negative - it has no inode.
2597  *
2598  * If the create succeeds, we fill in the inode information
2599  * with d_instantiate().
2600  */
2601 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2602                        bool excl)
2603 {
2604         handle_t *handle;
2605         struct inode *inode;
2606         int err, credits, retries = 0;
2607
2608         err = dquot_initialize(dir);
2609         if (err)
2610                 return err;
2611
2612         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2613                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2614 retry:
2615         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2616                                             NULL, EXT4_HT_DIR, credits);
2617         handle = ext4_journal_current_handle();
2618         err = PTR_ERR(inode);
2619         if (!IS_ERR(inode)) {
2620                 inode->i_op = &ext4_file_inode_operations;
2621                 inode->i_fop = &ext4_file_operations;
2622                 ext4_set_aops(inode);
2623                 err = ext4_add_nondir(handle, dentry, &inode);
2624                 if (!err)
2625                         ext4_fc_track_create(handle, dentry);
2626         }
2627         if (handle)
2628                 ext4_journal_stop(handle);
2629         if (!IS_ERR_OR_NULL(inode))
2630                 iput(inode);
2631         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2632                 goto retry;
2633         return err;
2634 }
2635
2636 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2637                       umode_t mode, dev_t rdev)
2638 {
2639         handle_t *handle;
2640         struct inode *inode;
2641         int err, credits, retries = 0;
2642
2643         err = dquot_initialize(dir);
2644         if (err)
2645                 return err;
2646
2647         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2648                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2649 retry:
2650         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2651                                             NULL, EXT4_HT_DIR, credits);
2652         handle = ext4_journal_current_handle();
2653         err = PTR_ERR(inode);
2654         if (!IS_ERR(inode)) {
2655                 init_special_inode(inode, inode->i_mode, rdev);
2656                 inode->i_op = &ext4_special_inode_operations;
2657                 err = ext4_add_nondir(handle, dentry, &inode);
2658                 if (!err)
2659                         ext4_fc_track_create(handle, dentry);
2660         }
2661         if (handle)
2662                 ext4_journal_stop(handle);
2663         if (!IS_ERR_OR_NULL(inode))
2664                 iput(inode);
2665         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2666                 goto retry;
2667         return err;
2668 }
2669
2670 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2671 {
2672         handle_t *handle;
2673         struct inode *inode;
2674         int err, retries = 0;
2675
2676         err = dquot_initialize(dir);
2677         if (err)
2678                 return err;
2679
2680 retry:
2681         inode = ext4_new_inode_start_handle(dir, mode,
2682                                             NULL, 0, NULL,
2683                                             EXT4_HT_DIR,
2684                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2685                           4 + EXT4_XATTR_TRANS_BLOCKS);
2686         handle = ext4_journal_current_handle();
2687         err = PTR_ERR(inode);
2688         if (!IS_ERR(inode)) {
2689                 inode->i_op = &ext4_file_inode_operations;
2690                 inode->i_fop = &ext4_file_operations;
2691                 ext4_set_aops(inode);
2692                 d_tmpfile(dentry, inode);
2693                 err = ext4_orphan_add(handle, inode);
2694                 if (err)
2695                         goto err_unlock_inode;
2696                 mark_inode_dirty(inode);
2697                 unlock_new_inode(inode);
2698         }
2699         if (handle)
2700                 ext4_journal_stop(handle);
2701         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2702                 goto retry;
2703         return err;
2704 err_unlock_inode:
2705         ext4_journal_stop(handle);
2706         unlock_new_inode(inode);
2707         return err;
2708 }
2709
2710 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2711                           struct ext4_dir_entry_2 *de,
2712                           int blocksize, int csum_size,
2713                           unsigned int parent_ino, int dotdot_real_len)
2714 {
2715         de->inode = cpu_to_le32(inode->i_ino);
2716         de->name_len = 1;
2717         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2718                                            blocksize);
2719         strcpy(de->name, ".");
2720         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2721
2722         de = ext4_next_entry(de, blocksize);
2723         de->inode = cpu_to_le32(parent_ino);
2724         de->name_len = 2;
2725         if (!dotdot_real_len)
2726                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2727                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2728                                         blocksize);
2729         else
2730                 de->rec_len = ext4_rec_len_to_disk(
2731                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2732         strcpy(de->name, "..");
2733         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2734
2735         return ext4_next_entry(de, blocksize);
2736 }
2737
2738 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2739                              struct inode *inode)
2740 {
2741         struct buffer_head *dir_block = NULL;
2742         struct ext4_dir_entry_2 *de;
2743         ext4_lblk_t block = 0;
2744         unsigned int blocksize = dir->i_sb->s_blocksize;
2745         int csum_size = 0;
2746         int err;
2747
2748         if (ext4_has_metadata_csum(dir->i_sb))
2749                 csum_size = sizeof(struct ext4_dir_entry_tail);
2750
2751         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2752                 err = ext4_try_create_inline_dir(handle, dir, inode);
2753                 if (err < 0 && err != -ENOSPC)
2754                         goto out;
2755                 if (!err)
2756                         goto out;
2757         }
2758
2759         inode->i_size = 0;
2760         dir_block = ext4_append(handle, inode, &block);
2761         if (IS_ERR(dir_block))
2762                 return PTR_ERR(dir_block);
2763         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2764         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2765         set_nlink(inode, 2);
2766         if (csum_size)
2767                 ext4_initialize_dirent_tail(dir_block, blocksize);
2768
2769         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2770         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2771         if (err)
2772                 goto out;
2773         set_buffer_verified(dir_block);
2774 out:
2775         brelse(dir_block);
2776         return err;
2777 }
2778
2779 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2780 {
2781         handle_t *handle;
2782         struct inode *inode;
2783         int err, err2 = 0, credits, retries = 0;
2784
2785         if (EXT4_DIR_LINK_MAX(dir))
2786                 return -EMLINK;
2787
2788         err = dquot_initialize(dir);
2789         if (err)
2790                 return err;
2791
2792         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2793                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2794 retry:
2795         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2796                                             &dentry->d_name,
2797                                             0, NULL, EXT4_HT_DIR, credits);
2798         handle = ext4_journal_current_handle();
2799         err = PTR_ERR(inode);
2800         if (IS_ERR(inode))
2801                 goto out_stop;
2802
2803         inode->i_op = &ext4_dir_inode_operations;
2804         inode->i_fop = &ext4_dir_operations;
2805         err = ext4_init_new_dir(handle, dir, inode);
2806         if (err)
2807                 goto out_clear_inode;
2808         err = ext4_mark_inode_dirty(handle, inode);
2809         if (!err)
2810                 err = ext4_add_entry(handle, dentry, inode);
2811         if (err) {
2812 out_clear_inode:
2813                 clear_nlink(inode);
2814                 ext4_orphan_add(handle, inode);
2815                 unlock_new_inode(inode);
2816                 err2 = ext4_mark_inode_dirty(handle, inode);
2817                 if (unlikely(err2))
2818                         err = err2;
2819                 ext4_journal_stop(handle);
2820                 iput(inode);
2821                 goto out_retry;
2822         }
2823         ext4_inc_count(dir);
2824
2825         ext4_update_dx_flag(dir);
2826         err = ext4_mark_inode_dirty(handle, dir);
2827         if (err)
2828                 goto out_clear_inode;
2829         d_instantiate_new(dentry, inode);
2830         ext4_fc_track_create(handle, dentry);
2831         if (IS_DIRSYNC(dir))
2832                 ext4_handle_sync(handle);
2833
2834 out_stop:
2835         if (handle)
2836                 ext4_journal_stop(handle);
2837 out_retry:
2838         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2839                 goto retry;
2840         return err;
2841 }
2842
2843 /*
2844  * routine to check that the specified directory is empty (for rmdir)
2845  */
2846 bool ext4_empty_dir(struct inode *inode)
2847 {
2848         unsigned int offset;
2849         struct buffer_head *bh;
2850         struct ext4_dir_entry_2 *de;
2851         struct super_block *sb;
2852
2853         if (ext4_has_inline_data(inode)) {
2854                 int has_inline_data = 1;
2855                 int ret;
2856
2857                 ret = empty_inline_dir(inode, &has_inline_data);
2858                 if (has_inline_data)
2859                         return ret;
2860         }
2861
2862         sb = inode->i_sb;
2863         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2864                 EXT4_ERROR_INODE(inode, "invalid size");
2865                 return true;
2866         }
2867         /* The first directory block must not be a hole,
2868          * so treat it as DIRENT_HTREE
2869          */
2870         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2871         if (IS_ERR(bh))
2872                 return true;
2873
2874         de = (struct ext4_dir_entry_2 *) bh->b_data;
2875         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2876                                  0) ||
2877             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2878                 ext4_warning_inode(inode, "directory missing '.'");
2879                 brelse(bh);
2880                 return true;
2881         }
2882         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2883         de = ext4_next_entry(de, sb->s_blocksize);
2884         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2885                                  offset) ||
2886             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2887                 ext4_warning_inode(inode, "directory missing '..'");
2888                 brelse(bh);
2889                 return true;
2890         }
2891         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2892         while (offset < inode->i_size) {
2893                 if (!(offset & (sb->s_blocksize - 1))) {
2894                         unsigned int lblock;
2895                         brelse(bh);
2896                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2897                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2898                         if (bh == NULL) {
2899                                 offset += sb->s_blocksize;
2900                                 continue;
2901                         }
2902                         if (IS_ERR(bh))
2903                                 return true;
2904                 }
2905                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2906                                         (offset & (sb->s_blocksize - 1)));
2907                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2908                                          bh->b_data, bh->b_size, offset)) {
2909                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2910                         continue;
2911                 }
2912                 if (le32_to_cpu(de->inode)) {
2913                         brelse(bh);
2914                         return false;
2915                 }
2916                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2917         }
2918         brelse(bh);
2919         return true;
2920 }
2921
2922 /*
2923  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2924  * such inodes, starting at the superblock, in case we crash before the
2925  * file is closed/deleted, or in case the inode truncate spans multiple
2926  * transactions and the last transaction is not recovered after a crash.
2927  *
2928  * At filesystem recovery time, we walk this list deleting unlinked
2929  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2930  *
2931  * Orphan list manipulation functions must be called under i_mutex unless
2932  * we are just creating the inode or deleting it.
2933  */
2934 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2935 {
2936         struct super_block *sb = inode->i_sb;
2937         struct ext4_sb_info *sbi = EXT4_SB(sb);
2938         struct ext4_iloc iloc;
2939         int err = 0, rc;
2940         bool dirty = false;
2941
2942         if (!sbi->s_journal || is_bad_inode(inode))
2943                 return 0;
2944
2945         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2946                      !inode_is_locked(inode));
2947         /*
2948          * Exit early if inode already is on orphan list. This is a big speedup
2949          * since we don't have to contend on the global s_orphan_lock.
2950          */
2951         if (!list_empty(&EXT4_I(inode)->i_orphan))
2952                 return 0;
2953
2954         /*
2955          * Orphan handling is only valid for files with data blocks
2956          * being truncated, or files being unlinked. Note that we either
2957          * hold i_mutex, or the inode can not be referenced from outside,
2958          * so i_nlink should not be bumped due to race
2959          */
2960         ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2961                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2962
2963         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2964         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2965         if (err)
2966                 goto out;
2967
2968         err = ext4_reserve_inode_write(handle, inode, &iloc);
2969         if (err)
2970                 goto out;
2971
2972         mutex_lock(&sbi->s_orphan_lock);
2973         /*
2974          * Due to previous errors inode may be already a part of on-disk
2975          * orphan list. If so skip on-disk list modification.
2976          */
2977         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2978             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2979                 /* Insert this inode at the head of the on-disk orphan list */
2980                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2981                 lock_buffer(sbi->s_sbh);
2982                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2983                 ext4_superblock_csum_set(sb);
2984                 unlock_buffer(sbi->s_sbh);
2985                 dirty = true;
2986         }
2987         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2988         mutex_unlock(&sbi->s_orphan_lock);
2989
2990         if (dirty) {
2991                 err = ext4_handle_dirty_super(handle, sb);
2992                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2993                 if (!err)
2994                         err = rc;
2995                 if (err) {
2996                         /*
2997                          * We have to remove inode from in-memory list if
2998                          * addition to on disk orphan list failed. Stray orphan
2999                          * list entries can cause panics at unmount time.
3000                          */
3001                         mutex_lock(&sbi->s_orphan_lock);
3002                         list_del_init(&EXT4_I(inode)->i_orphan);
3003                         mutex_unlock(&sbi->s_orphan_lock);
3004                 }
3005         } else
3006                 brelse(iloc.bh);
3007
3008         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3009         jbd_debug(4, "orphan inode %lu will point to %d\n",
3010                         inode->i_ino, NEXT_ORPHAN(inode));
3011 out:
3012         ext4_std_error(sb, err);
3013         return err;
3014 }
3015
3016 /*
3017  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3018  * of such inodes stored on disk, because it is finally being cleaned up.
3019  */
3020 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3021 {
3022         struct list_head *prev;
3023         struct ext4_inode_info *ei = EXT4_I(inode);
3024         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3025         __u32 ino_next;
3026         struct ext4_iloc iloc;
3027         int err = 0;
3028
3029         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3030                 return 0;
3031
3032         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3033                      !inode_is_locked(inode));
3034         /* Do this quick check before taking global s_orphan_lock. */
3035         if (list_empty(&ei->i_orphan))
3036                 return 0;
3037
3038         if (handle) {
3039                 /* Grab inode buffer early before taking global s_orphan_lock */
3040                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3041         }
3042
3043         mutex_lock(&sbi->s_orphan_lock);
3044         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3045
3046         prev = ei->i_orphan.prev;
3047         list_del_init(&ei->i_orphan);
3048
3049         /* If we're on an error path, we may not have a valid
3050          * transaction handle with which to update the orphan list on
3051          * disk, but we still need to remove the inode from the linked
3052          * list in memory. */
3053         if (!handle || err) {
3054                 mutex_unlock(&sbi->s_orphan_lock);
3055                 goto out_err;
3056         }
3057
3058         ino_next = NEXT_ORPHAN(inode);
3059         if (prev == &sbi->s_orphan) {
3060                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3061                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3062                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3063                 if (err) {
3064                         mutex_unlock(&sbi->s_orphan_lock);
3065                         goto out_brelse;
3066                 }
3067                 lock_buffer(sbi->s_sbh);
3068                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3069                 ext4_superblock_csum_set(inode->i_sb);
3070                 unlock_buffer(sbi->s_sbh);
3071                 mutex_unlock(&sbi->s_orphan_lock);
3072                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3073         } else {
3074                 struct ext4_iloc iloc2;
3075                 struct inode *i_prev =
3076                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3077
3078                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3079                           i_prev->i_ino, ino_next);
3080                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3081                 if (err) {
3082                         mutex_unlock(&sbi->s_orphan_lock);
3083                         goto out_brelse;
3084                 }
3085                 NEXT_ORPHAN(i_prev) = ino_next;
3086                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3087                 mutex_unlock(&sbi->s_orphan_lock);
3088         }
3089         if (err)
3090                 goto out_brelse;
3091         NEXT_ORPHAN(inode) = 0;
3092         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3093 out_err:
3094         ext4_std_error(inode->i_sb, err);
3095         return err;
3096
3097 out_brelse:
3098         brelse(iloc.bh);
3099         goto out_err;
3100 }
3101
3102 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3103 {
3104         int retval;
3105         struct inode *inode;
3106         struct buffer_head *bh;
3107         struct ext4_dir_entry_2 *de;
3108         handle_t *handle = NULL;
3109
3110         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3111                 return -EIO;
3112
3113         /* Initialize quotas before so that eventual writes go in
3114          * separate transaction */
3115         retval = dquot_initialize(dir);
3116         if (retval)
3117                 return retval;
3118         retval = dquot_initialize(d_inode(dentry));
3119         if (retval)
3120                 return retval;
3121
3122         retval = -ENOENT;
3123         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3124         if (IS_ERR(bh))
3125                 return PTR_ERR(bh);
3126         if (!bh)
3127                 goto end_rmdir;
3128
3129         inode = d_inode(dentry);
3130
3131         retval = -EFSCORRUPTED;
3132         if (le32_to_cpu(de->inode) != inode->i_ino)
3133                 goto end_rmdir;
3134
3135         retval = -ENOTEMPTY;
3136         if (!ext4_empty_dir(inode))
3137                 goto end_rmdir;
3138
3139         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3140                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3141         if (IS_ERR(handle)) {
3142                 retval = PTR_ERR(handle);
3143                 handle = NULL;
3144                 goto end_rmdir;
3145         }
3146
3147         if (IS_DIRSYNC(dir))
3148                 ext4_handle_sync(handle);
3149
3150         retval = ext4_delete_entry(handle, dir, de, bh);
3151         if (retval)
3152                 goto end_rmdir;
3153         if (!EXT4_DIR_LINK_EMPTY(inode))
3154                 ext4_warning_inode(inode,
3155                              "empty directory '%.*s' has too many links (%u)",
3156                              dentry->d_name.len, dentry->d_name.name,
3157                              inode->i_nlink);
3158         inode_inc_iversion(inode);
3159         clear_nlink(inode);
3160         /* There's no need to set i_disksize: the fact that i_nlink is
3161          * zero will ensure that the right thing happens during any
3162          * recovery. */
3163         inode->i_size = 0;
3164         ext4_orphan_add(handle, inode);
3165         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3166         retval = ext4_mark_inode_dirty(handle, inode);
3167         if (retval)
3168                 goto end_rmdir;
3169         ext4_dec_count(dir);
3170         ext4_update_dx_flag(dir);
3171         ext4_fc_track_unlink(handle, dentry);
3172         retval = ext4_mark_inode_dirty(handle, dir);
3173
3174 #ifdef CONFIG_UNICODE
3175         /* VFS negative dentries are incompatible with Encoding and
3176          * Case-insensitiveness. Eventually we'll want avoid
3177          * invalidating the dentries here, alongside with returning the
3178          * negative dentries at ext4_lookup(), when it is better
3179          * supported by the VFS for the CI case.
3180          */
3181         if (IS_CASEFOLDED(dir))
3182                 d_invalidate(dentry);
3183 #endif
3184
3185 end_rmdir:
3186         brelse(bh);
3187         if (handle)
3188                 ext4_journal_stop(handle);
3189         return retval;
3190 }
3191
3192 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3193                   struct inode *inode)
3194 {
3195         int retval = -ENOENT;
3196         struct buffer_head *bh;
3197         struct ext4_dir_entry_2 *de;
3198         int skip_remove_dentry = 0;
3199
3200         bh = ext4_find_entry(dir, d_name, &de, NULL);
3201         if (IS_ERR(bh))
3202                 return PTR_ERR(bh);
3203
3204         if (!bh)
3205                 return -ENOENT;
3206
3207         if (le32_to_cpu(de->inode) != inode->i_ino) {
3208                 /*
3209                  * It's okay if we find dont find dentry which matches
3210                  * the inode. That's because it might have gotten
3211                  * renamed to a different inode number
3212                  */
3213                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3214                         skip_remove_dentry = 1;
3215                 else
3216                         goto out;
3217         }
3218
3219         if (IS_DIRSYNC(dir))
3220                 ext4_handle_sync(handle);
3221
3222         if (!skip_remove_dentry) {
3223                 retval = ext4_delete_entry(handle, dir, de, bh);
3224                 if (retval)
3225                         goto out;
3226                 dir->i_ctime = dir->i_mtime = current_time(dir);
3227                 ext4_update_dx_flag(dir);
3228                 retval = ext4_mark_inode_dirty(handle, dir);
3229                 if (retval)
3230                         goto out;
3231         } else {
3232                 retval = 0;
3233         }
3234         if (inode->i_nlink == 0)
3235                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3236                                    d_name->len, d_name->name);
3237         else
3238                 drop_nlink(inode);
3239         if (!inode->i_nlink)
3240                 ext4_orphan_add(handle, inode);
3241         inode->i_ctime = current_time(inode);
3242         retval = ext4_mark_inode_dirty(handle, inode);
3243
3244 out:
3245         brelse(bh);
3246         return retval;
3247 }
3248
3249 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3250 {
3251         handle_t *handle;
3252         int retval;
3253
3254         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3255                 return -EIO;
3256
3257         trace_ext4_unlink_enter(dir, dentry);
3258         /*
3259          * Initialize quotas before so that eventual writes go
3260          * in separate transaction
3261          */
3262         retval = dquot_initialize(dir);
3263         if (retval)
3264                 goto out_trace;
3265         retval = dquot_initialize(d_inode(dentry));
3266         if (retval)
3267                 goto out_trace;
3268
3269         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3270                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3271         if (IS_ERR(handle)) {
3272                 retval = PTR_ERR(handle);
3273                 goto out_trace;
3274         }
3275
3276         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3277         if (!retval)
3278                 ext4_fc_track_unlink(handle, dentry);
3279 #ifdef CONFIG_UNICODE
3280         /* VFS negative dentries are incompatible with Encoding and
3281          * Case-insensitiveness. Eventually we'll want avoid
3282          * invalidating the dentries here, alongside with returning the
3283          * negative dentries at ext4_lookup(), when it is  better
3284          * supported by the VFS for the CI case.
3285          */
3286         if (IS_CASEFOLDED(dir))
3287                 d_invalidate(dentry);
3288 #endif
3289         if (handle)
3290                 ext4_journal_stop(handle);
3291
3292 out_trace:
3293         trace_ext4_unlink_exit(dentry, retval);
3294         return retval;
3295 }
3296
3297 static int ext4_symlink(struct inode *dir,
3298                         struct dentry *dentry, const char *symname)
3299 {
3300         handle_t *handle;
3301         struct inode *inode;
3302         int err, len = strlen(symname);
3303         int credits;
3304         struct fscrypt_str disk_link;
3305
3306         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3307                 return -EIO;
3308
3309         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3310                                       &disk_link);
3311         if (err)
3312                 return err;
3313
3314         err = dquot_initialize(dir);
3315         if (err)
3316                 return err;
3317
3318         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3319                 /*
3320                  * For non-fast symlinks, we just allocate inode and put it on
3321                  * orphan list in the first transaction => we need bitmap,
3322                  * group descriptor, sb, inode block, quota blocks, and
3323                  * possibly selinux xattr blocks.
3324                  */
3325                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3326                           EXT4_XATTR_TRANS_BLOCKS;
3327         } else {
3328                 /*
3329                  * Fast symlink. We have to add entry to directory
3330                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3331                  * allocate new inode (bitmap, group descriptor, inode block,
3332                  * quota blocks, sb is already counted in previous macros).
3333                  */
3334                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3335                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3336         }
3337
3338         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3339                                             &dentry->d_name, 0, NULL,
3340                                             EXT4_HT_DIR, credits);
3341         handle = ext4_journal_current_handle();
3342         if (IS_ERR(inode)) {
3343                 if (handle)
3344                         ext4_journal_stop(handle);
3345                 return PTR_ERR(inode);
3346         }
3347
3348         if (IS_ENCRYPTED(inode)) {
3349                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3350                 if (err)
3351                         goto err_drop_inode;
3352                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3353         }
3354
3355         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3356                 if (!IS_ENCRYPTED(inode))
3357                         inode->i_op = &ext4_symlink_inode_operations;
3358                 inode_nohighmem(inode);
3359                 ext4_set_aops(inode);
3360                 /*
3361                  * We cannot call page_symlink() with transaction started
3362                  * because it calls into ext4_write_begin() which can wait
3363                  * for transaction commit if we are running out of space
3364                  * and thus we deadlock. So we have to stop transaction now
3365                  * and restart it when symlink contents is written.
3366                  *
3367                  * To keep fs consistent in case of crash, we have to put inode
3368                  * to orphan list in the mean time.
3369                  */
3370                 drop_nlink(inode);
3371                 err = ext4_orphan_add(handle, inode);
3372                 if (handle)
3373                         ext4_journal_stop(handle);
3374                 handle = NULL;
3375                 if (err)
3376                         goto err_drop_inode;
3377                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3378                 if (err)
3379                         goto err_drop_inode;
3380                 /*
3381                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3382                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3383                  */
3384                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3385                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3386                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3387                 if (IS_ERR(handle)) {
3388                         err = PTR_ERR(handle);
3389                         handle = NULL;
3390                         goto err_drop_inode;
3391                 }
3392                 set_nlink(inode, 1);
3393                 err = ext4_orphan_del(handle, inode);
3394                 if (err)
3395                         goto err_drop_inode;
3396         } else {
3397                 /* clear the extent format for fast symlink */
3398                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3399                 if (!IS_ENCRYPTED(inode)) {
3400                         inode->i_op = &ext4_fast_symlink_inode_operations;
3401                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3402                 }
3403                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3404                        disk_link.len);
3405                 inode->i_size = disk_link.len - 1;
3406         }
3407         EXT4_I(inode)->i_disksize = inode->i_size;
3408         err = ext4_add_nondir(handle, dentry, &inode);
3409         if (handle)
3410                 ext4_journal_stop(handle);
3411         if (inode)
3412                 iput(inode);
3413         goto out_free_encrypted_link;
3414
3415 err_drop_inode:
3416         if (handle)
3417                 ext4_journal_stop(handle);
3418         clear_nlink(inode);
3419         unlock_new_inode(inode);
3420         iput(inode);
3421 out_free_encrypted_link:
3422         if (disk_link.name != (unsigned char *)symname)
3423                 kfree(disk_link.name);
3424         return err;
3425 }
3426
3427 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3428 {
3429         handle_t *handle;
3430         int err, retries = 0;
3431 retry:
3432         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3433                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3434                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3435         if (IS_ERR(handle))
3436                 return PTR_ERR(handle);
3437
3438         if (IS_DIRSYNC(dir))
3439                 ext4_handle_sync(handle);
3440
3441         inode->i_ctime = current_time(inode);
3442         ext4_inc_count(inode);
3443         ihold(inode);
3444
3445         err = ext4_add_entry(handle, dentry, inode);
3446         if (!err) {
3447                 err = ext4_mark_inode_dirty(handle, inode);
3448                 /* this can happen only for tmpfile being
3449                  * linked the first time
3450                  */
3451                 if (inode->i_nlink == 1)
3452                         ext4_orphan_del(handle, inode);
3453                 d_instantiate(dentry, inode);
3454                 ext4_fc_track_link(handle, dentry);
3455         } else {
3456                 drop_nlink(inode);
3457                 iput(inode);
3458         }
3459         ext4_journal_stop(handle);
3460         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3461                 goto retry;
3462         return err;
3463 }
3464
3465 static int ext4_link(struct dentry *old_dentry,
3466                      struct inode *dir, struct dentry *dentry)
3467 {
3468         struct inode *inode = d_inode(old_dentry);
3469         int err;
3470
3471         if (inode->i_nlink >= EXT4_LINK_MAX)
3472                 return -EMLINK;
3473
3474         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3475         if (err)
3476                 return err;
3477
3478         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3479             (!projid_eq(EXT4_I(dir)->i_projid,
3480                         EXT4_I(old_dentry->d_inode)->i_projid)))
3481                 return -EXDEV;
3482
3483         err = dquot_initialize(dir);
3484         if (err)
3485                 return err;
3486         return __ext4_link(dir, inode, dentry);
3487 }
3488
3489 /*
3490  * Try to find buffer head where contains the parent block.
3491  * It should be the inode block if it is inlined or the 1st block
3492  * if it is a normal dir.
3493  */
3494 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3495                                         struct inode *inode,
3496                                         int *retval,
3497                                         struct ext4_dir_entry_2 **parent_de,
3498                                         int *inlined)
3499 {
3500         struct buffer_head *bh;
3501
3502         if (!ext4_has_inline_data(inode)) {
3503                 /* The first directory block must not be a hole, so
3504                  * treat it as DIRENT_HTREE
3505                  */
3506                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3507                 if (IS_ERR(bh)) {
3508                         *retval = PTR_ERR(bh);
3509                         return NULL;
3510                 }
3511                 *parent_de = ext4_next_entry(
3512                                         (struct ext4_dir_entry_2 *)bh->b_data,
3513                                         inode->i_sb->s_blocksize);
3514                 return bh;
3515         }
3516
3517         *inlined = 1;
3518         return ext4_get_first_inline_block(inode, parent_de, retval);
3519 }
3520
3521 struct ext4_renament {
3522         struct inode *dir;
3523         struct dentry *dentry;
3524         struct inode *inode;
3525         bool is_dir;
3526         int dir_nlink_delta;
3527
3528         /* entry for "dentry" */
3529         struct buffer_head *bh;
3530         struct ext4_dir_entry_2 *de;
3531         int inlined;
3532
3533         /* entry for ".." in inode if it's a directory */
3534         struct buffer_head *dir_bh;
3535         struct ext4_dir_entry_2 *parent_de;
3536         int dir_inlined;
3537 };
3538
3539 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3540 {
3541         int retval;
3542
3543         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3544                                               &retval, &ent->parent_de,
3545                                               &ent->dir_inlined);
3546         if (!ent->dir_bh)
3547                 return retval;
3548         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3549                 return -EFSCORRUPTED;
3550         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3551         return ext4_journal_get_write_access(handle, ent->dir_bh);
3552 }
3553
3554 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3555                                   unsigned dir_ino)
3556 {
3557         int retval;
3558
3559         ent->parent_de->inode = cpu_to_le32(dir_ino);
3560         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3561         if (!ent->dir_inlined) {
3562                 if (is_dx(ent->inode)) {
3563                         retval = ext4_handle_dirty_dx_node(handle,
3564                                                            ent->inode,
3565                                                            ent->dir_bh);
3566                 } else {
3567                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3568                                                             ent->dir_bh);
3569                 }
3570         } else {
3571                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3572         }
3573         if (retval) {
3574                 ext4_std_error(ent->dir->i_sb, retval);
3575                 return retval;
3576         }
3577         return 0;
3578 }
3579
3580 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3581                        unsigned ino, unsigned file_type)
3582 {
3583         int retval, retval2;
3584
3585         BUFFER_TRACE(ent->bh, "get write access");
3586         retval = ext4_journal_get_write_access(handle, ent->bh);
3587         if (retval)
3588                 return retval;
3589         ent->de->inode = cpu_to_le32(ino);
3590         if (ext4_has_feature_filetype(ent->dir->i_sb))
3591                 ent->de->file_type = file_type;
3592         inode_inc_iversion(ent->dir);
3593         ent->dir->i_ctime = ent->dir->i_mtime =
3594                 current_time(ent->dir);
3595         retval = ext4_mark_inode_dirty(handle, ent->dir);
3596         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3597         if (!ent->inlined) {
3598                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3599                 if (unlikely(retval2)) {
3600                         ext4_std_error(ent->dir->i_sb, retval2);
3601                         return retval2;
3602                 }
3603         }
3604         brelse(ent->bh);
3605         ent->bh = NULL;
3606
3607         return retval;
3608 }
3609
3610 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3611                                   const struct qstr *d_name)
3612 {
3613         int retval = -ENOENT;
3614         struct buffer_head *bh;
3615         struct ext4_dir_entry_2 *de;
3616
3617         bh = ext4_find_entry(dir, d_name, &de, NULL);
3618         if (IS_ERR(bh))
3619                 return PTR_ERR(bh);
3620         if (bh) {
3621                 retval = ext4_delete_entry(handle, dir, de, bh);
3622                 brelse(bh);
3623         }
3624         return retval;
3625 }
3626
3627 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3628                                int force_reread)
3629 {
3630         int retval;
3631         /*
3632          * ent->de could have moved from under us during htree split, so make
3633          * sure that we are deleting the right entry.  We might also be pointing
3634          * to a stale entry in the unused part of ent->bh so just checking inum
3635          * and the name isn't enough.
3636          */
3637         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3638             ent->de->name_len != ent->dentry->d_name.len ||
3639             strncmp(ent->de->name, ent->dentry->d_name.name,
3640                     ent->de->name_len) ||
3641             force_reread) {
3642                 retval = ext4_find_delete_entry(handle, ent->dir,
3643                                                 &ent->dentry->d_name);
3644         } else {
3645                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3646                 if (retval == -ENOENT) {
3647                         retval = ext4_find_delete_entry(handle, ent->dir,
3648                                                         &ent->dentry->d_name);
3649                 }
3650         }
3651
3652         if (retval) {
3653                 ext4_warning_inode(ent->dir,
3654                                    "Deleting old file: nlink %d, error=%d",
3655                                    ent->dir->i_nlink, retval);
3656         }
3657 }
3658
3659 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3660 {
3661         if (ent->dir_nlink_delta) {
3662                 if (ent->dir_nlink_delta == -1)
3663                         ext4_dec_count(ent->dir);
3664                 else
3665                         ext4_inc_count(ent->dir);
3666                 ext4_mark_inode_dirty(handle, ent->dir);
3667         }
3668 }
3669
3670 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3671                                               int credits, handle_t **h)
3672 {
3673         struct inode *wh;
3674         handle_t *handle;
3675         int retries = 0;
3676
3677         /*
3678          * for inode block, sb block, group summaries,
3679          * and inode bitmap
3680          */
3681         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3682                     EXT4_XATTR_TRANS_BLOCKS + 4);
3683 retry:
3684         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3685                                          &ent->dentry->d_name, 0, NULL,
3686                                          EXT4_HT_DIR, credits);
3687
3688         handle = ext4_journal_current_handle();
3689         if (IS_ERR(wh)) {
3690                 if (handle)
3691                         ext4_journal_stop(handle);
3692                 if (PTR_ERR(wh) == -ENOSPC &&
3693                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3694                         goto retry;
3695         } else {
3696                 *h = handle;
3697                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3698                 wh->i_op = &ext4_special_inode_operations;
3699         }
3700         return wh;
3701 }
3702
3703 /*
3704  * Anybody can rename anything with this: the permission checks are left to the
3705  * higher-level routines.
3706  *
3707  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3708  * while new_{dentry,inode) refers to the destination dentry/inode
3709  * This comes from rename(const char *oldpath, const char *newpath)
3710  */
3711 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3712                        struct inode *new_dir, struct dentry *new_dentry,
3713                        unsigned int flags)
3714 {
3715         handle_t *handle = NULL;
3716         struct ext4_renament old = {
3717                 .dir = old_dir,
3718                 .dentry = old_dentry,
3719                 .inode = d_inode(old_dentry),
3720         };
3721         struct ext4_renament new = {
3722                 .dir = new_dir,
3723                 .dentry = new_dentry,
3724                 .inode = d_inode(new_dentry),
3725         };
3726         int force_reread;
3727         int retval;
3728         struct inode *whiteout = NULL;
3729         int credits;
3730         u8 old_file_type;
3731
3732         if (new.inode && new.inode->i_nlink == 0) {
3733                 EXT4_ERROR_INODE(new.inode,
3734                                  "target of rename is already freed");
3735                 return -EFSCORRUPTED;
3736         }
3737
3738         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3739             (!projid_eq(EXT4_I(new_dir)->i_projid,
3740                         EXT4_I(old_dentry->d_inode)->i_projid)))
3741                 return -EXDEV;
3742
3743         retval = dquot_initialize(old.dir);
3744         if (retval)
3745                 return retval;
3746         retval = dquot_initialize(new.dir);
3747         if (retval)
3748                 return retval;
3749
3750         /* Initialize quotas before so that eventual writes go
3751          * in separate transaction */
3752         if (new.inode) {
3753                 retval = dquot_initialize(new.inode);
3754                 if (retval)
3755                         return retval;
3756         }
3757
3758         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3759         if (IS_ERR(old.bh))
3760                 return PTR_ERR(old.bh);
3761         /*
3762          *  Check for inode number is _not_ due to possible IO errors.
3763          *  We might rmdir the source, keep it as pwd of some process
3764          *  and merrily kill the link to whatever was created under the
3765          *  same name. Goodbye sticky bit ;-<
3766          */
3767         retval = -ENOENT;
3768         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3769                 goto end_rename;
3770
3771         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3772                                  &new.de, &new.inlined);
3773         if (IS_ERR(new.bh)) {
3774                 retval = PTR_ERR(new.bh);
3775                 new.bh = NULL;
3776                 goto end_rename;
3777         }
3778         if (new.bh) {
3779                 if (!new.inode) {
3780                         brelse(new.bh);
3781                         new.bh = NULL;
3782                 }
3783         }
3784         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3785                 ext4_alloc_da_blocks(old.inode);
3786
3787         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3788                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3789         if (!(flags & RENAME_WHITEOUT)) {
3790                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3791                 if (IS_ERR(handle)) {
3792                         retval = PTR_ERR(handle);
3793                         handle = NULL;
3794                         goto end_rename;
3795                 }
3796         } else {
3797                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3798                 if (IS_ERR(whiteout)) {
3799                         retval = PTR_ERR(whiteout);
3800                         whiteout = NULL;
3801                         goto end_rename;
3802                 }
3803         }
3804
3805         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3806                 ext4_handle_sync(handle);
3807
3808         if (S_ISDIR(old.inode->i_mode)) {
3809                 if (new.inode) {
3810                         retval = -ENOTEMPTY;
3811                         if (!ext4_empty_dir(new.inode))
3812                                 goto end_rename;
3813                 } else {
3814                         retval = -EMLINK;
3815                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3816                                 goto end_rename;
3817                 }
3818                 retval = ext4_rename_dir_prepare(handle, &old);
3819                 if (retval)
3820                         goto end_rename;
3821         }
3822         /*
3823          * If we're renaming a file within an inline_data dir and adding or
3824          * setting the new dirent causes a conversion from inline_data to
3825          * extents/blockmap, we need to force the dirent delete code to
3826          * re-read the directory, or else we end up trying to delete a dirent
3827          * from what is now the extent tree root (or a block map).
3828          */
3829         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3830                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3831
3832         old_file_type = old.de->file_type;
3833         if (whiteout) {
3834                 /*
3835                  * Do this before adding a new entry, so the old entry is sure
3836                  * to be still pointing to the valid old entry.
3837                  */
3838                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3839                                      EXT4_FT_CHRDEV);
3840                 if (retval)
3841                         goto end_rename;
3842                 retval = ext4_mark_inode_dirty(handle, whiteout);
3843                 if (unlikely(retval))
3844                         goto end_rename;
3845         }
3846         if (!new.bh) {
3847                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3848                 if (retval)
3849                         goto end_rename;
3850         } else {
3851                 retval = ext4_setent(handle, &new,
3852                                      old.inode->i_ino, old_file_type);
3853                 if (retval)
3854                         goto end_rename;
3855         }
3856         if (force_reread)
3857                 force_reread = !ext4_test_inode_flag(new.dir,
3858                                                      EXT4_INODE_INLINE_DATA);
3859
3860         /*
3861          * Like most other Unix systems, set the ctime for inodes on a
3862          * rename.
3863          */
3864         old.inode->i_ctime = current_time(old.inode);
3865         retval = ext4_mark_inode_dirty(handle, old.inode);
3866         if (unlikely(retval))
3867                 goto end_rename;
3868
3869         if (!whiteout) {
3870                 /*
3871                  * ok, that's it
3872                  */
3873                 ext4_rename_delete(handle, &old, force_reread);
3874         }
3875
3876         if (new.inode) {
3877                 ext4_dec_count(new.inode);
3878                 new.inode->i_ctime = current_time(new.inode);
3879         }
3880         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3881         ext4_update_dx_flag(old.dir);
3882         if (old.dir_bh) {
3883                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3884                 if (retval)
3885                         goto end_rename;
3886
3887                 ext4_dec_count(old.dir);
3888                 if (new.inode) {
3889                         /* checked ext4_empty_dir above, can't have another
3890                          * parent, ext4_dec_count() won't work for many-linked
3891                          * dirs */
3892                         clear_nlink(new.inode);
3893                 } else {
3894                         ext4_inc_count(new.dir);
3895                         ext4_update_dx_flag(new.dir);
3896                         retval = ext4_mark_inode_dirty(handle, new.dir);
3897                         if (unlikely(retval))
3898                                 goto end_rename;
3899                 }
3900         }
3901         retval = ext4_mark_inode_dirty(handle, old.dir);
3902         if (unlikely(retval))
3903                 goto end_rename;
3904
3905         if (S_ISDIR(old.inode->i_mode)) {
3906                 /*
3907                  * We disable fast commits here that's because the
3908                  * replay code is not yet capable of changing dot dot
3909                  * dirents in directories.
3910                  */
3911                 ext4_fc_mark_ineligible(old.inode->i_sb,
3912                         EXT4_FC_REASON_RENAME_DIR);
3913         } else {
3914                 if (new.inode)
3915                         ext4_fc_track_unlink(handle, new.dentry);
3916                 __ext4_fc_track_link(handle, old.inode, new.dentry);
3917                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3918         }
3919
3920         if (new.inode) {
3921                 retval = ext4_mark_inode_dirty(handle, new.inode);
3922                 if (unlikely(retval))
3923                         goto end_rename;
3924                 if (!new.inode->i_nlink)
3925                         ext4_orphan_add(handle, new.inode);
3926         }
3927         retval = 0;
3928
3929 end_rename:
3930         brelse(old.dir_bh);
3931         brelse(old.bh);
3932         brelse(new.bh);
3933         if (whiteout) {
3934                 if (retval)
3935                         drop_nlink(whiteout);
3936                 unlock_new_inode(whiteout);
3937                 iput(whiteout);
3938         }
3939         if (handle)
3940                 ext4_journal_stop(handle);
3941         return retval;
3942 }
3943
3944 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3945                              struct inode *new_dir, struct dentry *new_dentry)
3946 {
3947         handle_t *handle = NULL;
3948         struct ext4_renament old = {
3949                 .dir = old_dir,
3950                 .dentry = old_dentry,
3951                 .inode = d_inode(old_dentry),
3952         };
3953         struct ext4_renament new = {
3954                 .dir = new_dir,
3955                 .dentry = new_dentry,
3956                 .inode = d_inode(new_dentry),
3957         };
3958         u8 new_file_type;
3959         int retval;
3960         struct timespec64 ctime;
3961
3962         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3963              !projid_eq(EXT4_I(new_dir)->i_projid,
3964                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3965             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3966              !projid_eq(EXT4_I(old_dir)->i_projid,
3967                         EXT4_I(new_dentry->d_inode)->i_projid)))
3968                 return -EXDEV;
3969
3970         retval = dquot_initialize(old.dir);
3971         if (retval)
3972                 return retval;
3973         retval = dquot_initialize(new.dir);
3974         if (retval)
3975                 return retval;
3976
3977         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3978                                  &old.de, &old.inlined);
3979         if (IS_ERR(old.bh))
3980                 return PTR_ERR(old.bh);
3981         /*
3982          *  Check for inode number is _not_ due to possible IO errors.
3983          *  We might rmdir the source, keep it as pwd of some process
3984          *  and merrily kill the link to whatever was created under the
3985          *  same name. Goodbye sticky bit ;-<
3986          */
3987         retval = -ENOENT;
3988         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3989                 goto end_rename;
3990
3991         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3992                                  &new.de, &new.inlined);
3993         if (IS_ERR(new.bh)) {
3994                 retval = PTR_ERR(new.bh);
3995                 new.bh = NULL;
3996                 goto end_rename;
3997         }
3998
3999         /* RENAME_EXCHANGE case: old *and* new must both exist */
4000         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4001                 goto end_rename;
4002
4003         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4004                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4005                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4006         if (IS_ERR(handle)) {
4007                 retval = PTR_ERR(handle);
4008                 handle = NULL;
4009                 goto end_rename;
4010         }
4011
4012         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4013                 ext4_handle_sync(handle);
4014
4015         if (S_ISDIR(old.inode->i_mode)) {
4016                 old.is_dir = true;
4017                 retval = ext4_rename_dir_prepare(handle, &old);
4018                 if (retval)
4019                         goto end_rename;
4020         }
4021         if (S_ISDIR(new.inode->i_mode)) {
4022                 new.is_dir = true;
4023                 retval = ext4_rename_dir_prepare(handle, &new);
4024                 if (retval)
4025                         goto end_rename;
4026         }
4027
4028         /*
4029          * Other than the special case of overwriting a directory, parents'
4030          * nlink only needs to be modified if this is a cross directory rename.
4031          */
4032         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4033                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4034                 new.dir_nlink_delta = -old.dir_nlink_delta;
4035                 retval = -EMLINK;
4036                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4037                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4038                         goto end_rename;
4039         }
4040
4041         new_file_type = new.de->file_type;
4042         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4043         if (retval)
4044                 goto end_rename;
4045
4046         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4047         if (retval)
4048                 goto end_rename;
4049
4050         /*
4051          * Like most other Unix systems, set the ctime for inodes on a
4052          * rename.
4053          */
4054         ctime = current_time(old.inode);
4055         old.inode->i_ctime = ctime;
4056         new.inode->i_ctime = ctime;
4057         retval = ext4_mark_inode_dirty(handle, old.inode);
4058         if (unlikely(retval))
4059                 goto end_rename;
4060         retval = ext4_mark_inode_dirty(handle, new.inode);
4061         if (unlikely(retval))
4062                 goto end_rename;
4063         ext4_fc_mark_ineligible(new.inode->i_sb,
4064                                 EXT4_FC_REASON_CROSS_RENAME);
4065         if (old.dir_bh) {
4066                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4067                 if (retval)
4068                         goto end_rename;
4069         }
4070         if (new.dir_bh) {
4071                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4072                 if (retval)
4073                         goto end_rename;
4074         }
4075         ext4_update_dir_count(handle, &old);
4076         ext4_update_dir_count(handle, &new);
4077         retval = 0;
4078
4079 end_rename:
4080         brelse(old.dir_bh);
4081         brelse(new.dir_bh);
4082         brelse(old.bh);
4083         brelse(new.bh);
4084         if (handle)
4085                 ext4_journal_stop(handle);
4086         return retval;
4087 }
4088
4089 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4090                         struct inode *new_dir, struct dentry *new_dentry,
4091                         unsigned int flags)
4092 {
4093         int err;
4094
4095         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4096                 return -EIO;
4097
4098         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4099                 return -EINVAL;
4100
4101         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4102                                      flags);
4103         if (err)
4104                 return err;
4105
4106         if (flags & RENAME_EXCHANGE) {
4107                 return ext4_cross_rename(old_dir, old_dentry,
4108                                          new_dir, new_dentry);
4109         }
4110
4111         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4112 }
4113
4114 /*
4115  * directories can handle most operations...
4116  */
4117 const struct inode_operations ext4_dir_inode_operations = {
4118         .create         = ext4_create,
4119         .lookup         = ext4_lookup,
4120         .link           = ext4_link,
4121         .unlink         = ext4_unlink,
4122         .symlink        = ext4_symlink,
4123         .mkdir          = ext4_mkdir,
4124         .rmdir          = ext4_rmdir,
4125         .mknod          = ext4_mknod,
4126         .tmpfile        = ext4_tmpfile,
4127         .rename         = ext4_rename2,
4128         .setattr        = ext4_setattr,
4129         .getattr        = ext4_getattr,
4130         .listxattr      = ext4_listxattr,
4131         .get_acl        = ext4_get_acl,
4132         .set_acl        = ext4_set_acl,
4133         .fiemap         = ext4_fiemap,
4134 };
4135
4136 const struct inode_operations ext4_special_inode_operations = {
4137         .setattr        = ext4_setattr,
4138         .getattr        = ext4_getattr,
4139         .listxattr      = ext4_listxattr,
4140         .get_acl        = ext4_get_acl,
4141         .set_acl        = ext4_set_acl,
4142 };