fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 typedef enum {
  86         EITHER, INDEX, DIRENT
  87 } dirblock_type_t;
  88
  89 #define ext4_read_dirblock(inode, block, type) \
  90         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
  91
  92 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
  93                                                 ext4_lblk_t block,
  94                                                 dirblock_type_t type,
  95                                                 const char *func,
  96                                                 unsigned int line)
  97 {
  98         struct buffer_head *bh;
  99         struct ext4_dir_entry *dirent;
 100         int is_dx_block = 0;
 101
 102         bh = ext4_bread(NULL, inode, block, 0);
 103         if (IS_ERR(bh)) {
 104                 __ext4_warning(inode->i_sb, func, line,
 105                                "inode #%lu: lblock %lu: comm %s: "
 106                                "error %ld reading directory block",
 107                                inode->i_ino, (unsigned long)block,
 108                                current->comm, PTR_ERR(bh));
 109
 110                 return bh;
 111         }
 112         if (!bh) {
 113                 ext4_error_inode(inode, func, line, block,
 114                                  "Directory hole found");
 115                 return ERR_PTR(-EFSCORRUPTED);
 116         }
 117         dirent = (struct ext4_dir_entry *) bh->b_data;
 118         /* Determine whether or not we have an index block */
 119         if (is_dx(inode)) {
 120                 if (block == 0)
 121                         is_dx_block = 1;
 122                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 123                                                 inode->i_sb->s_blocksize) ==
 124                          inode->i_sb->s_blocksize)
 125                         is_dx_block = 1;
 126         }
 127         if (!is_dx_block && type == INDEX) {
 128                 ext4_error_inode(inode, func, line, block,
 129                        "directory leaf block found instead of index block");
 130                 brelse(bh);
 131                 return ERR_PTR(-EFSCORRUPTED);
 132         }
 133         if (!ext4_has_metadata_csum(inode->i_sb) ||
 134             buffer_verified(bh))
 135                 return bh;
 136
 137         /*
 138          * An empty leaf block can get mistaken for a index block; for
 139          * this reason, we can only check the index checksum when the
 140          * caller is sure it should be an index block.
 141          */
 142         if (is_dx_block && type == INDEX) {
 143                 if (ext4_dx_csum_verify(inode, dirent))
 144                         set_buffer_verified(bh);
 145                 else {
 146                         ext4_error_inode(inode, func, line, block,
 147                                          "Directory index failed checksum");
 148                         brelse(bh);
 149                         return ERR_PTR(-EFSBADCRC);
 150                 }
 151         }
 152         if (!is_dx_block) {
 153                 if (ext4_dirent_csum_verify(inode, dirent))
 154                         set_buffer_verified(bh);
 155                 else {
 156                         ext4_error_inode(inode, func, line, block,
 157                                          "Directory block failed checksum");
 158                         brelse(bh);
 159                         return ERR_PTR(-EFSBADCRC);
 160                 }
 161         }
 162         return bh;
 163 }
 164
 165 #ifndef assert
 166 #define assert(test) J_ASSERT(test)
 167 #endif
 168
 169 #ifdef DX_DEBUG
 170 #define dxtrace(command) command
 171 #else
 172 #define dxtrace(command)
 173 #endif
 174
 175 struct fake_dirent
 176 {
 177         __le32 inode;
 178         __le16 rec_len;
 179         u8 name_len;
 180         u8 file_type;
 181 };
 182
 183 struct dx_countlimit
 184 {
 185         __le16 limit;
 186         __le16 count;
 187 };
 188
 189 struct dx_entry
 190 {
 191         __le32 hash;
 192         __le32 block;
 193 };
 194
 195 /*
 196  * dx_root_info is laid out so that if it should somehow get overlaid by a
 197  * dirent the two low bits of the hash version will be zero.  Therefore, the
 198  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 199  */
 200
 201 struct dx_root
 202 {
 203         struct fake_dirent dot;
 204         char dot_name[4];
 205         struct fake_dirent dotdot;
 206         char dotdot_name[4];
 207         struct dx_root_info
 208         {
 209                 __le32 reserved_zero;
 210                 u8 hash_version;
 211                 u8 info_length; /* 8 */
 212                 u8 indirect_levels;
 213                 u8 unused_flags;
 214         }
 215         info;
 216         struct dx_entry entries[0];
 217 };
 218
 219 struct dx_node
 220 {
 221         struct fake_dirent fake;
 222         struct dx_entry entries[0];
 223 };
 224
 225
 226 struct dx_frame
 227 {
 228         struct buffer_head *bh;
 229         struct dx_entry *entries;
 230         struct dx_entry *at;
 231 };
 232
 233 struct dx_map_entry
 234 {
 235         u32 hash;
 236         u16 offs;
 237         u16 size;
 238 };
 239
 240 /*
 241  * This goes at the end of each htree block.
 242  */
 243 struct dx_tail {
 244         u32 dt_reserved;
 245         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 246 };
 247
 248 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 249 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 250 static inline unsigned dx_get_hash(struct dx_entry *entry);
 251 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 252 static unsigned dx_get_count(struct dx_entry *entries);
 253 static unsigned dx_get_limit(struct dx_entry *entries);
 254 static void dx_set_count(struct dx_entry *entries, unsigned value);
 255 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 256 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 257 static unsigned dx_node_limit(struct inode *dir);
 258 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 259                                  struct inode *dir,
 260                                  struct dx_hash_info *hinfo,
 261                                  struct dx_frame *frame);
 262 static void dx_release(struct dx_frame *frames);
 263 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 264                        unsigned blocksize, struct dx_hash_info *hinfo,
 265                        struct dx_map_entry map[]);
 266 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 267 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 268                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 269 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 270 static void dx_insert_block(struct dx_frame *frame,
 271                                         u32 hash, ext4_lblk_t block);
 272 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 273                                  struct dx_frame *frame,
 274                                  struct dx_frame *frames,
 275                                  __u32 *start_hash);
 276 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 277                 struct ext4_filename *fname,
 278                 struct ext4_dir_entry_2 **res_dir);
 279 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 280                              struct inode *dir, struct inode *inode);
 281
 282 /* checksumming functions */
 283 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 284                             unsigned int blocksize)
 285 {
 286         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 287         t->det_rec_len = ext4_rec_len_to_disk(
 288                         sizeof(struct ext4_dir_entry_tail), blocksize);
 289         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 290 }
 291
 292 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 293 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 294                                                    struct ext4_dir_entry *de)
 295 {
 296         struct ext4_dir_entry_tail *t;
 297
 298 #ifdef PARANOID
 299         struct ext4_dir_entry *d, *top;
 300
 301         d = de;
 302         top = (struct ext4_dir_entry *)(((void *)de) +
 303                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 304                 sizeof(struct ext4_dir_entry_tail)));
 305         while (d < top && d->rec_len)
 306                 d = (struct ext4_dir_entry *)(((void *)d) +
 307                     le16_to_cpu(d->rec_len));
 308
 309         if (d != top)
 310                 return NULL;
 311
 312         t = (struct ext4_dir_entry_tail *)d;
 313 #else
 314         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 315 #endif
 316
 317         if (t->det_reserved_zero1 ||
 318             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 319             t->det_reserved_zero2 ||
 320             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 321                 return NULL;
 322
 323         return t;
 324 }
 325
 326 static __le32 ext4_dirent_csum(struct inode *inode,
 327                                struct ext4_dir_entry *dirent, int size)
 328 {
 329         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 330         struct ext4_inode_info *ei = EXT4_I(inode);
 331         __u32 csum;
 332
 333         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 334         return cpu_to_le32(csum);
 335 }
 336
 337 #define warn_no_space_for_csum(inode)                                   \
 338         __warn_no_space_for_csum((inode), __func__, __LINE__)
 339
 340 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 341                                      unsigned int line)
 342 {
 343         __ext4_warning_inode(inode, func, line,
 344                 "No space for directory leaf checksum. Please run e2fsck -D.");
 345 }
 346
 347 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 348 {
 349         struct ext4_dir_entry_tail *t;
 350
 351         if (!ext4_has_metadata_csum(inode->i_sb))
 352                 return 1;
 353
 354         t = get_dirent_tail(inode, dirent);
 355         if (!t) {
 356                 warn_no_space_for_csum(inode);
 357                 return 0;
 358         }
 359
 360         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 361                                                 (void *)t - (void *)dirent))
 362                 return 0;
 363
 364         return 1;
 365 }
 366
 367 static void ext4_dirent_csum_set(struct inode *inode,
 368                                  struct ext4_dir_entry *dirent)
 369 {
 370         struct ext4_dir_entry_tail *t;
 371
 372         if (!ext4_has_metadata_csum(inode->i_sb))
 373                 return;
 374
 375         t = get_dirent_tail(inode, dirent);
 376         if (!t) {
 377                 warn_no_space_for_csum(inode);
 378                 return;
 379         }
 380
 381         t->det_checksum = ext4_dirent_csum(inode, dirent,
 382                                            (void *)t - (void *)dirent);
 383 }
 384
 385 int ext4_handle_dirty_dirent_node(handle_t *handle,
 386                                   struct inode *inode,
 387                                   struct buffer_head *bh)
 388 {
 389         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 390         return ext4_handle_dirty_metadata(handle, inode, bh);
 391 }
 392
 393 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 394                                                struct ext4_dir_entry *dirent,
 395                                                int *offset)
 396 {
 397         struct ext4_dir_entry *dp;
 398         struct dx_root_info *root;
 399         int count_offset;
 400
 401         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 402                 count_offset = 8;
 403         else if (le16_to_cpu(dirent->rec_len) == 12) {
 404                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 405                 if (le16_to_cpu(dp->rec_len) !=
 406                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 407                         return NULL;
 408                 root = (struct dx_root_info *)(((void *)dp + 12));
 409                 if (root->reserved_zero ||
 410                     root->info_length != sizeof(struct dx_root_info))
 411                         return NULL;
 412                 count_offset = 32;
 413         } else
 414                 return NULL;
 415
 416         if (offset)
 417                 *offset = count_offset;
 418         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 419 }
 420
 421 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 422                            int count_offset, int count, struct dx_tail *t)
 423 {
 424         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 425         struct ext4_inode_info *ei = EXT4_I(inode);
 426         __u32 csum;
 427         int size;
 428         __u32 dummy_csum = 0;
 429         int offset = offsetof(struct dx_tail, dt_checksum);
 430
 431         size = count_offset + (count * sizeof(struct dx_entry));
 432         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 433         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 434         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 435
 436         return cpu_to_le32(csum);
 437 }
 438
 439 static int ext4_dx_csum_verify(struct inode *inode,
 440                                struct ext4_dir_entry *dirent)
 441 {
 442         struct dx_countlimit *c;
 443         struct dx_tail *t;
 444         int count_offset, limit, count;
 445
 446         if (!ext4_has_metadata_csum(inode->i_sb))
 447                 return 1;
 448
 449         c = get_dx_countlimit(inode, dirent, &count_offset);
 450         if (!c) {
 451                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 452                 return 0;
 453         }
 454         limit = le16_to_cpu(c->limit);
 455         count = le16_to_cpu(c->count);
 456         if (count_offset + (limit * sizeof(struct dx_entry)) >
 457             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 458                 warn_no_space_for_csum(inode);
 459                 return 0;
 460         }
 461         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 462
 463         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 464                                             count, t))
 465                 return 0;
 466         return 1;
 467 }
 468
 469 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 470 {
 471         struct dx_countlimit *c;
 472         struct dx_tail *t;
 473         int count_offset, limit, count;
 474
 475         if (!ext4_has_metadata_csum(inode->i_sb))
 476                 return;
 477
 478         c = get_dx_countlimit(inode, dirent, &count_offset);
 479         if (!c) {
 480                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 481                 return;
 482         }
 483         limit = le16_to_cpu(c->limit);
 484         count = le16_to_cpu(c->count);
 485         if (count_offset + (limit * sizeof(struct dx_entry)) >
 486             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 487                 warn_no_space_for_csum(inode);
 488                 return;
 489         }
 490         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 491
 492         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 493 }
 494
 495 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 496                                             struct inode *inode,
 497                                             struct buffer_head *bh)
 498 {
 499         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 500         return ext4_handle_dirty_metadata(handle, inode, bh);
 501 }
 502
 503 /*
 504  * p is at least 6 bytes before the end of page
 505  */
 506 static inline struct ext4_dir_entry_2 *
 507 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 508 {
 509         return (struct ext4_dir_entry_2 *)((char *)p +
 510                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 511 }
 512
 513 /*
 514  * Future: use high four bits of block for coalesce-on-delete flags
 515  * Mask them off for now.
 516  */
 517
 518 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 519 {
 520         return le32_to_cpu(entry->block) & 0x0fffffff;
 521 }
 522
 523 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 524 {
 525         entry->block = cpu_to_le32(value);
 526 }
 527
 528 static inline unsigned dx_get_hash(struct dx_entry *entry)
 529 {
 530         return le32_to_cpu(entry->hash);
 531 }
 532
 533 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 534 {
 535         entry->hash = cpu_to_le32(value);
 536 }
 537
 538 static inline unsigned dx_get_count(struct dx_entry *entries)
 539 {
 540         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 541 }
 542
 543 static inline unsigned dx_get_limit(struct dx_entry *entries)
 544 {
 545         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 546 }
 547
 548 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 549 {
 550         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 551 }
 552
 553 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 554 {
 555         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 556 }
 557
 558 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 559 {
 560         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 561                 EXT4_DIR_REC_LEN(2) - infosize;
 562
 563         if (ext4_has_metadata_csum(dir->i_sb))
 564                 entry_space -= sizeof(struct dx_tail);
 565         return entry_space / sizeof(struct dx_entry);
 566 }
 567
 568 static inline unsigned dx_node_limit(struct inode *dir)
 569 {
 570         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 571
 572         if (ext4_has_metadata_csum(dir->i_sb))
 573                 entry_space -= sizeof(struct dx_tail);
 574         return entry_space / sizeof(struct dx_entry);
 575 }
 576
 577 /*
 578  * Debug
 579  */
 580 #ifdef DX_DEBUG
 581 static void dx_show_index(char * label, struct dx_entry *entries)
 582 {
 583         int i, n = dx_get_count (entries);
 584         printk(KERN_DEBUG "%s index", label);
 585         for (i = 0; i < n; i++) {
 586                 printk(KERN_CONT " %x->%lu",
 587                        i ? dx_get_hash(entries + i) : 0,
 588                        (unsigned long)dx_get_block(entries + i));
 589         }
 590         printk(KERN_CONT "\n");
 591 }
 592
 593 struct stats
 594 {
 595         unsigned names;
 596         unsigned space;
 597         unsigned bcount;
 598 };
 599
 600 static struct stats dx_show_leaf(struct inode *dir,
 601                                 struct dx_hash_info *hinfo,
 602                                 struct ext4_dir_entry_2 *de,
 603                                 int size, int show_names)
 604 {
 605         unsigned names = 0, space = 0;
 606         char *base = (char *) de;
 607         struct dx_hash_info h = *hinfo;
 608
 609         printk("names: ");
 610         while ((char *) de < base + size)
 611         {
 612                 if (de->inode)
 613                 {
 614                         if (show_names)
 615                         {
 616 #ifdef CONFIG_FS_ENCRYPTION
 617                                 int len;
 618                                 char *name;
 619                                 struct fscrypt_str fname_crypto_str =
 620                                         FSTR_INIT(NULL, 0);
 621                                 int res = 0;
 622
 623                                 name  = de->name;
 624                                 len = de->name_len;
 625                                 if (IS_ENCRYPTED(dir))
 626                                         res = fscrypt_get_encryption_info(dir);
 627                                 if (res) {
 628                                         printk(KERN_WARNING "Error setting up"
 629                                                " fname crypto: %d\n", res);
 630                                 }
 631                                 if (!fscrypt_has_encryption_key(dir)) {
 632                                         /* Directory is not encrypted */
 633                                         ext4fs_dirhash(dir, de->name,
 634                                                 de->name_len, &h);
 635                                         printk("%*.s:(U)%x.%u ", len,
 636                                                name, h.hash,
 637                                                (unsigned) ((char *) de
 638                                                            - base));
 639                                 } else {
 640                                         struct fscrypt_str de_name =
 641                                                 FSTR_INIT(name, len);
 642
 643                                         /* Directory is encrypted */
 644                                         res = fscrypt_fname_alloc_buffer(
 645                                                 dir, len,
 646                                                 &fname_crypto_str);
 647                                         if (res)
 648                                                 printk(KERN_WARNING "Error "
 649                                                         "allocating crypto "
 650                                                         "buffer--skipping "
 651                                                         "crypto\n");
 652                                         res = fscrypt_fname_disk_to_usr(dir,
 653                                                 0, 0, &de_name,
 654                                                 &fname_crypto_str);
 655                                         if (res) {
 656                                                 printk(KERN_WARNING "Error "
 657                                                         "converting filename "
 658                                                         "from disk to usr"
 659                                                         "\n");
 660                                                 name = "??";
 661                                                 len = 2;
 662                                         } else {
 663                                                 name = fname_crypto_str.name;
 664                                                 len = fname_crypto_str.len;
 665                                         }
 666                                         ext4fs_dirhash(dir, de->name,
 667                                                        de->name_len, &h);
 668                                         printk("%*.s:(E)%x.%u ", len, name,
 669                                                h.hash, (unsigned) ((char *) de
 670                                                                    - base));
 671                                         fscrypt_fname_free_buffer(
 672                                                         &fname_crypto_str);
 673                                 }
 674 #else
 675                                 int len = de->name_len;
 676                                 char *name = de->name;
 677                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 678                                 printk("%*.s:%x.%u ", len, name, h.hash,
 679                                        (unsigned) ((char *) de - base));
 680 #endif
 681                         }
 682                         space += EXT4_DIR_REC_LEN(de->name_len);
 683                         names++;
 684                 }
 685                 de = ext4_next_entry(de, size);
 686         }
 687         printk(KERN_CONT "(%i)\n", names);
 688         return (struct stats) { names, space, 1 };
 689 }
 690
 691 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 692                              struct dx_entry *entries, int levels)
 693 {
 694         unsigned blocksize = dir->i_sb->s_blocksize;
 695         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 696         unsigned bcount = 0;
 697         struct buffer_head *bh;
 698         printk("%i indexed blocks...\n", count);
 699         for (i = 0; i < count; i++, entries++)
 700         {
 701                 ext4_lblk_t block = dx_get_block(entries);
 702                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 703                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 704                 struct stats stats;
 705                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 706                 bh = ext4_bread(NULL,dir, block, 0);
 707                 if (!bh || IS_ERR(bh))
 708                         continue;
 709                 stats = levels?
 710                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 711                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 712                         bh->b_data, blocksize, 0);
 713                 names += stats.names;
 714                 space += stats.space;
 715                 bcount += stats.bcount;
 716                 brelse(bh);
 717         }
 718         if (bcount)
 719                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 720                        levels ? "" : "   ", names, space/bcount,
 721                        (space/bcount)*100/blocksize);
 722         return (struct stats) { names, space, bcount};
 723 }
 724 #endif /* DX_DEBUG */
 725
 726 /*
 727  * Probe for a directory leaf block to search.
 728  *
 729  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 730  * error in the directory index, and the caller should fall back to
 731  * searching the directory normally.  The callers of dx_probe **MUST**
 732  * check for this error code, and make sure it never gets reflected
 733  * back to userspace.
 734  */
 735 static struct dx_frame *
 736 dx_probe(struct ext4_filename *fname, struct inode *dir,
 737          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 738 {
 739         unsigned count, indirect;
 740         struct dx_entry *at, *entries, *p, *q, *m;
 741         struct dx_root *root;
 742         struct dx_frame *frame = frame_in;
 743         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 744         u32 hash;
 745
 746         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 747         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 748         if (IS_ERR(frame->bh))
 749                 return (struct dx_frame *) frame->bh;
 750
 751         root = (struct dx_root *) frame->bh->b_data;
 752         if (root->info.hash_version != DX_HASH_TEA &&
 753             root->info.hash_version != DX_HASH_HALF_MD4 &&
 754             root->info.hash_version != DX_HASH_LEGACY) {
 755                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 756                                    root->info.hash_version);
 757                 goto fail;
 758         }
 759         if (fname)
 760                 hinfo = &fname->hinfo;
 761         hinfo->hash_version = root->info.hash_version;
 762         if (hinfo->hash_version <= DX_HASH_TEA)
 763                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 764         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 765         if (fname && fname_name(fname))
 766                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 767         hash = hinfo->hash;
 768
 769         if (root->info.unused_flags & 1) {
 770                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 771                                    root->info.unused_flags);
 772                 goto fail;
 773         }
 774
 775         indirect = root->info.indirect_levels;
 776         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 777                 ext4_warning(dir->i_sb,
 778                              "Directory (ino: %lu) htree depth %#06x exceed"
 779                              "supported value", dir->i_ino,
 780                              ext4_dir_htree_level(dir->i_sb));
 781                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 782                         ext4_warning(dir->i_sb, "Enable large directory "
 783                                                 "feature to access it");
 784                 }
 785                 goto fail;
 786         }
 787
 788         entries = (struct dx_entry *)(((char *)&root->info) +
 789                                       root->info.info_length);
 790
 791         if (dx_get_limit(entries) != dx_root_limit(dir,
 792                                                    root->info.info_length)) {
 793                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 794                                    dx_get_limit(entries),
 795                                    dx_root_limit(dir, root->info.info_length));
 796                 goto fail;
 797         }
 798
 799         dxtrace(printk("Look up %x", hash));
 800         while (1) {
 801                 count = dx_get_count(entries);
 802                 if (!count || count > dx_get_limit(entries)) {
 803                         ext4_warning_inode(dir,
 804                                            "dx entry: count %u beyond limit %u",
 805                                            count, dx_get_limit(entries));
 806                         goto fail;
 807                 }
 808
 809                 p = entries + 1;
 810                 q = entries + count - 1;
 811                 while (p <= q) {
 812                         m = p + (q - p) / 2;
 813                         dxtrace(printk(KERN_CONT "."));
 814                         if (dx_get_hash(m) > hash)
 815                                 q = m - 1;
 816                         else
 817                                 p = m + 1;
 818                 }
 819
 820                 if (0) { // linear search cross check
 821                         unsigned n = count - 1;
 822                         at = entries;
 823                         while (n--)
 824                         {
 825                                 dxtrace(printk(KERN_CONT ","));
 826                                 if (dx_get_hash(++at) > hash)
 827                                 {
 828                                         at--;
 829                                         break;
 830                                 }
 831                         }
 832                         assert (at == p - 1);
 833                 }
 834
 835                 at = p - 1;
 836                 dxtrace(printk(KERN_CONT " %x->%u\n",
 837                                at == entries ? 0 : dx_get_hash(at),
 838                                dx_get_block(at)));
 839                 frame->entries = entries;
 840                 frame->at = at;
 841                 if (!indirect--)
 842                         return frame;
 843                 frame++;
 844                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 845                 if (IS_ERR(frame->bh)) {
 846                         ret_err = (struct dx_frame *) frame->bh;
 847                         frame->bh = NULL;
 848                         goto fail;
 849                 }
 850                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 851
 852                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 853                         ext4_warning_inode(dir,
 854                                 "dx entry: limit %u != node limit %u",
 855                                 dx_get_limit(entries), dx_node_limit(dir));
 856                         goto fail;
 857                 }
 858         }
 859 fail:
 860         while (frame >= frame_in) {
 861                 brelse(frame->bh);
 862                 frame--;
 863         }
 864
 865         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 866                 ext4_warning_inode(dir,
 867                         "Corrupt directory, running e2fsck is recommended");
 868         return ret_err;
 869 }
 870
 871 static void dx_release(struct dx_frame *frames)
 872 {
 873         struct dx_root_info *info;
 874         int i;
 875         unsigned int indirect_levels;
 876
 877         if (frames[0].bh == NULL)
 878                 return;
 879
 880         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 881         /* save local copy, "info" may be freed after brelse() */
 882         indirect_levels = info->indirect_levels;
 883         for (i = 0; i <= indirect_levels; i++) {
 884                 if (frames[i].bh == NULL)
 885                         break;
 886                 brelse(frames[i].bh);
 887                 frames[i].bh = NULL;
 888         }
 889 }
 890
 891 /*
 892  * This function increments the frame pointer to search the next leaf
 893  * block, and reads in the necessary intervening nodes if the search
 894  * should be necessary.  Whether or not the search is necessary is
 895  * controlled by the hash parameter.  If the hash value is even, then
 896  * the search is only continued if the next block starts with that
 897  * hash value.  This is used if we are searching for a specific file.
 898  *
 899  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 900  *
 901  * This function returns 1 if the caller should continue to search,
 902  * or 0 if it should not.  If there is an error reading one of the
 903  * index blocks, it will a negative error code.
 904  *
 905  * If start_hash is non-null, it will be filled in with the starting
 906  * hash of the next page.
 907  */
 908 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 909                                  struct dx_frame *frame,
 910                                  struct dx_frame *frames,
 911                                  __u32 *start_hash)
 912 {
 913         struct dx_frame *p;
 914         struct buffer_head *bh;
 915         int num_frames = 0;
 916         __u32 bhash;
 917
 918         p = frame;
 919         /*
 920          * Find the next leaf page by incrementing the frame pointer.
 921          * If we run out of entries in the interior node, loop around and
 922          * increment pointer in the parent node.  When we break out of
 923          * this loop, num_frames indicates the number of interior
 924          * nodes need to be read.
 925          */
 926         while (1) {
 927                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 928                         break;
 929                 if (p == frames)
 930                         return 0;
 931                 num_frames++;
 932                 p--;
 933         }
 934
 935         /*
 936          * If the hash is 1, then continue only if the next page has a
 937          * continuation hash of any value.  This is used for readdir
 938          * handling.  Otherwise, check to see if the hash matches the
 939          * desired contiuation hash.  If it doesn't, return since
 940          * there's no point to read in the successive index pages.
 941          */
 942         bhash = dx_get_hash(p->at);
 943         if (start_hash)
 944                 *start_hash = bhash;
 945         if ((hash & 1) == 0) {
 946                 if ((bhash & ~1) != hash)
 947                         return 0;
 948         }
 949         /*
 950          * If the hash is HASH_NB_ALWAYS, we always go to the next
 951          * block so no check is necessary
 952          */
 953         while (num_frames--) {
 954                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 955                 if (IS_ERR(bh))
 956                         return PTR_ERR(bh);
 957                 p++;
 958                 brelse(p->bh);
 959                 p->bh = bh;
 960                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 961         }
 962         return 1;
 963 }
 964
 965
 966 /*
 967  * This function fills a red-black tree with information from a
 968  * directory block.  It returns the number directory entries loaded
 969  * into the tree.  If there is an error it is returned in err.
 970  */
 971 static int htree_dirblock_to_tree(struct file *dir_file,
 972                                   struct inode *dir, ext4_lblk_t block,
 973                                   struct dx_hash_info *hinfo,
 974                                   __u32 start_hash, __u32 start_minor_hash)
 975 {
 976         struct buffer_head *bh;
 977         struct ext4_dir_entry_2 *de, *top;
 978         int err = 0, count = 0;
 979         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 980
 981         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 982                                                         (unsigned long)block));
 983         bh = ext4_read_dirblock(dir, block, DIRENT);
 984         if (IS_ERR(bh))
 985                 return PTR_ERR(bh);
 986
 987         de = (struct ext4_dir_entry_2 *) bh->b_data;
 988         top = (struct ext4_dir_entry_2 *) ((char *) de +
 989                                            dir->i_sb->s_blocksize -
 990                                            EXT4_DIR_REC_LEN(0));
 991 #ifdef CONFIG_FS_ENCRYPTION
 992         /* Check if the directory is encrypted */
 993         if (IS_ENCRYPTED(dir)) {
 994                 err = fscrypt_get_encryption_info(dir);
 995                 if (err < 0) {
 996                         brelse(bh);
 997                         return err;
 998                 }
 999                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1000                                                      &fname_crypto_str);
1001                 if (err < 0) {
1002                         brelse(bh);
1003                         return err;
1004                 }
1005         }
1006 #endif
1007         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1008                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1009                                 bh->b_data, bh->b_size,
1010                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1011                                          + ((char *)de - bh->b_data))) {
1012                         /* silently ignore the rest of the block */
1013                         break;
1014                 }
1015                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1016                 if ((hinfo->hash < start_hash) ||
1017                     ((hinfo->hash == start_hash) &&
1018                      (hinfo->minor_hash < start_minor_hash)))
1019                         continue;
1020                 if (de->inode == 0)
1021                         continue;
1022                 if (!IS_ENCRYPTED(dir)) {
1023                         tmp_str.name = de->name;
1024                         tmp_str.len = de->name_len;
1025                         err = ext4_htree_store_dirent(dir_file,
1026                                    hinfo->hash, hinfo->minor_hash, de,
1027                                    &tmp_str);
1028                 } else {
1029                         int save_len = fname_crypto_str.len;
1030                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1031                                                                 de->name_len);
1032
1033                         /* Directory is encrypted */
1034                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1035                                         hinfo->minor_hash, &de_name,
1036                                         &fname_crypto_str);
1037                         if (err) {
1038                                 count = err;
1039                                 goto errout;
1040                         }
1041                         err = ext4_htree_store_dirent(dir_file,
1042                                    hinfo->hash, hinfo->minor_hash, de,
1043                                         &fname_crypto_str);
1044                         fname_crypto_str.len = save_len;
1045                 }
1046                 if (err != 0) {
1047                         count = err;
1048                         goto errout;
1049                 }
1050                 count++;
1051         }
1052 errout:
1053         brelse(bh);
1054 #ifdef CONFIG_FS_ENCRYPTION
1055         fscrypt_fname_free_buffer(&fname_crypto_str);
1056 #endif
1057         return count;
1058 }
1059
1060
1061 /*
1062  * This function fills a red-black tree with information from a
1063  * directory.  We start scanning the directory in hash order, starting
1064  * at start_hash and start_minor_hash.
1065  *
1066  * This function returns the number of entries inserted into the tree,
1067  * or a negative error code.
1068  */
1069 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1070                          __u32 start_minor_hash, __u32 *next_hash)
1071 {
1072         struct dx_hash_info hinfo;
1073         struct ext4_dir_entry_2 *de;
1074         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1075         struct inode *dir;
1076         ext4_lblk_t block;
1077         int count = 0;
1078         int ret, err;
1079         __u32 hashval;
1080         struct fscrypt_str tmp_str;
1081
1082         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1083                        start_hash, start_minor_hash));
1084         dir = file_inode(dir_file);
1085         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1086                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1087                 if (hinfo.hash_version <= DX_HASH_TEA)
1088                         hinfo.hash_version +=
1089                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1090                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1091                 if (ext4_has_inline_data(dir)) {
1092                         int has_inline_data = 1;
1093                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1094                                                         &hinfo, start_hash,
1095                                                         start_minor_hash,
1096                                                         &has_inline_data);
1097                         if (has_inline_data) {
1098                                 *next_hash = ~0;
1099                                 return count;
1100                         }
1101                 }
1102                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1103                                                start_hash, start_minor_hash);
1104                 *next_hash = ~0;
1105                 return count;
1106         }
1107         hinfo.hash = start_hash;
1108         hinfo.minor_hash = 0;
1109         frame = dx_probe(NULL, dir, &hinfo, frames);
1110         if (IS_ERR(frame))
1111                 return PTR_ERR(frame);
1112
1113         /* Add '.' and '..' from the htree header */
1114         if (!start_hash && !start_minor_hash) {
1115                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1116                 tmp_str.name = de->name;
1117                 tmp_str.len = de->name_len;
1118                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1119                                               de, &tmp_str);
1120                 if (err != 0)
1121                         goto errout;
1122                 count++;
1123         }
1124         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1125                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1126                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1127                 tmp_str.name = de->name;
1128                 tmp_str.len = de->name_len;
1129                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1130                                               de, &tmp_str);
1131                 if (err != 0)
1132                         goto errout;
1133                 count++;
1134         }
1135
1136         while (1) {
1137                 if (fatal_signal_pending(current)) {
1138                         err = -ERESTARTSYS;
1139                         goto errout;
1140                 }
1141                 cond_resched();
1142                 block = dx_get_block(frame->at);
1143                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1144                                              start_hash, start_minor_hash);
1145                 if (ret < 0) {
1146                         err = ret;
1147                         goto errout;
1148                 }
1149                 count += ret;
1150                 hashval = ~0;
1151                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1152                                             frame, frames, &hashval);
1153                 *next_hash = hashval;
1154                 if (ret < 0) {
1155                         err = ret;
1156                         goto errout;
1157                 }
1158                 /*
1159                  * Stop if:  (a) there are no more entries, or
1160                  * (b) we have inserted at least one entry and the
1161                  * next hash value is not a continuation
1162                  */
1163                 if ((ret == 0) ||
1164                     (count && ((hashval & 1) == 0)))
1165                         break;
1166         }
1167         dx_release(frames);
1168         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1169                        "next hash: %x\n", count, *next_hash));
1170         return count;
1171 errout:
1172         dx_release(frames);
1173         return (err);
1174 }
1175
1176 static inline int search_dirblock(struct buffer_head *bh,
1177                                   struct inode *dir,
1178                                   struct ext4_filename *fname,
1179                                   unsigned int offset,
1180                                   struct ext4_dir_entry_2 **res_dir)
1181 {
1182         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1183                                fname, offset, res_dir);
1184 }
1185
1186 /*
1187  * Directory block splitting, compacting
1188  */
1189
1190 /*
1191  * Create map of hash values, offsets, and sizes, stored at end of block.
1192  * Returns number of entries mapped.
1193  */
1194 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1195                        unsigned blocksize, struct dx_hash_info *hinfo,
1196                        struct dx_map_entry *map_tail)
1197 {
1198         int count = 0;
1199         char *base = (char *) de;
1200         struct dx_hash_info h = *hinfo;
1201
1202         while ((char *) de < base + blocksize) {
1203                 if (de->name_len && de->inode) {
1204                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1205                         map_tail--;
1206                         map_tail->hash = h.hash;
1207                         map_tail->offs = ((char *) de - base)>>2;
1208                         map_tail->size = le16_to_cpu(de->rec_len);
1209                         count++;
1210                         cond_resched();
1211                 }
1212                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1213                 de = ext4_next_entry(de, blocksize);
1214         }
1215         return count;
1216 }
1217
1218 /* Sort map by hash value */
1219 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1220 {
1221         struct dx_map_entry *p, *q, *top = map + count - 1;
1222         int more;
1223         /* Combsort until bubble sort doesn't suck */
1224         while (count > 2) {
1225                 count = count*10/13;
1226                 if (count - 9 < 2) /* 9, 10 -> 11 */
1227                         count = 11;
1228                 for (p = top, q = p - count; q >= map; p--, q--)
1229                         if (p->hash < q->hash)
1230                                 swap(*p, *q);
1231         }
1232         /* Garden variety bubble sort */
1233         do {
1234                 more = 0;
1235                 q = top;
1236                 while (q-- > map) {
1237                         if (q[1].hash >= q[0].hash)
1238                                 continue;
1239                         swap(*(q+1), *q);
1240                         more = 1;
1241                 }
1242         } while(more);
1243 }
1244
1245 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1246 {
1247         struct dx_entry *entries = frame->entries;
1248         struct dx_entry *old = frame->at, *new = old + 1;
1249         int count = dx_get_count(entries);
1250
1251         assert(count < dx_get_limit(entries));
1252         assert(old < entries + count);
1253         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1254         dx_set_hash(new, hash);
1255         dx_set_block(new, block);
1256         dx_set_count(entries, count + 1);
1257 }
1258
1259 #ifdef CONFIG_UNICODE
1260 /*
1261  * Test whether a case-insensitive directory entry matches the filename
1262  * being searched for.
1263  *
1264  * Returns: 0 if the directory entry matches, more than 0 if it
1265  * doesn't match or less than zero on error.
1266  */
1267 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1268                     const struct qstr *entry)
1269 {
1270         const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1271         const struct unicode_map *um = sbi->s_encoding;
1272         int ret;
1273
1274         ret = utf8_strncasecmp(um, name, entry);
1275         if (ret < 0) {
1276                 /* Handle invalid character sequence as either an error
1277                  * or as an opaque byte sequence.
1278                  */
1279                 if (ext4_has_strict_mode(sbi))
1280                         return -EINVAL;
1281
1282                 if (name->len != entry->len)
1283                         return 1;
1284
1285                 return !!memcmp(name->name, entry->name, name->len);
1286         }
1287
1288         return ret;
1289 }
1290 #endif
1291
1292 /*
1293  * Test whether a directory entry matches the filename being searched for.
1294  *
1295  * Return: %true if the directory entry matches, otherwise %false.
1296  */
1297 static inline bool ext4_match(const struct inode *parent,
1298                               const struct ext4_filename *fname,
1299                               const struct ext4_dir_entry_2 *de)
1300 {
1301         struct fscrypt_name f;
1302 #ifdef CONFIG_UNICODE
1303         const struct qstr entry = {.name = de->name, .len = de->name_len};
1304 #endif
1305
1306         if (!de->inode)
1307                 return false;
1308
1309         f.usr_fname = fname->usr_fname;
1310         f.disk_name = fname->disk_name;
1311 #ifdef CONFIG_FS_ENCRYPTION
1312         f.crypto_buf = fname->crypto_buf;
1313 #endif
1314
1315 #ifdef CONFIG_UNICODE
1316         if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent))
1317                 return (ext4_ci_compare(parent, fname->usr_fname, &entry) == 0);
1318 #endif
1319
1320         return fscrypt_match_name(&f, de->name, de->name_len);
1321 }
1322
1323 /*
1324  * Returns 0 if not found, -1 on failure, and 1 on success
1325  */
1326 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1327                     struct inode *dir, struct ext4_filename *fname,
1328                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1329 {
1330         struct ext4_dir_entry_2 * de;
1331         char * dlimit;
1332         int de_len;
1333
1334         de = (struct ext4_dir_entry_2 *)search_buf;
1335         dlimit = search_buf + buf_size;
1336         while ((char *) de < dlimit) {
1337                 /* this code is executed quadratically often */
1338                 /* do minimal checking `by hand' */
1339                 if ((char *) de + de->name_len <= dlimit &&
1340                     ext4_match(dir, fname, de)) {
1341                         /* found a match - just to be sure, do
1342                          * a full check */
1343                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1344                                                  bh->b_size, offset))
1345                                 return -1;
1346                         *res_dir = de;
1347                         return 1;
1348                 }
1349                 /* prevent looping on a bad block */
1350                 de_len = ext4_rec_len_from_disk(de->rec_len,
1351                                                 dir->i_sb->s_blocksize);
1352                 if (de_len <= 0)
1353                         return -1;
1354                 offset += de_len;
1355                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1356         }
1357         return 0;
1358 }
1359
1360 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1361                                struct ext4_dir_entry *de)
1362 {
1363         struct super_block *sb = dir->i_sb;
1364
1365         if (!is_dx(dir))
1366                 return 0;
1367         if (block == 0)
1368                 return 1;
1369         if (de->inode == 0 &&
1370             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1371                         sb->s_blocksize)
1372                 return 1;
1373         return 0;
1374 }
1375
1376 /*
1377  *      __ext4_find_entry()
1378  *
1379  * finds an entry in the specified directory with the wanted name. It
1380  * returns the cache buffer in which the entry was found, and the entry
1381  * itself (as a parameter - res_dir). It does NOT read the inode of the
1382  * entry - you'll have to do that yourself if you want to.
1383  *
1384  * The returned buffer_head has ->b_count elevated.  The caller is expected
1385  * to brelse() it when appropriate.
1386  */
1387 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1388                                              struct ext4_filename *fname,
1389                                              struct ext4_dir_entry_2 **res_dir,
1390                                              int *inlined)
1391 {
1392         struct super_block *sb;
1393         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1394         struct buffer_head *bh, *ret = NULL;
1395         ext4_lblk_t start, block;
1396         const u8 *name = fname->usr_fname->name;
1397         size_t ra_max = 0;      /* Number of bh's in the readahead
1398                                    buffer, bh_use[] */
1399         size_t ra_ptr = 0;      /* Current index into readahead
1400                                    buffer */
1401         ext4_lblk_t  nblocks;
1402         int i, namelen, retval;
1403
1404         *res_dir = NULL;
1405         sb = dir->i_sb;
1406         namelen = fname->usr_fname->len;
1407         if (namelen > EXT4_NAME_LEN)
1408                 return NULL;
1409
1410         if (ext4_has_inline_data(dir)) {
1411                 int has_inline_data = 1;
1412                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1413                                              &has_inline_data);
1414                 if (has_inline_data) {
1415                         if (inlined)
1416                                 *inlined = 1;
1417                         goto cleanup_and_exit;
1418                 }
1419         }
1420
1421         if ((namelen <= 2) && (name[0] == '.') &&
1422             (name[1] == '.' || name[1] == '\0')) {
1423                 /*
1424                  * "." or ".." will only be in the first block
1425                  * NFS may look up ".."; "." should be handled by the VFS
1426                  */
1427                 block = start = 0;
1428                 nblocks = 1;
1429                 goto restart;
1430         }
1431         if (is_dx(dir)) {
1432                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1433                 /*
1434                  * On success, or if the error was file not found,
1435                  * return.  Otherwise, fall back to doing a search the
1436                  * old fashioned way.
1437                  */
1438                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1439                         goto cleanup_and_exit;
1440                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1441                                "falling back\n"));
1442                 ret = NULL;
1443         }
1444         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1445         if (!nblocks) {
1446                 ret = NULL;
1447                 goto cleanup_and_exit;
1448         }
1449         start = EXT4_I(dir)->i_dir_start_lookup;
1450         if (start >= nblocks)
1451                 start = 0;
1452         block = start;
1453 restart:
1454         do {
1455                 /*
1456                  * We deal with the read-ahead logic here.
1457                  */
1458                 if (ra_ptr >= ra_max) {
1459                         /* Refill the readahead buffer */
1460                         ra_ptr = 0;
1461                         if (block < start)
1462                                 ra_max = start - block;
1463                         else
1464                                 ra_max = nblocks - block;
1465                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1466                         retval = ext4_bread_batch(dir, block, ra_max,
1467                                                   false /* wait */, bh_use);
1468                         if (retval) {
1469                                 ret = ERR_PTR(retval);
1470                                 ra_max = 0;
1471                                 goto cleanup_and_exit;
1472                         }
1473                 }
1474                 if ((bh = bh_use[ra_ptr++]) == NULL)
1475                         goto next;
1476                 wait_on_buffer(bh);
1477                 if (!buffer_uptodate(bh)) {
1478                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1479                                          (unsigned long) block);
1480                         brelse(bh);
1481                         ret = ERR_PTR(-EIO);
1482                         goto cleanup_and_exit;
1483                 }
1484                 if (!buffer_verified(bh) &&
1485                     !is_dx_internal_node(dir, block,
1486                                          (struct ext4_dir_entry *)bh->b_data) &&
1487                     !ext4_dirent_csum_verify(dir,
1488                                 (struct ext4_dir_entry *)bh->b_data)) {
1489                         EXT4_ERROR_INODE(dir, "checksumming directory "
1490                                          "block %lu", (unsigned long)block);
1491                         brelse(bh);
1492                         ret = ERR_PTR(-EFSBADCRC);
1493                         goto cleanup_and_exit;
1494                 }
1495                 set_buffer_verified(bh);
1496                 i = search_dirblock(bh, dir, fname,
1497                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1498                 if (i == 1) {
1499                         EXT4_I(dir)->i_dir_start_lookup = block;
1500                         ret = bh;
1501                         goto cleanup_and_exit;
1502                 } else {
1503                         brelse(bh);
1504                         if (i < 0)
1505                                 goto cleanup_and_exit;
1506                 }
1507         next:
1508                 if (++block >= nblocks)
1509                         block = 0;
1510         } while (block != start);
1511
1512         /*
1513          * If the directory has grown while we were searching, then
1514          * search the last part of the directory before giving up.
1515          */
1516         block = nblocks;
1517         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1518         if (block < nblocks) {
1519                 start = 0;
1520                 goto restart;
1521         }
1522
1523 cleanup_and_exit:
1524         /* Clean up the read-ahead blocks */
1525         for (; ra_ptr < ra_max; ra_ptr++)
1526                 brelse(bh_use[ra_ptr]);
1527         return ret;
1528 }
1529
1530 static struct buffer_head *ext4_find_entry(struct inode *dir,
1531                                            const struct qstr *d_name,
1532                                            struct ext4_dir_entry_2 **res_dir,
1533                                            int *inlined)
1534 {
1535         int err;
1536         struct ext4_filename fname;
1537         struct buffer_head *bh;
1538
1539         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1540         if (err == -ENOENT)
1541                 return NULL;
1542         if (err)
1543                 return ERR_PTR(err);
1544
1545         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1546
1547         ext4_fname_free_filename(&fname);
1548         return bh;
1549 }
1550
1551 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1552                                              struct dentry *dentry,
1553                                              struct ext4_dir_entry_2 **res_dir)
1554 {
1555         int err;
1556         struct ext4_filename fname;
1557         struct buffer_head *bh;
1558
1559         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1560         if (err == -ENOENT)
1561                 return NULL;
1562         if (err)
1563                 return ERR_PTR(err);
1564
1565         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1566
1567         ext4_fname_free_filename(&fname);
1568         return bh;
1569 }
1570
1571 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1572                         struct ext4_filename *fname,
1573                         struct ext4_dir_entry_2 **res_dir)
1574 {
1575         struct super_block * sb = dir->i_sb;
1576         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1577         struct buffer_head *bh;
1578         ext4_lblk_t block;
1579         int retval;
1580
1581 #ifdef CONFIG_FS_ENCRYPTION
1582         *res_dir = NULL;
1583 #endif
1584         frame = dx_probe(fname, dir, NULL, frames);
1585         if (IS_ERR(frame))
1586                 return (struct buffer_head *) frame;
1587         do {
1588                 block = dx_get_block(frame->at);
1589                 bh = ext4_read_dirblock(dir, block, DIRENT);
1590                 if (IS_ERR(bh))
1591                         goto errout;
1592
1593                 retval = search_dirblock(bh, dir, fname,
1594                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1595                                          res_dir);
1596                 if (retval == 1)
1597                         goto success;
1598                 brelse(bh);
1599                 if (retval == -1) {
1600                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1601                         goto errout;
1602                 }
1603
1604                 /* Check to see if we should continue to search */
1605                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1606                                                frames, NULL);
1607                 if (retval < 0) {
1608                         ext4_warning_inode(dir,
1609                                 "error %d reading directory index block",
1610                                 retval);
1611                         bh = ERR_PTR(retval);
1612                         goto errout;
1613                 }
1614         } while (retval == 1);
1615
1616         bh = NULL;
1617 errout:
1618         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1619 success:
1620         dx_release(frames);
1621         return bh;
1622 }
1623
1624 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1625 {
1626         struct inode *inode;
1627         struct ext4_dir_entry_2 *de;
1628         struct buffer_head *bh;
1629
1630         if (dentry->d_name.len > EXT4_NAME_LEN)
1631                 return ERR_PTR(-ENAMETOOLONG);
1632
1633         bh = ext4_lookup_entry(dir, dentry, &de);
1634         if (IS_ERR(bh))
1635                 return ERR_CAST(bh);
1636         inode = NULL;
1637         if (bh) {
1638                 __u32 ino = le32_to_cpu(de->inode);
1639                 brelse(bh);
1640                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1641                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1642                         return ERR_PTR(-EFSCORRUPTED);
1643                 }
1644                 if (unlikely(ino == dir->i_ino)) {
1645                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1646                                          dentry);
1647                         return ERR_PTR(-EFSCORRUPTED);
1648                 }
1649                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1650                 if (inode == ERR_PTR(-ESTALE)) {
1651                         EXT4_ERROR_INODE(dir,
1652                                          "deleted inode referenced: %u",
1653                                          ino);
1654                         return ERR_PTR(-EFSCORRUPTED);
1655                 }
1656                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1657                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1658                     !fscrypt_has_permitted_context(dir, inode)) {
1659                         ext4_warning(inode->i_sb,
1660                                      "Inconsistent encryption contexts: %lu/%lu",
1661                                      dir->i_ino, inode->i_ino);
1662                         iput(inode);
1663                         return ERR_PTR(-EPERM);
1664                 }
1665         }
1666
1667 #ifdef CONFIG_UNICODE
1668         if (!inode && IS_CASEFOLDED(dir)) {
1669                 /* Eventually we want to call d_add_ci(dentry, NULL)
1670                  * for negative dentries in the encoding case as
1671                  * well.  For now, prevent the negative dentry
1672                  * from being cached.
1673                  */
1674                 return NULL;
1675         }
1676 #endif
1677         return d_splice_alias(inode, dentry);
1678 }
1679
1680
1681 struct dentry *ext4_get_parent(struct dentry *child)
1682 {
1683         __u32 ino;
1684         static const struct qstr dotdot = QSTR_INIT("..", 2);
1685         struct ext4_dir_entry_2 * de;
1686         struct buffer_head *bh;
1687
1688         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1689         if (IS_ERR(bh))
1690                 return ERR_CAST(bh);
1691         if (!bh)
1692                 return ERR_PTR(-ENOENT);
1693         ino = le32_to_cpu(de->inode);
1694         brelse(bh);
1695
1696         if (!ext4_valid_inum(child->d_sb, ino)) {
1697                 EXT4_ERROR_INODE(d_inode(child),
1698                                  "bad parent inode number: %u", ino);
1699                 return ERR_PTR(-EFSCORRUPTED);
1700         }
1701
1702         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1703 }
1704
1705 /*
1706  * Move count entries from end of map between two memory locations.
1707  * Returns pointer to last entry moved.
1708  */
1709 static struct ext4_dir_entry_2 *
1710 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1711                 unsigned blocksize)
1712 {
1713         unsigned rec_len = 0;
1714
1715         while (count--) {
1716                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1717                                                 (from + (map->offs<<2));
1718                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1719                 memcpy (to, de, rec_len);
1720                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1721                                 ext4_rec_len_to_disk(rec_len, blocksize);
1722                 de->inode = 0;
1723                 map++;
1724                 to += rec_len;
1725         }
1726         return (struct ext4_dir_entry_2 *) (to - rec_len);
1727 }
1728
1729 /*
1730  * Compact each dir entry in the range to the minimal rec_len.
1731  * Returns pointer to last entry in range.
1732  */
1733 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1734 {
1735         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1736         unsigned rec_len = 0;
1737
1738         prev = to = de;
1739         while ((char*)de < base + blocksize) {
1740                 next = ext4_next_entry(de, blocksize);
1741                 if (de->inode && de->name_len) {
1742                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1743                         if (de > to)
1744                                 memmove(to, de, rec_len);
1745                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1746                         prev = to;
1747                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1748                 }
1749                 de = next;
1750         }
1751         return prev;
1752 }
1753
1754 /*
1755  * Split a full leaf block to make room for a new dir entry.
1756  * Allocate a new block, and move entries so that they are approx. equally full.
1757  * Returns pointer to de in block into which the new entry will be inserted.
1758  */
1759 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1760                         struct buffer_head **bh,struct dx_frame *frame,
1761                         struct dx_hash_info *hinfo)
1762 {
1763         unsigned blocksize = dir->i_sb->s_blocksize;
1764         unsigned count, continued;
1765         struct buffer_head *bh2;
1766         ext4_lblk_t newblock;
1767         u32 hash2;
1768         struct dx_map_entry *map;
1769         char *data1 = (*bh)->b_data, *data2;
1770         unsigned split, move, size;
1771         struct ext4_dir_entry_2 *de = NULL, *de2;
1772         struct ext4_dir_entry_tail *t;
1773         int     csum_size = 0;
1774         int     err = 0, i;
1775
1776         if (ext4_has_metadata_csum(dir->i_sb))
1777                 csum_size = sizeof(struct ext4_dir_entry_tail);
1778
1779         bh2 = ext4_append(handle, dir, &newblock);
1780         if (IS_ERR(bh2)) {
1781                 brelse(*bh);
1782                 *bh = NULL;
1783                 return (struct ext4_dir_entry_2 *) bh2;
1784         }
1785
1786         BUFFER_TRACE(*bh, "get_write_access");
1787         err = ext4_journal_get_write_access(handle, *bh);
1788         if (err)
1789                 goto journal_error;
1790
1791         BUFFER_TRACE(frame->bh, "get_write_access");
1792         err = ext4_journal_get_write_access(handle, frame->bh);
1793         if (err)
1794                 goto journal_error;
1795
1796         data2 = bh2->b_data;
1797
1798         /* create map in the end of data2 block */
1799         map = (struct dx_map_entry *) (data2 + blocksize);
1800         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1801                              blocksize, hinfo, map);
1802         map -= count;
1803         dx_sort_map(map, count);
1804         /* Split the existing block in the middle, size-wise */
1805         size = 0;
1806         move = 0;
1807         for (i = count-1; i >= 0; i--) {
1808                 /* is more than half of this entry in 2nd half of the block? */
1809                 if (size + map[i].size/2 > blocksize/2)
1810                         break;
1811                 size += map[i].size;
1812                 move++;
1813         }
1814         /* map index at which we will split */
1815         split = count - move;
1816         hash2 = map[split].hash;
1817         continued = hash2 == map[split - 1].hash;
1818         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1819                         (unsigned long)dx_get_block(frame->at),
1820                                         hash2, split, count-split));
1821
1822         /* Fancy dance to stay within two buffers */
1823         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1824                               blocksize);
1825         de = dx_pack_dirents(data1, blocksize);
1826         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1827                                            (char *) de,
1828                                            blocksize);
1829         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1830                                             (char *) de2,
1831                                             blocksize);
1832         if (csum_size) {
1833                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1834                 initialize_dirent_tail(t, blocksize);
1835
1836                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1837                 initialize_dirent_tail(t, blocksize);
1838         }
1839
1840         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1841                         blocksize, 1));
1842         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1843                         blocksize, 1));
1844
1845         /* Which block gets the new entry? */
1846         if (hinfo->hash >= hash2) {
1847                 swap(*bh, bh2);
1848                 de = de2;
1849         }
1850         dx_insert_block(frame, hash2 + continued, newblock);
1851         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1852         if (err)
1853                 goto journal_error;
1854         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1855         if (err)
1856                 goto journal_error;
1857         brelse(bh2);
1858         dxtrace(dx_show_index("frame", frame->entries));
1859         return de;
1860
1861 journal_error:
1862         brelse(*bh);
1863         brelse(bh2);
1864         *bh = NULL;
1865         ext4_std_error(dir->i_sb, err);
1866         return ERR_PTR(err);
1867 }
1868
1869 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1870                       struct buffer_head *bh,
1871                       void *buf, int buf_size,
1872                       struct ext4_filename *fname,
1873                       struct ext4_dir_entry_2 **dest_de)
1874 {
1875         struct ext4_dir_entry_2 *de;
1876         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1877         int nlen, rlen;
1878         unsigned int offset = 0;
1879         char *top;
1880
1881         de = (struct ext4_dir_entry_2 *)buf;
1882         top = buf + buf_size - reclen;
1883         while ((char *) de <= top) {
1884                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1885                                          buf, buf_size, offset))
1886                         return -EFSCORRUPTED;
1887                 if (ext4_match(dir, fname, de))
1888                         return -EEXIST;
1889                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1890                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1891                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1892                         break;
1893                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1894                 offset += rlen;
1895         }
1896         if ((char *) de > top)
1897                 return -ENOSPC;
1898
1899         *dest_de = de;
1900         return 0;
1901 }
1902
1903 void ext4_insert_dentry(struct inode *inode,
1904                         struct ext4_dir_entry_2 *de,
1905                         int buf_size,
1906                         struct ext4_filename *fname)
1907 {
1908
1909         int nlen, rlen;
1910
1911         nlen = EXT4_DIR_REC_LEN(de->name_len);
1912         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1913         if (de->inode) {
1914                 struct ext4_dir_entry_2 *de1 =
1915                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1916                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1917                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1918                 de = de1;
1919         }
1920         de->file_type = EXT4_FT_UNKNOWN;
1921         de->inode = cpu_to_le32(inode->i_ino);
1922         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1923         de->name_len = fname_len(fname);
1924         memcpy(de->name, fname_name(fname), fname_len(fname));
1925 }
1926
1927 /*
1928  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1929  * it points to a directory entry which is guaranteed to be large
1930  * enough for new directory entry.  If de is NULL, then
1931  * add_dirent_to_buf will attempt search the directory block for
1932  * space.  It will return -ENOSPC if no space is available, and -EIO
1933  * and -EEXIST if directory entry already exists.
1934  */
1935 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1936                              struct inode *dir,
1937                              struct inode *inode, struct ext4_dir_entry_2 *de,
1938                              struct buffer_head *bh)
1939 {
1940         unsigned int    blocksize = dir->i_sb->s_blocksize;
1941         int             csum_size = 0;
1942         int             err;
1943
1944         if (ext4_has_metadata_csum(inode->i_sb))
1945                 csum_size = sizeof(struct ext4_dir_entry_tail);
1946
1947         if (!de) {
1948                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1949                                         blocksize - csum_size, fname, &de);
1950                 if (err)
1951                         return err;
1952         }
1953         BUFFER_TRACE(bh, "get_write_access");
1954         err = ext4_journal_get_write_access(handle, bh);
1955         if (err) {
1956                 ext4_std_error(dir->i_sb, err);
1957                 return err;
1958         }
1959
1960         /* By now the buffer is marked for journaling */
1961         ext4_insert_dentry(inode, de, blocksize, fname);
1962
1963         /*
1964          * XXX shouldn't update any times until successful
1965          * completion of syscall, but too many callers depend
1966          * on this.
1967          *
1968          * XXX similarly, too many callers depend on
1969          * ext4_new_inode() setting the times, but error
1970          * recovery deletes the inode, so the worst that can
1971          * happen is that the times are slightly out of date
1972          * and/or different from the directory change time.
1973          */
1974         dir->i_mtime = dir->i_ctime = current_time(dir);
1975         ext4_update_dx_flag(dir);
1976         inode_inc_iversion(dir);
1977         ext4_mark_inode_dirty(handle, dir);
1978         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1979         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1980         if (err)
1981                 ext4_std_error(dir->i_sb, err);
1982         return 0;
1983 }
1984
1985 /*
1986  * This converts a one block unindexed directory to a 3 block indexed
1987  * directory, and adds the dentry to the indexed directory.
1988  */
1989 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1990                             struct inode *dir,
1991                             struct inode *inode, struct buffer_head *bh)
1992 {
1993         struct buffer_head *bh2;
1994         struct dx_root  *root;
1995         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1996         struct dx_entry *entries;
1997         struct ext4_dir_entry_2 *de, *de2;
1998         struct ext4_dir_entry_tail *t;
1999         char            *data1, *top;
2000         unsigned        len;
2001         int             retval;
2002         unsigned        blocksize;
2003         ext4_lblk_t  block;
2004         struct fake_dirent *fde;
2005         int csum_size = 0;
2006
2007         if (ext4_has_metadata_csum(inode->i_sb))
2008                 csum_size = sizeof(struct ext4_dir_entry_tail);
2009
2010         blocksize =  dir->i_sb->s_blocksize;
2011         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2012         BUFFER_TRACE(bh, "get_write_access");
2013         retval = ext4_journal_get_write_access(handle, bh);
2014         if (retval) {
2015                 ext4_std_error(dir->i_sb, retval);
2016                 brelse(bh);
2017                 return retval;
2018         }
2019         root = (struct dx_root *) bh->b_data;
2020
2021         /* The 0th block becomes the root, move the dirents out */
2022         fde = &root->dotdot;
2023         de = (struct ext4_dir_entry_2 *)((char *)fde +
2024                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2025         if ((char *) de >= (((char *) root) + blocksize)) {
2026                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2027                 brelse(bh);
2028                 return -EFSCORRUPTED;
2029         }
2030         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2031
2032         /* Allocate new block for the 0th block's dirents */
2033         bh2 = ext4_append(handle, dir, &block);
2034         if (IS_ERR(bh2)) {
2035                 brelse(bh);
2036                 return PTR_ERR(bh2);
2037         }
2038         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2039         data1 = bh2->b_data;
2040
2041         memcpy (data1, de, len);
2042         de = (struct ext4_dir_entry_2 *) data1;
2043         top = data1 + len;
2044         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2045                 de = de2;
2046         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2047                                            (char *) de,
2048                                            blocksize);
2049
2050         if (csum_size) {
2051                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2052                 initialize_dirent_tail(t, blocksize);
2053         }
2054
2055         /* Initialize the root; the dot dirents already exist */
2056         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2057         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2058                                            blocksize);
2059         memset (&root->info, 0, sizeof(root->info));
2060         root->info.info_length = sizeof(root->info);
2061         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2062         entries = root->entries;
2063         dx_set_block(entries, 1);
2064         dx_set_count(entries, 1);
2065         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2066
2067         /* Initialize as for dx_probe */
2068         fname->hinfo.hash_version = root->info.hash_version;
2069         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2070                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2071         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2072         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2073
2074         memset(frames, 0, sizeof(frames));
2075         frame = frames;
2076         frame->entries = entries;
2077         frame->at = entries;
2078         frame->bh = bh;
2079
2080         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2081         if (retval)
2082                 goto out_frames;
2083         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2084         if (retval)
2085                 goto out_frames;
2086
2087         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2088         if (IS_ERR(de)) {
2089                 retval = PTR_ERR(de);
2090                 goto out_frames;
2091         }
2092
2093         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2094 out_frames:
2095         /*
2096          * Even if the block split failed, we have to properly write
2097          * out all the changes we did so far. Otherwise we can end up
2098          * with corrupted filesystem.
2099          */
2100         if (retval)
2101                 ext4_mark_inode_dirty(handle, dir);
2102         dx_release(frames);
2103         brelse(bh2);
2104         return retval;
2105 }
2106
2107 /*
2108  *      ext4_add_entry()
2109  *
2110  * adds a file entry to the specified directory, using the same
2111  * semantics as ext4_find_entry(). It returns NULL if it failed.
2112  *
2113  * NOTE!! The inode part of 'de' is left at 0 - which means you
2114  * may not sleep between calling this and putting something into
2115  * the entry, as someone else might have used it while you slept.
2116  */
2117 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2118                           struct inode *inode)
2119 {
2120         struct inode *dir = d_inode(dentry->d_parent);
2121         struct buffer_head *bh = NULL;
2122         struct ext4_dir_entry_2 *de;
2123         struct ext4_dir_entry_tail *t;
2124         struct super_block *sb;
2125         struct ext4_sb_info *sbi;
2126         struct ext4_filename fname;
2127         int     retval;
2128         int     dx_fallback=0;
2129         unsigned blocksize;
2130         ext4_lblk_t block, blocks;
2131         int     csum_size = 0;
2132
2133         if (ext4_has_metadata_csum(inode->i_sb))
2134                 csum_size = sizeof(struct ext4_dir_entry_tail);
2135
2136         sb = dir->i_sb;
2137         sbi = EXT4_SB(sb);
2138         blocksize = sb->s_blocksize;
2139         if (!dentry->d_name.len)
2140                 return -EINVAL;
2141
2142 #ifdef CONFIG_UNICODE
2143         if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2144             utf8_validate(sbi->s_encoding, &dentry->d_name))
2145                 return -EINVAL;
2146 #endif
2147
2148         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2149         if (retval)
2150                 return retval;
2151
2152         if (ext4_has_inline_data(dir)) {
2153                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2154                 if (retval < 0)
2155                         goto out;
2156                 if (retval == 1) {
2157                         retval = 0;
2158                         goto out;
2159                 }
2160         }
2161
2162         if (is_dx(dir)) {
2163                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2164                 if (!retval || (retval != ERR_BAD_DX_DIR))
2165                         goto out;
2166                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2167                 dx_fallback++;
2168                 ext4_mark_inode_dirty(handle, dir);
2169         }
2170         blocks = dir->i_size >> sb->s_blocksize_bits;
2171         for (block = 0; block < blocks; block++) {
2172                 bh = ext4_read_dirblock(dir, block, DIRENT);
2173                 if (IS_ERR(bh)) {
2174                         retval = PTR_ERR(bh);
2175                         bh = NULL;
2176                         goto out;
2177                 }
2178                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2179                                            NULL, bh);
2180                 if (retval != -ENOSPC)
2181                         goto out;
2182
2183                 if (blocks == 1 && !dx_fallback &&
2184                     ext4_has_feature_dir_index(sb)) {
2185                         retval = make_indexed_dir(handle, &fname, dir,
2186                                                   inode, bh);
2187                         bh = NULL; /* make_indexed_dir releases bh */
2188                         goto out;
2189                 }
2190                 brelse(bh);
2191         }
2192         bh = ext4_append(handle, dir, &block);
2193         if (IS_ERR(bh)) {
2194                 retval = PTR_ERR(bh);
2195                 bh = NULL;
2196                 goto out;
2197         }
2198         de = (struct ext4_dir_entry_2 *) bh->b_data;
2199         de->inode = 0;
2200         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2201
2202         if (csum_size) {
2203                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2204                 initialize_dirent_tail(t, blocksize);
2205         }
2206
2207         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2208 out:
2209         ext4_fname_free_filename(&fname);
2210         brelse(bh);
2211         if (retval == 0)
2212                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2213         return retval;
2214 }
2215
2216 /*
2217  * Returns 0 for success, or a negative error value
2218  */
2219 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2220                              struct inode *dir, struct inode *inode)
2221 {
2222         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2223         struct dx_entry *entries, *at;
2224         struct buffer_head *bh;
2225         struct super_block *sb = dir->i_sb;
2226         struct ext4_dir_entry_2 *de;
2227         int restart;
2228         int err;
2229
2230 again:
2231         restart = 0;
2232         frame = dx_probe(fname, dir, NULL, frames);
2233         if (IS_ERR(frame))
2234                 return PTR_ERR(frame);
2235         entries = frame->entries;
2236         at = frame->at;
2237         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2238         if (IS_ERR(bh)) {
2239                 err = PTR_ERR(bh);
2240                 bh = NULL;
2241                 goto cleanup;
2242         }
2243
2244         BUFFER_TRACE(bh, "get_write_access");
2245         err = ext4_journal_get_write_access(handle, bh);
2246         if (err)
2247                 goto journal_error;
2248
2249         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2250         if (err != -ENOSPC)
2251                 goto cleanup;
2252
2253         err = 0;
2254         /* Block full, should compress but for now just split */
2255         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2256                        dx_get_count(entries), dx_get_limit(entries)));
2257         /* Need to split index? */
2258         if (dx_get_count(entries) == dx_get_limit(entries)) {
2259                 ext4_lblk_t newblock;
2260                 int levels = frame - frames + 1;
2261                 unsigned int icount;
2262                 int add_level = 1;
2263                 struct dx_entry *entries2;
2264                 struct dx_node *node2;
2265                 struct buffer_head *bh2;
2266
2267                 while (frame > frames) {
2268                         if (dx_get_count((frame - 1)->entries) <
2269                             dx_get_limit((frame - 1)->entries)) {
2270                                 add_level = 0;
2271                                 break;
2272                         }
2273                         frame--; /* split higher index block */
2274                         at = frame->at;
2275                         entries = frame->entries;
2276                         restart = 1;
2277                 }
2278                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2279                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2280                                          "reach max htree level :%d",
2281                                          dir->i_ino, levels);
2282                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2283                                 ext4_warning(sb, "Large directory feature is "
2284                                                  "not enabled on this "
2285                                                  "filesystem");
2286                         }
2287                         err = -ENOSPC;
2288                         goto cleanup;
2289                 }
2290                 icount = dx_get_count(entries);
2291                 bh2 = ext4_append(handle, dir, &newblock);
2292                 if (IS_ERR(bh2)) {
2293                         err = PTR_ERR(bh2);
2294                         goto cleanup;
2295                 }
2296                 node2 = (struct dx_node *)(bh2->b_data);
2297                 entries2 = node2->entries;
2298                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2299                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2300                                                            sb->s_blocksize);
2301                 BUFFER_TRACE(frame->bh, "get_write_access");
2302                 err = ext4_journal_get_write_access(handle, frame->bh);
2303                 if (err)
2304                         goto journal_error;
2305                 if (!add_level) {
2306                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2307                         unsigned hash2 = dx_get_hash(entries + icount1);
2308                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2309                                        icount1, icount2));
2310
2311                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2312                         err = ext4_journal_get_write_access(handle,
2313                                                              (frame - 1)->bh);
2314                         if (err)
2315                                 goto journal_error;
2316
2317                         memcpy((char *) entries2, (char *) (entries + icount1),
2318                                icount2 * sizeof(struct dx_entry));
2319                         dx_set_count(entries, icount1);
2320                         dx_set_count(entries2, icount2);
2321                         dx_set_limit(entries2, dx_node_limit(dir));
2322
2323                         /* Which index block gets the new entry? */
2324                         if (at - entries >= icount1) {
2325                                 frame->at = at = at - entries - icount1 + entries2;
2326                                 frame->entries = entries = entries2;
2327                                 swap(frame->bh, bh2);
2328                         }
2329                         dx_insert_block((frame - 1), hash2, newblock);
2330                         dxtrace(dx_show_index("node", frame->entries));
2331                         dxtrace(dx_show_index("node",
2332                                ((struct dx_node *) bh2->b_data)->entries));
2333                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2334                         if (err)
2335                                 goto journal_error;
2336                         brelse (bh2);
2337                         err = ext4_handle_dirty_dx_node(handle, dir,
2338                                                    (frame - 1)->bh);
2339                         if (err)
2340                                 goto journal_error;
2341                         if (restart) {
2342                                 err = ext4_handle_dirty_dx_node(handle, dir,
2343                                                            frame->bh);
2344                                 goto journal_error;
2345                         }
2346                 } else {
2347                         struct dx_root *dxroot;
2348                         memcpy((char *) entries2, (char *) entries,
2349                                icount * sizeof(struct dx_entry));
2350                         dx_set_limit(entries2, dx_node_limit(dir));
2351
2352                         /* Set up root */
2353                         dx_set_count(entries, 1);
2354                         dx_set_block(entries + 0, newblock);
2355                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2356                         dxroot->info.indirect_levels += 1;
2357                         dxtrace(printk(KERN_DEBUG
2358                                        "Creating %d level index...\n",
2359                                        dxroot->info.indirect_levels));
2360                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2361                         if (err)
2362                                 goto journal_error;
2363                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2364                         brelse(bh2);
2365                         restart = 1;
2366                         goto journal_error;
2367                 }
2368         }
2369         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2370         if (IS_ERR(de)) {
2371                 err = PTR_ERR(de);
2372                 goto cleanup;
2373         }
2374         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2375         goto cleanup;
2376
2377 journal_error:
2378         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2379 cleanup:
2380         brelse(bh);
2381         dx_release(frames);
2382         /* @restart is true means htree-path has been changed, we need to
2383          * repeat dx_probe() to find out valid htree-path
2384          */
2385         if (restart && err == 0)
2386                 goto again;
2387         return err;
2388 }
2389
2390 /*
2391  * ext4_generic_delete_entry deletes a directory entry by merging it
2392  * with the previous entry
2393  */
2394 int ext4_generic_delete_entry(handle_t *handle,
2395                               struct inode *dir,
2396                               struct ext4_dir_entry_2 *de_del,
2397                               struct buffer_head *bh,
2398                               void *entry_buf,
2399                               int buf_size,
2400                               int csum_size)
2401 {
2402         struct ext4_dir_entry_2 *de, *pde;
2403         unsigned int blocksize = dir->i_sb->s_blocksize;
2404         int i;
2405
2406         i = 0;
2407         pde = NULL;
2408         de = (struct ext4_dir_entry_2 *)entry_buf;
2409         while (i < buf_size - csum_size) {
2410                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2411                                          bh->b_data, bh->b_size, i))
2412                         return -EFSCORRUPTED;
2413                 if (de == de_del)  {
2414                         if (pde)
2415                                 pde->rec_len = ext4_rec_len_to_disk(
2416                                         ext4_rec_len_from_disk(pde->rec_len,
2417                                                                blocksize) +
2418                                         ext4_rec_len_from_disk(de->rec_len,
2419                                                                blocksize),
2420                                         blocksize);
2421                         else
2422                                 de->inode = 0;
2423                         inode_inc_iversion(dir);
2424                         return 0;
2425                 }
2426                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2427                 pde = de;
2428                 de = ext4_next_entry(de, blocksize);
2429         }
2430         return -ENOENT;
2431 }
2432
2433 static int ext4_delete_entry(handle_t *handle,
2434                              struct inode *dir,
2435                              struct ext4_dir_entry_2 *de_del,
2436                              struct buffer_head *bh)
2437 {
2438         int err, csum_size = 0;
2439
2440         if (ext4_has_inline_data(dir)) {
2441                 int has_inline_data = 1;
2442                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2443                                                &has_inline_data);
2444                 if (has_inline_data)
2445                         return err;
2446         }
2447
2448         if (ext4_has_metadata_csum(dir->i_sb))
2449                 csum_size = sizeof(struct ext4_dir_entry_tail);
2450
2451         BUFFER_TRACE(bh, "get_write_access");
2452         err = ext4_journal_get_write_access(handle, bh);
2453         if (unlikely(err))
2454                 goto out;
2455
2456         err = ext4_generic_delete_entry(handle, dir, de_del,
2457                                         bh, bh->b_data,
2458                                         dir->i_sb->s_blocksize, csum_size);
2459         if (err)
2460                 goto out;
2461
2462         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2463         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2464         if (unlikely(err))
2465                 goto out;
2466
2467         return 0;
2468 out:
2469         if (err != -ENOENT)
2470                 ext4_std_error(dir->i_sb, err);
2471         return err;
2472 }
2473
2474 /*
2475  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2476  * since this indicates that nlinks count was previously 1 to avoid overflowing
2477  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2478  * that subdirectory link counts are not being maintained accurately.
2479  *
2480  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2481  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2482  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2483  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2484  */
2485 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2486 {
2487         inc_nlink(inode);
2488         if (is_dx(inode) &&
2489             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2490                 set_nlink(inode, 1);
2491 }
2492
2493 /*
2494  * If a directory had nlink == 1, then we should let it be 1. This indicates
2495  * directory has >EXT4_LINK_MAX subdirs.
2496  */
2497 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2498 {
2499         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2500                 drop_nlink(inode);
2501 }
2502
2503
2504 static int ext4_add_nondir(handle_t *handle,
2505                 struct dentry *dentry, struct inode *inode)
2506 {
2507         int err = ext4_add_entry(handle, dentry, inode);
2508         if (!err) {
2509                 ext4_mark_inode_dirty(handle, inode);
2510                 d_instantiate_new(dentry, inode);
2511                 return 0;
2512         }
2513         drop_nlink(inode);
2514         unlock_new_inode(inode);
2515         iput(inode);
2516         return err;
2517 }
2518
2519 /*
2520  * By the time this is called, we already have created
2521  * the directory cache entry for the new file, but it
2522  * is so far negative - it has no inode.
2523  *
2524  * If the create succeeds, we fill in the inode information
2525  * with d_instantiate().
2526  */
2527 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2528                        bool excl)
2529 {
2530         handle_t *handle;
2531         struct inode *inode;
2532         int err, credits, retries = 0;
2533
2534         err = dquot_initialize(dir);
2535         if (err)
2536                 return err;
2537
2538         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2539                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2540 retry:
2541         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2542                                             NULL, EXT4_HT_DIR, credits);
2543         handle = ext4_journal_current_handle();
2544         err = PTR_ERR(inode);
2545         if (!IS_ERR(inode)) {
2546                 inode->i_op = &ext4_file_inode_operations;
2547                 inode->i_fop = &ext4_file_operations;
2548                 ext4_set_aops(inode);
2549                 err = ext4_add_nondir(handle, dentry, inode);
2550                 if (!err && IS_DIRSYNC(dir))
2551                         ext4_handle_sync(handle);
2552         }
2553         if (handle)
2554                 ext4_journal_stop(handle);
2555         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2556                 goto retry;
2557         return err;
2558 }
2559
2560 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2561                       umode_t mode, dev_t rdev)
2562 {
2563         handle_t *handle;
2564         struct inode *inode;
2565         int err, credits, retries = 0;
2566
2567         err = dquot_initialize(dir);
2568         if (err)
2569                 return err;
2570
2571         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2572                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2573 retry:
2574         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2575                                             NULL, EXT4_HT_DIR, credits);
2576         handle = ext4_journal_current_handle();
2577         err = PTR_ERR(inode);
2578         if (!IS_ERR(inode)) {
2579                 init_special_inode(inode, inode->i_mode, rdev);
2580                 inode->i_op = &ext4_special_inode_operations;
2581                 err = ext4_add_nondir(handle, dentry, inode);
2582                 if (!err && IS_DIRSYNC(dir))
2583                         ext4_handle_sync(handle);
2584         }
2585         if (handle)
2586                 ext4_journal_stop(handle);
2587         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2588                 goto retry;
2589         return err;
2590 }
2591
2592 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2593 {
2594         handle_t *handle;
2595         struct inode *inode;
2596         int err, retries = 0;
2597
2598         err = dquot_initialize(dir);
2599         if (err)
2600                 return err;
2601
2602 retry:
2603         inode = ext4_new_inode_start_handle(dir, mode,
2604                                             NULL, 0, NULL,
2605                                             EXT4_HT_DIR,
2606                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2607                           4 + EXT4_XATTR_TRANS_BLOCKS);
2608         handle = ext4_journal_current_handle();
2609         err = PTR_ERR(inode);
2610         if (!IS_ERR(inode)) {
2611                 inode->i_op = &ext4_file_inode_operations;
2612                 inode->i_fop = &ext4_file_operations;
2613                 ext4_set_aops(inode);
2614                 d_tmpfile(dentry, inode);
2615                 err = ext4_orphan_add(handle, inode);
2616                 if (err)
2617                         goto err_unlock_inode;
2618                 mark_inode_dirty(inode);
2619                 unlock_new_inode(inode);
2620         }
2621         if (handle)
2622                 ext4_journal_stop(handle);
2623         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2624                 goto retry;
2625         return err;
2626 err_unlock_inode:
2627         ext4_journal_stop(handle);
2628         unlock_new_inode(inode);
2629         return err;
2630 }
2631
2632 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2633                           struct ext4_dir_entry_2 *de,
2634                           int blocksize, int csum_size,
2635                           unsigned int parent_ino, int dotdot_real_len)
2636 {
2637         de->inode = cpu_to_le32(inode->i_ino);
2638         de->name_len = 1;
2639         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2640                                            blocksize);
2641         strcpy(de->name, ".");
2642         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2643
2644         de = ext4_next_entry(de, blocksize);
2645         de->inode = cpu_to_le32(parent_ino);
2646         de->name_len = 2;
2647         if (!dotdot_real_len)
2648                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2649                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2650                                         blocksize);
2651         else
2652                 de->rec_len = ext4_rec_len_to_disk(
2653                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2654         strcpy(de->name, "..");
2655         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2656
2657         return ext4_next_entry(de, blocksize);
2658 }
2659
2660 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2661                              struct inode *inode)
2662 {
2663         struct buffer_head *dir_block = NULL;
2664         struct ext4_dir_entry_2 *de;
2665         struct ext4_dir_entry_tail *t;
2666         ext4_lblk_t block = 0;
2667         unsigned int blocksize = dir->i_sb->s_blocksize;
2668         int csum_size = 0;
2669         int err;
2670
2671         if (ext4_has_metadata_csum(dir->i_sb))
2672                 csum_size = sizeof(struct ext4_dir_entry_tail);
2673
2674         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2675                 err = ext4_try_create_inline_dir(handle, dir, inode);
2676                 if (err < 0 && err != -ENOSPC)
2677                         goto out;
2678                 if (!err)
2679                         goto out;
2680         }
2681
2682         inode->i_size = 0;
2683         dir_block = ext4_append(handle, inode, &block);
2684         if (IS_ERR(dir_block))
2685                 return PTR_ERR(dir_block);
2686         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2687         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2688         set_nlink(inode, 2);
2689         if (csum_size) {
2690                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2691                 initialize_dirent_tail(t, blocksize);
2692         }
2693
2694         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2695         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2696         if (err)
2697                 goto out;
2698         set_buffer_verified(dir_block);
2699 out:
2700         brelse(dir_block);
2701         return err;
2702 }
2703
2704 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2705 {
2706         handle_t *handle;
2707         struct inode *inode;
2708         int err, credits, retries = 0;
2709
2710         if (EXT4_DIR_LINK_MAX(dir))
2711                 return -EMLINK;
2712
2713         err = dquot_initialize(dir);
2714         if (err)
2715                 return err;
2716
2717         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2718                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2719 retry:
2720         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2721                                             &dentry->d_name,
2722                                             0, NULL, EXT4_HT_DIR, credits);
2723         handle = ext4_journal_current_handle();
2724         err = PTR_ERR(inode);
2725         if (IS_ERR(inode))
2726                 goto out_stop;
2727
2728         inode->i_op = &ext4_dir_inode_operations;
2729         inode->i_fop = &ext4_dir_operations;
2730         err = ext4_init_new_dir(handle, dir, inode);
2731         if (err)
2732                 goto out_clear_inode;
2733         err = ext4_mark_inode_dirty(handle, inode);
2734         if (!err)
2735                 err = ext4_add_entry(handle, dentry, inode);
2736         if (err) {
2737 out_clear_inode:
2738                 clear_nlink(inode);
2739                 unlock_new_inode(inode);
2740                 ext4_mark_inode_dirty(handle, inode);
2741                 iput(inode);
2742                 goto out_stop;
2743         }
2744         ext4_inc_count(handle, dir);
2745         ext4_update_dx_flag(dir);
2746         err = ext4_mark_inode_dirty(handle, dir);
2747         if (err)
2748                 goto out_clear_inode;
2749         d_instantiate_new(dentry, inode);
2750         if (IS_DIRSYNC(dir))
2751                 ext4_handle_sync(handle);
2752
2753 out_stop:
2754         if (handle)
2755                 ext4_journal_stop(handle);
2756         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2757                 goto retry;
2758         return err;
2759 }
2760
2761 /*
2762  * routine to check that the specified directory is empty (for rmdir)
2763  */
2764 bool ext4_empty_dir(struct inode *inode)
2765 {
2766         unsigned int offset;
2767         struct buffer_head *bh;
2768         struct ext4_dir_entry_2 *de, *de1;
2769         struct super_block *sb;
2770
2771         if (ext4_has_inline_data(inode)) {
2772                 int has_inline_data = 1;
2773                 int ret;
2774
2775                 ret = empty_inline_dir(inode, &has_inline_data);
2776                 if (has_inline_data)
2777                         return ret;
2778         }
2779
2780         sb = inode->i_sb;
2781         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2782                 EXT4_ERROR_INODE(inode, "invalid size");
2783                 return true;
2784         }
2785         bh = ext4_read_dirblock(inode, 0, EITHER);
2786         if (IS_ERR(bh))
2787                 return true;
2788
2789         de = (struct ext4_dir_entry_2 *) bh->b_data;
2790         de1 = ext4_next_entry(de, sb->s_blocksize);
2791         if (le32_to_cpu(de->inode) != inode->i_ino ||
2792                         le32_to_cpu(de1->inode) == 0 ||
2793                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2794                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2795                 brelse(bh);
2796                 return true;
2797         }
2798         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2799                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2800         de = ext4_next_entry(de1, sb->s_blocksize);
2801         while (offset < inode->i_size) {
2802                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2803                         unsigned int lblock;
2804                         brelse(bh);
2805                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2806                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2807                         if (IS_ERR(bh))
2808                                 return true;
2809                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2810                 }
2811                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2812                                          bh->b_data, bh->b_size, offset)) {
2813                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2814                                                          sb->s_blocksize);
2815                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2816                         continue;
2817                 }
2818                 if (le32_to_cpu(de->inode)) {
2819                         brelse(bh);
2820                         return false;
2821                 }
2822                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2823                 de = ext4_next_entry(de, sb->s_blocksize);
2824         }
2825         brelse(bh);
2826         return true;
2827 }
2828
2829 /*
2830  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2831  * such inodes, starting at the superblock, in case we crash before the
2832  * file is closed/deleted, or in case the inode truncate spans multiple
2833  * transactions and the last transaction is not recovered after a crash.
2834  *
2835  * At filesystem recovery time, we walk this list deleting unlinked
2836  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2837  *
2838  * Orphan list manipulation functions must be called under i_mutex unless
2839  * we are just creating the inode or deleting it.
2840  */
2841 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2842 {
2843         struct super_block *sb = inode->i_sb;
2844         struct ext4_sb_info *sbi = EXT4_SB(sb);
2845         struct ext4_iloc iloc;
2846         int err = 0, rc;
2847         bool dirty = false;
2848
2849         if (!sbi->s_journal || is_bad_inode(inode))
2850                 return 0;
2851
2852         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2853                      !inode_is_locked(inode));
2854         /*
2855          * Exit early if inode already is on orphan list. This is a big speedup
2856          * since we don't have to contend on the global s_orphan_lock.
2857          */
2858         if (!list_empty(&EXT4_I(inode)->i_orphan))
2859                 return 0;
2860
2861         /*
2862          * Orphan handling is only valid for files with data blocks
2863          * being truncated, or files being unlinked. Note that we either
2864          * hold i_mutex, or the inode can not be referenced from outside,
2865          * so i_nlink should not be bumped due to race
2866          */
2867         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2868                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2869
2870         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2871         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2872         if (err)
2873                 goto out;
2874
2875         err = ext4_reserve_inode_write(handle, inode, &iloc);
2876         if (err)
2877                 goto out;
2878
2879         mutex_lock(&sbi->s_orphan_lock);
2880         /*
2881          * Due to previous errors inode may be already a part of on-disk
2882          * orphan list. If so skip on-disk list modification.
2883          */
2884         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2885             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2886                 /* Insert this inode at the head of the on-disk orphan list */
2887                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2888                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2889                 dirty = true;
2890         }
2891         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2892         mutex_unlock(&sbi->s_orphan_lock);
2893
2894         if (dirty) {
2895                 err = ext4_handle_dirty_super(handle, sb);
2896                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2897                 if (!err)
2898                         err = rc;
2899                 if (err) {
2900                         /*
2901                          * We have to remove inode from in-memory list if
2902                          * addition to on disk orphan list failed. Stray orphan
2903                          * list entries can cause panics at unmount time.
2904                          */
2905                         mutex_lock(&sbi->s_orphan_lock);
2906                         list_del_init(&EXT4_I(inode)->i_orphan);
2907                         mutex_unlock(&sbi->s_orphan_lock);
2908                 }
2909         } else
2910                 brelse(iloc.bh);
2911
2912         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2913         jbd_debug(4, "orphan inode %lu will point to %d\n",
2914                         inode->i_ino, NEXT_ORPHAN(inode));
2915 out:
2916         ext4_std_error(sb, err);
2917         return err;
2918 }
2919
2920 /*
2921  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2922  * of such inodes stored on disk, because it is finally being cleaned up.
2923  */
2924 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2925 {
2926         struct list_head *prev;
2927         struct ext4_inode_info *ei = EXT4_I(inode);
2928         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2929         __u32 ino_next;
2930         struct ext4_iloc iloc;
2931         int err = 0;
2932
2933         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2934                 return 0;
2935
2936         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2937                      !inode_is_locked(inode));
2938         /* Do this quick check before taking global s_orphan_lock. */
2939         if (list_empty(&ei->i_orphan))
2940                 return 0;
2941
2942         if (handle) {
2943                 /* Grab inode buffer early before taking global s_orphan_lock */
2944                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2945         }
2946
2947         mutex_lock(&sbi->s_orphan_lock);
2948         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2949
2950         prev = ei->i_orphan.prev;
2951         list_del_init(&ei->i_orphan);
2952
2953         /* If we're on an error path, we may not have a valid
2954          * transaction handle with which to update the orphan list on
2955          * disk, but we still need to remove the inode from the linked
2956          * list in memory. */
2957         if (!handle || err) {
2958                 mutex_unlock(&sbi->s_orphan_lock);
2959                 goto out_err;
2960         }
2961
2962         ino_next = NEXT_ORPHAN(inode);
2963         if (prev == &sbi->s_orphan) {
2964                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2965                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2966                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2967                 if (err) {
2968                         mutex_unlock(&sbi->s_orphan_lock);
2969                         goto out_brelse;
2970                 }
2971                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2972                 mutex_unlock(&sbi->s_orphan_lock);
2973                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2974         } else {
2975                 struct ext4_iloc iloc2;
2976                 struct inode *i_prev =
2977                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2978
2979                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2980                           i_prev->i_ino, ino_next);
2981                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2982                 if (err) {
2983                         mutex_unlock(&sbi->s_orphan_lock);
2984                         goto out_brelse;
2985                 }
2986                 NEXT_ORPHAN(i_prev) = ino_next;
2987                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2988                 mutex_unlock(&sbi->s_orphan_lock);
2989         }
2990         if (err)
2991                 goto out_brelse;
2992         NEXT_ORPHAN(inode) = 0;
2993         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2994 out_err:
2995         ext4_std_error(inode->i_sb, err);
2996         return err;
2997
2998 out_brelse:
2999         brelse(iloc.bh);
3000         goto out_err;
3001 }
3002
3003 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3004 {
3005         int retval;
3006         struct inode *inode;
3007         struct buffer_head *bh;
3008         struct ext4_dir_entry_2 *de;
3009         handle_t *handle = NULL;
3010
3011         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3012                 return -EIO;
3013
3014         /* Initialize quotas before so that eventual writes go in
3015          * separate transaction */
3016         retval = dquot_initialize(dir);
3017         if (retval)
3018                 return retval;
3019         retval = dquot_initialize(d_inode(dentry));
3020         if (retval)
3021                 return retval;
3022
3023         retval = -ENOENT;
3024         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3025         if (IS_ERR(bh))
3026                 return PTR_ERR(bh);
3027         if (!bh)
3028                 goto end_rmdir;
3029
3030         inode = d_inode(dentry);
3031
3032         retval = -EFSCORRUPTED;
3033         if (le32_to_cpu(de->inode) != inode->i_ino)
3034                 goto end_rmdir;
3035
3036         retval = -ENOTEMPTY;
3037         if (!ext4_empty_dir(inode))
3038                 goto end_rmdir;
3039
3040         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3041                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3042         if (IS_ERR(handle)) {
3043                 retval = PTR_ERR(handle);
3044                 handle = NULL;
3045                 goto end_rmdir;
3046         }
3047
3048         if (IS_DIRSYNC(dir))
3049                 ext4_handle_sync(handle);
3050
3051         retval = ext4_delete_entry(handle, dir, de, bh);
3052         if (retval)
3053                 goto end_rmdir;
3054         if (!EXT4_DIR_LINK_EMPTY(inode))
3055                 ext4_warning_inode(inode,
3056                              "empty directory '%.*s' has too many links (%u)",
3057                              dentry->d_name.len, dentry->d_name.name,
3058                              inode->i_nlink);
3059         inode_inc_iversion(inode);
3060         clear_nlink(inode);
3061         /* There's no need to set i_disksize: the fact that i_nlink is
3062          * zero will ensure that the right thing happens during any
3063          * recovery. */
3064         inode->i_size = 0;
3065         ext4_orphan_add(handle, inode);
3066         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3067         ext4_mark_inode_dirty(handle, inode);
3068         ext4_dec_count(handle, dir);
3069         ext4_update_dx_flag(dir);
3070         ext4_mark_inode_dirty(handle, dir);
3071
3072 #ifdef CONFIG_UNICODE
3073         /* VFS negative dentries are incompatible with Encoding and
3074          * Case-insensitiveness. Eventually we'll want avoid
3075          * invalidating the dentries here, alongside with returning the
3076          * negative dentries at ext4_lookup(), when it is better
3077          * supported by the VFS for the CI case.
3078          */
3079         if (IS_CASEFOLDED(dir))
3080                 d_invalidate(dentry);
3081 #endif
3082
3083 end_rmdir:
3084         brelse(bh);
3085         if (handle)
3086                 ext4_journal_stop(handle);
3087         return retval;
3088 }
3089
3090 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3091 {
3092         int retval;
3093         struct inode *inode;
3094         struct buffer_head *bh;
3095         struct ext4_dir_entry_2 *de;
3096         handle_t *handle = NULL;
3097
3098         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3099                 return -EIO;
3100
3101         trace_ext4_unlink_enter(dir, dentry);
3102         /* Initialize quotas before so that eventual writes go
3103          * in separate transaction */
3104         retval = dquot_initialize(dir);
3105         if (retval)
3106                 return retval;
3107         retval = dquot_initialize(d_inode(dentry));
3108         if (retval)
3109                 return retval;
3110
3111         retval = -ENOENT;
3112         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3113         if (IS_ERR(bh))
3114                 return PTR_ERR(bh);
3115         if (!bh)
3116                 goto end_unlink;
3117
3118         inode = d_inode(dentry);
3119
3120         retval = -EFSCORRUPTED;
3121         if (le32_to_cpu(de->inode) != inode->i_ino)
3122                 goto end_unlink;
3123
3124         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3125                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3126         if (IS_ERR(handle)) {
3127                 retval = PTR_ERR(handle);
3128                 handle = NULL;
3129                 goto end_unlink;
3130         }
3131
3132         if (IS_DIRSYNC(dir))
3133                 ext4_handle_sync(handle);
3134
3135         if (inode->i_nlink == 0) {
3136                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3137                                    dentry->d_name.len, dentry->d_name.name);
3138                 set_nlink(inode, 1);
3139         }
3140         retval = ext4_delete_entry(handle, dir, de, bh);
3141         if (retval)
3142                 goto end_unlink;
3143         dir->i_ctime = dir->i_mtime = current_time(dir);
3144         ext4_update_dx_flag(dir);
3145         ext4_mark_inode_dirty(handle, dir);
3146         drop_nlink(inode);
3147         if (!inode->i_nlink)
3148                 ext4_orphan_add(handle, inode);
3149         inode->i_ctime = current_time(inode);
3150         ext4_mark_inode_dirty(handle, inode);
3151
3152 #ifdef CONFIG_UNICODE
3153         /* VFS negative dentries are incompatible with Encoding and
3154          * Case-insensitiveness. Eventually we'll want avoid
3155          * invalidating the dentries here, alongside with returning the
3156          * negative dentries at ext4_lookup(), when it is  better
3157          * supported by the VFS for the CI case.
3158          */
3159         if (IS_CASEFOLDED(dir))
3160                 d_invalidate(dentry);
3161 #endif
3162
3163 end_unlink:
3164         brelse(bh);
3165         if (handle)
3166                 ext4_journal_stop(handle);
3167         trace_ext4_unlink_exit(dentry, retval);
3168         return retval;
3169 }
3170
3171 static int ext4_symlink(struct inode *dir,
3172                         struct dentry *dentry, const char *symname)
3173 {
3174         handle_t *handle;
3175         struct inode *inode;
3176         int err, len = strlen(symname);
3177         int credits;
3178         struct fscrypt_str disk_link;
3179
3180         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3181                 return -EIO;
3182
3183         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3184                                       &disk_link);
3185         if (err)
3186                 return err;
3187
3188         err = dquot_initialize(dir);
3189         if (err)
3190                 return err;
3191
3192         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3193                 /*
3194                  * For non-fast symlinks, we just allocate inode and put it on
3195                  * orphan list in the first transaction => we need bitmap,
3196                  * group descriptor, sb, inode block, quota blocks, and
3197                  * possibly selinux xattr blocks.
3198                  */
3199                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3200                           EXT4_XATTR_TRANS_BLOCKS;
3201         } else {
3202                 /*
3203                  * Fast symlink. We have to add entry to directory
3204                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3205                  * allocate new inode (bitmap, group descriptor, inode block,
3206                  * quota blocks, sb is already counted in previous macros).
3207                  */
3208                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3209                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3210         }
3211
3212         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3213                                             &dentry->d_name, 0, NULL,
3214                                             EXT4_HT_DIR, credits);
3215         handle = ext4_journal_current_handle();
3216         if (IS_ERR(inode)) {
3217                 if (handle)
3218                         ext4_journal_stop(handle);
3219                 return PTR_ERR(inode);
3220         }
3221
3222         if (IS_ENCRYPTED(inode)) {
3223                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3224                 if (err)
3225                         goto err_drop_inode;
3226                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3227         }
3228
3229         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3230                 if (!IS_ENCRYPTED(inode))
3231                         inode->i_op = &ext4_symlink_inode_operations;
3232                 inode_nohighmem(inode);
3233                 ext4_set_aops(inode);
3234                 /*
3235                  * We cannot call page_symlink() with transaction started
3236                  * because it calls into ext4_write_begin() which can wait
3237                  * for transaction commit if we are running out of space
3238                  * and thus we deadlock. So we have to stop transaction now
3239                  * and restart it when symlink contents is written.
3240                  *
3241                  * To keep fs consistent in case of crash, we have to put inode
3242                  * to orphan list in the mean time.
3243                  */
3244                 drop_nlink(inode);
3245                 err = ext4_orphan_add(handle, inode);
3246                 ext4_journal_stop(handle);
3247                 handle = NULL;
3248                 if (err)
3249                         goto err_drop_inode;
3250                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3251                 if (err)
3252                         goto err_drop_inode;
3253                 /*
3254                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3255                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3256                  */
3257                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3258                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3259                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3260                 if (IS_ERR(handle)) {
3261                         err = PTR_ERR(handle);
3262                         handle = NULL;
3263                         goto err_drop_inode;
3264                 }
3265                 set_nlink(inode, 1);
3266                 err = ext4_orphan_del(handle, inode);
3267                 if (err)
3268                         goto err_drop_inode;
3269         } else {
3270                 /* clear the extent format for fast symlink */
3271                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3272                 if (!IS_ENCRYPTED(inode)) {
3273                         inode->i_op = &ext4_fast_symlink_inode_operations;
3274                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3275                 }
3276                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3277                        disk_link.len);
3278                 inode->i_size = disk_link.len - 1;
3279         }
3280         EXT4_I(inode)->i_disksize = inode->i_size;
3281         err = ext4_add_nondir(handle, dentry, inode);
3282         if (!err && IS_DIRSYNC(dir))
3283                 ext4_handle_sync(handle);
3284
3285         if (handle)
3286                 ext4_journal_stop(handle);
3287         goto out_free_encrypted_link;
3288
3289 err_drop_inode:
3290         if (handle)
3291                 ext4_journal_stop(handle);
3292         clear_nlink(inode);
3293         unlock_new_inode(inode);
3294         iput(inode);
3295 out_free_encrypted_link:
3296         if (disk_link.name != (unsigned char *)symname)
3297                 kfree(disk_link.name);
3298         return err;
3299 }
3300
3301 static int ext4_link(struct dentry *old_dentry,
3302                      struct inode *dir, struct dentry *dentry)
3303 {
3304         handle_t *handle;
3305         struct inode *inode = d_inode(old_dentry);
3306         int err, retries = 0;
3307
3308         if (inode->i_nlink >= EXT4_LINK_MAX)
3309                 return -EMLINK;
3310
3311         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3312         if (err)
3313                 return err;
3314
3315         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3316             (!projid_eq(EXT4_I(dir)->i_projid,
3317                         EXT4_I(old_dentry->d_inode)->i_projid)))
3318                 return -EXDEV;
3319
3320         err = dquot_initialize(dir);
3321         if (err)
3322                 return err;
3323
3324 retry:
3325         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3326                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3327                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3328         if (IS_ERR(handle))
3329                 return PTR_ERR(handle);
3330
3331         if (IS_DIRSYNC(dir))
3332                 ext4_handle_sync(handle);
3333
3334         inode->i_ctime = current_time(inode);
3335         ext4_inc_count(handle, inode);
3336         ihold(inode);
3337
3338         err = ext4_add_entry(handle, dentry, inode);
3339         if (!err) {
3340                 ext4_mark_inode_dirty(handle, inode);
3341                 /* this can happen only for tmpfile being
3342                  * linked the first time
3343                  */
3344                 if (inode->i_nlink == 1)
3345                         ext4_orphan_del(handle, inode);
3346                 d_instantiate(dentry, inode);
3347         } else {
3348                 drop_nlink(inode);
3349                 iput(inode);
3350         }
3351         ext4_journal_stop(handle);
3352         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3353                 goto retry;
3354         return err;
3355 }
3356
3357
3358 /*
3359  * Try to find buffer head where contains the parent block.
3360  * It should be the inode block if it is inlined or the 1st block
3361  * if it is a normal dir.
3362  */
3363 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3364                                         struct inode *inode,
3365                                         int *retval,
3366                                         struct ext4_dir_entry_2 **parent_de,
3367                                         int *inlined)
3368 {
3369         struct buffer_head *bh;
3370
3371         if (!ext4_has_inline_data(inode)) {
3372                 bh = ext4_read_dirblock(inode, 0, EITHER);
3373                 if (IS_ERR(bh)) {
3374                         *retval = PTR_ERR(bh);
3375                         return NULL;
3376                 }
3377                 *parent_de = ext4_next_entry(
3378                                         (struct ext4_dir_entry_2 *)bh->b_data,
3379                                         inode->i_sb->s_blocksize);
3380                 return bh;
3381         }
3382
3383         *inlined = 1;
3384         return ext4_get_first_inline_block(inode, parent_de, retval);
3385 }
3386
3387 struct ext4_renament {
3388         struct inode *dir;
3389         struct dentry *dentry;
3390         struct inode *inode;
3391         bool is_dir;
3392         int dir_nlink_delta;
3393
3394         /* entry for "dentry" */
3395         struct buffer_head *bh;
3396         struct ext4_dir_entry_2 *de;
3397         int inlined;
3398
3399         /* entry for ".." in inode if it's a directory */
3400         struct buffer_head *dir_bh;
3401         struct ext4_dir_entry_2 *parent_de;
3402         int dir_inlined;
3403 };
3404
3405 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3406 {
3407         int retval;
3408
3409         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3410                                               &retval, &ent->parent_de,
3411                                               &ent->dir_inlined);
3412         if (!ent->dir_bh)
3413                 return retval;
3414         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3415                 return -EFSCORRUPTED;
3416         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3417         return ext4_journal_get_write_access(handle, ent->dir_bh);
3418 }
3419
3420 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3421                                   unsigned dir_ino)
3422 {
3423         int retval;
3424
3425         ent->parent_de->inode = cpu_to_le32(dir_ino);
3426         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3427         if (!ent->dir_inlined) {
3428                 if (is_dx(ent->inode)) {
3429                         retval = ext4_handle_dirty_dx_node(handle,
3430                                                            ent->inode,
3431                                                            ent->dir_bh);
3432                 } else {
3433                         retval = ext4_handle_dirty_dirent_node(handle,
3434                                                                ent->inode,
3435                                                                ent->dir_bh);
3436                 }
3437         } else {
3438                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3439         }
3440         if (retval) {
3441                 ext4_std_error(ent->dir->i_sb, retval);
3442                 return retval;
3443         }
3444         return 0;
3445 }
3446
3447 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3448                        unsigned ino, unsigned file_type)
3449 {
3450         int retval;
3451
3452         BUFFER_TRACE(ent->bh, "get write access");
3453         retval = ext4_journal_get_write_access(handle, ent->bh);
3454         if (retval)
3455                 return retval;
3456         ent->de->inode = cpu_to_le32(ino);
3457         if (ext4_has_feature_filetype(ent->dir->i_sb))
3458                 ent->de->file_type = file_type;
3459         inode_inc_iversion(ent->dir);
3460         ent->dir->i_ctime = ent->dir->i_mtime =
3461                 current_time(ent->dir);
3462         ext4_mark_inode_dirty(handle, ent->dir);
3463         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3464         if (!ent->inlined) {
3465                 retval = ext4_handle_dirty_dirent_node(handle,
3466                                                        ent->dir, ent->bh);
3467                 if (unlikely(retval)) {
3468                         ext4_std_error(ent->dir->i_sb, retval);
3469                         return retval;
3470                 }
3471         }
3472         brelse(ent->bh);
3473         ent->bh = NULL;
3474
3475         return 0;
3476 }
3477
3478 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3479                                   const struct qstr *d_name)
3480 {
3481         int retval = -ENOENT;
3482         struct buffer_head *bh;
3483         struct ext4_dir_entry_2 *de;
3484
3485         bh = ext4_find_entry(dir, d_name, &de, NULL);
3486         if (IS_ERR(bh))
3487                 return PTR_ERR(bh);
3488         if (bh) {
3489                 retval = ext4_delete_entry(handle, dir, de, bh);
3490                 brelse(bh);
3491         }
3492         return retval;
3493 }
3494
3495 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3496                                int force_reread)
3497 {
3498         int retval;
3499         /*
3500          * ent->de could have moved from under us during htree split, so make
3501          * sure that we are deleting the right entry.  We might also be pointing
3502          * to a stale entry in the unused part of ent->bh so just checking inum
3503          * and the name isn't enough.
3504          */
3505         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3506             ent->de->name_len != ent->dentry->d_name.len ||
3507             strncmp(ent->de->name, ent->dentry->d_name.name,
3508                     ent->de->name_len) ||
3509             force_reread) {
3510                 retval = ext4_find_delete_entry(handle, ent->dir,
3511                                                 &ent->dentry->d_name);
3512         } else {
3513                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3514                 if (retval == -ENOENT) {
3515                         retval = ext4_find_delete_entry(handle, ent->dir,
3516                                                         &ent->dentry->d_name);
3517                 }
3518         }
3519
3520         if (retval) {
3521                 ext4_warning_inode(ent->dir,
3522                                    "Deleting old file: nlink %d, error=%d",
3523                                    ent->dir->i_nlink, retval);
3524         }
3525 }
3526
3527 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3528 {
3529         if (ent->dir_nlink_delta) {
3530                 if (ent->dir_nlink_delta == -1)
3531                         ext4_dec_count(handle, ent->dir);
3532                 else
3533                         ext4_inc_count(handle, ent->dir);
3534                 ext4_mark_inode_dirty(handle, ent->dir);
3535         }
3536 }
3537
3538 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3539                                               int credits, handle_t **h)
3540 {
3541         struct inode *wh;
3542         handle_t *handle;
3543         int retries = 0;
3544
3545         /*
3546          * for inode block, sb block, group summaries,
3547          * and inode bitmap
3548          */
3549         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3550                     EXT4_XATTR_TRANS_BLOCKS + 4);
3551 retry:
3552         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3553                                          &ent->dentry->d_name, 0, NULL,
3554                                          EXT4_HT_DIR, credits);
3555
3556         handle = ext4_journal_current_handle();
3557         if (IS_ERR(wh)) {
3558                 if (handle)
3559                         ext4_journal_stop(handle);
3560                 if (PTR_ERR(wh) == -ENOSPC &&
3561                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3562                         goto retry;
3563         } else {
3564                 *h = handle;
3565                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3566                 wh->i_op = &ext4_special_inode_operations;
3567         }
3568         return wh;
3569 }
3570
3571 /*
3572  * Anybody can rename anything with this: the permission checks are left to the
3573  * higher-level routines.
3574  *
3575  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3576  * while new_{dentry,inode) refers to the destination dentry/inode
3577  * This comes from rename(const char *oldpath, const char *newpath)
3578  */
3579 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3580                        struct inode *new_dir, struct dentry *new_dentry,
3581                        unsigned int flags)
3582 {
3583         handle_t *handle = NULL;
3584         struct ext4_renament old = {
3585                 .dir = old_dir,
3586                 .dentry = old_dentry,
3587                 .inode = d_inode(old_dentry),
3588         };
3589         struct ext4_renament new = {
3590                 .dir = new_dir,
3591                 .dentry = new_dentry,
3592                 .inode = d_inode(new_dentry),
3593         };
3594         int force_reread;
3595         int retval;
3596         struct inode *whiteout = NULL;
3597         int credits;
3598         u8 old_file_type;
3599
3600         if (new.inode && new.inode->i_nlink == 0) {
3601                 EXT4_ERROR_INODE(new.inode,
3602                                  "target of rename is already freed");
3603                 return -EFSCORRUPTED;
3604         }
3605
3606         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3607             (!projid_eq(EXT4_I(new_dir)->i_projid,
3608                         EXT4_I(old_dentry->d_inode)->i_projid)))
3609                 return -EXDEV;
3610
3611         retval = dquot_initialize(old.dir);
3612         if (retval)
3613                 return retval;
3614         retval = dquot_initialize(new.dir);
3615         if (retval)
3616                 return retval;
3617
3618         /* Initialize quotas before so that eventual writes go
3619          * in separate transaction */
3620         if (new.inode) {
3621                 retval = dquot_initialize(new.inode);
3622                 if (retval)
3623                         return retval;
3624         }
3625
3626         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3627         if (IS_ERR(old.bh))
3628                 return PTR_ERR(old.bh);
3629         /*
3630          *  Check for inode number is _not_ due to possible IO errors.
3631          *  We might rmdir the source, keep it as pwd of some process
3632          *  and merrily kill the link to whatever was created under the
3633          *  same name. Goodbye sticky bit ;-<
3634          */
3635         retval = -ENOENT;
3636         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3637                 goto end_rename;
3638
3639         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3640                                  &new.de, &new.inlined);
3641         if (IS_ERR(new.bh)) {
3642                 retval = PTR_ERR(new.bh);
3643                 new.bh = NULL;
3644                 goto end_rename;
3645         }
3646         if (new.bh) {
3647                 if (!new.inode) {
3648                         brelse(new.bh);
3649                         new.bh = NULL;
3650                 }
3651         }
3652         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3653                 ext4_alloc_da_blocks(old.inode);
3654
3655         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3656                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3657         if (!(flags & RENAME_WHITEOUT)) {
3658                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3659                 if (IS_ERR(handle)) {
3660                         retval = PTR_ERR(handle);
3661                         handle = NULL;
3662                         goto end_rename;
3663                 }
3664         } else {
3665                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3666                 if (IS_ERR(whiteout)) {
3667                         retval = PTR_ERR(whiteout);
3668                         whiteout = NULL;
3669                         goto end_rename;
3670                 }
3671         }
3672
3673         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3674                 ext4_handle_sync(handle);
3675
3676         if (S_ISDIR(old.inode->i_mode)) {
3677                 if (new.inode) {
3678                         retval = -ENOTEMPTY;
3679                         if (!ext4_empty_dir(new.inode))
3680                                 goto end_rename;
3681                 } else {
3682                         retval = -EMLINK;
3683                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3684                                 goto end_rename;
3685                 }
3686                 retval = ext4_rename_dir_prepare(handle, &old);
3687                 if (retval)
3688                         goto end_rename;
3689         }
3690         /*
3691          * If we're renaming a file within an inline_data dir and adding or
3692          * setting the new dirent causes a conversion from inline_data to
3693          * extents/blockmap, we need to force the dirent delete code to
3694          * re-read the directory, or else we end up trying to delete a dirent
3695          * from what is now the extent tree root (or a block map).
3696          */
3697         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3698                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3699
3700         old_file_type = old.de->file_type;
3701         if (whiteout) {
3702                 /*
3703                  * Do this before adding a new entry, so the old entry is sure
3704                  * to be still pointing to the valid old entry.
3705                  */
3706                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3707                                      EXT4_FT_CHRDEV);
3708                 if (retval)
3709                         goto end_rename;
3710                 ext4_mark_inode_dirty(handle, whiteout);
3711         }
3712         if (!new.bh) {
3713                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3714                 if (retval)
3715                         goto end_rename;
3716         } else {
3717                 retval = ext4_setent(handle, &new,
3718                                      old.inode->i_ino, old_file_type);
3719                 if (retval)
3720                         goto end_rename;
3721         }
3722         if (force_reread)
3723                 force_reread = !ext4_test_inode_flag(new.dir,
3724                                                      EXT4_INODE_INLINE_DATA);
3725
3726         /*
3727          * Like most other Unix systems, set the ctime for inodes on a
3728          * rename.
3729          */
3730         old.inode->i_ctime = current_time(old.inode);
3731         ext4_mark_inode_dirty(handle, old.inode);
3732
3733         if (!whiteout) {
3734                 /*
3735                  * ok, that's it
3736                  */
3737                 ext4_rename_delete(handle, &old, force_reread);
3738         }
3739
3740         if (new.inode) {
3741                 ext4_dec_count(handle, new.inode);
3742                 new.inode->i_ctime = current_time(new.inode);
3743         }
3744         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3745         ext4_update_dx_flag(old.dir);
3746         if (old.dir_bh) {
3747                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3748                 if (retval)
3749                         goto end_rename;
3750
3751                 ext4_dec_count(handle, old.dir);
3752                 if (new.inode) {
3753                         /* checked ext4_empty_dir above, can't have another
3754                          * parent, ext4_dec_count() won't work for many-linked
3755                          * dirs */
3756                         clear_nlink(new.inode);
3757                 } else {
3758                         ext4_inc_count(handle, new.dir);
3759                         ext4_update_dx_flag(new.dir);
3760                         ext4_mark_inode_dirty(handle, new.dir);
3761                 }
3762         }
3763         ext4_mark_inode_dirty(handle, old.dir);
3764         if (new.inode) {
3765                 ext4_mark_inode_dirty(handle, new.inode);
3766                 if (!new.inode->i_nlink)
3767                         ext4_orphan_add(handle, new.inode);
3768         }
3769         retval = 0;
3770
3771 end_rename:
3772         brelse(old.dir_bh);
3773         brelse(old.bh);
3774         brelse(new.bh);
3775         if (whiteout) {
3776                 if (retval)
3777                         drop_nlink(whiteout);
3778                 unlock_new_inode(whiteout);
3779                 iput(whiteout);
3780         }
3781         if (handle)
3782                 ext4_journal_stop(handle);
3783         return retval;
3784 }
3785
3786 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3787                              struct inode *new_dir, struct dentry *new_dentry)
3788 {
3789         handle_t *handle = NULL;
3790         struct ext4_renament old = {
3791                 .dir = old_dir,
3792                 .dentry = old_dentry,
3793                 .inode = d_inode(old_dentry),
3794         };
3795         struct ext4_renament new = {
3796                 .dir = new_dir,
3797                 .dentry = new_dentry,
3798                 .inode = d_inode(new_dentry),
3799         };
3800         u8 new_file_type;
3801         int retval;
3802         struct timespec64 ctime;
3803
3804         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3805              !projid_eq(EXT4_I(new_dir)->i_projid,
3806                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3807             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3808              !projid_eq(EXT4_I(old_dir)->i_projid,
3809                         EXT4_I(new_dentry->d_inode)->i_projid)))
3810                 return -EXDEV;
3811
3812         retval = dquot_initialize(old.dir);
3813         if (retval)
3814                 return retval;
3815         retval = dquot_initialize(new.dir);
3816         if (retval)
3817                 return retval;
3818
3819         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3820                                  &old.de, &old.inlined);
3821         if (IS_ERR(old.bh))
3822                 return PTR_ERR(old.bh);
3823         /*
3824          *  Check for inode number is _not_ due to possible IO errors.
3825          *  We might rmdir the source, keep it as pwd of some process
3826          *  and merrily kill the link to whatever was created under the
3827          *  same name. Goodbye sticky bit ;-<
3828          */
3829         retval = -ENOENT;
3830         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3831                 goto end_rename;
3832
3833         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3834                                  &new.de, &new.inlined);
3835         if (IS_ERR(new.bh)) {
3836                 retval = PTR_ERR(new.bh);
3837                 new.bh = NULL;
3838                 goto end_rename;
3839         }
3840
3841         /* RENAME_EXCHANGE case: old *and* new must both exist */
3842         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3843                 goto end_rename;
3844
3845         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3846                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3847                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3848         if (IS_ERR(handle)) {
3849                 retval = PTR_ERR(handle);
3850                 handle = NULL;
3851                 goto end_rename;
3852         }
3853
3854         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3855                 ext4_handle_sync(handle);
3856
3857         if (S_ISDIR(old.inode->i_mode)) {
3858                 old.is_dir = true;
3859                 retval = ext4_rename_dir_prepare(handle, &old);
3860                 if (retval)
3861                         goto end_rename;
3862         }
3863         if (S_ISDIR(new.inode->i_mode)) {
3864                 new.is_dir = true;
3865                 retval = ext4_rename_dir_prepare(handle, &new);
3866                 if (retval)
3867                         goto end_rename;
3868         }
3869
3870         /*
3871          * Other than the special case of overwriting a directory, parents'
3872          * nlink only needs to be modified if this is a cross directory rename.
3873          */
3874         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3875                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3876                 new.dir_nlink_delta = -old.dir_nlink_delta;
3877                 retval = -EMLINK;
3878                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3879                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3880                         goto end_rename;
3881         }
3882
3883         new_file_type = new.de->file_type;
3884         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3885         if (retval)
3886                 goto end_rename;
3887
3888         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3889         if (retval)
3890                 goto end_rename;
3891
3892         /*
3893          * Like most other Unix systems, set the ctime for inodes on a
3894          * rename.
3895          */
3896         ctime = current_time(old.inode);
3897         old.inode->i_ctime = ctime;
3898         new.inode->i_ctime = ctime;
3899         ext4_mark_inode_dirty(handle, old.inode);
3900         ext4_mark_inode_dirty(handle, new.inode);
3901
3902         if (old.dir_bh) {
3903                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3904                 if (retval)
3905                         goto end_rename;
3906         }
3907         if (new.dir_bh) {
3908                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3909                 if (retval)
3910                         goto end_rename;
3911         }
3912         ext4_update_dir_count(handle, &old);
3913         ext4_update_dir_count(handle, &new);
3914         retval = 0;
3915
3916 end_rename:
3917         brelse(old.dir_bh);
3918         brelse(new.dir_bh);
3919         brelse(old.bh);
3920         brelse(new.bh);
3921         if (handle)
3922                 ext4_journal_stop(handle);
3923         return retval;
3924 }
3925
3926 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3927                         struct inode *new_dir, struct dentry *new_dentry,
3928                         unsigned int flags)
3929 {
3930         int err;
3931
3932         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3933                 return -EIO;
3934
3935         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3936                 return -EINVAL;
3937
3938         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3939                                      flags);
3940         if (err)
3941                 return err;
3942
3943         if (flags & RENAME_EXCHANGE) {
3944                 return ext4_cross_rename(old_dir, old_dentry,
3945                                          new_dir, new_dentry);
3946         }
3947
3948         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3949 }
3950
3951 /*
3952  * directories can handle most operations...
3953  */
3954 const struct inode_operations ext4_dir_inode_operations = {
3955         .create         = ext4_create,
3956         .lookup         = ext4_lookup,
3957         .link           = ext4_link,
3958         .unlink         = ext4_unlink,
3959         .symlink        = ext4_symlink,
3960         .mkdir          = ext4_mkdir,
3961         .rmdir          = ext4_rmdir,
3962         .mknod          = ext4_mknod,
3963         .tmpfile        = ext4_tmpfile,
3964         .rename         = ext4_rename2,
3965         .setattr        = ext4_setattr,
3966         .getattr        = ext4_getattr,
3967         .listxattr      = ext4_listxattr,
3968         .get_acl        = ext4_get_acl,
3969         .set_acl        = ext4_set_acl,
3970         .fiemap         = ext4_fiemap,
3971 };
3972
3973 const struct inode_operations ext4_special_inode_operations = {
3974         .setattr        = ext4_setattr,
3975         .getattr        = ext4_getattr,
3976         .listxattr      = ext4_listxattr,
3977         .get_acl        = ext4_get_acl,
3978         .set_acl        = ext4_set_acl,
3979 };