fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 113                 bh = ERR_PTR(-EIO);
 114         else
 115                 bh = ext4_bread(NULL, inode, block, 0);
 116         if (IS_ERR(bh)) {
 117                 __ext4_warning(inode->i_sb, func, line,
 118                                "inode #%lu: lblock %lu: comm %s: "
 119                                "error %ld reading directory block",
 120                                inode->i_ino, (unsigned long)block,
 121                                current->comm, PTR_ERR(bh));
 122
 123                 return bh;
 124         }
 125         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 126                 ext4_error_inode(inode, func, line, block,
 127                                  "Directory hole found for htree %s block",
 128                                  (type == INDEX) ? "index" : "leaf");
 129                 return ERR_PTR(-EFSCORRUPTED);
 130         }
 131         if (!bh)
 132                 return NULL;
 133         dirent = (struct ext4_dir_entry *) bh->b_data;
 134         /* Determine whether or not we have an index block */
 135         if (is_dx(inode)) {
 136                 if (block == 0)
 137                         is_dx_block = 1;
 138                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 139                                                 inode->i_sb->s_blocksize) ==
 140                          inode->i_sb->s_blocksize)
 141                         is_dx_block = 1;
 142         }
 143         if (!is_dx_block && type == INDEX) {
 144                 ext4_error_inode(inode, func, line, block,
 145                        "directory leaf block found instead of index block");
 146                 brelse(bh);
 147                 return ERR_PTR(-EFSCORRUPTED);
 148         }
 149         if (!ext4_has_metadata_csum(inode->i_sb) ||
 150             buffer_verified(bh))
 151                 return bh;
 152
 153         /*
 154          * An empty leaf block can get mistaken for a index block; for
 155          * this reason, we can only check the index checksum when the
 156          * caller is sure it should be an index block.
 157          */
 158         if (is_dx_block && type == INDEX) {
 159                 if (ext4_dx_csum_verify(inode, dirent) &&
 160                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 161                         set_buffer_verified(bh);
 162                 else {
 163                         ext4_error_inode_err(inode, func, line, block,
 164                                              EFSBADCRC,
 165                                              "Directory index failed checksum");
 166                         brelse(bh);
 167                         return ERR_PTR(-EFSBADCRC);
 168                 }
 169         }
 170         if (!is_dx_block) {
 171                 if (ext4_dirblock_csum_verify(inode, bh) &&
 172                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 173                         set_buffer_verified(bh);
 174                 else {
 175                         ext4_error_inode_err(inode, func, line, block,
 176                                              EFSBADCRC,
 177                                              "Directory block failed checksum");
 178                         brelse(bh);
 179                         return ERR_PTR(-EFSBADCRC);
 180                 }
 181         }
 182         return bh;
 183 }
 184
 185 #ifdef DX_DEBUG
 186 #define dxtrace(command) command
 187 #else
 188 #define dxtrace(command)
 189 #endif
 190
 191 struct fake_dirent
 192 {
 193         __le32 inode;
 194         __le16 rec_len;
 195         u8 name_len;
 196         u8 file_type;
 197 };
 198
 199 struct dx_countlimit
 200 {
 201         __le16 limit;
 202         __le16 count;
 203 };
 204
 205 struct dx_entry
 206 {
 207         __le32 hash;
 208         __le32 block;
 209 };
 210
 211 /*
 212  * dx_root_info is laid out so that if it should somehow get overlaid by a
 213  * dirent the two low bits of the hash version will be zero.  Therefore, the
 214  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 215  */
 216
 217 struct dx_root
 218 {
 219         struct fake_dirent dot;
 220         char dot_name[4];
 221         struct fake_dirent dotdot;
 222         char dotdot_name[4];
 223         struct dx_root_info
 224         {
 225                 __le32 reserved_zero;
 226                 u8 hash_version;
 227                 u8 info_length; /* 8 */
 228                 u8 indirect_levels;
 229                 u8 unused_flags;
 230         }
 231         info;
 232         struct dx_entry entries[];
 233 };
 234
 235 struct dx_node
 236 {
 237         struct fake_dirent fake;
 238         struct dx_entry entries[];
 239 };
 240
 241
 242 struct dx_frame
 243 {
 244         struct buffer_head *bh;
 245         struct dx_entry *entries;
 246         struct dx_entry *at;
 247 };
 248
 249 struct dx_map_entry
 250 {
 251         u32 hash;
 252         u16 offs;
 253         u16 size;
 254 };
 255
 256 /*
 257  * This goes at the end of each htree block.
 258  */
 259 struct dx_tail {
 260         u32 dt_reserved;
 261         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 262 };
 263
 264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 266 static inline unsigned dx_get_hash(struct dx_entry *entry);
 267 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 268 static unsigned dx_get_count(struct dx_entry *entries);
 269 static unsigned dx_get_limit(struct dx_entry *entries);
 270 static void dx_set_count(struct dx_entry *entries, unsigned value);
 271 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 273 static unsigned dx_node_limit(struct inode *dir);
 274 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 275                                  struct inode *dir,
 276                                  struct dx_hash_info *hinfo,
 277                                  struct dx_frame *frame);
 278 static void dx_release(struct dx_frame *frames);
 279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 280                        unsigned blocksize, struct dx_hash_info *hinfo,
 281                        struct dx_map_entry map[]);
 282 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 283 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 284                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 285 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 286 static void dx_insert_block(struct dx_frame *frame,
 287                                         u32 hash, ext4_lblk_t block);
 288 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 289                                  struct dx_frame *frame,
 290                                  struct dx_frame *frames,
 291                                  __u32 *start_hash);
 292 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 293                 struct ext4_filename *fname,
 294                 struct ext4_dir_entry_2 **res_dir);
 295 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 296                              struct inode *dir, struct inode *inode);
 297
 298 /* checksumming functions */
 299 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 300                                  unsigned int blocksize)
 301 {
 302         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 303
 304         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 305         t->det_rec_len = ext4_rec_len_to_disk(
 306                         sizeof(struct ext4_dir_entry_tail), blocksize);
 307         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 308 }
 309
 310 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 311 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 312                                                    struct buffer_head *bh)
 313 {
 314         struct ext4_dir_entry_tail *t;
 315
 316 #ifdef PARANOID
 317         struct ext4_dir_entry *d, *top;
 318
 319         d = (struct ext4_dir_entry *)bh->b_data;
 320         top = (struct ext4_dir_entry *)(bh->b_data +
 321                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 322                  sizeof(struct ext4_dir_entry_tail)));
 323         while (d < top && d->rec_len)
 324                 d = (struct ext4_dir_entry *)(((void *)d) +
 325                     le16_to_cpu(d->rec_len));
 326
 327         if (d != top)
 328                 return NULL;
 329
 330         t = (struct ext4_dir_entry_tail *)d;
 331 #else
 332         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 333 #endif
 334
 335         if (t->det_reserved_zero1 ||
 336             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 337             t->det_reserved_zero2 ||
 338             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 339                 return NULL;
 340
 341         return t;
 342 }
 343
 344 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 345 {
 346         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 347         struct ext4_inode_info *ei = EXT4_I(inode);
 348         __u32 csum;
 349
 350         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 351         return cpu_to_le32(csum);
 352 }
 353
 354 #define warn_no_space_for_csum(inode)                                   \
 355         __warn_no_space_for_csum((inode), __func__, __LINE__)
 356
 357 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 358                                      unsigned int line)
 359 {
 360         __ext4_warning_inode(inode, func, line,
 361                 "No space for directory leaf checksum. Please run e2fsck -D.");
 362 }
 363
 364 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 365 {
 366         struct ext4_dir_entry_tail *t;
 367
 368         if (!ext4_has_metadata_csum(inode->i_sb))
 369                 return 1;
 370
 371         t = get_dirent_tail(inode, bh);
 372         if (!t) {
 373                 warn_no_space_for_csum(inode);
 374                 return 0;
 375         }
 376
 377         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 378                                                   (char *)t - bh->b_data))
 379                 return 0;
 380
 381         return 1;
 382 }
 383
 384 static void ext4_dirblock_csum_set(struct inode *inode,
 385                                  struct buffer_head *bh)
 386 {
 387         struct ext4_dir_entry_tail *t;
 388
 389         if (!ext4_has_metadata_csum(inode->i_sb))
 390                 return;
 391
 392         t = get_dirent_tail(inode, bh);
 393         if (!t) {
 394                 warn_no_space_for_csum(inode);
 395                 return;
 396         }
 397
 398         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 399                                              (char *)t - bh->b_data);
 400 }
 401
 402 int ext4_handle_dirty_dirblock(handle_t *handle,
 403                                struct inode *inode,
 404                                struct buffer_head *bh)
 405 {
 406         ext4_dirblock_csum_set(inode, bh);
 407         return ext4_handle_dirty_metadata(handle, inode, bh);
 408 }
 409
 410 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 411                                                struct ext4_dir_entry *dirent,
 412                                                int *offset)
 413 {
 414         struct ext4_dir_entry *dp;
 415         struct dx_root_info *root;
 416         int count_offset;
 417
 418         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 419                 count_offset = 8;
 420         else if (le16_to_cpu(dirent->rec_len) == 12) {
 421                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 422                 if (le16_to_cpu(dp->rec_len) !=
 423                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 424                         return NULL;
 425                 root = (struct dx_root_info *)(((void *)dp + 12));
 426                 if (root->reserved_zero ||
 427                     root->info_length != sizeof(struct dx_root_info))
 428                         return NULL;
 429                 count_offset = 32;
 430         } else
 431                 return NULL;
 432
 433         if (offset)
 434                 *offset = count_offset;
 435         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 436 }
 437
 438 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 439                            int count_offset, int count, struct dx_tail *t)
 440 {
 441         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 442         struct ext4_inode_info *ei = EXT4_I(inode);
 443         __u32 csum;
 444         int size;
 445         __u32 dummy_csum = 0;
 446         int offset = offsetof(struct dx_tail, dt_checksum);
 447
 448         size = count_offset + (count * sizeof(struct dx_entry));
 449         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 450         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 451         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 452
 453         return cpu_to_le32(csum);
 454 }
 455
 456 static int ext4_dx_csum_verify(struct inode *inode,
 457                                struct ext4_dir_entry *dirent)
 458 {
 459         struct dx_countlimit *c;
 460         struct dx_tail *t;
 461         int count_offset, limit, count;
 462
 463         if (!ext4_has_metadata_csum(inode->i_sb))
 464                 return 1;
 465
 466         c = get_dx_countlimit(inode, dirent, &count_offset);
 467         if (!c) {
 468                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 469                 return 0;
 470         }
 471         limit = le16_to_cpu(c->limit);
 472         count = le16_to_cpu(c->count);
 473         if (count_offset + (limit * sizeof(struct dx_entry)) >
 474             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 475                 warn_no_space_for_csum(inode);
 476                 return 0;
 477         }
 478         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 479
 480         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 481                                             count, t))
 482                 return 0;
 483         return 1;
 484 }
 485
 486 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 487 {
 488         struct dx_countlimit *c;
 489         struct dx_tail *t;
 490         int count_offset, limit, count;
 491
 492         if (!ext4_has_metadata_csum(inode->i_sb))
 493                 return;
 494
 495         c = get_dx_countlimit(inode, dirent, &count_offset);
 496         if (!c) {
 497                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 498                 return;
 499         }
 500         limit = le16_to_cpu(c->limit);
 501         count = le16_to_cpu(c->count);
 502         if (count_offset + (limit * sizeof(struct dx_entry)) >
 503             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 504                 warn_no_space_for_csum(inode);
 505                 return;
 506         }
 507         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 508
 509         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 510 }
 511
 512 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 513                                             struct inode *inode,
 514                                             struct buffer_head *bh)
 515 {
 516         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 517         return ext4_handle_dirty_metadata(handle, inode, bh);
 518 }
 519
 520 /*
 521  * p is at least 6 bytes before the end of page
 522  */
 523 static inline struct ext4_dir_entry_2 *
 524 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 525 {
 526         return (struct ext4_dir_entry_2 *)((char *)p +
 527                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 528 }
 529
 530 /*
 531  * Future: use high four bits of block for coalesce-on-delete flags
 532  * Mask them off for now.
 533  */
 534
 535 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 536 {
 537         return le32_to_cpu(entry->block) & 0x0fffffff;
 538 }
 539
 540 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 541 {
 542         entry->block = cpu_to_le32(value);
 543 }
 544
 545 static inline unsigned dx_get_hash(struct dx_entry *entry)
 546 {
 547         return le32_to_cpu(entry->hash);
 548 }
 549
 550 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 551 {
 552         entry->hash = cpu_to_le32(value);
 553 }
 554
 555 static inline unsigned dx_get_count(struct dx_entry *entries)
 556 {
 557         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 558 }
 559
 560 static inline unsigned dx_get_limit(struct dx_entry *entries)
 561 {
 562         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 563 }
 564
 565 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 566 {
 567         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 568 }
 569
 570 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 571 {
 572         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 573 }
 574
 575 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 576 {
 577         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 578                 EXT4_DIR_REC_LEN(2) - infosize;
 579
 580         if (ext4_has_metadata_csum(dir->i_sb))
 581                 entry_space -= sizeof(struct dx_tail);
 582         return entry_space / sizeof(struct dx_entry);
 583 }
 584
 585 static inline unsigned dx_node_limit(struct inode *dir)
 586 {
 587         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 588
 589         if (ext4_has_metadata_csum(dir->i_sb))
 590                 entry_space -= sizeof(struct dx_tail);
 591         return entry_space / sizeof(struct dx_entry);
 592 }
 593
 594 /*
 595  * Debug
 596  */
 597 #ifdef DX_DEBUG
 598 static void dx_show_index(char * label, struct dx_entry *entries)
 599 {
 600         int i, n = dx_get_count (entries);
 601         printk(KERN_DEBUG "%s index", label);
 602         for (i = 0; i < n; i++) {
 603                 printk(KERN_CONT " %x->%lu",
 604                        i ? dx_get_hash(entries + i) : 0,
 605                        (unsigned long)dx_get_block(entries + i));
 606         }
 607         printk(KERN_CONT "\n");
 608 }
 609
 610 struct stats
 611 {
 612         unsigned names;
 613         unsigned space;
 614         unsigned bcount;
 615 };
 616
 617 static struct stats dx_show_leaf(struct inode *dir,
 618                                 struct dx_hash_info *hinfo,
 619                                 struct ext4_dir_entry_2 *de,
 620                                 int size, int show_names)
 621 {
 622         unsigned names = 0, space = 0;
 623         char *base = (char *) de;
 624         struct dx_hash_info h = *hinfo;
 625
 626         printk("names: ");
 627         while ((char *) de < base + size)
 628         {
 629                 if (de->inode)
 630                 {
 631                         if (show_names)
 632                         {
 633 #ifdef CONFIG_FS_ENCRYPTION
 634                                 int len;
 635                                 char *name;
 636                                 struct fscrypt_str fname_crypto_str =
 637                                         FSTR_INIT(NULL, 0);
 638                                 int res = 0;
 639
 640                                 name  = de->name;
 641                                 len = de->name_len;
 642                                 if (!IS_ENCRYPTED(dir)) {
 643                                         /* Directory is not encrypted */
 644                                         ext4fs_dirhash(dir, de->name,
 645                                                 de->name_len, &h);
 646                                         printk("%*.s:(U)%x.%u ", len,
 647                                                name, h.hash,
 648                                                (unsigned) ((char *) de
 649                                                            - base));
 650                                 } else {
 651                                         struct fscrypt_str de_name =
 652                                                 FSTR_INIT(name, len);
 653
 654                                         /* Directory is encrypted */
 655                                         res = fscrypt_fname_alloc_buffer(
 656                                                 len, &fname_crypto_str);
 657                                         if (res)
 658                                                 printk(KERN_WARNING "Error "
 659                                                         "allocating crypto "
 660                                                         "buffer--skipping "
 661                                                         "crypto\n");
 662                                         res = fscrypt_fname_disk_to_usr(dir,
 663                                                 0, 0, &de_name,
 664                                                 &fname_crypto_str);
 665                                         if (res) {
 666                                                 printk(KERN_WARNING "Error "
 667                                                         "converting filename "
 668                                                         "from disk to usr"
 669                                                         "\n");
 670                                                 name = "??";
 671                                                 len = 2;
 672                                         } else {
 673                                                 name = fname_crypto_str.name;
 674                                                 len = fname_crypto_str.len;
 675                                         }
 676                                         ext4fs_dirhash(dir, de->name,
 677                                                        de->name_len, &h);
 678                                         printk("%*.s:(E)%x.%u ", len, name,
 679                                                h.hash, (unsigned) ((char *) de
 680                                                                    - base));
 681                                         fscrypt_fname_free_buffer(
 682                                                         &fname_crypto_str);
 683                                 }
 684 #else
 685                                 int len = de->name_len;
 686                                 char *name = de->name;
 687                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 688                                 printk("%*.s:%x.%u ", len, name, h.hash,
 689                                        (unsigned) ((char *) de - base));
 690 #endif
 691                         }
 692                         space += EXT4_DIR_REC_LEN(de->name_len);
 693                         names++;
 694                 }
 695                 de = ext4_next_entry(de, size);
 696         }
 697         printk(KERN_CONT "(%i)\n", names);
 698         return (struct stats) { names, space, 1 };
 699 }
 700
 701 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 702                              struct dx_entry *entries, int levels)
 703 {
 704         unsigned blocksize = dir->i_sb->s_blocksize;
 705         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 706         unsigned bcount = 0;
 707         struct buffer_head *bh;
 708         printk("%i indexed blocks...\n", count);
 709         for (i = 0; i < count; i++, entries++)
 710         {
 711                 ext4_lblk_t block = dx_get_block(entries);
 712                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 713                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 714                 struct stats stats;
 715                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 716                 bh = ext4_bread(NULL,dir, block, 0);
 717                 if (!bh || IS_ERR(bh))
 718                         continue;
 719                 stats = levels?
 720                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 721                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 722                         bh->b_data, blocksize, 0);
 723                 names += stats.names;
 724                 space += stats.space;
 725                 bcount += stats.bcount;
 726                 brelse(bh);
 727         }
 728         if (bcount)
 729                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 730                        levels ? "" : "   ", names, space/bcount,
 731                        (space/bcount)*100/blocksize);
 732         return (struct stats) { names, space, bcount};
 733 }
 734
 735 /*
 736  * Linear search cross check
 737  */
 738 static inline void htree_rep_invariant_check(struct dx_entry *at,
 739                                              struct dx_entry *target,
 740                                              u32 hash, unsigned int n)
 741 {
 742         while (n--) {
 743                 dxtrace(printk(KERN_CONT ","));
 744                 if (dx_get_hash(++at) > hash) {
 745                         at--;
 746                         break;
 747                 }
 748         }
 749         ASSERT(at == target - 1);
 750 }
 751 #else /* DX_DEBUG */
 752 static inline void htree_rep_invariant_check(struct dx_entry *at,
 753                                              struct dx_entry *target,
 754                                              u32 hash, unsigned int n)
 755 {
 756 }
 757 #endif /* DX_DEBUG */
 758
 759 /*
 760  * Probe for a directory leaf block to search.
 761  *
 762  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 763  * error in the directory index, and the caller should fall back to
 764  * searching the directory normally.  The callers of dx_probe **MUST**
 765  * check for this error code, and make sure it never gets reflected
 766  * back to userspace.
 767  */
 768 static struct dx_frame *
 769 dx_probe(struct ext4_filename *fname, struct inode *dir,
 770          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 771 {
 772         unsigned count, indirect;
 773         struct dx_entry *at, *entries, *p, *q, *m;
 774         struct dx_root *root;
 775         struct dx_frame *frame = frame_in;
 776         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 777         u32 hash;
 778
 779         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 780         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 781         if (IS_ERR(frame->bh))
 782                 return (struct dx_frame *) frame->bh;
 783
 784         root = (struct dx_root *) frame->bh->b_data;
 785         if (root->info.hash_version != DX_HASH_TEA &&
 786             root->info.hash_version != DX_HASH_HALF_MD4 &&
 787             root->info.hash_version != DX_HASH_LEGACY) {
 788                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 789                                    root->info.hash_version);
 790                 goto fail;
 791         }
 792         if (fname)
 793                 hinfo = &fname->hinfo;
 794         hinfo->hash_version = root->info.hash_version;
 795         if (hinfo->hash_version <= DX_HASH_TEA)
 796                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 797         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 798         if (fname && fname_name(fname))
 799                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 800         hash = hinfo->hash;
 801
 802         if (root->info.unused_flags & 1) {
 803                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 804                                    root->info.unused_flags);
 805                 goto fail;
 806         }
 807
 808         indirect = root->info.indirect_levels;
 809         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 810                 ext4_warning(dir->i_sb,
 811                              "Directory (ino: %lu) htree depth %#06x exceed"
 812                              "supported value", dir->i_ino,
 813                              ext4_dir_htree_level(dir->i_sb));
 814                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 815                         ext4_warning(dir->i_sb, "Enable large directory "
 816                                                 "feature to access it");
 817                 }
 818                 goto fail;
 819         }
 820
 821         entries = (struct dx_entry *)(((char *)&root->info) +
 822                                       root->info.info_length);
 823
 824         if (dx_get_limit(entries) != dx_root_limit(dir,
 825                                                    root->info.info_length)) {
 826                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 827                                    dx_get_limit(entries),
 828                                    dx_root_limit(dir, root->info.info_length));
 829                 goto fail;
 830         }
 831
 832         dxtrace(printk("Look up %x", hash));
 833         while (1) {
 834                 count = dx_get_count(entries);
 835                 if (!count || count > dx_get_limit(entries)) {
 836                         ext4_warning_inode(dir,
 837                                            "dx entry: count %u beyond limit %u",
 838                                            count, dx_get_limit(entries));
 839                         goto fail;
 840                 }
 841
 842                 p = entries + 1;
 843                 q = entries + count - 1;
 844                 while (p <= q) {
 845                         m = p + (q - p) / 2;
 846                         dxtrace(printk(KERN_CONT "."));
 847                         if (dx_get_hash(m) > hash)
 848                                 q = m - 1;
 849                         else
 850                                 p = m + 1;
 851                 }
 852
 853                 htree_rep_invariant_check(entries, p, hash, count - 1);
 854
 855                 at = p - 1;
 856                 dxtrace(printk(KERN_CONT " %x->%u\n",
 857                                at == entries ? 0 : dx_get_hash(at),
 858                                dx_get_block(at)));
 859                 frame->entries = entries;
 860                 frame->at = at;
 861                 if (!indirect--)
 862                         return frame;
 863                 frame++;
 864                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 865                 if (IS_ERR(frame->bh)) {
 866                         ret_err = (struct dx_frame *) frame->bh;
 867                         frame->bh = NULL;
 868                         goto fail;
 869                 }
 870                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 871
 872                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 873                         ext4_warning_inode(dir,
 874                                 "dx entry: limit %u != node limit %u",
 875                                 dx_get_limit(entries), dx_node_limit(dir));
 876                         goto fail;
 877                 }
 878         }
 879 fail:
 880         while (frame >= frame_in) {
 881                 brelse(frame->bh);
 882                 frame--;
 883         }
 884
 885         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 886                 ext4_warning_inode(dir,
 887                         "Corrupt directory, running e2fsck is recommended");
 888         return ret_err;
 889 }
 890
 891 static void dx_release(struct dx_frame *frames)
 892 {
 893         struct dx_root_info *info;
 894         int i;
 895         unsigned int indirect_levels;
 896
 897         if (frames[0].bh == NULL)
 898                 return;
 899
 900         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 901         /* save local copy, "info" may be freed after brelse() */
 902         indirect_levels = info->indirect_levels;
 903         for (i = 0; i <= indirect_levels; i++) {
 904                 if (frames[i].bh == NULL)
 905                         break;
 906                 brelse(frames[i].bh);
 907                 frames[i].bh = NULL;
 908         }
 909 }
 910
 911 /*
 912  * This function increments the frame pointer to search the next leaf
 913  * block, and reads in the necessary intervening nodes if the search
 914  * should be necessary.  Whether or not the search is necessary is
 915  * controlled by the hash parameter.  If the hash value is even, then
 916  * the search is only continued if the next block starts with that
 917  * hash value.  This is used if we are searching for a specific file.
 918  *
 919  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 920  *
 921  * This function returns 1 if the caller should continue to search,
 922  * or 0 if it should not.  If there is an error reading one of the
 923  * index blocks, it will a negative error code.
 924  *
 925  * If start_hash is non-null, it will be filled in with the starting
 926  * hash of the next page.
 927  */
 928 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 929                                  struct dx_frame *frame,
 930                                  struct dx_frame *frames,
 931                                  __u32 *start_hash)
 932 {
 933         struct dx_frame *p;
 934         struct buffer_head *bh;
 935         int num_frames = 0;
 936         __u32 bhash;
 937
 938         p = frame;
 939         /*
 940          * Find the next leaf page by incrementing the frame pointer.
 941          * If we run out of entries in the interior node, loop around and
 942          * increment pointer in the parent node.  When we break out of
 943          * this loop, num_frames indicates the number of interior
 944          * nodes need to be read.
 945          */
 946         while (1) {
 947                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 948                         break;
 949                 if (p == frames)
 950                         return 0;
 951                 num_frames++;
 952                 p--;
 953         }
 954
 955         /*
 956          * If the hash is 1, then continue only if the next page has a
 957          * continuation hash of any value.  This is used for readdir
 958          * handling.  Otherwise, check to see if the hash matches the
 959          * desired contiuation hash.  If it doesn't, return since
 960          * there's no point to read in the successive index pages.
 961          */
 962         bhash = dx_get_hash(p->at);
 963         if (start_hash)
 964                 *start_hash = bhash;
 965         if ((hash & 1) == 0) {
 966                 if ((bhash & ~1) != hash)
 967                         return 0;
 968         }
 969         /*
 970          * If the hash is HASH_NB_ALWAYS, we always go to the next
 971          * block so no check is necessary
 972          */
 973         while (num_frames--) {
 974                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 975                 if (IS_ERR(bh))
 976                         return PTR_ERR(bh);
 977                 p++;
 978                 brelse(p->bh);
 979                 p->bh = bh;
 980                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 981         }
 982         return 1;
 983 }
 984
 985
 986 /*
 987  * This function fills a red-black tree with information from a
 988  * directory block.  It returns the number directory entries loaded
 989  * into the tree.  If there is an error it is returned in err.
 990  */
 991 static int htree_dirblock_to_tree(struct file *dir_file,
 992                                   struct inode *dir, ext4_lblk_t block,
 993                                   struct dx_hash_info *hinfo,
 994                                   __u32 start_hash, __u32 start_minor_hash)
 995 {
 996         struct buffer_head *bh;
 997         struct ext4_dir_entry_2 *de, *top;
 998         int err = 0, count = 0;
 999         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1000
1001         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1002                                                         (unsigned long)block));
1003         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1004         if (IS_ERR(bh))
1005                 return PTR_ERR(bh);
1006
1007         de = (struct ext4_dir_entry_2 *) bh->b_data;
1008         top = (struct ext4_dir_entry_2 *) ((char *) de +
1009                                            dir->i_sb->s_blocksize -
1010                                            EXT4_DIR_REC_LEN(0));
1011         /* Check if the directory is encrypted */
1012         if (IS_ENCRYPTED(dir)) {
1013                 err = fscrypt_prepare_readdir(dir);
1014                 if (err < 0) {
1015                         brelse(bh);
1016                         return err;
1017                 }
1018                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1019                                                  &fname_crypto_str);
1020                 if (err < 0) {
1021                         brelse(bh);
1022                         return err;
1023                 }
1024         }
1025
1026         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1027                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1028                                 bh->b_data, bh->b_size,
1029                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1030                                          + ((char *)de - bh->b_data))) {
1031                         /* silently ignore the rest of the block */
1032                         break;
1033                 }
1034                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1035                 if ((hinfo->hash < start_hash) ||
1036                     ((hinfo->hash == start_hash) &&
1037                      (hinfo->minor_hash < start_minor_hash)))
1038                         continue;
1039                 if (de->inode == 0)
1040                         continue;
1041                 if (!IS_ENCRYPTED(dir)) {
1042                         tmp_str.name = de->name;
1043                         tmp_str.len = de->name_len;
1044                         err = ext4_htree_store_dirent(dir_file,
1045                                    hinfo->hash, hinfo->minor_hash, de,
1046                                    &tmp_str);
1047                 } else {
1048                         int save_len = fname_crypto_str.len;
1049                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1050                                                                 de->name_len);
1051
1052                         /* Directory is encrypted */
1053                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1054                                         hinfo->minor_hash, &de_name,
1055                                         &fname_crypto_str);
1056                         if (err) {
1057                                 count = err;
1058                                 goto errout;
1059                         }
1060                         err = ext4_htree_store_dirent(dir_file,
1061                                    hinfo->hash, hinfo->minor_hash, de,
1062                                         &fname_crypto_str);
1063                         fname_crypto_str.len = save_len;
1064                 }
1065                 if (err != 0) {
1066                         count = err;
1067                         goto errout;
1068                 }
1069                 count++;
1070         }
1071 errout:
1072         brelse(bh);
1073         fscrypt_fname_free_buffer(&fname_crypto_str);
1074         return count;
1075 }
1076
1077
1078 /*
1079  * This function fills a red-black tree with information from a
1080  * directory.  We start scanning the directory in hash order, starting
1081  * at start_hash and start_minor_hash.
1082  *
1083  * This function returns the number of entries inserted into the tree,
1084  * or a negative error code.
1085  */
1086 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1087                          __u32 start_minor_hash, __u32 *next_hash)
1088 {
1089         struct dx_hash_info hinfo;
1090         struct ext4_dir_entry_2 *de;
1091         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1092         struct inode *dir;
1093         ext4_lblk_t block;
1094         int count = 0;
1095         int ret, err;
1096         __u32 hashval;
1097         struct fscrypt_str tmp_str;
1098
1099         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1100                        start_hash, start_minor_hash));
1101         dir = file_inode(dir_file);
1102         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1103                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1104                 if (hinfo.hash_version <= DX_HASH_TEA)
1105                         hinfo.hash_version +=
1106                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1107                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1108                 if (ext4_has_inline_data(dir)) {
1109                         int has_inline_data = 1;
1110                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1111                                                        &hinfo, start_hash,
1112                                                        start_minor_hash,
1113                                                        &has_inline_data);
1114                         if (has_inline_data) {
1115                                 *next_hash = ~0;
1116                                 return count;
1117                         }
1118                 }
1119                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1120                                                start_hash, start_minor_hash);
1121                 *next_hash = ~0;
1122                 return count;
1123         }
1124         hinfo.hash = start_hash;
1125         hinfo.minor_hash = 0;
1126         frame = dx_probe(NULL, dir, &hinfo, frames);
1127         if (IS_ERR(frame))
1128                 return PTR_ERR(frame);
1129
1130         /* Add '.' and '..' from the htree header */
1131         if (!start_hash && !start_minor_hash) {
1132                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1133                 tmp_str.name = de->name;
1134                 tmp_str.len = de->name_len;
1135                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1136                                               de, &tmp_str);
1137                 if (err != 0)
1138                         goto errout;
1139                 count++;
1140         }
1141         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1142                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1143                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1144                 tmp_str.name = de->name;
1145                 tmp_str.len = de->name_len;
1146                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1147                                               de, &tmp_str);
1148                 if (err != 0)
1149                         goto errout;
1150                 count++;
1151         }
1152
1153         while (1) {
1154                 if (fatal_signal_pending(current)) {
1155                         err = -ERESTARTSYS;
1156                         goto errout;
1157                 }
1158                 cond_resched();
1159                 block = dx_get_block(frame->at);
1160                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1161                                              start_hash, start_minor_hash);
1162                 if (ret < 0) {
1163                         err = ret;
1164                         goto errout;
1165                 }
1166                 count += ret;
1167                 hashval = ~0;
1168                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1169                                             frame, frames, &hashval);
1170                 *next_hash = hashval;
1171                 if (ret < 0) {
1172                         err = ret;
1173                         goto errout;
1174                 }
1175                 /*
1176                  * Stop if:  (a) there are no more entries, or
1177                  * (b) we have inserted at least one entry and the
1178                  * next hash value is not a continuation
1179                  */
1180                 if ((ret == 0) ||
1181                     (count && ((hashval & 1) == 0)))
1182                         break;
1183         }
1184         dx_release(frames);
1185         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1186                        "next hash: %x\n", count, *next_hash));
1187         return count;
1188 errout:
1189         dx_release(frames);
1190         return (err);
1191 }
1192
1193 static inline int search_dirblock(struct buffer_head *bh,
1194                                   struct inode *dir,
1195                                   struct ext4_filename *fname,
1196                                   unsigned int offset,
1197                                   struct ext4_dir_entry_2 **res_dir)
1198 {
1199         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1200                                fname, offset, res_dir);
1201 }
1202
1203 /*
1204  * Directory block splitting, compacting
1205  */
1206
1207 /*
1208  * Create map of hash values, offsets, and sizes, stored at end of block.
1209  * Returns number of entries mapped.
1210  */
1211 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1212                        unsigned blocksize, struct dx_hash_info *hinfo,
1213                        struct dx_map_entry *map_tail)
1214 {
1215         int count = 0;
1216         char *base = (char *) de;
1217         struct dx_hash_info h = *hinfo;
1218
1219         while ((char *) de < base + blocksize) {
1220                 if (de->name_len && de->inode) {
1221                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1222                         map_tail--;
1223                         map_tail->hash = h.hash;
1224                         map_tail->offs = ((char *) de - base)>>2;
1225                         map_tail->size = le16_to_cpu(de->rec_len);
1226                         count++;
1227                         cond_resched();
1228                 }
1229                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1230                 de = ext4_next_entry(de, blocksize);
1231         }
1232         return count;
1233 }
1234
1235 /* Sort map by hash value */
1236 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1237 {
1238         struct dx_map_entry *p, *q, *top = map + count - 1;
1239         int more;
1240         /* Combsort until bubble sort doesn't suck */
1241         while (count > 2) {
1242                 count = count*10/13;
1243                 if (count - 9 < 2) /* 9, 10 -> 11 */
1244                         count = 11;
1245                 for (p = top, q = p - count; q >= map; p--, q--)
1246                         if (p->hash < q->hash)
1247                                 swap(*p, *q);
1248         }
1249         /* Garden variety bubble sort */
1250         do {
1251                 more = 0;
1252                 q = top;
1253                 while (q-- > map) {
1254                         if (q[1].hash >= q[0].hash)
1255                                 continue;
1256                         swap(*(q+1), *q);
1257                         more = 1;
1258                 }
1259         } while(more);
1260 }
1261
1262 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1263 {
1264         struct dx_entry *entries = frame->entries;
1265         struct dx_entry *old = frame->at, *new = old + 1;
1266         int count = dx_get_count(entries);
1267
1268         ASSERT(count < dx_get_limit(entries));
1269         ASSERT(old < entries + count);
1270         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1271         dx_set_hash(new, hash);
1272         dx_set_block(new, block);
1273         dx_set_count(entries, count + 1);
1274 }
1275
1276 #ifdef CONFIG_UNICODE
1277 /*
1278  * Test whether a case-insensitive directory entry matches the filename
1279  * being searched for.  If quick is set, assume the name being looked up
1280  * is already in the casefolded form.
1281  *
1282  * Returns: 0 if the directory entry matches, more than 0 if it
1283  * doesn't match or less than zero on error.
1284  */
1285 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1286                     const struct qstr *entry, bool quick)
1287 {
1288         const struct super_block *sb = parent->i_sb;
1289         const struct unicode_map *um = sb->s_encoding;
1290         int ret;
1291
1292         if (quick)
1293                 ret = utf8_strncasecmp_folded(um, name, entry);
1294         else
1295                 ret = utf8_strncasecmp(um, name, entry);
1296
1297         if (ret < 0) {
1298                 /* Handle invalid character sequence as either an error
1299                  * or as an opaque byte sequence.
1300                  */
1301                 if (sb_has_strict_encoding(sb))
1302                         return -EINVAL;
1303
1304                 if (name->len != entry->len)
1305                         return 1;
1306
1307                 return !!memcmp(name->name, entry->name, name->len);
1308         }
1309
1310         return ret;
1311 }
1312
1313 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1314                                   struct fscrypt_str *cf_name)
1315 {
1316         int len;
1317
1318         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1319                 cf_name->name = NULL;
1320                 return;
1321         }
1322
1323         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1324         if (!cf_name->name)
1325                 return;
1326
1327         len = utf8_casefold(dir->i_sb->s_encoding,
1328                             iname, cf_name->name,
1329                             EXT4_NAME_LEN);
1330         if (len <= 0) {
1331                 kfree(cf_name->name);
1332                 cf_name->name = NULL;
1333                 return;
1334         }
1335         cf_name->len = (unsigned) len;
1336
1337 }
1338 #endif
1339
1340 /*
1341  * Test whether a directory entry matches the filename being searched for.
1342  *
1343  * Return: %true if the directory entry matches, otherwise %false.
1344  */
1345 static inline bool ext4_match(const struct inode *parent,
1346                               const struct ext4_filename *fname,
1347                               const struct ext4_dir_entry_2 *de)
1348 {
1349         struct fscrypt_name f;
1350 #ifdef CONFIG_UNICODE
1351         const struct qstr entry = {.name = de->name, .len = de->name_len};
1352 #endif
1353
1354         if (!de->inode)
1355                 return false;
1356
1357         f.usr_fname = fname->usr_fname;
1358         f.disk_name = fname->disk_name;
1359 #ifdef CONFIG_FS_ENCRYPTION
1360         f.crypto_buf = fname->crypto_buf;
1361 #endif
1362
1363 #ifdef CONFIG_UNICODE
1364         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1365                 if (fname->cf_name.name) {
1366                         struct qstr cf = {.name = fname->cf_name.name,
1367                                           .len = fname->cf_name.len};
1368                         return !ext4_ci_compare(parent, &cf, &entry, true);
1369                 }
1370                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1371                                         false);
1372         }
1373 #endif
1374
1375         return fscrypt_match_name(&f, de->name, de->name_len);
1376 }
1377
1378 /*
1379  * Returns 0 if not found, -1 on failure, and 1 on success
1380  */
1381 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1382                     struct inode *dir, struct ext4_filename *fname,
1383                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1384 {
1385         struct ext4_dir_entry_2 * de;
1386         char * dlimit;
1387         int de_len;
1388
1389         de = (struct ext4_dir_entry_2 *)search_buf;
1390         dlimit = search_buf + buf_size;
1391         while ((char *) de < dlimit) {
1392                 /* this code is executed quadratically often */
1393                 /* do minimal checking `by hand' */
1394                 if ((char *) de + de->name_len <= dlimit &&
1395                     ext4_match(dir, fname, de)) {
1396                         /* found a match - just to be sure, do
1397                          * a full check */
1398                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1399                                                  buf_size, offset))
1400                                 return -1;
1401                         *res_dir = de;
1402                         return 1;
1403                 }
1404                 /* prevent looping on a bad block */
1405                 de_len = ext4_rec_len_from_disk(de->rec_len,
1406                                                 dir->i_sb->s_blocksize);
1407                 if (de_len <= 0)
1408                         return -1;
1409                 offset += de_len;
1410                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1411         }
1412         return 0;
1413 }
1414
1415 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1416                                struct ext4_dir_entry *de)
1417 {
1418         struct super_block *sb = dir->i_sb;
1419
1420         if (!is_dx(dir))
1421                 return 0;
1422         if (block == 0)
1423                 return 1;
1424         if (de->inode == 0 &&
1425             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1426                         sb->s_blocksize)
1427                 return 1;
1428         return 0;
1429 }
1430
1431 /*
1432  *      __ext4_find_entry()
1433  *
1434  * finds an entry in the specified directory with the wanted name. It
1435  * returns the cache buffer in which the entry was found, and the entry
1436  * itself (as a parameter - res_dir). It does NOT read the inode of the
1437  * entry - you'll have to do that yourself if you want to.
1438  *
1439  * The returned buffer_head has ->b_count elevated.  The caller is expected
1440  * to brelse() it when appropriate.
1441  */
1442 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1443                                              struct ext4_filename *fname,
1444                                              struct ext4_dir_entry_2 **res_dir,
1445                                              int *inlined)
1446 {
1447         struct super_block *sb;
1448         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1449         struct buffer_head *bh, *ret = NULL;
1450         ext4_lblk_t start, block;
1451         const u8 *name = fname->usr_fname->name;
1452         size_t ra_max = 0;      /* Number of bh's in the readahead
1453                                    buffer, bh_use[] */
1454         size_t ra_ptr = 0;      /* Current index into readahead
1455                                    buffer */
1456         ext4_lblk_t  nblocks;
1457         int i, namelen, retval;
1458
1459         *res_dir = NULL;
1460         sb = dir->i_sb;
1461         namelen = fname->usr_fname->len;
1462         if (namelen > EXT4_NAME_LEN)
1463                 return NULL;
1464
1465         if (ext4_has_inline_data(dir)) {
1466                 int has_inline_data = 1;
1467                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1468                                              &has_inline_data);
1469                 if (has_inline_data) {
1470                         if (inlined)
1471                                 *inlined = 1;
1472                         goto cleanup_and_exit;
1473                 }
1474         }
1475
1476         if ((namelen <= 2) && (name[0] == '.') &&
1477             (name[1] == '.' || name[1] == '\0')) {
1478                 /*
1479                  * "." or ".." will only be in the first block
1480                  * NFS may look up ".."; "." should be handled by the VFS
1481                  */
1482                 block = start = 0;
1483                 nblocks = 1;
1484                 goto restart;
1485         }
1486         if (is_dx(dir)) {
1487                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1488                 /*
1489                  * On success, or if the error was file not found,
1490                  * return.  Otherwise, fall back to doing a search the
1491                  * old fashioned way.
1492                  */
1493                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1494                         goto cleanup_and_exit;
1495                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1496                                "falling back\n"));
1497                 ret = NULL;
1498         }
1499         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1500         if (!nblocks) {
1501                 ret = NULL;
1502                 goto cleanup_and_exit;
1503         }
1504         start = EXT4_I(dir)->i_dir_start_lookup;
1505         if (start >= nblocks)
1506                 start = 0;
1507         block = start;
1508 restart:
1509         do {
1510                 /*
1511                  * We deal with the read-ahead logic here.
1512                  */
1513                 cond_resched();
1514                 if (ra_ptr >= ra_max) {
1515                         /* Refill the readahead buffer */
1516                         ra_ptr = 0;
1517                         if (block < start)
1518                                 ra_max = start - block;
1519                         else
1520                                 ra_max = nblocks - block;
1521                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1522                         retval = ext4_bread_batch(dir, block, ra_max,
1523                                                   false /* wait */, bh_use);
1524                         if (retval) {
1525                                 ret = ERR_PTR(retval);
1526                                 ra_max = 0;
1527                                 goto cleanup_and_exit;
1528                         }
1529                 }
1530                 if ((bh = bh_use[ra_ptr++]) == NULL)
1531                         goto next;
1532                 wait_on_buffer(bh);
1533                 if (!buffer_uptodate(bh)) {
1534                         EXT4_ERROR_INODE_ERR(dir, EIO,
1535                                              "reading directory lblock %lu",
1536                                              (unsigned long) block);
1537                         brelse(bh);
1538                         ret = ERR_PTR(-EIO);
1539                         goto cleanup_and_exit;
1540                 }
1541                 if (!buffer_verified(bh) &&
1542                     !is_dx_internal_node(dir, block,
1543                                          (struct ext4_dir_entry *)bh->b_data) &&
1544                     !ext4_dirblock_csum_verify(dir, bh)) {
1545                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1546                                              "checksumming directory "
1547                                              "block %lu", (unsigned long)block);
1548                         brelse(bh);
1549                         ret = ERR_PTR(-EFSBADCRC);
1550                         goto cleanup_and_exit;
1551                 }
1552                 set_buffer_verified(bh);
1553                 i = search_dirblock(bh, dir, fname,
1554                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1555                 if (i == 1) {
1556                         EXT4_I(dir)->i_dir_start_lookup = block;
1557                         ret = bh;
1558                         goto cleanup_and_exit;
1559                 } else {
1560                         brelse(bh);
1561                         if (i < 0)
1562                                 goto cleanup_and_exit;
1563                 }
1564         next:
1565                 if (++block >= nblocks)
1566                         block = 0;
1567         } while (block != start);
1568
1569         /*
1570          * If the directory has grown while we were searching, then
1571          * search the last part of the directory before giving up.
1572          */
1573         block = nblocks;
1574         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1575         if (block < nblocks) {
1576                 start = 0;
1577                 goto restart;
1578         }
1579
1580 cleanup_and_exit:
1581         /* Clean up the read-ahead blocks */
1582         for (; ra_ptr < ra_max; ra_ptr++)
1583                 brelse(bh_use[ra_ptr]);
1584         return ret;
1585 }
1586
1587 static struct buffer_head *ext4_find_entry(struct inode *dir,
1588                                            const struct qstr *d_name,
1589                                            struct ext4_dir_entry_2 **res_dir,
1590                                            int *inlined)
1591 {
1592         int err;
1593         struct ext4_filename fname;
1594         struct buffer_head *bh;
1595
1596         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1597         if (err == -ENOENT)
1598                 return NULL;
1599         if (err)
1600                 return ERR_PTR(err);
1601
1602         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1603
1604         ext4_fname_free_filename(&fname);
1605         return bh;
1606 }
1607
1608 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1609                                              struct dentry *dentry,
1610                                              struct ext4_dir_entry_2 **res_dir)
1611 {
1612         int err;
1613         struct ext4_filename fname;
1614         struct buffer_head *bh;
1615
1616         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1617         generic_set_encrypted_ci_d_ops(dentry);
1618         if (err == -ENOENT)
1619                 return NULL;
1620         if (err)
1621                 return ERR_PTR(err);
1622
1623         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1624
1625         ext4_fname_free_filename(&fname);
1626         return bh;
1627 }
1628
1629 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1630                         struct ext4_filename *fname,
1631                         struct ext4_dir_entry_2 **res_dir)
1632 {
1633         struct super_block * sb = dir->i_sb;
1634         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1635         struct buffer_head *bh;
1636         ext4_lblk_t block;
1637         int retval;
1638
1639 #ifdef CONFIG_FS_ENCRYPTION
1640         *res_dir = NULL;
1641 #endif
1642         frame = dx_probe(fname, dir, NULL, frames);
1643         if (IS_ERR(frame))
1644                 return (struct buffer_head *) frame;
1645         do {
1646                 block = dx_get_block(frame->at);
1647                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1648                 if (IS_ERR(bh))
1649                         goto errout;
1650
1651                 retval = search_dirblock(bh, dir, fname,
1652                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1653                                          res_dir);
1654                 if (retval == 1)
1655                         goto success;
1656                 brelse(bh);
1657                 if (retval == -1) {
1658                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1659                         goto errout;
1660                 }
1661
1662                 /* Check to see if we should continue to search */
1663                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1664                                                frames, NULL);
1665                 if (retval < 0) {
1666                         ext4_warning_inode(dir,
1667                                 "error %d reading directory index block",
1668                                 retval);
1669                         bh = ERR_PTR(retval);
1670                         goto errout;
1671                 }
1672         } while (retval == 1);
1673
1674         bh = NULL;
1675 errout:
1676         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1677 success:
1678         dx_release(frames);
1679         return bh;
1680 }
1681
1682 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1683 {
1684         struct inode *inode;
1685         struct ext4_dir_entry_2 *de;
1686         struct buffer_head *bh;
1687
1688         if (dentry->d_name.len > EXT4_NAME_LEN)
1689                 return ERR_PTR(-ENAMETOOLONG);
1690
1691         bh = ext4_lookup_entry(dir, dentry, &de);
1692         if (IS_ERR(bh))
1693                 return ERR_CAST(bh);
1694         inode = NULL;
1695         if (bh) {
1696                 __u32 ino = le32_to_cpu(de->inode);
1697                 brelse(bh);
1698                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1699                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1700                         return ERR_PTR(-EFSCORRUPTED);
1701                 }
1702                 if (unlikely(ino == dir->i_ino)) {
1703                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1704                                          dentry);
1705                         return ERR_PTR(-EFSCORRUPTED);
1706                 }
1707                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1708                 if (inode == ERR_PTR(-ESTALE)) {
1709                         EXT4_ERROR_INODE(dir,
1710                                          "deleted inode referenced: %u",
1711                                          ino);
1712                         return ERR_PTR(-EFSCORRUPTED);
1713                 }
1714                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1715                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1716                     !fscrypt_has_permitted_context(dir, inode)) {
1717                         ext4_warning(inode->i_sb,
1718                                      "Inconsistent encryption contexts: %lu/%lu",
1719                                      dir->i_ino, inode->i_ino);
1720                         iput(inode);
1721                         return ERR_PTR(-EPERM);
1722                 }
1723         }
1724
1725 #ifdef CONFIG_UNICODE
1726         if (!inode && IS_CASEFOLDED(dir)) {
1727                 /* Eventually we want to call d_add_ci(dentry, NULL)
1728                  * for negative dentries in the encoding case as
1729                  * well.  For now, prevent the negative dentry
1730                  * from being cached.
1731                  */
1732                 return NULL;
1733         }
1734 #endif
1735         return d_splice_alias(inode, dentry);
1736 }
1737
1738
1739 struct dentry *ext4_get_parent(struct dentry *child)
1740 {
1741         __u32 ino;
1742         static const struct qstr dotdot = QSTR_INIT("..", 2);
1743         struct ext4_dir_entry_2 * de;
1744         struct buffer_head *bh;
1745
1746         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1747         if (IS_ERR(bh))
1748                 return ERR_CAST(bh);
1749         if (!bh)
1750                 return ERR_PTR(-ENOENT);
1751         ino = le32_to_cpu(de->inode);
1752         brelse(bh);
1753
1754         if (!ext4_valid_inum(child->d_sb, ino)) {
1755                 EXT4_ERROR_INODE(d_inode(child),
1756                                  "bad parent inode number: %u", ino);
1757                 return ERR_PTR(-EFSCORRUPTED);
1758         }
1759
1760         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1761 }
1762
1763 /*
1764  * Move count entries from end of map between two memory locations.
1765  * Returns pointer to last entry moved.
1766  */
1767 static struct ext4_dir_entry_2 *
1768 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1769                 unsigned blocksize)
1770 {
1771         unsigned rec_len = 0;
1772
1773         while (count--) {
1774                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1775                                                 (from + (map->offs<<2));
1776                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1777                 memcpy (to, de, rec_len);
1778                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1779                                 ext4_rec_len_to_disk(rec_len, blocksize);
1780                 de->inode = 0;
1781                 map++;
1782                 to += rec_len;
1783         }
1784         return (struct ext4_dir_entry_2 *) (to - rec_len);
1785 }
1786
1787 /*
1788  * Compact each dir entry in the range to the minimal rec_len.
1789  * Returns pointer to last entry in range.
1790  */
1791 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1792 {
1793         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1794         unsigned rec_len = 0;
1795
1796         prev = to = de;
1797         while ((char*)de < base + blocksize) {
1798                 next = ext4_next_entry(de, blocksize);
1799                 if (de->inode && de->name_len) {
1800                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1801                         if (de > to)
1802                                 memmove(to, de, rec_len);
1803                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1804                         prev = to;
1805                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1806                 }
1807                 de = next;
1808         }
1809         return prev;
1810 }
1811
1812 /*
1813  * Split a full leaf block to make room for a new dir entry.
1814  * Allocate a new block, and move entries so that they are approx. equally full.
1815  * Returns pointer to de in block into which the new entry will be inserted.
1816  */
1817 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1818                         struct buffer_head **bh,struct dx_frame *frame,
1819                         struct dx_hash_info *hinfo)
1820 {
1821         unsigned blocksize = dir->i_sb->s_blocksize;
1822         unsigned count, continued;
1823         struct buffer_head *bh2;
1824         ext4_lblk_t newblock;
1825         u32 hash2;
1826         struct dx_map_entry *map;
1827         char *data1 = (*bh)->b_data, *data2;
1828         unsigned split, move, size;
1829         struct ext4_dir_entry_2 *de = NULL, *de2;
1830         int     csum_size = 0;
1831         int     err = 0, i;
1832
1833         if (ext4_has_metadata_csum(dir->i_sb))
1834                 csum_size = sizeof(struct ext4_dir_entry_tail);
1835
1836         bh2 = ext4_append(handle, dir, &newblock);
1837         if (IS_ERR(bh2)) {
1838                 brelse(*bh);
1839                 *bh = NULL;
1840                 return (struct ext4_dir_entry_2 *) bh2;
1841         }
1842
1843         BUFFER_TRACE(*bh, "get_write_access");
1844         err = ext4_journal_get_write_access(handle, *bh);
1845         if (err)
1846                 goto journal_error;
1847
1848         BUFFER_TRACE(frame->bh, "get_write_access");
1849         err = ext4_journal_get_write_access(handle, frame->bh);
1850         if (err)
1851                 goto journal_error;
1852
1853         data2 = bh2->b_data;
1854
1855         /* create map in the end of data2 block */
1856         map = (struct dx_map_entry *) (data2 + blocksize);
1857         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1858                              blocksize, hinfo, map);
1859         map -= count;
1860         dx_sort_map(map, count);
1861         /* Ensure that neither split block is over half full */
1862         size = 0;
1863         move = 0;
1864         for (i = count-1; i >= 0; i--) {
1865                 /* is more than half of this entry in 2nd half of the block? */
1866                 if (size + map[i].size/2 > blocksize/2)
1867                         break;
1868                 size += map[i].size;
1869                 move++;
1870         }
1871         /*
1872          * map index at which we will split
1873          *
1874          * If the sum of active entries didn't exceed half the block size, just
1875          * split it in half by count; each resulting block will have at least
1876          * half the space free.
1877          */
1878         if (i > 0)
1879                 split = count - move;
1880         else
1881                 split = count/2;
1882
1883         hash2 = map[split].hash;
1884         continued = hash2 == map[split - 1].hash;
1885         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1886                         (unsigned long)dx_get_block(frame->at),
1887                                         hash2, split, count-split));
1888
1889         /* Fancy dance to stay within two buffers */
1890         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1891                               blocksize);
1892         de = dx_pack_dirents(data1, blocksize);
1893         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1894                                            (char *) de,
1895                                            blocksize);
1896         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1897                                             (char *) de2,
1898                                             blocksize);
1899         if (csum_size) {
1900                 ext4_initialize_dirent_tail(*bh, blocksize);
1901                 ext4_initialize_dirent_tail(bh2, blocksize);
1902         }
1903
1904         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1905                         blocksize, 1));
1906         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1907                         blocksize, 1));
1908
1909         /* Which block gets the new entry? */
1910         if (hinfo->hash >= hash2) {
1911                 swap(*bh, bh2);
1912                 de = de2;
1913         }
1914         dx_insert_block(frame, hash2 + continued, newblock);
1915         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1916         if (err)
1917                 goto journal_error;
1918         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1919         if (err)
1920                 goto journal_error;
1921         brelse(bh2);
1922         dxtrace(dx_show_index("frame", frame->entries));
1923         return de;
1924
1925 journal_error:
1926         brelse(*bh);
1927         brelse(bh2);
1928         *bh = NULL;
1929         ext4_std_error(dir->i_sb, err);
1930         return ERR_PTR(err);
1931 }
1932
1933 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1934                       struct buffer_head *bh,
1935                       void *buf, int buf_size,
1936                       struct ext4_filename *fname,
1937                       struct ext4_dir_entry_2 **dest_de)
1938 {
1939         struct ext4_dir_entry_2 *de;
1940         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1941         int nlen, rlen;
1942         unsigned int offset = 0;
1943         char *top;
1944
1945         de = (struct ext4_dir_entry_2 *)buf;
1946         top = buf + buf_size - reclen;
1947         while ((char *) de <= top) {
1948                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1949                                          buf, buf_size, offset))
1950                         return -EFSCORRUPTED;
1951                 if (ext4_match(dir, fname, de))
1952                         return -EEXIST;
1953                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1954                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1955                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1956                         break;
1957                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1958                 offset += rlen;
1959         }
1960         if ((char *) de > top)
1961                 return -ENOSPC;
1962
1963         *dest_de = de;
1964         return 0;
1965 }
1966
1967 void ext4_insert_dentry(struct inode *inode,
1968                         struct ext4_dir_entry_2 *de,
1969                         int buf_size,
1970                         struct ext4_filename *fname)
1971 {
1972
1973         int nlen, rlen;
1974
1975         nlen = EXT4_DIR_REC_LEN(de->name_len);
1976         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1977         if (de->inode) {
1978                 struct ext4_dir_entry_2 *de1 =
1979                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1980                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1981                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1982                 de = de1;
1983         }
1984         de->file_type = EXT4_FT_UNKNOWN;
1985         de->inode = cpu_to_le32(inode->i_ino);
1986         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1987         de->name_len = fname_len(fname);
1988         memcpy(de->name, fname_name(fname), fname_len(fname));
1989 }
1990
1991 /*
1992  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1993  * it points to a directory entry which is guaranteed to be large
1994  * enough for new directory entry.  If de is NULL, then
1995  * add_dirent_to_buf will attempt search the directory block for
1996  * space.  It will return -ENOSPC if no space is available, and -EIO
1997  * and -EEXIST if directory entry already exists.
1998  */
1999 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2000                              struct inode *dir,
2001                              struct inode *inode, struct ext4_dir_entry_2 *de,
2002                              struct buffer_head *bh)
2003 {
2004         unsigned int    blocksize = dir->i_sb->s_blocksize;
2005         int             csum_size = 0;
2006         int             err, err2;
2007
2008         if (ext4_has_metadata_csum(inode->i_sb))
2009                 csum_size = sizeof(struct ext4_dir_entry_tail);
2010
2011         if (!de) {
2012                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2013                                         blocksize - csum_size, fname, &de);
2014                 if (err)
2015                         return err;
2016         }
2017         BUFFER_TRACE(bh, "get_write_access");
2018         err = ext4_journal_get_write_access(handle, bh);
2019         if (err) {
2020                 ext4_std_error(dir->i_sb, err);
2021                 return err;
2022         }
2023
2024         /* By now the buffer is marked for journaling */
2025         ext4_insert_dentry(inode, de, blocksize, fname);
2026
2027         /*
2028          * XXX shouldn't update any times until successful
2029          * completion of syscall, but too many callers depend
2030          * on this.
2031          *
2032          * XXX similarly, too many callers depend on
2033          * ext4_new_inode() setting the times, but error
2034          * recovery deletes the inode, so the worst that can
2035          * happen is that the times are slightly out of date
2036          * and/or different from the directory change time.
2037          */
2038         dir->i_mtime = dir->i_ctime = current_time(dir);
2039         ext4_update_dx_flag(dir);
2040         inode_inc_iversion(dir);
2041         err2 = ext4_mark_inode_dirty(handle, dir);
2042         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2043         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2044         if (err)
2045                 ext4_std_error(dir->i_sb, err);
2046         return err ? err : err2;
2047 }
2048
2049 /*
2050  * This converts a one block unindexed directory to a 3 block indexed
2051  * directory, and adds the dentry to the indexed directory.
2052  */
2053 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2054                             struct inode *dir,
2055                             struct inode *inode, struct buffer_head *bh)
2056 {
2057         struct buffer_head *bh2;
2058         struct dx_root  *root;
2059         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2060         struct dx_entry *entries;
2061         struct ext4_dir_entry_2 *de, *de2;
2062         char            *data2, *top;
2063         unsigned        len;
2064         int             retval;
2065         unsigned        blocksize;
2066         ext4_lblk_t  block;
2067         struct fake_dirent *fde;
2068         int csum_size = 0;
2069
2070         if (ext4_has_metadata_csum(inode->i_sb))
2071                 csum_size = sizeof(struct ext4_dir_entry_tail);
2072
2073         blocksize =  dir->i_sb->s_blocksize;
2074         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2075         BUFFER_TRACE(bh, "get_write_access");
2076         retval = ext4_journal_get_write_access(handle, bh);
2077         if (retval) {
2078                 ext4_std_error(dir->i_sb, retval);
2079                 brelse(bh);
2080                 return retval;
2081         }
2082         root = (struct dx_root *) bh->b_data;
2083
2084         /* The 0th block becomes the root, move the dirents out */
2085         fde = &root->dotdot;
2086         de = (struct ext4_dir_entry_2 *)((char *)fde +
2087                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2088         if ((char *) de >= (((char *) root) + blocksize)) {
2089                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2090                 brelse(bh);
2091                 return -EFSCORRUPTED;
2092         }
2093         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2094
2095         /* Allocate new block for the 0th block's dirents */
2096         bh2 = ext4_append(handle, dir, &block);
2097         if (IS_ERR(bh2)) {
2098                 brelse(bh);
2099                 return PTR_ERR(bh2);
2100         }
2101         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2102         data2 = bh2->b_data;
2103
2104         memcpy(data2, de, len);
2105         de = (struct ext4_dir_entry_2 *) data2;
2106         top = data2 + len;
2107         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2108                 de = de2;
2109         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2110                                            (char *) de, blocksize);
2111
2112         if (csum_size)
2113                 ext4_initialize_dirent_tail(bh2, blocksize);
2114
2115         /* Initialize the root; the dot dirents already exist */
2116         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2117         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2118                                            blocksize);
2119         memset (&root->info, 0, sizeof(root->info));
2120         root->info.info_length = sizeof(root->info);
2121         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2122         entries = root->entries;
2123         dx_set_block(entries, 1);
2124         dx_set_count(entries, 1);
2125         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2126
2127         /* Initialize as for dx_probe */
2128         fname->hinfo.hash_version = root->info.hash_version;
2129         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2130                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2131         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2132         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2133
2134         memset(frames, 0, sizeof(frames));
2135         frame = frames;
2136         frame->entries = entries;
2137         frame->at = entries;
2138         frame->bh = bh;
2139
2140         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2141         if (retval)
2142                 goto out_frames;
2143         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2144         if (retval)
2145                 goto out_frames;
2146
2147         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2148         if (IS_ERR(de)) {
2149                 retval = PTR_ERR(de);
2150                 goto out_frames;
2151         }
2152
2153         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2154 out_frames:
2155         /*
2156          * Even if the block split failed, we have to properly write
2157          * out all the changes we did so far. Otherwise we can end up
2158          * with corrupted filesystem.
2159          */
2160         if (retval)
2161                 ext4_mark_inode_dirty(handle, dir);
2162         dx_release(frames);
2163         brelse(bh2);
2164         return retval;
2165 }
2166
2167 /*
2168  *      ext4_add_entry()
2169  *
2170  * adds a file entry to the specified directory, using the same
2171  * semantics as ext4_find_entry(). It returns NULL if it failed.
2172  *
2173  * NOTE!! The inode part of 'de' is left at 0 - which means you
2174  * may not sleep between calling this and putting something into
2175  * the entry, as someone else might have used it while you slept.
2176  */
2177 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2178                           struct inode *inode)
2179 {
2180         struct inode *dir = d_inode(dentry->d_parent);
2181         struct buffer_head *bh = NULL;
2182         struct ext4_dir_entry_2 *de;
2183         struct super_block *sb;
2184         struct ext4_filename fname;
2185         int     retval;
2186         int     dx_fallback=0;
2187         unsigned blocksize;
2188         ext4_lblk_t block, blocks;
2189         int     csum_size = 0;
2190
2191         if (ext4_has_metadata_csum(inode->i_sb))
2192                 csum_size = sizeof(struct ext4_dir_entry_tail);
2193
2194         sb = dir->i_sb;
2195         blocksize = sb->s_blocksize;
2196         if (!dentry->d_name.len)
2197                 return -EINVAL;
2198
2199         if (fscrypt_is_nokey_name(dentry))
2200                 return -ENOKEY;
2201
2202 #ifdef CONFIG_UNICODE
2203         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2204             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2205                 return -EINVAL;
2206 #endif
2207
2208         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2209         if (retval)
2210                 return retval;
2211
2212         if (ext4_has_inline_data(dir)) {
2213                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2214                 if (retval < 0)
2215                         goto out;
2216                 if (retval == 1) {
2217                         retval = 0;
2218                         goto out;
2219                 }
2220         }
2221
2222         if (is_dx(dir)) {
2223                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2224                 if (!retval || (retval != ERR_BAD_DX_DIR))
2225                         goto out;
2226                 /* Can we just ignore htree data? */
2227                 if (ext4_has_metadata_csum(sb)) {
2228                         EXT4_ERROR_INODE(dir,
2229                                 "Directory has corrupted htree index.");
2230                         retval = -EFSCORRUPTED;
2231                         goto out;
2232                 }
2233                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2234                 dx_fallback++;
2235                 retval = ext4_mark_inode_dirty(handle, dir);
2236                 if (unlikely(retval))
2237                         goto out;
2238         }
2239         blocks = dir->i_size >> sb->s_blocksize_bits;
2240         for (block = 0; block < blocks; block++) {
2241                 bh = ext4_read_dirblock(dir, block, DIRENT);
2242                 if (bh == NULL) {
2243                         bh = ext4_bread(handle, dir, block,
2244                                         EXT4_GET_BLOCKS_CREATE);
2245                         goto add_to_new_block;
2246                 }
2247                 if (IS_ERR(bh)) {
2248                         retval = PTR_ERR(bh);
2249                         bh = NULL;
2250                         goto out;
2251                 }
2252                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2253                                            NULL, bh);
2254                 if (retval != -ENOSPC)
2255                         goto out;
2256
2257                 if (blocks == 1 && !dx_fallback &&
2258                     ext4_has_feature_dir_index(sb)) {
2259                         retval = make_indexed_dir(handle, &fname, dir,
2260                                                   inode, bh);
2261                         bh = NULL; /* make_indexed_dir releases bh */
2262                         goto out;
2263                 }
2264                 brelse(bh);
2265         }
2266         bh = ext4_append(handle, dir, &block);
2267 add_to_new_block:
2268         if (IS_ERR(bh)) {
2269                 retval = PTR_ERR(bh);
2270                 bh = NULL;
2271                 goto out;
2272         }
2273         de = (struct ext4_dir_entry_2 *) bh->b_data;
2274         de->inode = 0;
2275         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2276
2277         if (csum_size)
2278                 ext4_initialize_dirent_tail(bh, blocksize);
2279
2280         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2281 out:
2282         ext4_fname_free_filename(&fname);
2283         brelse(bh);
2284         if (retval == 0)
2285                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2286         return retval;
2287 }
2288
2289 /*
2290  * Returns 0 for success, or a negative error value
2291  */
2292 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2293                              struct inode *dir, struct inode *inode)
2294 {
2295         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2296         struct dx_entry *entries, *at;
2297         struct buffer_head *bh;
2298         struct super_block *sb = dir->i_sb;
2299         struct ext4_dir_entry_2 *de;
2300         int restart;
2301         int err;
2302
2303 again:
2304         restart = 0;
2305         frame = dx_probe(fname, dir, NULL, frames);
2306         if (IS_ERR(frame))
2307                 return PTR_ERR(frame);
2308         entries = frame->entries;
2309         at = frame->at;
2310         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2311         if (IS_ERR(bh)) {
2312                 err = PTR_ERR(bh);
2313                 bh = NULL;
2314                 goto cleanup;
2315         }
2316
2317         BUFFER_TRACE(bh, "get_write_access");
2318         err = ext4_journal_get_write_access(handle, bh);
2319         if (err)
2320                 goto journal_error;
2321
2322         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2323         if (err != -ENOSPC)
2324                 goto cleanup;
2325
2326         err = 0;
2327         /* Block full, should compress but for now just split */
2328         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2329                        dx_get_count(entries), dx_get_limit(entries)));
2330         /* Need to split index? */
2331         if (dx_get_count(entries) == dx_get_limit(entries)) {
2332                 ext4_lblk_t newblock;
2333                 int levels = frame - frames + 1;
2334                 unsigned int icount;
2335                 int add_level = 1;
2336                 struct dx_entry *entries2;
2337                 struct dx_node *node2;
2338                 struct buffer_head *bh2;
2339
2340                 while (frame > frames) {
2341                         if (dx_get_count((frame - 1)->entries) <
2342                             dx_get_limit((frame - 1)->entries)) {
2343                                 add_level = 0;
2344                                 break;
2345                         }
2346                         frame--; /* split higher index block */
2347                         at = frame->at;
2348                         entries = frame->entries;
2349                         restart = 1;
2350                 }
2351                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2352                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2353                                          "reach max htree level :%d",
2354                                          dir->i_ino, levels);
2355                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2356                                 ext4_warning(sb, "Large directory feature is "
2357                                                  "not enabled on this "
2358                                                  "filesystem");
2359                         }
2360                         err = -ENOSPC;
2361                         goto cleanup;
2362                 }
2363                 icount = dx_get_count(entries);
2364                 bh2 = ext4_append(handle, dir, &newblock);
2365                 if (IS_ERR(bh2)) {
2366                         err = PTR_ERR(bh2);
2367                         goto cleanup;
2368                 }
2369                 node2 = (struct dx_node *)(bh2->b_data);
2370                 entries2 = node2->entries;
2371                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2372                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2373                                                            sb->s_blocksize);
2374                 BUFFER_TRACE(frame->bh, "get_write_access");
2375                 err = ext4_journal_get_write_access(handle, frame->bh);
2376                 if (err)
2377                         goto journal_error;
2378                 if (!add_level) {
2379                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2380                         unsigned hash2 = dx_get_hash(entries + icount1);
2381                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2382                                        icount1, icount2));
2383
2384                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2385                         err = ext4_journal_get_write_access(handle,
2386                                                              (frame - 1)->bh);
2387                         if (err)
2388                                 goto journal_error;
2389
2390                         memcpy((char *) entries2, (char *) (entries + icount1),
2391                                icount2 * sizeof(struct dx_entry));
2392                         dx_set_count(entries, icount1);
2393                         dx_set_count(entries2, icount2);
2394                         dx_set_limit(entries2, dx_node_limit(dir));
2395
2396                         /* Which index block gets the new entry? */
2397                         if (at - entries >= icount1) {
2398                                 frame->at = at = at - entries - icount1 + entries2;
2399                                 frame->entries = entries = entries2;
2400                                 swap(frame->bh, bh2);
2401                         }
2402                         dx_insert_block((frame - 1), hash2, newblock);
2403                         dxtrace(dx_show_index("node", frame->entries));
2404                         dxtrace(dx_show_index("node",
2405                                ((struct dx_node *) bh2->b_data)->entries));
2406                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2407                         if (err)
2408                                 goto journal_error;
2409                         brelse (bh2);
2410                         err = ext4_handle_dirty_dx_node(handle, dir,
2411                                                    (frame - 1)->bh);
2412                         if (err)
2413                                 goto journal_error;
2414                         err = ext4_handle_dirty_dx_node(handle, dir,
2415                                                         frame->bh);
2416                         if (err)
2417                                 goto journal_error;
2418                 } else {
2419                         struct dx_root *dxroot;
2420                         memcpy((char *) entries2, (char *) entries,
2421                                icount * sizeof(struct dx_entry));
2422                         dx_set_limit(entries2, dx_node_limit(dir));
2423
2424                         /* Set up root */
2425                         dx_set_count(entries, 1);
2426                         dx_set_block(entries + 0, newblock);
2427                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2428                         dxroot->info.indirect_levels += 1;
2429                         dxtrace(printk(KERN_DEBUG
2430                                        "Creating %d level index...\n",
2431                                        dxroot->info.indirect_levels));
2432                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2433                         if (err)
2434                                 goto journal_error;
2435                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2436                         brelse(bh2);
2437                         restart = 1;
2438                         goto journal_error;
2439                 }
2440         }
2441         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2442         if (IS_ERR(de)) {
2443                 err = PTR_ERR(de);
2444                 goto cleanup;
2445         }
2446         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2447         goto cleanup;
2448
2449 journal_error:
2450         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2451 cleanup:
2452         brelse(bh);
2453         dx_release(frames);
2454         /* @restart is true means htree-path has been changed, we need to
2455          * repeat dx_probe() to find out valid htree-path
2456          */
2457         if (restart && err == 0)
2458                 goto again;
2459         return err;
2460 }
2461
2462 /*
2463  * ext4_generic_delete_entry deletes a directory entry by merging it
2464  * with the previous entry
2465  */
2466 int ext4_generic_delete_entry(struct inode *dir,
2467                               struct ext4_dir_entry_2 *de_del,
2468                               struct buffer_head *bh,
2469                               void *entry_buf,
2470                               int buf_size,
2471                               int csum_size)
2472 {
2473         struct ext4_dir_entry_2 *de, *pde;
2474         unsigned int blocksize = dir->i_sb->s_blocksize;
2475         int i;
2476
2477         i = 0;
2478         pde = NULL;
2479         de = (struct ext4_dir_entry_2 *)entry_buf;
2480         while (i < buf_size - csum_size) {
2481                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2482                                          entry_buf, buf_size, i))
2483                         return -EFSCORRUPTED;
2484                 if (de == de_del)  {
2485                         if (pde)
2486                                 pde->rec_len = ext4_rec_len_to_disk(
2487                                         ext4_rec_len_from_disk(pde->rec_len,
2488                                                                blocksize) +
2489                                         ext4_rec_len_from_disk(de->rec_len,
2490                                                                blocksize),
2491                                         blocksize);
2492                         else
2493                                 de->inode = 0;
2494                         inode_inc_iversion(dir);
2495                         return 0;
2496                 }
2497                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2498                 pde = de;
2499                 de = ext4_next_entry(de, blocksize);
2500         }
2501         return -ENOENT;
2502 }
2503
2504 static int ext4_delete_entry(handle_t *handle,
2505                              struct inode *dir,
2506                              struct ext4_dir_entry_2 *de_del,
2507                              struct buffer_head *bh)
2508 {
2509         int err, csum_size = 0;
2510
2511         if (ext4_has_inline_data(dir)) {
2512                 int has_inline_data = 1;
2513                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2514                                                &has_inline_data);
2515                 if (has_inline_data)
2516                         return err;
2517         }
2518
2519         if (ext4_has_metadata_csum(dir->i_sb))
2520                 csum_size = sizeof(struct ext4_dir_entry_tail);
2521
2522         BUFFER_TRACE(bh, "get_write_access");
2523         err = ext4_journal_get_write_access(handle, bh);
2524         if (unlikely(err))
2525                 goto out;
2526
2527         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2528                                         dir->i_sb->s_blocksize, csum_size);
2529         if (err)
2530                 goto out;
2531
2532         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2533         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2534         if (unlikely(err))
2535                 goto out;
2536
2537         return 0;
2538 out:
2539         if (err != -ENOENT)
2540                 ext4_std_error(dir->i_sb, err);
2541         return err;
2542 }
2543
2544 /*
2545  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2546  * since this indicates that nlinks count was previously 1 to avoid overflowing
2547  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2548  * that subdirectory link counts are not being maintained accurately.
2549  *
2550  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2551  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2552  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2553  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2554  */
2555 static void ext4_inc_count(struct inode *inode)
2556 {
2557         inc_nlink(inode);
2558         if (is_dx(inode) &&
2559             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2560                 set_nlink(inode, 1);
2561 }
2562
2563 /*
2564  * If a directory had nlink == 1, then we should let it be 1. This indicates
2565  * directory has >EXT4_LINK_MAX subdirs.
2566  */
2567 static void ext4_dec_count(struct inode *inode)
2568 {
2569         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2570                 drop_nlink(inode);
2571 }
2572
2573
2574 /*
2575  * Add non-directory inode to a directory. On success, the inode reference is
2576  * consumed by dentry is instantiation. This is also indicated by clearing of
2577  * *inodep pointer. On failure, the caller is responsible for dropping the
2578  * inode reference in the safe context.
2579  */
2580 static int ext4_add_nondir(handle_t *handle,
2581                 struct dentry *dentry, struct inode **inodep)
2582 {
2583         struct inode *dir = d_inode(dentry->d_parent);
2584         struct inode *inode = *inodep;
2585         int err = ext4_add_entry(handle, dentry, inode);
2586         if (!err) {
2587                 err = ext4_mark_inode_dirty(handle, inode);
2588                 if (IS_DIRSYNC(dir))
2589                         ext4_handle_sync(handle);
2590                 d_instantiate_new(dentry, inode);
2591                 *inodep = NULL;
2592                 return err;
2593         }
2594         drop_nlink(inode);
2595         ext4_orphan_add(handle, inode);
2596         unlock_new_inode(inode);
2597         return err;
2598 }
2599
2600 /*
2601  * By the time this is called, we already have created
2602  * the directory cache entry for the new file, but it
2603  * is so far negative - it has no inode.
2604  *
2605  * If the create succeeds, we fill in the inode information
2606  * with d_instantiate().
2607  */
2608 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2609                        struct dentry *dentry, umode_t mode, bool excl)
2610 {
2611         handle_t *handle;
2612         struct inode *inode;
2613         int err, credits, retries = 0;
2614
2615         err = dquot_initialize(dir);
2616         if (err)
2617                 return err;
2618
2619         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2620                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2621 retry:
2622         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2623                                             0, NULL, EXT4_HT_DIR, credits);
2624         handle = ext4_journal_current_handle();
2625         err = PTR_ERR(inode);
2626         if (!IS_ERR(inode)) {
2627                 inode->i_op = &ext4_file_inode_operations;
2628                 inode->i_fop = &ext4_file_operations;
2629                 ext4_set_aops(inode);
2630                 err = ext4_add_nondir(handle, dentry, &inode);
2631                 if (!err)
2632                         ext4_fc_track_create(handle, dentry);
2633         }
2634         if (handle)
2635                 ext4_journal_stop(handle);
2636         if (!IS_ERR_OR_NULL(inode))
2637                 iput(inode);
2638         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2639                 goto retry;
2640         return err;
2641 }
2642
2643 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2644                       struct dentry *dentry, umode_t mode, dev_t rdev)
2645 {
2646         handle_t *handle;
2647         struct inode *inode;
2648         int err, credits, retries = 0;
2649
2650         err = dquot_initialize(dir);
2651         if (err)
2652                 return err;
2653
2654         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2655                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2656 retry:
2657         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2658                                             0, NULL, EXT4_HT_DIR, credits);
2659         handle = ext4_journal_current_handle();
2660         err = PTR_ERR(inode);
2661         if (!IS_ERR(inode)) {
2662                 init_special_inode(inode, inode->i_mode, rdev);
2663                 inode->i_op = &ext4_special_inode_operations;
2664                 err = ext4_add_nondir(handle, dentry, &inode);
2665                 if (!err)
2666                         ext4_fc_track_create(handle, dentry);
2667         }
2668         if (handle)
2669                 ext4_journal_stop(handle);
2670         if (!IS_ERR_OR_NULL(inode))
2671                 iput(inode);
2672         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2673                 goto retry;
2674         return err;
2675 }
2676
2677 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2678                         struct dentry *dentry, umode_t mode)
2679 {
2680         handle_t *handle;
2681         struct inode *inode;
2682         int err, retries = 0;
2683
2684         err = dquot_initialize(dir);
2685         if (err)
2686                 return err;
2687
2688 retry:
2689         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2690                                             NULL, 0, NULL,
2691                                             EXT4_HT_DIR,
2692                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2693                           4 + EXT4_XATTR_TRANS_BLOCKS);
2694         handle = ext4_journal_current_handle();
2695         err = PTR_ERR(inode);
2696         if (!IS_ERR(inode)) {
2697                 inode->i_op = &ext4_file_inode_operations;
2698                 inode->i_fop = &ext4_file_operations;
2699                 ext4_set_aops(inode);
2700                 d_tmpfile(dentry, inode);
2701                 err = ext4_orphan_add(handle, inode);
2702                 if (err)
2703                         goto err_unlock_inode;
2704                 mark_inode_dirty(inode);
2705                 unlock_new_inode(inode);
2706         }
2707         if (handle)
2708                 ext4_journal_stop(handle);
2709         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2710                 goto retry;
2711         return err;
2712 err_unlock_inode:
2713         ext4_journal_stop(handle);
2714         unlock_new_inode(inode);
2715         return err;
2716 }
2717
2718 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2719                           struct ext4_dir_entry_2 *de,
2720                           int blocksize, int csum_size,
2721                           unsigned int parent_ino, int dotdot_real_len)
2722 {
2723         de->inode = cpu_to_le32(inode->i_ino);
2724         de->name_len = 1;
2725         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2726                                            blocksize);
2727         strcpy(de->name, ".");
2728         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2729
2730         de = ext4_next_entry(de, blocksize);
2731         de->inode = cpu_to_le32(parent_ino);
2732         de->name_len = 2;
2733         if (!dotdot_real_len)
2734                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2735                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2736                                         blocksize);
2737         else
2738                 de->rec_len = ext4_rec_len_to_disk(
2739                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2740         strcpy(de->name, "..");
2741         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2742
2743         return ext4_next_entry(de, blocksize);
2744 }
2745
2746 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2747                              struct inode *inode)
2748 {
2749         struct buffer_head *dir_block = NULL;
2750         struct ext4_dir_entry_2 *de;
2751         ext4_lblk_t block = 0;
2752         unsigned int blocksize = dir->i_sb->s_blocksize;
2753         int csum_size = 0;
2754         int err;
2755
2756         if (ext4_has_metadata_csum(dir->i_sb))
2757                 csum_size = sizeof(struct ext4_dir_entry_tail);
2758
2759         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2760                 err = ext4_try_create_inline_dir(handle, dir, inode);
2761                 if (err < 0 && err != -ENOSPC)
2762                         goto out;
2763                 if (!err)
2764                         goto out;
2765         }
2766
2767         inode->i_size = 0;
2768         dir_block = ext4_append(handle, inode, &block);
2769         if (IS_ERR(dir_block))
2770                 return PTR_ERR(dir_block);
2771         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2772         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2773         set_nlink(inode, 2);
2774         if (csum_size)
2775                 ext4_initialize_dirent_tail(dir_block, blocksize);
2776
2777         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2778         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2779         if (err)
2780                 goto out;
2781         set_buffer_verified(dir_block);
2782 out:
2783         brelse(dir_block);
2784         return err;
2785 }
2786
2787 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2788                       struct dentry *dentry, umode_t mode)
2789 {
2790         handle_t *handle;
2791         struct inode *inode;
2792         int err, err2 = 0, credits, retries = 0;
2793
2794         if (EXT4_DIR_LINK_MAX(dir))
2795                 return -EMLINK;
2796
2797         err = dquot_initialize(dir);
2798         if (err)
2799                 return err;
2800
2801         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2802                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2803 retry:
2804         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2805                                             &dentry->d_name,
2806                                             0, NULL, EXT4_HT_DIR, credits);
2807         handle = ext4_journal_current_handle();
2808         err = PTR_ERR(inode);
2809         if (IS_ERR(inode))
2810                 goto out_stop;
2811
2812         inode->i_op = &ext4_dir_inode_operations;
2813         inode->i_fop = &ext4_dir_operations;
2814         err = ext4_init_new_dir(handle, dir, inode);
2815         if (err)
2816                 goto out_clear_inode;
2817         err = ext4_mark_inode_dirty(handle, inode);
2818         if (!err)
2819                 err = ext4_add_entry(handle, dentry, inode);
2820         if (err) {
2821 out_clear_inode:
2822                 clear_nlink(inode);
2823                 ext4_orphan_add(handle, inode);
2824                 unlock_new_inode(inode);
2825                 err2 = ext4_mark_inode_dirty(handle, inode);
2826                 if (unlikely(err2))
2827                         err = err2;
2828                 ext4_journal_stop(handle);
2829                 iput(inode);
2830                 goto out_retry;
2831         }
2832         ext4_inc_count(dir);
2833
2834         ext4_update_dx_flag(dir);
2835         err = ext4_mark_inode_dirty(handle, dir);
2836         if (err)
2837                 goto out_clear_inode;
2838         d_instantiate_new(dentry, inode);
2839         ext4_fc_track_create(handle, dentry);
2840         if (IS_DIRSYNC(dir))
2841                 ext4_handle_sync(handle);
2842
2843 out_stop:
2844         if (handle)
2845                 ext4_journal_stop(handle);
2846 out_retry:
2847         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2848                 goto retry;
2849         return err;
2850 }
2851
2852 /*
2853  * routine to check that the specified directory is empty (for rmdir)
2854  */
2855 bool ext4_empty_dir(struct inode *inode)
2856 {
2857         unsigned int offset;
2858         struct buffer_head *bh;
2859         struct ext4_dir_entry_2 *de;
2860         struct super_block *sb;
2861
2862         if (ext4_has_inline_data(inode)) {
2863                 int has_inline_data = 1;
2864                 int ret;
2865
2866                 ret = empty_inline_dir(inode, &has_inline_data);
2867                 if (has_inline_data)
2868                         return ret;
2869         }
2870
2871         sb = inode->i_sb;
2872         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2873                 EXT4_ERROR_INODE(inode, "invalid size");
2874                 return true;
2875         }
2876         /* The first directory block must not be a hole,
2877          * so treat it as DIRENT_HTREE
2878          */
2879         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2880         if (IS_ERR(bh))
2881                 return true;
2882
2883         de = (struct ext4_dir_entry_2 *) bh->b_data;
2884         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2885                                  0) ||
2886             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2887                 ext4_warning_inode(inode, "directory missing '.'");
2888                 brelse(bh);
2889                 return true;
2890         }
2891         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2892         de = ext4_next_entry(de, sb->s_blocksize);
2893         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2894                                  offset) ||
2895             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2896                 ext4_warning_inode(inode, "directory missing '..'");
2897                 brelse(bh);
2898                 return true;
2899         }
2900         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2901         while (offset < inode->i_size) {
2902                 if (!(offset & (sb->s_blocksize - 1))) {
2903                         unsigned int lblock;
2904                         brelse(bh);
2905                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2906                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2907                         if (bh == NULL) {
2908                                 offset += sb->s_blocksize;
2909                                 continue;
2910                         }
2911                         if (IS_ERR(bh))
2912                                 return true;
2913                 }
2914                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2915                                         (offset & (sb->s_blocksize - 1)));
2916                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2917                                          bh->b_data, bh->b_size, offset)) {
2918                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2919                         continue;
2920                 }
2921                 if (le32_to_cpu(de->inode)) {
2922                         brelse(bh);
2923                         return false;
2924                 }
2925                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2926         }
2927         brelse(bh);
2928         return true;
2929 }
2930
2931 /*
2932  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2933  * such inodes, starting at the superblock, in case we crash before the
2934  * file is closed/deleted, or in case the inode truncate spans multiple
2935  * transactions and the last transaction is not recovered after a crash.
2936  *
2937  * At filesystem recovery time, we walk this list deleting unlinked
2938  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2939  *
2940  * Orphan list manipulation functions must be called under i_mutex unless
2941  * we are just creating the inode or deleting it.
2942  */
2943 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2944 {
2945         struct super_block *sb = inode->i_sb;
2946         struct ext4_sb_info *sbi = EXT4_SB(sb);
2947         struct ext4_iloc iloc;
2948         int err = 0, rc;
2949         bool dirty = false;
2950
2951         if (!sbi->s_journal || is_bad_inode(inode))
2952                 return 0;
2953
2954         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2955                      !inode_is_locked(inode));
2956         /*
2957          * Exit early if inode already is on orphan list. This is a big speedup
2958          * since we don't have to contend on the global s_orphan_lock.
2959          */
2960         if (!list_empty(&EXT4_I(inode)->i_orphan))
2961                 return 0;
2962
2963         /*
2964          * Orphan handling is only valid for files with data blocks
2965          * being truncated, or files being unlinked. Note that we either
2966          * hold i_mutex, or the inode can not be referenced from outside,
2967          * so i_nlink should not be bumped due to race
2968          */
2969         ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2970                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2971
2972         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2973         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2974         if (err)
2975                 goto out;
2976
2977         err = ext4_reserve_inode_write(handle, inode, &iloc);
2978         if (err)
2979                 goto out;
2980
2981         mutex_lock(&sbi->s_orphan_lock);
2982         /*
2983          * Due to previous errors inode may be already a part of on-disk
2984          * orphan list. If so skip on-disk list modification.
2985          */
2986         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2987             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2988                 /* Insert this inode at the head of the on-disk orphan list */
2989                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2990                 lock_buffer(sbi->s_sbh);
2991                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2992                 ext4_superblock_csum_set(sb);
2993                 unlock_buffer(sbi->s_sbh);
2994                 dirty = true;
2995         }
2996         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2997         mutex_unlock(&sbi->s_orphan_lock);
2998
2999         if (dirty) {
3000                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3001                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3002                 if (!err)
3003                         err = rc;
3004                 if (err) {
3005                         /*
3006                          * We have to remove inode from in-memory list if
3007                          * addition to on disk orphan list failed. Stray orphan
3008                          * list entries can cause panics at unmount time.
3009                          */
3010                         mutex_lock(&sbi->s_orphan_lock);
3011                         list_del_init(&EXT4_I(inode)->i_orphan);
3012                         mutex_unlock(&sbi->s_orphan_lock);
3013                 }
3014         } else
3015                 brelse(iloc.bh);
3016
3017         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3018         jbd_debug(4, "orphan inode %lu will point to %d\n",
3019                         inode->i_ino, NEXT_ORPHAN(inode));
3020 out:
3021         ext4_std_error(sb, err);
3022         return err;
3023 }
3024
3025 /*
3026  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3027  * of such inodes stored on disk, because it is finally being cleaned up.
3028  */
3029 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3030 {
3031         struct list_head *prev;
3032         struct ext4_inode_info *ei = EXT4_I(inode);
3033         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3034         __u32 ino_next;
3035         struct ext4_iloc iloc;
3036         int err = 0;
3037
3038         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3039                 return 0;
3040
3041         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3042                      !inode_is_locked(inode));
3043         /* Do this quick check before taking global s_orphan_lock. */
3044         if (list_empty(&ei->i_orphan))
3045                 return 0;
3046
3047         if (handle) {
3048                 /* Grab inode buffer early before taking global s_orphan_lock */
3049                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3050         }
3051
3052         mutex_lock(&sbi->s_orphan_lock);
3053         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3054
3055         prev = ei->i_orphan.prev;
3056         list_del_init(&ei->i_orphan);
3057
3058         /* If we're on an error path, we may not have a valid
3059          * transaction handle with which to update the orphan list on
3060          * disk, but we still need to remove the inode from the linked
3061          * list in memory. */
3062         if (!handle || err) {
3063                 mutex_unlock(&sbi->s_orphan_lock);
3064                 goto out_err;
3065         }
3066
3067         ino_next = NEXT_ORPHAN(inode);
3068         if (prev == &sbi->s_orphan) {
3069                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3070                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3071                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3072                 if (err) {
3073                         mutex_unlock(&sbi->s_orphan_lock);
3074                         goto out_brelse;
3075                 }
3076                 lock_buffer(sbi->s_sbh);
3077                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3078                 ext4_superblock_csum_set(inode->i_sb);
3079                 unlock_buffer(sbi->s_sbh);
3080                 mutex_unlock(&sbi->s_orphan_lock);
3081                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3082         } else {
3083                 struct ext4_iloc iloc2;
3084                 struct inode *i_prev =
3085                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3086
3087                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3088                           i_prev->i_ino, ino_next);
3089                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3090                 if (err) {
3091                         mutex_unlock(&sbi->s_orphan_lock);
3092                         goto out_brelse;
3093                 }
3094                 NEXT_ORPHAN(i_prev) = ino_next;
3095                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3096                 mutex_unlock(&sbi->s_orphan_lock);
3097         }
3098         if (err)
3099                 goto out_brelse;
3100         NEXT_ORPHAN(inode) = 0;
3101         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3102 out_err:
3103         ext4_std_error(inode->i_sb, err);
3104         return err;
3105
3106 out_brelse:
3107         brelse(iloc.bh);
3108         goto out_err;
3109 }
3110
3111 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3112 {
3113         int retval;
3114         struct inode *inode;
3115         struct buffer_head *bh;
3116         struct ext4_dir_entry_2 *de;
3117         handle_t *handle = NULL;
3118
3119         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3120                 return -EIO;
3121
3122         /* Initialize quotas before so that eventual writes go in
3123          * separate transaction */
3124         retval = dquot_initialize(dir);
3125         if (retval)
3126                 return retval;
3127         retval = dquot_initialize(d_inode(dentry));
3128         if (retval)
3129                 return retval;
3130
3131         retval = -ENOENT;
3132         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3133         if (IS_ERR(bh))
3134                 return PTR_ERR(bh);
3135         if (!bh)
3136                 goto end_rmdir;
3137
3138         inode = d_inode(dentry);
3139
3140         retval = -EFSCORRUPTED;
3141         if (le32_to_cpu(de->inode) != inode->i_ino)
3142                 goto end_rmdir;
3143
3144         retval = -ENOTEMPTY;
3145         if (!ext4_empty_dir(inode))
3146                 goto end_rmdir;
3147
3148         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3149                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3150         if (IS_ERR(handle)) {
3151                 retval = PTR_ERR(handle);
3152                 handle = NULL;
3153                 goto end_rmdir;
3154         }
3155
3156         if (IS_DIRSYNC(dir))
3157                 ext4_handle_sync(handle);
3158
3159         retval = ext4_delete_entry(handle, dir, de, bh);
3160         if (retval)
3161                 goto end_rmdir;
3162         if (!EXT4_DIR_LINK_EMPTY(inode))
3163                 ext4_warning_inode(inode,
3164                              "empty directory '%.*s' has too many links (%u)",
3165                              dentry->d_name.len, dentry->d_name.name,
3166                              inode->i_nlink);
3167         inode_inc_iversion(inode);
3168         clear_nlink(inode);
3169         /* There's no need to set i_disksize: the fact that i_nlink is
3170          * zero will ensure that the right thing happens during any
3171          * recovery. */
3172         inode->i_size = 0;
3173         ext4_orphan_add(handle, inode);
3174         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3175         retval = ext4_mark_inode_dirty(handle, inode);
3176         if (retval)
3177                 goto end_rmdir;
3178         ext4_dec_count(dir);
3179         ext4_update_dx_flag(dir);
3180         ext4_fc_track_unlink(handle, dentry);
3181         retval = ext4_mark_inode_dirty(handle, dir);
3182
3183 #ifdef CONFIG_UNICODE
3184         /* VFS negative dentries are incompatible with Encoding and
3185          * Case-insensitiveness. Eventually we'll want avoid
3186          * invalidating the dentries here, alongside with returning the
3187          * negative dentries at ext4_lookup(), when it is better
3188          * supported by the VFS for the CI case.
3189          */
3190         if (IS_CASEFOLDED(dir))
3191                 d_invalidate(dentry);
3192 #endif
3193
3194 end_rmdir:
3195         brelse(bh);
3196         if (handle)
3197                 ext4_journal_stop(handle);
3198         return retval;
3199 }
3200
3201 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3202                   struct inode *inode)
3203 {
3204         int retval = -ENOENT;
3205         struct buffer_head *bh;
3206         struct ext4_dir_entry_2 *de;
3207         int skip_remove_dentry = 0;
3208
3209         bh = ext4_find_entry(dir, d_name, &de, NULL);
3210         if (IS_ERR(bh))
3211                 return PTR_ERR(bh);
3212
3213         if (!bh)
3214                 return -ENOENT;
3215
3216         if (le32_to_cpu(de->inode) != inode->i_ino) {
3217                 /*
3218                  * It's okay if we find dont find dentry which matches
3219                  * the inode. That's because it might have gotten
3220                  * renamed to a different inode number
3221                  */
3222                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3223                         skip_remove_dentry = 1;
3224                 else
3225                         goto out;
3226         }
3227
3228         if (IS_DIRSYNC(dir))
3229                 ext4_handle_sync(handle);
3230
3231         if (!skip_remove_dentry) {
3232                 retval = ext4_delete_entry(handle, dir, de, bh);
3233                 if (retval)
3234                         goto out;
3235                 dir->i_ctime = dir->i_mtime = current_time(dir);
3236                 ext4_update_dx_flag(dir);
3237                 retval = ext4_mark_inode_dirty(handle, dir);
3238                 if (retval)
3239                         goto out;
3240         } else {
3241                 retval = 0;
3242         }
3243         if (inode->i_nlink == 0)
3244                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3245                                    d_name->len, d_name->name);
3246         else
3247                 drop_nlink(inode);
3248         if (!inode->i_nlink)
3249                 ext4_orphan_add(handle, inode);
3250         inode->i_ctime = current_time(inode);
3251         retval = ext4_mark_inode_dirty(handle, inode);
3252
3253 out:
3254         brelse(bh);
3255         return retval;
3256 }
3257
3258 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3259 {
3260         handle_t *handle;
3261         int retval;
3262
3263         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3264                 return -EIO;
3265
3266         trace_ext4_unlink_enter(dir, dentry);
3267         /*
3268          * Initialize quotas before so that eventual writes go
3269          * in separate transaction
3270          */
3271         retval = dquot_initialize(dir);
3272         if (retval)
3273                 goto out_trace;
3274         retval = dquot_initialize(d_inode(dentry));
3275         if (retval)
3276                 goto out_trace;
3277
3278         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3279                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3280         if (IS_ERR(handle)) {
3281                 retval = PTR_ERR(handle);
3282                 goto out_trace;
3283         }
3284
3285         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3286         if (!retval)
3287                 ext4_fc_track_unlink(handle, dentry);
3288 #ifdef CONFIG_UNICODE
3289         /* VFS negative dentries are incompatible with Encoding and
3290          * Case-insensitiveness. Eventually we'll want avoid
3291          * invalidating the dentries here, alongside with returning the
3292          * negative dentries at ext4_lookup(), when it is  better
3293          * supported by the VFS for the CI case.
3294          */
3295         if (IS_CASEFOLDED(dir))
3296                 d_invalidate(dentry);
3297 #endif
3298         if (handle)
3299                 ext4_journal_stop(handle);
3300
3301 out_trace:
3302         trace_ext4_unlink_exit(dentry, retval);
3303         return retval;
3304 }
3305
3306 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3307                         struct dentry *dentry, const char *symname)
3308 {
3309         handle_t *handle;
3310         struct inode *inode;
3311         int err, len = strlen(symname);
3312         int credits;
3313         struct fscrypt_str disk_link;
3314
3315         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3316                 return -EIO;
3317
3318         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3319                                       &disk_link);
3320         if (err)
3321                 return err;
3322
3323         err = dquot_initialize(dir);
3324         if (err)
3325                 return err;
3326
3327         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3328                 /*
3329                  * For non-fast symlinks, we just allocate inode and put it on
3330                  * orphan list in the first transaction => we need bitmap,
3331                  * group descriptor, sb, inode block, quota blocks, and
3332                  * possibly selinux xattr blocks.
3333                  */
3334                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3335                           EXT4_XATTR_TRANS_BLOCKS;
3336         } else {
3337                 /*
3338                  * Fast symlink. We have to add entry to directory
3339                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3340                  * allocate new inode (bitmap, group descriptor, inode block,
3341                  * quota blocks, sb is already counted in previous macros).
3342                  */
3343                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3344                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3345         }
3346
3347         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3348                                             &dentry->d_name, 0, NULL,
3349                                             EXT4_HT_DIR, credits);
3350         handle = ext4_journal_current_handle();
3351         if (IS_ERR(inode)) {
3352                 if (handle)
3353                         ext4_journal_stop(handle);
3354                 return PTR_ERR(inode);
3355         }
3356
3357         if (IS_ENCRYPTED(inode)) {
3358                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3359                 if (err)
3360                         goto err_drop_inode;
3361                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3362         }
3363
3364         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3365                 if (!IS_ENCRYPTED(inode))
3366                         inode->i_op = &ext4_symlink_inode_operations;
3367                 inode_nohighmem(inode);
3368                 ext4_set_aops(inode);
3369                 /*
3370                  * We cannot call page_symlink() with transaction started
3371                  * because it calls into ext4_write_begin() which can wait
3372                  * for transaction commit if we are running out of space
3373                  * and thus we deadlock. So we have to stop transaction now
3374                  * and restart it when symlink contents is written.
3375                  *
3376                  * To keep fs consistent in case of crash, we have to put inode
3377                  * to orphan list in the mean time.
3378                  */
3379                 drop_nlink(inode);
3380                 err = ext4_orphan_add(handle, inode);
3381                 if (handle)
3382                         ext4_journal_stop(handle);
3383                 handle = NULL;
3384                 if (err)
3385                         goto err_drop_inode;
3386                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3387                 if (err)
3388                         goto err_drop_inode;
3389                 /*
3390                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3391                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3392                  */
3393                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3394                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3395                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3396                 if (IS_ERR(handle)) {
3397                         err = PTR_ERR(handle);
3398                         handle = NULL;
3399                         goto err_drop_inode;
3400                 }
3401                 set_nlink(inode, 1);
3402                 err = ext4_orphan_del(handle, inode);
3403                 if (err)
3404                         goto err_drop_inode;
3405         } else {
3406                 /* clear the extent format for fast symlink */
3407                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3408                 if (!IS_ENCRYPTED(inode)) {
3409                         inode->i_op = &ext4_fast_symlink_inode_operations;
3410                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3411                 }
3412                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3413                        disk_link.len);
3414                 inode->i_size = disk_link.len - 1;
3415         }
3416         EXT4_I(inode)->i_disksize = inode->i_size;
3417         err = ext4_add_nondir(handle, dentry, &inode);
3418         if (handle)
3419                 ext4_journal_stop(handle);
3420         if (inode)
3421                 iput(inode);
3422         goto out_free_encrypted_link;
3423
3424 err_drop_inode:
3425         if (handle)
3426                 ext4_journal_stop(handle);
3427         clear_nlink(inode);
3428         unlock_new_inode(inode);
3429         iput(inode);
3430 out_free_encrypted_link:
3431         if (disk_link.name != (unsigned char *)symname)
3432                 kfree(disk_link.name);
3433         return err;
3434 }
3435
3436 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3437 {
3438         handle_t *handle;
3439         int err, retries = 0;
3440 retry:
3441         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3442                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3443                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3444         if (IS_ERR(handle))
3445                 return PTR_ERR(handle);
3446
3447         if (IS_DIRSYNC(dir))
3448                 ext4_handle_sync(handle);
3449
3450         inode->i_ctime = current_time(inode);
3451         ext4_inc_count(inode);
3452         ihold(inode);
3453
3454         err = ext4_add_entry(handle, dentry, inode);
3455         if (!err) {
3456                 err = ext4_mark_inode_dirty(handle, inode);
3457                 /* this can happen only for tmpfile being
3458                  * linked the first time
3459                  */
3460                 if (inode->i_nlink == 1)
3461                         ext4_orphan_del(handle, inode);
3462                 d_instantiate(dentry, inode);
3463                 ext4_fc_track_link(handle, dentry);
3464         } else {
3465                 drop_nlink(inode);
3466                 iput(inode);
3467         }
3468         ext4_journal_stop(handle);
3469         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3470                 goto retry;
3471         return err;
3472 }
3473
3474 static int ext4_link(struct dentry *old_dentry,
3475                      struct inode *dir, struct dentry *dentry)
3476 {
3477         struct inode *inode = d_inode(old_dentry);
3478         int err;
3479
3480         if (inode->i_nlink >= EXT4_LINK_MAX)
3481                 return -EMLINK;
3482
3483         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3484         if (err)
3485                 return err;
3486
3487         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3488             (!projid_eq(EXT4_I(dir)->i_projid,
3489                         EXT4_I(old_dentry->d_inode)->i_projid)))
3490                 return -EXDEV;
3491
3492         err = dquot_initialize(dir);
3493         if (err)
3494                 return err;
3495         return __ext4_link(dir, inode, dentry);
3496 }
3497
3498 /*
3499  * Try to find buffer head where contains the parent block.
3500  * It should be the inode block if it is inlined or the 1st block
3501  * if it is a normal dir.
3502  */
3503 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3504                                         struct inode *inode,
3505                                         int *retval,
3506                                         struct ext4_dir_entry_2 **parent_de,
3507                                         int *inlined)
3508 {
3509         struct buffer_head *bh;
3510
3511         if (!ext4_has_inline_data(inode)) {
3512                 /* The first directory block must not be a hole, so
3513                  * treat it as DIRENT_HTREE
3514                  */
3515                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3516                 if (IS_ERR(bh)) {
3517                         *retval = PTR_ERR(bh);
3518                         return NULL;
3519                 }
3520                 *parent_de = ext4_next_entry(
3521                                         (struct ext4_dir_entry_2 *)bh->b_data,
3522                                         inode->i_sb->s_blocksize);
3523                 return bh;
3524         }
3525
3526         *inlined = 1;
3527         return ext4_get_first_inline_block(inode, parent_de, retval);
3528 }
3529
3530 struct ext4_renament {
3531         struct inode *dir;
3532         struct dentry *dentry;
3533         struct inode *inode;
3534         bool is_dir;
3535         int dir_nlink_delta;
3536
3537         /* entry for "dentry" */
3538         struct buffer_head *bh;
3539         struct ext4_dir_entry_2 *de;
3540         int inlined;
3541
3542         /* entry for ".." in inode if it's a directory */
3543         struct buffer_head *dir_bh;
3544         struct ext4_dir_entry_2 *parent_de;
3545         int dir_inlined;
3546 };
3547
3548 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3549 {
3550         int retval;
3551
3552         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3553                                               &retval, &ent->parent_de,
3554                                               &ent->dir_inlined);
3555         if (!ent->dir_bh)
3556                 return retval;
3557         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3558                 return -EFSCORRUPTED;
3559         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3560         return ext4_journal_get_write_access(handle, ent->dir_bh);
3561 }
3562
3563 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3564                                   unsigned dir_ino)
3565 {
3566         int retval;
3567
3568         ent->parent_de->inode = cpu_to_le32(dir_ino);
3569         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3570         if (!ent->dir_inlined) {
3571                 if (is_dx(ent->inode)) {
3572                         retval = ext4_handle_dirty_dx_node(handle,
3573                                                            ent->inode,
3574                                                            ent->dir_bh);
3575                 } else {
3576                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3577                                                             ent->dir_bh);
3578                 }
3579         } else {
3580                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3581         }
3582         if (retval) {
3583                 ext4_std_error(ent->dir->i_sb, retval);
3584                 return retval;
3585         }
3586         return 0;
3587 }
3588
3589 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3590                        unsigned ino, unsigned file_type)
3591 {
3592         int retval, retval2;
3593
3594         BUFFER_TRACE(ent->bh, "get write access");
3595         retval = ext4_journal_get_write_access(handle, ent->bh);
3596         if (retval)
3597                 return retval;
3598         ent->de->inode = cpu_to_le32(ino);
3599         if (ext4_has_feature_filetype(ent->dir->i_sb))
3600                 ent->de->file_type = file_type;
3601         inode_inc_iversion(ent->dir);
3602         ent->dir->i_ctime = ent->dir->i_mtime =
3603                 current_time(ent->dir);
3604         retval = ext4_mark_inode_dirty(handle, ent->dir);
3605         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3606         if (!ent->inlined) {
3607                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3608                 if (unlikely(retval2)) {
3609                         ext4_std_error(ent->dir->i_sb, retval2);
3610                         return retval2;
3611                 }
3612         }
3613         return retval;
3614 }
3615
3616 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3617                                   const struct qstr *d_name)
3618 {
3619         int retval = -ENOENT;
3620         struct buffer_head *bh;
3621         struct ext4_dir_entry_2 *de;
3622
3623         bh = ext4_find_entry(dir, d_name, &de, NULL);
3624         if (IS_ERR(bh))
3625                 return PTR_ERR(bh);
3626         if (bh) {
3627                 retval = ext4_delete_entry(handle, dir, de, bh);
3628                 brelse(bh);
3629         }
3630         return retval;
3631 }
3632
3633 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3634                                int force_reread)
3635 {
3636         int retval;
3637         /*
3638          * ent->de could have moved from under us during htree split, so make
3639          * sure that we are deleting the right entry.  We might also be pointing
3640          * to a stale entry in the unused part of ent->bh so just checking inum
3641          * and the name isn't enough.
3642          */
3643         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3644             ent->de->name_len != ent->dentry->d_name.len ||
3645             strncmp(ent->de->name, ent->dentry->d_name.name,
3646                     ent->de->name_len) ||
3647             force_reread) {
3648                 retval = ext4_find_delete_entry(handle, ent->dir,
3649                                                 &ent->dentry->d_name);
3650         } else {
3651                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3652                 if (retval == -ENOENT) {
3653                         retval = ext4_find_delete_entry(handle, ent->dir,
3654                                                         &ent->dentry->d_name);
3655                 }
3656         }
3657
3658         if (retval) {
3659                 ext4_warning_inode(ent->dir,
3660                                    "Deleting old file: nlink %d, error=%d",
3661                                    ent->dir->i_nlink, retval);
3662         }
3663 }
3664
3665 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3666 {
3667         if (ent->dir_nlink_delta) {
3668                 if (ent->dir_nlink_delta == -1)
3669                         ext4_dec_count(ent->dir);
3670                 else
3671                         ext4_inc_count(ent->dir);
3672                 ext4_mark_inode_dirty(handle, ent->dir);
3673         }
3674 }
3675
3676 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3677                                               struct ext4_renament *ent,
3678                                               int credits, handle_t **h)
3679 {
3680         struct inode *wh;
3681         handle_t *handle;
3682         int retries = 0;
3683
3684         /*
3685          * for inode block, sb block, group summaries,
3686          * and inode bitmap
3687          */
3688         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3689                     EXT4_XATTR_TRANS_BLOCKS + 4);
3690 retry:
3691         wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3692                                          S_IFCHR | WHITEOUT_MODE,
3693                                          &ent->dentry->d_name, 0, NULL,
3694                                          EXT4_HT_DIR, credits);
3695
3696         handle = ext4_journal_current_handle();
3697         if (IS_ERR(wh)) {
3698                 if (handle)
3699                         ext4_journal_stop(handle);
3700                 if (PTR_ERR(wh) == -ENOSPC &&
3701                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3702                         goto retry;
3703         } else {
3704                 *h = handle;
3705                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3706                 wh->i_op = &ext4_special_inode_operations;
3707         }
3708         return wh;
3709 }
3710
3711 /*
3712  * Anybody can rename anything with this: the permission checks are left to the
3713  * higher-level routines.
3714  *
3715  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3716  * while new_{dentry,inode) refers to the destination dentry/inode
3717  * This comes from rename(const char *oldpath, const char *newpath)
3718  */
3719 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3720                        struct dentry *old_dentry, struct inode *new_dir,
3721                        struct dentry *new_dentry, unsigned int flags)
3722 {
3723         handle_t *handle = NULL;
3724         struct ext4_renament old = {
3725                 .dir = old_dir,
3726                 .dentry = old_dentry,
3727                 .inode = d_inode(old_dentry),
3728         };
3729         struct ext4_renament new = {
3730                 .dir = new_dir,
3731                 .dentry = new_dentry,
3732                 .inode = d_inode(new_dentry),
3733         };
3734         int force_reread;
3735         int retval;
3736         struct inode *whiteout = NULL;
3737         int credits;
3738         u8 old_file_type;
3739
3740         if (new.inode && new.inode->i_nlink == 0) {
3741                 EXT4_ERROR_INODE(new.inode,
3742                                  "target of rename is already freed");
3743                 return -EFSCORRUPTED;
3744         }
3745
3746         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3747             (!projid_eq(EXT4_I(new_dir)->i_projid,
3748                         EXT4_I(old_dentry->d_inode)->i_projid)))
3749                 return -EXDEV;
3750
3751         retval = dquot_initialize(old.dir);
3752         if (retval)
3753                 return retval;
3754         retval = dquot_initialize(new.dir);
3755         if (retval)
3756                 return retval;
3757
3758         /* Initialize quotas before so that eventual writes go
3759          * in separate transaction */
3760         if (new.inode) {
3761                 retval = dquot_initialize(new.inode);
3762                 if (retval)
3763                         return retval;
3764         }
3765
3766         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3767         if (IS_ERR(old.bh))
3768                 return PTR_ERR(old.bh);
3769         /*
3770          *  Check for inode number is _not_ due to possible IO errors.
3771          *  We might rmdir the source, keep it as pwd of some process
3772          *  and merrily kill the link to whatever was created under the
3773          *  same name. Goodbye sticky bit ;-<
3774          */
3775         retval = -ENOENT;
3776         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3777                 goto end_rename;
3778
3779         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3780                                  &new.de, &new.inlined);
3781         if (IS_ERR(new.bh)) {
3782                 retval = PTR_ERR(new.bh);
3783                 new.bh = NULL;
3784                 goto end_rename;
3785         }
3786         if (new.bh) {
3787                 if (!new.inode) {
3788                         brelse(new.bh);
3789                         new.bh = NULL;
3790                 }
3791         }
3792         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3793                 ext4_alloc_da_blocks(old.inode);
3794
3795         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3796                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3797         if (!(flags & RENAME_WHITEOUT)) {
3798                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3799                 if (IS_ERR(handle)) {
3800                         retval = PTR_ERR(handle);
3801                         handle = NULL;
3802                         goto end_rename;
3803                 }
3804         } else {
3805                 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3806                 if (IS_ERR(whiteout)) {
3807                         retval = PTR_ERR(whiteout);
3808                         whiteout = NULL;
3809                         goto end_rename;
3810                 }
3811         }
3812
3813         old_file_type = old.de->file_type;
3814         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3815                 ext4_handle_sync(handle);
3816
3817         if (S_ISDIR(old.inode->i_mode)) {
3818                 if (new.inode) {
3819                         retval = -ENOTEMPTY;
3820                         if (!ext4_empty_dir(new.inode))
3821                                 goto end_rename;
3822                 } else {
3823                         retval = -EMLINK;
3824                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3825                                 goto end_rename;
3826                 }
3827                 retval = ext4_rename_dir_prepare(handle, &old);
3828                 if (retval)
3829                         goto end_rename;
3830         }
3831         /*
3832          * If we're renaming a file within an inline_data dir and adding or
3833          * setting the new dirent causes a conversion from inline_data to
3834          * extents/blockmap, we need to force the dirent delete code to
3835          * re-read the directory, or else we end up trying to delete a dirent
3836          * from what is now the extent tree root (or a block map).
3837          */
3838         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3839                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3840
3841         if (whiteout) {
3842                 /*
3843                  * Do this before adding a new entry, so the old entry is sure
3844                  * to be still pointing to the valid old entry.
3845                  */
3846                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3847                                      EXT4_FT_CHRDEV);
3848                 if (retval)
3849                         goto end_rename;
3850                 retval = ext4_mark_inode_dirty(handle, whiteout);
3851                 if (unlikely(retval))
3852                         goto end_rename;
3853         }
3854         if (!new.bh) {
3855                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3856                 if (retval)
3857                         goto end_rename;
3858         } else {
3859                 retval = ext4_setent(handle, &new,
3860                                      old.inode->i_ino, old_file_type);
3861                 if (retval)
3862                         goto end_rename;
3863         }
3864         if (force_reread)
3865                 force_reread = !ext4_test_inode_flag(new.dir,
3866                                                      EXT4_INODE_INLINE_DATA);
3867
3868         /*
3869          * Like most other Unix systems, set the ctime for inodes on a
3870          * rename.
3871          */
3872         old.inode->i_ctime = current_time(old.inode);
3873         retval = ext4_mark_inode_dirty(handle, old.inode);
3874         if (unlikely(retval))
3875                 goto end_rename;
3876
3877         if (!whiteout) {
3878                 /*
3879                  * ok, that's it
3880                  */
3881                 ext4_rename_delete(handle, &old, force_reread);
3882         }
3883
3884         if (new.inode) {
3885                 ext4_dec_count(new.inode);
3886                 new.inode->i_ctime = current_time(new.inode);
3887         }
3888         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3889         ext4_update_dx_flag(old.dir);
3890         if (old.dir_bh) {
3891                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3892                 if (retval)
3893                         goto end_rename;
3894
3895                 ext4_dec_count(old.dir);
3896                 if (new.inode) {
3897                         /* checked ext4_empty_dir above, can't have another
3898                          * parent, ext4_dec_count() won't work for many-linked
3899                          * dirs */
3900                         clear_nlink(new.inode);
3901                 } else {
3902                         ext4_inc_count(new.dir);
3903                         ext4_update_dx_flag(new.dir);
3904                         retval = ext4_mark_inode_dirty(handle, new.dir);
3905                         if (unlikely(retval))
3906                                 goto end_rename;
3907                 }
3908         }
3909         retval = ext4_mark_inode_dirty(handle, old.dir);
3910         if (unlikely(retval))
3911                 goto end_rename;
3912
3913         if (S_ISDIR(old.inode->i_mode)) {
3914                 /*
3915                  * We disable fast commits here that's because the
3916                  * replay code is not yet capable of changing dot dot
3917                  * dirents in directories.
3918                  */
3919                 ext4_fc_mark_ineligible(old.inode->i_sb,
3920                         EXT4_FC_REASON_RENAME_DIR);
3921         } else {
3922                 if (new.inode)
3923                         ext4_fc_track_unlink(handle, new.dentry);
3924                 __ext4_fc_track_link(handle, old.inode, new.dentry);
3925                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3926         }
3927
3928         if (new.inode) {
3929                 retval = ext4_mark_inode_dirty(handle, new.inode);
3930                 if (unlikely(retval))
3931                         goto end_rename;
3932                 if (!new.inode->i_nlink)
3933                         ext4_orphan_add(handle, new.inode);
3934         }
3935         retval = 0;
3936
3937 end_rename:
3938         if (whiteout) {
3939                 if (retval) {
3940                         ext4_setent(handle, &old,
3941                                 old.inode->i_ino, old_file_type);
3942                         drop_nlink(whiteout);
3943                 }
3944                 unlock_new_inode(whiteout);
3945                 iput(whiteout);
3946
3947         }
3948         brelse(old.dir_bh);
3949         brelse(old.bh);
3950         brelse(new.bh);
3951         if (handle)
3952                 ext4_journal_stop(handle);
3953         return retval;
3954 }
3955
3956 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3957                              struct inode *new_dir, struct dentry *new_dentry)
3958 {
3959         handle_t *handle = NULL;
3960         struct ext4_renament old = {
3961                 .dir = old_dir,
3962                 .dentry = old_dentry,
3963                 .inode = d_inode(old_dentry),
3964         };
3965         struct ext4_renament new = {
3966                 .dir = new_dir,
3967                 .dentry = new_dentry,
3968                 .inode = d_inode(new_dentry),
3969         };
3970         u8 new_file_type;
3971         int retval;
3972         struct timespec64 ctime;
3973
3974         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3975              !projid_eq(EXT4_I(new_dir)->i_projid,
3976                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3977             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3978              !projid_eq(EXT4_I(old_dir)->i_projid,
3979                         EXT4_I(new_dentry->d_inode)->i_projid)))
3980                 return -EXDEV;
3981
3982         retval = dquot_initialize(old.dir);
3983         if (retval)
3984                 return retval;
3985         retval = dquot_initialize(new.dir);
3986         if (retval)
3987                 return retval;
3988
3989         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3990                                  &old.de, &old.inlined);
3991         if (IS_ERR(old.bh))
3992                 return PTR_ERR(old.bh);
3993         /*
3994          *  Check for inode number is _not_ due to possible IO errors.
3995          *  We might rmdir the source, keep it as pwd of some process
3996          *  and merrily kill the link to whatever was created under the
3997          *  same name. Goodbye sticky bit ;-<
3998          */
3999         retval = -ENOENT;
4000         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4001                 goto end_rename;
4002
4003         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4004                                  &new.de, &new.inlined);
4005         if (IS_ERR(new.bh)) {
4006                 retval = PTR_ERR(new.bh);
4007                 new.bh = NULL;
4008                 goto end_rename;
4009         }
4010
4011         /* RENAME_EXCHANGE case: old *and* new must both exist */
4012         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4013                 goto end_rename;
4014
4015         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4016                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4017                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4018         if (IS_ERR(handle)) {
4019                 retval = PTR_ERR(handle);
4020                 handle = NULL;
4021                 goto end_rename;
4022         }
4023
4024         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4025                 ext4_handle_sync(handle);
4026
4027         if (S_ISDIR(old.inode->i_mode)) {
4028                 old.is_dir = true;
4029                 retval = ext4_rename_dir_prepare(handle, &old);
4030                 if (retval)
4031                         goto end_rename;
4032         }
4033         if (S_ISDIR(new.inode->i_mode)) {
4034                 new.is_dir = true;
4035                 retval = ext4_rename_dir_prepare(handle, &new);
4036                 if (retval)
4037                         goto end_rename;
4038         }
4039
4040         /*
4041          * Other than the special case of overwriting a directory, parents'
4042          * nlink only needs to be modified if this is a cross directory rename.
4043          */
4044         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4045                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4046                 new.dir_nlink_delta = -old.dir_nlink_delta;
4047                 retval = -EMLINK;
4048                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4049                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4050                         goto end_rename;
4051         }
4052
4053         new_file_type = new.de->file_type;
4054         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4055         if (retval)
4056                 goto end_rename;
4057
4058         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4059         if (retval)
4060                 goto end_rename;
4061
4062         /*
4063          * Like most other Unix systems, set the ctime for inodes on a
4064          * rename.
4065          */
4066         ctime = current_time(old.inode);
4067         old.inode->i_ctime = ctime;
4068         new.inode->i_ctime = ctime;
4069         retval = ext4_mark_inode_dirty(handle, old.inode);
4070         if (unlikely(retval))
4071                 goto end_rename;
4072         retval = ext4_mark_inode_dirty(handle, new.inode);
4073         if (unlikely(retval))
4074                 goto end_rename;
4075         ext4_fc_mark_ineligible(new.inode->i_sb,
4076                                 EXT4_FC_REASON_CROSS_RENAME);
4077         if (old.dir_bh) {
4078                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4079                 if (retval)
4080                         goto end_rename;
4081         }
4082         if (new.dir_bh) {
4083                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4084                 if (retval)
4085                         goto end_rename;
4086         }
4087         ext4_update_dir_count(handle, &old);
4088         ext4_update_dir_count(handle, &new);
4089         retval = 0;
4090
4091 end_rename:
4092         brelse(old.dir_bh);
4093         brelse(new.dir_bh);
4094         brelse(old.bh);
4095         brelse(new.bh);
4096         if (handle)
4097                 ext4_journal_stop(handle);
4098         return retval;
4099 }
4100
4101 static int ext4_rename2(struct user_namespace *mnt_userns,
4102                         struct inode *old_dir, struct dentry *old_dentry,
4103                         struct inode *new_dir, struct dentry *new_dentry,
4104                         unsigned int flags)
4105 {
4106         int err;
4107
4108         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4109                 return -EIO;
4110
4111         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4112                 return -EINVAL;
4113
4114         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4115                                      flags);
4116         if (err)
4117                 return err;
4118
4119         if (flags & RENAME_EXCHANGE) {
4120                 return ext4_cross_rename(old_dir, old_dentry,
4121                                          new_dir, new_dentry);
4122         }
4123
4124         return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4125 }
4126
4127 /*
4128  * directories can handle most operations...
4129  */
4130 const struct inode_operations ext4_dir_inode_operations = {
4131         .create         = ext4_create,
4132         .lookup         = ext4_lookup,
4133         .link           = ext4_link,
4134         .unlink         = ext4_unlink,
4135         .symlink        = ext4_symlink,
4136         .mkdir          = ext4_mkdir,
4137         .rmdir          = ext4_rmdir,
4138         .mknod          = ext4_mknod,
4139         .tmpfile        = ext4_tmpfile,
4140         .rename         = ext4_rename2,
4141         .setattr        = ext4_setattr,
4142         .getattr        = ext4_getattr,
4143         .listxattr      = ext4_listxattr,
4144         .get_acl        = ext4_get_acl,
4145         .set_acl        = ext4_set_acl,
4146         .fiemap         = ext4_fiemap,
4147 };
4148
4149 const struct inode_operations ext4_special_inode_operations = {
4150         .setattr        = ext4_setattr,
4151         .getattr        = ext4_getattr,
4152         .listxattr      = ext4_listxattr,
4153         .get_acl        = ext4_get_acl,
4154         .set_acl        = ext4_set_acl,
4155 };