fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 113                 bh = ERR_PTR(-EIO);
 114         else
 115                 bh = ext4_bread(NULL, inode, block, 0);
 116         if (IS_ERR(bh)) {
 117                 __ext4_warning(inode->i_sb, func, line,
 118                                "inode #%lu: lblock %lu: comm %s: "
 119                                "error %ld reading directory block",
 120                                inode->i_ino, (unsigned long)block,
 121                                current->comm, PTR_ERR(bh));
 122
 123                 return bh;
 124         }
 125         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 126                 ext4_error_inode(inode, func, line, block,
 127                                  "Directory hole found for htree %s block",
 128                                  (type == INDEX) ? "index" : "leaf");
 129                 return ERR_PTR(-EFSCORRUPTED);
 130         }
 131         if (!bh)
 132                 return NULL;
 133         dirent = (struct ext4_dir_entry *) bh->b_data;
 134         /* Determine whether or not we have an index block */
 135         if (is_dx(inode)) {
 136                 if (block == 0)
 137                         is_dx_block = 1;
 138                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 139                                                 inode->i_sb->s_blocksize) ==
 140                          inode->i_sb->s_blocksize)
 141                         is_dx_block = 1;
 142         }
 143         if (!is_dx_block && type == INDEX) {
 144                 ext4_error_inode(inode, func, line, block,
 145                        "directory leaf block found instead of index block");
 146                 brelse(bh);
 147                 return ERR_PTR(-EFSCORRUPTED);
 148         }
 149         if (!ext4_has_metadata_csum(inode->i_sb) ||
 150             buffer_verified(bh))
 151                 return bh;
 152
 153         /*
 154          * An empty leaf block can get mistaken for a index block; for
 155          * this reason, we can only check the index checksum when the
 156          * caller is sure it should be an index block.
 157          */
 158         if (is_dx_block && type == INDEX) {
 159                 if (ext4_dx_csum_verify(inode, dirent) &&
 160                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 161                         set_buffer_verified(bh);
 162                 else {
 163                         ext4_error_inode_err(inode, func, line, block,
 164                                              EFSBADCRC,
 165                                              "Directory index failed checksum");
 166                         brelse(bh);
 167                         return ERR_PTR(-EFSBADCRC);
 168                 }
 169         }
 170         if (!is_dx_block) {
 171                 if (ext4_dirblock_csum_verify(inode, bh) &&
 172                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 173                         set_buffer_verified(bh);
 174                 else {
 175                         ext4_error_inode_err(inode, func, line, block,
 176                                              EFSBADCRC,
 177                                              "Directory block failed checksum");
 178                         brelse(bh);
 179                         return ERR_PTR(-EFSBADCRC);
 180                 }
 181         }
 182         return bh;
 183 }
 184
 185 #ifdef DX_DEBUG
 186 #define dxtrace(command) command
 187 #else
 188 #define dxtrace(command)
 189 #endif
 190
 191 struct fake_dirent
 192 {
 193         __le32 inode;
 194         __le16 rec_len;
 195         u8 name_len;
 196         u8 file_type;
 197 };
 198
 199 struct dx_countlimit
 200 {
 201         __le16 limit;
 202         __le16 count;
 203 };
 204
 205 struct dx_entry
 206 {
 207         __le32 hash;
 208         __le32 block;
 209 };
 210
 211 /*
 212  * dx_root_info is laid out so that if it should somehow get overlaid by a
 213  * dirent the two low bits of the hash version will be zero.  Therefore, the
 214  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 215  */
 216
 217 struct dx_root
 218 {
 219         struct fake_dirent dot;
 220         char dot_name[4];
 221         struct fake_dirent dotdot;
 222         char dotdot_name[4];
 223         struct dx_root_info
 224         {
 225                 __le32 reserved_zero;
 226                 u8 hash_version;
 227                 u8 info_length; /* 8 */
 228                 u8 indirect_levels;
 229                 u8 unused_flags;
 230         }
 231         info;
 232         struct dx_entry entries[];
 233 };
 234
 235 struct dx_node
 236 {
 237         struct fake_dirent fake;
 238         struct dx_entry entries[];
 239 };
 240
 241
 242 struct dx_frame
 243 {
 244         struct buffer_head *bh;
 245         struct dx_entry *entries;
 246         struct dx_entry *at;
 247 };
 248
 249 struct dx_map_entry
 250 {
 251         u32 hash;
 252         u16 offs;
 253         u16 size;
 254 };
 255
 256 /*
 257  * This goes at the end of each htree block.
 258  */
 259 struct dx_tail {
 260         u32 dt_reserved;
 261         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 262 };
 263
 264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 266 static inline unsigned dx_get_hash(struct dx_entry *entry);
 267 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 268 static unsigned dx_get_count(struct dx_entry *entries);
 269 static unsigned dx_get_limit(struct dx_entry *entries);
 270 static void dx_set_count(struct dx_entry *entries, unsigned value);
 271 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 273 static unsigned dx_node_limit(struct inode *dir);
 274 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 275                                  struct inode *dir,
 276                                  struct dx_hash_info *hinfo,
 277                                  struct dx_frame *frame);
 278 static void dx_release(struct dx_frame *frames);
 279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 280                        unsigned blocksize, struct dx_hash_info *hinfo,
 281                        struct dx_map_entry map[]);
 282 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 283 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
 284                                         char *to, struct dx_map_entry *offsets,
 285                                         int count, unsigned int blocksize);
 286 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 287                                                 unsigned int blocksize);
 288 static void dx_insert_block(struct dx_frame *frame,
 289                                         u32 hash, ext4_lblk_t block);
 290 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 291                                  struct dx_frame *frame,
 292                                  struct dx_frame *frames,
 293                                  __u32 *start_hash);
 294 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 295                 struct ext4_filename *fname,
 296                 struct ext4_dir_entry_2 **res_dir);
 297 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 298                              struct inode *dir, struct inode *inode);
 299
 300 /* checksumming functions */
 301 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 302                                  unsigned int blocksize)
 303 {
 304         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 305
 306         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 307         t->det_rec_len = ext4_rec_len_to_disk(
 308                         sizeof(struct ext4_dir_entry_tail), blocksize);
 309         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 310 }
 311
 312 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 313 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 314                                                    struct buffer_head *bh)
 315 {
 316         struct ext4_dir_entry_tail *t;
 317
 318 #ifdef PARANOID
 319         struct ext4_dir_entry *d, *top;
 320
 321         d = (struct ext4_dir_entry *)bh->b_data;
 322         top = (struct ext4_dir_entry *)(bh->b_data +
 323                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 324                  sizeof(struct ext4_dir_entry_tail)));
 325         while (d < top && d->rec_len)
 326                 d = (struct ext4_dir_entry *)(((void *)d) +
 327                     le16_to_cpu(d->rec_len));
 328
 329         if (d != top)
 330                 return NULL;
 331
 332         t = (struct ext4_dir_entry_tail *)d;
 333 #else
 334         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 335 #endif
 336
 337         if (t->det_reserved_zero1 ||
 338             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 339             t->det_reserved_zero2 ||
 340             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 341                 return NULL;
 342
 343         return t;
 344 }
 345
 346 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 347 {
 348         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 349         struct ext4_inode_info *ei = EXT4_I(inode);
 350         __u32 csum;
 351
 352         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 353         return cpu_to_le32(csum);
 354 }
 355
 356 #define warn_no_space_for_csum(inode)                                   \
 357         __warn_no_space_for_csum((inode), __func__, __LINE__)
 358
 359 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 360                                      unsigned int line)
 361 {
 362         __ext4_warning_inode(inode, func, line,
 363                 "No space for directory leaf checksum. Please run e2fsck -D.");
 364 }
 365
 366 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 367 {
 368         struct ext4_dir_entry_tail *t;
 369
 370         if (!ext4_has_metadata_csum(inode->i_sb))
 371                 return 1;
 372
 373         t = get_dirent_tail(inode, bh);
 374         if (!t) {
 375                 warn_no_space_for_csum(inode);
 376                 return 0;
 377         }
 378
 379         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 380                                                   (char *)t - bh->b_data))
 381                 return 0;
 382
 383         return 1;
 384 }
 385
 386 static void ext4_dirblock_csum_set(struct inode *inode,
 387                                  struct buffer_head *bh)
 388 {
 389         struct ext4_dir_entry_tail *t;
 390
 391         if (!ext4_has_metadata_csum(inode->i_sb))
 392                 return;
 393
 394         t = get_dirent_tail(inode, bh);
 395         if (!t) {
 396                 warn_no_space_for_csum(inode);
 397                 return;
 398         }
 399
 400         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 401                                              (char *)t - bh->b_data);
 402 }
 403
 404 int ext4_handle_dirty_dirblock(handle_t *handle,
 405                                struct inode *inode,
 406                                struct buffer_head *bh)
 407 {
 408         ext4_dirblock_csum_set(inode, bh);
 409         return ext4_handle_dirty_metadata(handle, inode, bh);
 410 }
 411
 412 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 413                                                struct ext4_dir_entry *dirent,
 414                                                int *offset)
 415 {
 416         struct ext4_dir_entry *dp;
 417         struct dx_root_info *root;
 418         int count_offset;
 419
 420         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 421                 count_offset = 8;
 422         else if (le16_to_cpu(dirent->rec_len) == 12) {
 423                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 424                 if (le16_to_cpu(dp->rec_len) !=
 425                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 426                         return NULL;
 427                 root = (struct dx_root_info *)(((void *)dp + 12));
 428                 if (root->reserved_zero ||
 429                     root->info_length != sizeof(struct dx_root_info))
 430                         return NULL;
 431                 count_offset = 32;
 432         } else
 433                 return NULL;
 434
 435         if (offset)
 436                 *offset = count_offset;
 437         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 438 }
 439
 440 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 441                            int count_offset, int count, struct dx_tail *t)
 442 {
 443         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 444         struct ext4_inode_info *ei = EXT4_I(inode);
 445         __u32 csum;
 446         int size;
 447         __u32 dummy_csum = 0;
 448         int offset = offsetof(struct dx_tail, dt_checksum);
 449
 450         size = count_offset + (count * sizeof(struct dx_entry));
 451         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 452         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 453         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 454
 455         return cpu_to_le32(csum);
 456 }
 457
 458 static int ext4_dx_csum_verify(struct inode *inode,
 459                                struct ext4_dir_entry *dirent)
 460 {
 461         struct dx_countlimit *c;
 462         struct dx_tail *t;
 463         int count_offset, limit, count;
 464
 465         if (!ext4_has_metadata_csum(inode->i_sb))
 466                 return 1;
 467
 468         c = get_dx_countlimit(inode, dirent, &count_offset);
 469         if (!c) {
 470                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 471                 return 0;
 472         }
 473         limit = le16_to_cpu(c->limit);
 474         count = le16_to_cpu(c->count);
 475         if (count_offset + (limit * sizeof(struct dx_entry)) >
 476             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 477                 warn_no_space_for_csum(inode);
 478                 return 0;
 479         }
 480         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 481
 482         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 483                                             count, t))
 484                 return 0;
 485         return 1;
 486 }
 487
 488 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 489 {
 490         struct dx_countlimit *c;
 491         struct dx_tail *t;
 492         int count_offset, limit, count;
 493
 494         if (!ext4_has_metadata_csum(inode->i_sb))
 495                 return;
 496
 497         c = get_dx_countlimit(inode, dirent, &count_offset);
 498         if (!c) {
 499                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 500                 return;
 501         }
 502         limit = le16_to_cpu(c->limit);
 503         count = le16_to_cpu(c->count);
 504         if (count_offset + (limit * sizeof(struct dx_entry)) >
 505             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 506                 warn_no_space_for_csum(inode);
 507                 return;
 508         }
 509         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 510
 511         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 512 }
 513
 514 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 515                                             struct inode *inode,
 516                                             struct buffer_head *bh)
 517 {
 518         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 519         return ext4_handle_dirty_metadata(handle, inode, bh);
 520 }
 521
 522 /*
 523  * p is at least 6 bytes before the end of page
 524  */
 525 static inline struct ext4_dir_entry_2 *
 526 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 527 {
 528         return (struct ext4_dir_entry_2 *)((char *)p +
 529                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 530 }
 531
 532 /*
 533  * Future: use high four bits of block for coalesce-on-delete flags
 534  * Mask them off for now.
 535  */
 536
 537 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 538 {
 539         return le32_to_cpu(entry->block) & 0x0fffffff;
 540 }
 541
 542 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 543 {
 544         entry->block = cpu_to_le32(value);
 545 }
 546
 547 static inline unsigned dx_get_hash(struct dx_entry *entry)
 548 {
 549         return le32_to_cpu(entry->hash);
 550 }
 551
 552 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 553 {
 554         entry->hash = cpu_to_le32(value);
 555 }
 556
 557 static inline unsigned dx_get_count(struct dx_entry *entries)
 558 {
 559         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 560 }
 561
 562 static inline unsigned dx_get_limit(struct dx_entry *entries)
 563 {
 564         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 565 }
 566
 567 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 568 {
 569         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 570 }
 571
 572 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 573 {
 574         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 575 }
 576
 577 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 578 {
 579         unsigned int entry_space = dir->i_sb->s_blocksize -
 580                         ext4_dir_rec_len(1, NULL) -
 581                         ext4_dir_rec_len(2, NULL) - infosize;
 582
 583         if (ext4_has_metadata_csum(dir->i_sb))
 584                 entry_space -= sizeof(struct dx_tail);
 585         return entry_space / sizeof(struct dx_entry);
 586 }
 587
 588 static inline unsigned dx_node_limit(struct inode *dir)
 589 {
 590         unsigned int entry_space = dir->i_sb->s_blocksize -
 591                         ext4_dir_rec_len(0, dir);
 592
 593         if (ext4_has_metadata_csum(dir->i_sb))
 594                 entry_space -= sizeof(struct dx_tail);
 595         return entry_space / sizeof(struct dx_entry);
 596 }
 597
 598 /*
 599  * Debug
 600  */
 601 #ifdef DX_DEBUG
 602 static void dx_show_index(char * label, struct dx_entry *entries)
 603 {
 604         int i, n = dx_get_count (entries);
 605         printk(KERN_DEBUG "%s index", label);
 606         for (i = 0; i < n; i++) {
 607                 printk(KERN_CONT " %x->%lu",
 608                        i ? dx_get_hash(entries + i) : 0,
 609                        (unsigned long)dx_get_block(entries + i));
 610         }
 611         printk(KERN_CONT "\n");
 612 }
 613
 614 struct stats
 615 {
 616         unsigned names;
 617         unsigned space;
 618         unsigned bcount;
 619 };
 620
 621 static struct stats dx_show_leaf(struct inode *dir,
 622                                 struct dx_hash_info *hinfo,
 623                                 struct ext4_dir_entry_2 *de,
 624                                 int size, int show_names)
 625 {
 626         unsigned names = 0, space = 0;
 627         char *base = (char *) de;
 628         struct dx_hash_info h = *hinfo;
 629
 630         printk("names: ");
 631         while ((char *) de < base + size)
 632         {
 633                 if (de->inode)
 634                 {
 635                         if (show_names)
 636                         {
 637 #ifdef CONFIG_FS_ENCRYPTION
 638                                 int len;
 639                                 char *name;
 640                                 struct fscrypt_str fname_crypto_str =
 641                                         FSTR_INIT(NULL, 0);
 642                                 int res = 0;
 643
 644                                 name  = de->name;
 645                                 len = de->name_len;
 646                                 if (!IS_ENCRYPTED(dir)) {
 647                                         /* Directory is not encrypted */
 648                                         ext4fs_dirhash(dir, de->name,
 649                                                 de->name_len, &h);
 650                                         printk("%*.s:(U)%x.%u ", len,
 651                                                name, h.hash,
 652                                                (unsigned) ((char *) de
 653                                                            - base));
 654                                 } else {
 655                                         struct fscrypt_str de_name =
 656                                                 FSTR_INIT(name, len);
 657
 658                                         /* Directory is encrypted */
 659                                         res = fscrypt_fname_alloc_buffer(
 660                                                 len, &fname_crypto_str);
 661                                         if (res)
 662                                                 printk(KERN_WARNING "Error "
 663                                                         "allocating crypto "
 664                                                         "buffer--skipping "
 665                                                         "crypto\n");
 666                                         res = fscrypt_fname_disk_to_usr(dir,
 667                                                 0, 0, &de_name,
 668                                                 &fname_crypto_str);
 669                                         if (res) {
 670                                                 printk(KERN_WARNING "Error "
 671                                                         "converting filename "
 672                                                         "from disk to usr"
 673                                                         "\n");
 674                                                 name = "??";
 675                                                 len = 2;
 676                                         } else {
 677                                                 name = fname_crypto_str.name;
 678                                                 len = fname_crypto_str.len;
 679                                         }
 680                                         if (IS_CASEFOLDED(dir))
 681                                                 h.hash = EXT4_DIRENT_HASH(de);
 682                                         else
 683                                                 ext4fs_dirhash(dir, de->name,
 684                                                        de->name_len, &h);
 685                                         printk("%*.s:(E)%x.%u ", len, name,
 686                                                h.hash, (unsigned) ((char *) de
 687                                                                    - base));
 688                                         fscrypt_fname_free_buffer(
 689                                                         &fname_crypto_str);
 690                                 }
 691 #else
 692                                 int len = de->name_len;
 693                                 char *name = de->name;
 694                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 695                                 printk("%*.s:%x.%u ", len, name, h.hash,
 696                                        (unsigned) ((char *) de - base));
 697 #endif
 698                         }
 699                         space += ext4_dir_rec_len(de->name_len, dir);
 700                         names++;
 701                 }
 702                 de = ext4_next_entry(de, size);
 703         }
 704         printk(KERN_CONT "(%i)\n", names);
 705         return (struct stats) { names, space, 1 };
 706 }
 707
 708 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 709                              struct dx_entry *entries, int levels)
 710 {
 711         unsigned blocksize = dir->i_sb->s_blocksize;
 712         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 713         unsigned bcount = 0;
 714         struct buffer_head *bh;
 715         printk("%i indexed blocks...\n", count);
 716         for (i = 0; i < count; i++, entries++)
 717         {
 718                 ext4_lblk_t block = dx_get_block(entries);
 719                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 720                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 721                 struct stats stats;
 722                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 723                 bh = ext4_bread(NULL,dir, block, 0);
 724                 if (!bh || IS_ERR(bh))
 725                         continue;
 726                 stats = levels?
 727                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 728                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 729                         bh->b_data, blocksize, 0);
 730                 names += stats.names;
 731                 space += stats.space;
 732                 bcount += stats.bcount;
 733                 brelse(bh);
 734         }
 735         if (bcount)
 736                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 737                        levels ? "" : "   ", names, space/bcount,
 738                        (space/bcount)*100/blocksize);
 739         return (struct stats) { names, space, bcount};
 740 }
 741
 742 /*
 743  * Linear search cross check
 744  */
 745 static inline void htree_rep_invariant_check(struct dx_entry *at,
 746                                              struct dx_entry *target,
 747                                              u32 hash, unsigned int n)
 748 {
 749         while (n--) {
 750                 dxtrace(printk(KERN_CONT ","));
 751                 if (dx_get_hash(++at) > hash) {
 752                         at--;
 753                         break;
 754                 }
 755         }
 756         ASSERT(at == target - 1);
 757 }
 758 #else /* DX_DEBUG */
 759 static inline void htree_rep_invariant_check(struct dx_entry *at,
 760                                              struct dx_entry *target,
 761                                              u32 hash, unsigned int n)
 762 {
 763 }
 764 #endif /* DX_DEBUG */
 765
 766 /*
 767  * Probe for a directory leaf block to search.
 768  *
 769  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 770  * error in the directory index, and the caller should fall back to
 771  * searching the directory normally.  The callers of dx_probe **MUST**
 772  * check for this error code, and make sure it never gets reflected
 773  * back to userspace.
 774  */
 775 static struct dx_frame *
 776 dx_probe(struct ext4_filename *fname, struct inode *dir,
 777          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 778 {
 779         unsigned count, indirect;
 780         struct dx_entry *at, *entries, *p, *q, *m;
 781         struct dx_root *root;
 782         struct dx_frame *frame = frame_in;
 783         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 784         u32 hash;
 785
 786         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 787         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 788         if (IS_ERR(frame->bh))
 789                 return (struct dx_frame *) frame->bh;
 790
 791         root = (struct dx_root *) frame->bh->b_data;
 792         if (root->info.hash_version != DX_HASH_TEA &&
 793             root->info.hash_version != DX_HASH_HALF_MD4 &&
 794             root->info.hash_version != DX_HASH_LEGACY &&
 795             root->info.hash_version != DX_HASH_SIPHASH) {
 796                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 797                                    root->info.hash_version);
 798                 goto fail;
 799         }
 800         if (ext4_hash_in_dirent(dir)) {
 801                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 802                         ext4_warning_inode(dir,
 803                                 "Hash in dirent, but hash is not SIPHASH");
 804                         goto fail;
 805                 }
 806         } else {
 807                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 808                         ext4_warning_inode(dir,
 809                                 "Hash code is SIPHASH, but hash not in dirent");
 810                         goto fail;
 811                 }
 812         }
 813         if (fname)
 814                 hinfo = &fname->hinfo;
 815         hinfo->hash_version = root->info.hash_version;
 816         if (hinfo->hash_version <= DX_HASH_TEA)
 817                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 818         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 819         /* hash is already computed for encrypted casefolded directory */
 820         if (fname && fname_name(fname) &&
 821                                 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
 822                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 823         hash = hinfo->hash;
 824
 825         if (root->info.unused_flags & 1) {
 826                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 827                                    root->info.unused_flags);
 828                 goto fail;
 829         }
 830
 831         indirect = root->info.indirect_levels;
 832         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 833                 ext4_warning(dir->i_sb,
 834                              "Directory (ino: %lu) htree depth %#06x exceed"
 835                              "supported value", dir->i_ino,
 836                              ext4_dir_htree_level(dir->i_sb));
 837                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 838                         ext4_warning(dir->i_sb, "Enable large directory "
 839                                                 "feature to access it");
 840                 }
 841                 goto fail;
 842         }
 843
 844         entries = (struct dx_entry *)(((char *)&root->info) +
 845                                       root->info.info_length);
 846
 847         if (dx_get_limit(entries) != dx_root_limit(dir,
 848                                                    root->info.info_length)) {
 849                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 850                                    dx_get_limit(entries),
 851                                    dx_root_limit(dir, root->info.info_length));
 852                 goto fail;
 853         }
 854
 855         dxtrace(printk("Look up %x", hash));
 856         while (1) {
 857                 count = dx_get_count(entries);
 858                 if (!count || count > dx_get_limit(entries)) {
 859                         ext4_warning_inode(dir,
 860                                            "dx entry: count %u beyond limit %u",
 861                                            count, dx_get_limit(entries));
 862                         goto fail;
 863                 }
 864
 865                 p = entries + 1;
 866                 q = entries + count - 1;
 867                 while (p <= q) {
 868                         m = p + (q - p) / 2;
 869                         dxtrace(printk(KERN_CONT "."));
 870                         if (dx_get_hash(m) > hash)
 871                                 q = m - 1;
 872                         else
 873                                 p = m + 1;
 874                 }
 875
 876                 htree_rep_invariant_check(entries, p, hash, count - 1);
 877
 878                 at = p - 1;
 879                 dxtrace(printk(KERN_CONT " %x->%u\n",
 880                                at == entries ? 0 : dx_get_hash(at),
 881                                dx_get_block(at)));
 882                 frame->entries = entries;
 883                 frame->at = at;
 884                 if (!indirect--)
 885                         return frame;
 886                 frame++;
 887                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 888                 if (IS_ERR(frame->bh)) {
 889                         ret_err = (struct dx_frame *) frame->bh;
 890                         frame->bh = NULL;
 891                         goto fail;
 892                 }
 893                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 894
 895                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 896                         ext4_warning_inode(dir,
 897                                 "dx entry: limit %u != node limit %u",
 898                                 dx_get_limit(entries), dx_node_limit(dir));
 899                         goto fail;
 900                 }
 901         }
 902 fail:
 903         while (frame >= frame_in) {
 904                 brelse(frame->bh);
 905                 frame--;
 906         }
 907
 908         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 909                 ext4_warning_inode(dir,
 910                         "Corrupt directory, running e2fsck is recommended");
 911         return ret_err;
 912 }
 913
 914 static void dx_release(struct dx_frame *frames)
 915 {
 916         struct dx_root_info *info;
 917         int i;
 918         unsigned int indirect_levels;
 919
 920         if (frames[0].bh == NULL)
 921                 return;
 922
 923         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 924         /* save local copy, "info" may be freed after brelse() */
 925         indirect_levels = info->indirect_levels;
 926         for (i = 0; i <= indirect_levels; i++) {
 927                 if (frames[i].bh == NULL)
 928                         break;
 929                 brelse(frames[i].bh);
 930                 frames[i].bh = NULL;
 931         }
 932 }
 933
 934 /*
 935  * This function increments the frame pointer to search the next leaf
 936  * block, and reads in the necessary intervening nodes if the search
 937  * should be necessary.  Whether or not the search is necessary is
 938  * controlled by the hash parameter.  If the hash value is even, then
 939  * the search is only continued if the next block starts with that
 940  * hash value.  This is used if we are searching for a specific file.
 941  *
 942  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 943  *
 944  * This function returns 1 if the caller should continue to search,
 945  * or 0 if it should not.  If there is an error reading one of the
 946  * index blocks, it will a negative error code.
 947  *
 948  * If start_hash is non-null, it will be filled in with the starting
 949  * hash of the next page.
 950  */
 951 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 952                                  struct dx_frame *frame,
 953                                  struct dx_frame *frames,
 954                                  __u32 *start_hash)
 955 {
 956         struct dx_frame *p;
 957         struct buffer_head *bh;
 958         int num_frames = 0;
 959         __u32 bhash;
 960
 961         p = frame;
 962         /*
 963          * Find the next leaf page by incrementing the frame pointer.
 964          * If we run out of entries in the interior node, loop around and
 965          * increment pointer in the parent node.  When we break out of
 966          * this loop, num_frames indicates the number of interior
 967          * nodes need to be read.
 968          */
 969         while (1) {
 970                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 971                         break;
 972                 if (p == frames)
 973                         return 0;
 974                 num_frames++;
 975                 p--;
 976         }
 977
 978         /*
 979          * If the hash is 1, then continue only if the next page has a
 980          * continuation hash of any value.  This is used for readdir
 981          * handling.  Otherwise, check to see if the hash matches the
 982          * desired continuation hash.  If it doesn't, return since
 983          * there's no point to read in the successive index pages.
 984          */
 985         bhash = dx_get_hash(p->at);
 986         if (start_hash)
 987                 *start_hash = bhash;
 988         if ((hash & 1) == 0) {
 989                 if ((bhash & ~1) != hash)
 990                         return 0;
 991         }
 992         /*
 993          * If the hash is HASH_NB_ALWAYS, we always go to the next
 994          * block so no check is necessary
 995          */
 996         while (num_frames--) {
 997                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 998                 if (IS_ERR(bh))
 999                         return PTR_ERR(bh);
1000                 p++;
1001                 brelse(p->bh);
1002                 p->bh = bh;
1003                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1004         }
1005         return 1;
1006 }
1007
1008
1009 /*
1010  * This function fills a red-black tree with information from a
1011  * directory block.  It returns the number directory entries loaded
1012  * into the tree.  If there is an error it is returned in err.
1013  */
1014 static int htree_dirblock_to_tree(struct file *dir_file,
1015                                   struct inode *dir, ext4_lblk_t block,
1016                                   struct dx_hash_info *hinfo,
1017                                   __u32 start_hash, __u32 start_minor_hash)
1018 {
1019         struct buffer_head *bh;
1020         struct ext4_dir_entry_2 *de, *top;
1021         int err = 0, count = 0;
1022         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1023         int csum = ext4_has_metadata_csum(dir->i_sb);
1024
1025         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1026                                                         (unsigned long)block));
1027         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1028         if (IS_ERR(bh))
1029                 return PTR_ERR(bh);
1030
1031         de = (struct ext4_dir_entry_2 *) bh->b_data;
1032         /* csum entries are not larger in the casefolded encrypted case */
1033         top = (struct ext4_dir_entry_2 *) ((char *) de +
1034                                            dir->i_sb->s_blocksize -
1035                                            ext4_dir_rec_len(0,
1036                                                            csum ? NULL : dir));
1037         /* Check if the directory is encrypted */
1038         if (IS_ENCRYPTED(dir)) {
1039                 err = fscrypt_prepare_readdir(dir);
1040                 if (err < 0) {
1041                         brelse(bh);
1042                         return err;
1043                 }
1044                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1045                                                  &fname_crypto_str);
1046                 if (err < 0) {
1047                         brelse(bh);
1048                         return err;
1049                 }
1050         }
1051
1052         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1053                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1054                                 bh->b_data, bh->b_size,
1055                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1056                                          + ((char *)de - bh->b_data))) {
1057                         /* silently ignore the rest of the block */
1058                         break;
1059                 }
1060                 if (ext4_hash_in_dirent(dir)) {
1061                         if (de->name_len && de->inode) {
1062                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1063                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1064                         } else {
1065                                 hinfo->hash = 0;
1066                                 hinfo->minor_hash = 0;
1067                         }
1068                 } else {
1069                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1070                 }
1071                 if ((hinfo->hash < start_hash) ||
1072                     ((hinfo->hash == start_hash) &&
1073                      (hinfo->minor_hash < start_minor_hash)))
1074                         continue;
1075                 if (de->inode == 0)
1076                         continue;
1077                 if (!IS_ENCRYPTED(dir)) {
1078                         tmp_str.name = de->name;
1079                         tmp_str.len = de->name_len;
1080                         err = ext4_htree_store_dirent(dir_file,
1081                                    hinfo->hash, hinfo->minor_hash, de,
1082                                    &tmp_str);
1083                 } else {
1084                         int save_len = fname_crypto_str.len;
1085                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1086                                                                 de->name_len);
1087
1088                         /* Directory is encrypted */
1089                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1090                                         hinfo->minor_hash, &de_name,
1091                                         &fname_crypto_str);
1092                         if (err) {
1093                                 count = err;
1094                                 goto errout;
1095                         }
1096                         err = ext4_htree_store_dirent(dir_file,
1097                                    hinfo->hash, hinfo->minor_hash, de,
1098                                         &fname_crypto_str);
1099                         fname_crypto_str.len = save_len;
1100                 }
1101                 if (err != 0) {
1102                         count = err;
1103                         goto errout;
1104                 }
1105                 count++;
1106         }
1107 errout:
1108         brelse(bh);
1109         fscrypt_fname_free_buffer(&fname_crypto_str);
1110         return count;
1111 }
1112
1113
1114 /*
1115  * This function fills a red-black tree with information from a
1116  * directory.  We start scanning the directory in hash order, starting
1117  * at start_hash and start_minor_hash.
1118  *
1119  * This function returns the number of entries inserted into the tree,
1120  * or a negative error code.
1121  */
1122 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1123                          __u32 start_minor_hash, __u32 *next_hash)
1124 {
1125         struct dx_hash_info hinfo;
1126         struct ext4_dir_entry_2 *de;
1127         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1128         struct inode *dir;
1129         ext4_lblk_t block;
1130         int count = 0;
1131         int ret, err;
1132         __u32 hashval;
1133         struct fscrypt_str tmp_str;
1134
1135         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1136                        start_hash, start_minor_hash));
1137         dir = file_inode(dir_file);
1138         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1139                 if (ext4_hash_in_dirent(dir))
1140                         hinfo.hash_version = DX_HASH_SIPHASH;
1141                 else
1142                         hinfo.hash_version =
1143                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1144                 if (hinfo.hash_version <= DX_HASH_TEA)
1145                         hinfo.hash_version +=
1146                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1147                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1148                 if (ext4_has_inline_data(dir)) {
1149                         int has_inline_data = 1;
1150                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1151                                                        &hinfo, start_hash,
1152                                                        start_minor_hash,
1153                                                        &has_inline_data);
1154                         if (has_inline_data) {
1155                                 *next_hash = ~0;
1156                                 return count;
1157                         }
1158                 }
1159                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1160                                                start_hash, start_minor_hash);
1161                 *next_hash = ~0;
1162                 return count;
1163         }
1164         hinfo.hash = start_hash;
1165         hinfo.minor_hash = 0;
1166         frame = dx_probe(NULL, dir, &hinfo, frames);
1167         if (IS_ERR(frame))
1168                 return PTR_ERR(frame);
1169
1170         /* Add '.' and '..' from the htree header */
1171         if (!start_hash && !start_minor_hash) {
1172                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1173                 tmp_str.name = de->name;
1174                 tmp_str.len = de->name_len;
1175                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1176                                               de, &tmp_str);
1177                 if (err != 0)
1178                         goto errout;
1179                 count++;
1180         }
1181         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1182                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1183                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1184                 tmp_str.name = de->name;
1185                 tmp_str.len = de->name_len;
1186                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1187                                               de, &tmp_str);
1188                 if (err != 0)
1189                         goto errout;
1190                 count++;
1191         }
1192
1193         while (1) {
1194                 if (fatal_signal_pending(current)) {
1195                         err = -ERESTARTSYS;
1196                         goto errout;
1197                 }
1198                 cond_resched();
1199                 block = dx_get_block(frame->at);
1200                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1201                                              start_hash, start_minor_hash);
1202                 if (ret < 0) {
1203                         err = ret;
1204                         goto errout;
1205                 }
1206                 count += ret;
1207                 hashval = ~0;
1208                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1209                                             frame, frames, &hashval);
1210                 *next_hash = hashval;
1211                 if (ret < 0) {
1212                         err = ret;
1213                         goto errout;
1214                 }
1215                 /*
1216                  * Stop if:  (a) there are no more entries, or
1217                  * (b) we have inserted at least one entry and the
1218                  * next hash value is not a continuation
1219                  */
1220                 if ((ret == 0) ||
1221                     (count && ((hashval & 1) == 0)))
1222                         break;
1223         }
1224         dx_release(frames);
1225         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1226                        "next hash: %x\n", count, *next_hash));
1227         return count;
1228 errout:
1229         dx_release(frames);
1230         return (err);
1231 }
1232
1233 static inline int search_dirblock(struct buffer_head *bh,
1234                                   struct inode *dir,
1235                                   struct ext4_filename *fname,
1236                                   unsigned int offset,
1237                                   struct ext4_dir_entry_2 **res_dir)
1238 {
1239         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1240                                fname, offset, res_dir);
1241 }
1242
1243 /*
1244  * Directory block splitting, compacting
1245  */
1246
1247 /*
1248  * Create map of hash values, offsets, and sizes, stored at end of block.
1249  * Returns number of entries mapped.
1250  */
1251 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1252                        unsigned blocksize, struct dx_hash_info *hinfo,
1253                        struct dx_map_entry *map_tail)
1254 {
1255         int count = 0;
1256         char *base = (char *) de;
1257         struct dx_hash_info h = *hinfo;
1258
1259         while ((char *) de < base + blocksize) {
1260                 if (de->name_len && de->inode) {
1261                         if (ext4_hash_in_dirent(dir))
1262                                 h.hash = EXT4_DIRENT_HASH(de);
1263                         else
1264                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1265                         map_tail--;
1266                         map_tail->hash = h.hash;
1267                         map_tail->offs = ((char *) de - base)>>2;
1268                         map_tail->size = le16_to_cpu(de->rec_len);
1269                         count++;
1270                         cond_resched();
1271                 }
1272                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1273                 de = ext4_next_entry(de, blocksize);
1274         }
1275         return count;
1276 }
1277
1278 /* Sort map by hash value */
1279 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1280 {
1281         struct dx_map_entry *p, *q, *top = map + count - 1;
1282         int more;
1283         /* Combsort until bubble sort doesn't suck */
1284         while (count > 2) {
1285                 count = count*10/13;
1286                 if (count - 9 < 2) /* 9, 10 -> 11 */
1287                         count = 11;
1288                 for (p = top, q = p - count; q >= map; p--, q--)
1289                         if (p->hash < q->hash)
1290                                 swap(*p, *q);
1291         }
1292         /* Garden variety bubble sort */
1293         do {
1294                 more = 0;
1295                 q = top;
1296                 while (q-- > map) {
1297                         if (q[1].hash >= q[0].hash)
1298                                 continue;
1299                         swap(*(q+1), *q);
1300                         more = 1;
1301                 }
1302         } while(more);
1303 }
1304
1305 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1306 {
1307         struct dx_entry *entries = frame->entries;
1308         struct dx_entry *old = frame->at, *new = old + 1;
1309         int count = dx_get_count(entries);
1310
1311         ASSERT(count < dx_get_limit(entries));
1312         ASSERT(old < entries + count);
1313         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1314         dx_set_hash(new, hash);
1315         dx_set_block(new, block);
1316         dx_set_count(entries, count + 1);
1317 }
1318
1319 #ifdef CONFIG_UNICODE
1320 /*
1321  * Test whether a case-insensitive directory entry matches the filename
1322  * being searched for.  If quick is set, assume the name being looked up
1323  * is already in the casefolded form.
1324  *
1325  * Returns: 0 if the directory entry matches, more than 0 if it
1326  * doesn't match or less than zero on error.
1327  */
1328 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1329                            u8 *de_name, size_t de_name_len, bool quick)
1330 {
1331         const struct super_block *sb = parent->i_sb;
1332         const struct unicode_map *um = sb->s_encoding;
1333         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1334         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1335         int ret;
1336
1337         if (IS_ENCRYPTED(parent)) {
1338                 const struct fscrypt_str encrypted_name =
1339                                 FSTR_INIT(de_name, de_name_len);
1340
1341                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1342                 if (!decrypted_name.name)
1343                         return -ENOMEM;
1344                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1345                                                 &decrypted_name);
1346                 if (ret < 0)
1347                         goto out;
1348                 entry.name = decrypted_name.name;
1349                 entry.len = decrypted_name.len;
1350         }
1351
1352         if (quick)
1353                 ret = utf8_strncasecmp_folded(um, name, &entry);
1354         else
1355                 ret = utf8_strncasecmp(um, name, &entry);
1356         if (ret < 0) {
1357                 /* Handle invalid character sequence as either an error
1358                  * or as an opaque byte sequence.
1359                  */
1360                 if (sb_has_strict_encoding(sb))
1361                         ret = -EINVAL;
1362                 else if (name->len != entry.len)
1363                         ret = 1;
1364                 else
1365                         ret = !!memcmp(name->name, entry.name, entry.len);
1366         }
1367 out:
1368         kfree(decrypted_name.name);
1369         return ret;
1370 }
1371
1372 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1373                                   struct ext4_filename *name)
1374 {
1375         struct fscrypt_str *cf_name = &name->cf_name;
1376         struct dx_hash_info *hinfo = &name->hinfo;
1377         int len;
1378
1379         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1380             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1381                 cf_name->name = NULL;
1382                 return 0;
1383         }
1384
1385         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1386         if (!cf_name->name)
1387                 return -ENOMEM;
1388
1389         len = utf8_casefold(dir->i_sb->s_encoding,
1390                             iname, cf_name->name,
1391                             EXT4_NAME_LEN);
1392         if (len <= 0) {
1393                 kfree(cf_name->name);
1394                 cf_name->name = NULL;
1395         }
1396         cf_name->len = (unsigned) len;
1397         if (!IS_ENCRYPTED(dir))
1398                 return 0;
1399
1400         hinfo->hash_version = DX_HASH_SIPHASH;
1401         hinfo->seed = NULL;
1402         if (cf_name->name)
1403                 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1404         else
1405                 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1406         return 0;
1407 }
1408 #endif
1409
1410 /*
1411  * Test whether a directory entry matches the filename being searched for.
1412  *
1413  * Return: %true if the directory entry matches, otherwise %false.
1414  */
1415 static bool ext4_match(struct inode *parent,
1416                               const struct ext4_filename *fname,
1417                               struct ext4_dir_entry_2 *de)
1418 {
1419         struct fscrypt_name f;
1420
1421         if (!de->inode)
1422                 return false;
1423
1424         f.usr_fname = fname->usr_fname;
1425         f.disk_name = fname->disk_name;
1426 #ifdef CONFIG_FS_ENCRYPTION
1427         f.crypto_buf = fname->crypto_buf;
1428 #endif
1429
1430 #ifdef CONFIG_UNICODE
1431         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1432             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1433                 if (fname->cf_name.name) {
1434                         struct qstr cf = {.name = fname->cf_name.name,
1435                                           .len = fname->cf_name.len};
1436                         if (IS_ENCRYPTED(parent)) {
1437                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1438                                         fname->hinfo.minor_hash !=
1439                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1440
1441                                         return 0;
1442                                 }
1443                         }
1444                         return !ext4_ci_compare(parent, &cf, de->name,
1445                                                         de->name_len, true);
1446                 }
1447                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1448                                                 de->name_len, false);
1449         }
1450 #endif
1451
1452         return fscrypt_match_name(&f, de->name, de->name_len);
1453 }
1454
1455 /*
1456  * Returns 0 if not found, -1 on failure, and 1 on success
1457  */
1458 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1459                     struct inode *dir, struct ext4_filename *fname,
1460                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1461 {
1462         struct ext4_dir_entry_2 * de;
1463         char * dlimit;
1464         int de_len;
1465
1466         de = (struct ext4_dir_entry_2 *)search_buf;
1467         dlimit = search_buf + buf_size;
1468         while ((char *) de < dlimit) {
1469                 /* this code is executed quadratically often */
1470                 /* do minimal checking `by hand' */
1471                 if ((char *) de + de->name_len <= dlimit &&
1472                     ext4_match(dir, fname, de)) {
1473                         /* found a match - just to be sure, do
1474                          * a full check */
1475                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1476                                                  buf_size, offset))
1477                                 return -1;
1478                         *res_dir = de;
1479                         return 1;
1480                 }
1481                 /* prevent looping on a bad block */
1482                 de_len = ext4_rec_len_from_disk(de->rec_len,
1483                                                 dir->i_sb->s_blocksize);
1484                 if (de_len <= 0)
1485                         return -1;
1486                 offset += de_len;
1487                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1488         }
1489         return 0;
1490 }
1491
1492 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1493                                struct ext4_dir_entry *de)
1494 {
1495         struct super_block *sb = dir->i_sb;
1496
1497         if (!is_dx(dir))
1498                 return 0;
1499         if (block == 0)
1500                 return 1;
1501         if (de->inode == 0 &&
1502             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1503                         sb->s_blocksize)
1504                 return 1;
1505         return 0;
1506 }
1507
1508 /*
1509  *      __ext4_find_entry()
1510  *
1511  * finds an entry in the specified directory with the wanted name. It
1512  * returns the cache buffer in which the entry was found, and the entry
1513  * itself (as a parameter - res_dir). It does NOT read the inode of the
1514  * entry - you'll have to do that yourself if you want to.
1515  *
1516  * The returned buffer_head has ->b_count elevated.  The caller is expected
1517  * to brelse() it when appropriate.
1518  */
1519 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1520                                              struct ext4_filename *fname,
1521                                              struct ext4_dir_entry_2 **res_dir,
1522                                              int *inlined)
1523 {
1524         struct super_block *sb;
1525         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1526         struct buffer_head *bh, *ret = NULL;
1527         ext4_lblk_t start, block;
1528         const u8 *name = fname->usr_fname->name;
1529         size_t ra_max = 0;      /* Number of bh's in the readahead
1530                                    buffer, bh_use[] */
1531         size_t ra_ptr = 0;      /* Current index into readahead
1532                                    buffer */
1533         ext4_lblk_t  nblocks;
1534         int i, namelen, retval;
1535
1536         *res_dir = NULL;
1537         sb = dir->i_sb;
1538         namelen = fname->usr_fname->len;
1539         if (namelen > EXT4_NAME_LEN)
1540                 return NULL;
1541
1542         if (ext4_has_inline_data(dir)) {
1543                 int has_inline_data = 1;
1544                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1545                                              &has_inline_data);
1546                 if (has_inline_data) {
1547                         if (inlined)
1548                                 *inlined = 1;
1549                         goto cleanup_and_exit;
1550                 }
1551         }
1552
1553         if ((namelen <= 2) && (name[0] == '.') &&
1554             (name[1] == '.' || name[1] == '\0')) {
1555                 /*
1556                  * "." or ".." will only be in the first block
1557                  * NFS may look up ".."; "." should be handled by the VFS
1558                  */
1559                 block = start = 0;
1560                 nblocks = 1;
1561                 goto restart;
1562         }
1563         if (is_dx(dir)) {
1564                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1565                 /*
1566                  * On success, or if the error was file not found,
1567                  * return.  Otherwise, fall back to doing a search the
1568                  * old fashioned way.
1569                  */
1570                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1571                         goto cleanup_and_exit;
1572                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1573                                "falling back\n"));
1574                 ret = NULL;
1575         }
1576         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1577         if (!nblocks) {
1578                 ret = NULL;
1579                 goto cleanup_and_exit;
1580         }
1581         start = EXT4_I(dir)->i_dir_start_lookup;
1582         if (start >= nblocks)
1583                 start = 0;
1584         block = start;
1585 restart:
1586         do {
1587                 /*
1588                  * We deal with the read-ahead logic here.
1589                  */
1590                 cond_resched();
1591                 if (ra_ptr >= ra_max) {
1592                         /* Refill the readahead buffer */
1593                         ra_ptr = 0;
1594                         if (block < start)
1595                                 ra_max = start - block;
1596                         else
1597                                 ra_max = nblocks - block;
1598                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1599                         retval = ext4_bread_batch(dir, block, ra_max,
1600                                                   false /* wait */, bh_use);
1601                         if (retval) {
1602                                 ret = ERR_PTR(retval);
1603                                 ra_max = 0;
1604                                 goto cleanup_and_exit;
1605                         }
1606                 }
1607                 if ((bh = bh_use[ra_ptr++]) == NULL)
1608                         goto next;
1609                 wait_on_buffer(bh);
1610                 if (!buffer_uptodate(bh)) {
1611                         EXT4_ERROR_INODE_ERR(dir, EIO,
1612                                              "reading directory lblock %lu",
1613                                              (unsigned long) block);
1614                         brelse(bh);
1615                         ret = ERR_PTR(-EIO);
1616                         goto cleanup_and_exit;
1617                 }
1618                 if (!buffer_verified(bh) &&
1619                     !is_dx_internal_node(dir, block,
1620                                          (struct ext4_dir_entry *)bh->b_data) &&
1621                     !ext4_dirblock_csum_verify(dir, bh)) {
1622                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1623                                              "checksumming directory "
1624                                              "block %lu", (unsigned long)block);
1625                         brelse(bh);
1626                         ret = ERR_PTR(-EFSBADCRC);
1627                         goto cleanup_and_exit;
1628                 }
1629                 set_buffer_verified(bh);
1630                 i = search_dirblock(bh, dir, fname,
1631                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1632                 if (i == 1) {
1633                         EXT4_I(dir)->i_dir_start_lookup = block;
1634                         ret = bh;
1635                         goto cleanup_and_exit;
1636                 } else {
1637                         brelse(bh);
1638                         if (i < 0)
1639                                 goto cleanup_and_exit;
1640                 }
1641         next:
1642                 if (++block >= nblocks)
1643                         block = 0;
1644         } while (block != start);
1645
1646         /*
1647          * If the directory has grown while we were searching, then
1648          * search the last part of the directory before giving up.
1649          */
1650         block = nblocks;
1651         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1652         if (block < nblocks) {
1653                 start = 0;
1654                 goto restart;
1655         }
1656
1657 cleanup_and_exit:
1658         /* Clean up the read-ahead blocks */
1659         for (; ra_ptr < ra_max; ra_ptr++)
1660                 brelse(bh_use[ra_ptr]);
1661         return ret;
1662 }
1663
1664 static struct buffer_head *ext4_find_entry(struct inode *dir,
1665                                            const struct qstr *d_name,
1666                                            struct ext4_dir_entry_2 **res_dir,
1667                                            int *inlined)
1668 {
1669         int err;
1670         struct ext4_filename fname;
1671         struct buffer_head *bh;
1672
1673         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1674         if (err == -ENOENT)
1675                 return NULL;
1676         if (err)
1677                 return ERR_PTR(err);
1678
1679         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1680
1681         ext4_fname_free_filename(&fname);
1682         return bh;
1683 }
1684
1685 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1686                                              struct dentry *dentry,
1687                                              struct ext4_dir_entry_2 **res_dir)
1688 {
1689         int err;
1690         struct ext4_filename fname;
1691         struct buffer_head *bh;
1692
1693         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1694         generic_set_encrypted_ci_d_ops(dentry);
1695         if (err == -ENOENT)
1696                 return NULL;
1697         if (err)
1698                 return ERR_PTR(err);
1699
1700         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1701
1702         ext4_fname_free_filename(&fname);
1703         return bh;
1704 }
1705
1706 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1707                         struct ext4_filename *fname,
1708                         struct ext4_dir_entry_2 **res_dir)
1709 {
1710         struct super_block * sb = dir->i_sb;
1711         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1712         struct buffer_head *bh;
1713         ext4_lblk_t block;
1714         int retval;
1715
1716 #ifdef CONFIG_FS_ENCRYPTION
1717         *res_dir = NULL;
1718 #endif
1719         frame = dx_probe(fname, dir, NULL, frames);
1720         if (IS_ERR(frame))
1721                 return (struct buffer_head *) frame;
1722         do {
1723                 block = dx_get_block(frame->at);
1724                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1725                 if (IS_ERR(bh))
1726                         goto errout;
1727
1728                 retval = search_dirblock(bh, dir, fname,
1729                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1730                                          res_dir);
1731                 if (retval == 1)
1732                         goto success;
1733                 brelse(bh);
1734                 if (retval == -1) {
1735                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1736                         goto errout;
1737                 }
1738
1739                 /* Check to see if we should continue to search */
1740                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1741                                                frames, NULL);
1742                 if (retval < 0) {
1743                         ext4_warning_inode(dir,
1744                                 "error %d reading directory index block",
1745                                 retval);
1746                         bh = ERR_PTR(retval);
1747                         goto errout;
1748                 }
1749         } while (retval == 1);
1750
1751         bh = NULL;
1752 errout:
1753         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1754 success:
1755         dx_release(frames);
1756         return bh;
1757 }
1758
1759 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1760 {
1761         struct inode *inode;
1762         struct ext4_dir_entry_2 *de;
1763         struct buffer_head *bh;
1764
1765         if (dentry->d_name.len > EXT4_NAME_LEN)
1766                 return ERR_PTR(-ENAMETOOLONG);
1767
1768         bh = ext4_lookup_entry(dir, dentry, &de);
1769         if (IS_ERR(bh))
1770                 return ERR_CAST(bh);
1771         inode = NULL;
1772         if (bh) {
1773                 __u32 ino = le32_to_cpu(de->inode);
1774                 brelse(bh);
1775                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1776                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1777                         return ERR_PTR(-EFSCORRUPTED);
1778                 }
1779                 if (unlikely(ino == dir->i_ino)) {
1780                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1781                                          dentry);
1782                         return ERR_PTR(-EFSCORRUPTED);
1783                 }
1784                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1785                 if (inode == ERR_PTR(-ESTALE)) {
1786                         EXT4_ERROR_INODE(dir,
1787                                          "deleted inode referenced: %u",
1788                                          ino);
1789                         return ERR_PTR(-EFSCORRUPTED);
1790                 }
1791                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1792                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1793                     !fscrypt_has_permitted_context(dir, inode)) {
1794                         ext4_warning(inode->i_sb,
1795                                      "Inconsistent encryption contexts: %lu/%lu",
1796                                      dir->i_ino, inode->i_ino);
1797                         iput(inode);
1798                         return ERR_PTR(-EPERM);
1799                 }
1800         }
1801
1802 #ifdef CONFIG_UNICODE
1803         if (!inode && IS_CASEFOLDED(dir)) {
1804                 /* Eventually we want to call d_add_ci(dentry, NULL)
1805                  * for negative dentries in the encoding case as
1806                  * well.  For now, prevent the negative dentry
1807                  * from being cached.
1808                  */
1809                 return NULL;
1810         }
1811 #endif
1812         return d_splice_alias(inode, dentry);
1813 }
1814
1815
1816 struct dentry *ext4_get_parent(struct dentry *child)
1817 {
1818         __u32 ino;
1819         struct ext4_dir_entry_2 * de;
1820         struct buffer_head *bh;
1821
1822         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1823         if (IS_ERR(bh))
1824                 return ERR_CAST(bh);
1825         if (!bh)
1826                 return ERR_PTR(-ENOENT);
1827         ino = le32_to_cpu(de->inode);
1828         brelse(bh);
1829
1830         if (!ext4_valid_inum(child->d_sb, ino)) {
1831                 EXT4_ERROR_INODE(d_inode(child),
1832                                  "bad parent inode number: %u", ino);
1833                 return ERR_PTR(-EFSCORRUPTED);
1834         }
1835
1836         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1837 }
1838
1839 /*
1840  * Move count entries from end of map between two memory locations.
1841  * Returns pointer to last entry moved.
1842  */
1843 static struct ext4_dir_entry_2 *
1844 dx_move_dirents(struct inode *dir, char *from, char *to,
1845                 struct dx_map_entry *map, int count,
1846                 unsigned blocksize)
1847 {
1848         unsigned rec_len = 0;
1849
1850         while (count--) {
1851                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1852                                                 (from + (map->offs<<2));
1853                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1854
1855                 memcpy (to, de, rec_len);
1856                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1857                                 ext4_rec_len_to_disk(rec_len, blocksize);
1858
1859                 /* wipe dir_entry excluding the rec_len field */
1860                 de->inode = 0;
1861                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1862                                                                 blocksize) -
1863                                          offsetof(struct ext4_dir_entry_2,
1864                                                                 name_len));
1865
1866                 map++;
1867                 to += rec_len;
1868         }
1869         return (struct ext4_dir_entry_2 *) (to - rec_len);
1870 }
1871
1872 /*
1873  * Compact each dir entry in the range to the minimal rec_len.
1874  * Returns pointer to last entry in range.
1875  */
1876 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1877                                                         unsigned int blocksize)
1878 {
1879         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1880         unsigned rec_len = 0;
1881
1882         prev = to = de;
1883         while ((char*)de < base + blocksize) {
1884                 next = ext4_next_entry(de, blocksize);
1885                 if (de->inode && de->name_len) {
1886                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1887                         if (de > to)
1888                                 memmove(to, de, rec_len);
1889                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1890                         prev = to;
1891                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1892                 }
1893                 de = next;
1894         }
1895         return prev;
1896 }
1897
1898 /*
1899  * Split a full leaf block to make room for a new dir entry.
1900  * Allocate a new block, and move entries so that they are approx. equally full.
1901  * Returns pointer to de in block into which the new entry will be inserted.
1902  */
1903 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1904                         struct buffer_head **bh,struct dx_frame *frame,
1905                         struct dx_hash_info *hinfo)
1906 {
1907         unsigned blocksize = dir->i_sb->s_blocksize;
1908         unsigned count, continued;
1909         struct buffer_head *bh2;
1910         ext4_lblk_t newblock;
1911         u32 hash2;
1912         struct dx_map_entry *map;
1913         char *data1 = (*bh)->b_data, *data2;
1914         unsigned split, move, size;
1915         struct ext4_dir_entry_2 *de = NULL, *de2;
1916         int     csum_size = 0;
1917         int     err = 0, i;
1918
1919         if (ext4_has_metadata_csum(dir->i_sb))
1920                 csum_size = sizeof(struct ext4_dir_entry_tail);
1921
1922         bh2 = ext4_append(handle, dir, &newblock);
1923         if (IS_ERR(bh2)) {
1924                 brelse(*bh);
1925                 *bh = NULL;
1926                 return (struct ext4_dir_entry_2 *) bh2;
1927         }
1928
1929         BUFFER_TRACE(*bh, "get_write_access");
1930         err = ext4_journal_get_write_access(handle, *bh);
1931         if (err)
1932                 goto journal_error;
1933
1934         BUFFER_TRACE(frame->bh, "get_write_access");
1935         err = ext4_journal_get_write_access(handle, frame->bh);
1936         if (err)
1937                 goto journal_error;
1938
1939         data2 = bh2->b_data;
1940
1941         /* create map in the end of data2 block */
1942         map = (struct dx_map_entry *) (data2 + blocksize);
1943         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1944                              blocksize, hinfo, map);
1945         map -= count;
1946         dx_sort_map(map, count);
1947         /* Ensure that neither split block is over half full */
1948         size = 0;
1949         move = 0;
1950         for (i = count-1; i >= 0; i--) {
1951                 /* is more than half of this entry in 2nd half of the block? */
1952                 if (size + map[i].size/2 > blocksize/2)
1953                         break;
1954                 size += map[i].size;
1955                 move++;
1956         }
1957         /*
1958          * map index at which we will split
1959          *
1960          * If the sum of active entries didn't exceed half the block size, just
1961          * split it in half by count; each resulting block will have at least
1962          * half the space free.
1963          */
1964         if (i > 0)
1965                 split = count - move;
1966         else
1967                 split = count/2;
1968
1969         hash2 = map[split].hash;
1970         continued = hash2 == map[split - 1].hash;
1971         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1972                         (unsigned long)dx_get_block(frame->at),
1973                                         hash2, split, count-split));
1974
1975         /* Fancy dance to stay within two buffers */
1976         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
1977                               blocksize);
1978         de = dx_pack_dirents(dir, data1, blocksize);
1979         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1980                                            (char *) de,
1981                                            blocksize);
1982         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1983                                             (char *) de2,
1984                                             blocksize);
1985         if (csum_size) {
1986                 ext4_initialize_dirent_tail(*bh, blocksize);
1987                 ext4_initialize_dirent_tail(bh2, blocksize);
1988         }
1989
1990         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1991                         blocksize, 1));
1992         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1993                         blocksize, 1));
1994
1995         /* Which block gets the new entry? */
1996         if (hinfo->hash >= hash2) {
1997                 swap(*bh, bh2);
1998                 de = de2;
1999         }
2000         dx_insert_block(frame, hash2 + continued, newblock);
2001         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2002         if (err)
2003                 goto journal_error;
2004         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2005         if (err)
2006                 goto journal_error;
2007         brelse(bh2);
2008         dxtrace(dx_show_index("frame", frame->entries));
2009         return de;
2010
2011 journal_error:
2012         brelse(*bh);
2013         brelse(bh2);
2014         *bh = NULL;
2015         ext4_std_error(dir->i_sb, err);
2016         return ERR_PTR(err);
2017 }
2018
2019 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2020                       struct buffer_head *bh,
2021                       void *buf, int buf_size,
2022                       struct ext4_filename *fname,
2023                       struct ext4_dir_entry_2 **dest_de)
2024 {
2025         struct ext4_dir_entry_2 *de;
2026         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2027         int nlen, rlen;
2028         unsigned int offset = 0;
2029         char *top;
2030
2031         de = (struct ext4_dir_entry_2 *)buf;
2032         top = buf + buf_size - reclen;
2033         while ((char *) de <= top) {
2034                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2035                                          buf, buf_size, offset))
2036                         return -EFSCORRUPTED;
2037                 if (ext4_match(dir, fname, de))
2038                         return -EEXIST;
2039                 nlen = ext4_dir_rec_len(de->name_len, dir);
2040                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2041                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2042                         break;
2043                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2044                 offset += rlen;
2045         }
2046         if ((char *) de > top)
2047                 return -ENOSPC;
2048
2049         *dest_de = de;
2050         return 0;
2051 }
2052
2053 void ext4_insert_dentry(struct inode *dir,
2054                         struct inode *inode,
2055                         struct ext4_dir_entry_2 *de,
2056                         int buf_size,
2057                         struct ext4_filename *fname)
2058 {
2059
2060         int nlen, rlen;
2061
2062         nlen = ext4_dir_rec_len(de->name_len, dir);
2063         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2064         if (de->inode) {
2065                 struct ext4_dir_entry_2 *de1 =
2066                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2067                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2068                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2069                 de = de1;
2070         }
2071         de->file_type = EXT4_FT_UNKNOWN;
2072         de->inode = cpu_to_le32(inode->i_ino);
2073         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2074         de->name_len = fname_len(fname);
2075         memcpy(de->name, fname_name(fname), fname_len(fname));
2076         if (ext4_hash_in_dirent(dir)) {
2077                 struct dx_hash_info *hinfo = &fname->hinfo;
2078
2079                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2080                 EXT4_DIRENT_HASHES(de)->minor_hash =
2081                                                 cpu_to_le32(hinfo->minor_hash);
2082         }
2083 }
2084
2085 /*
2086  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2087  * it points to a directory entry which is guaranteed to be large
2088  * enough for new directory entry.  If de is NULL, then
2089  * add_dirent_to_buf will attempt search the directory block for
2090  * space.  It will return -ENOSPC if no space is available, and -EIO
2091  * and -EEXIST if directory entry already exists.
2092  */
2093 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2094                              struct inode *dir,
2095                              struct inode *inode, struct ext4_dir_entry_2 *de,
2096                              struct buffer_head *bh)
2097 {
2098         unsigned int    blocksize = dir->i_sb->s_blocksize;
2099         int             csum_size = 0;
2100         int             err, err2;
2101
2102         if (ext4_has_metadata_csum(inode->i_sb))
2103                 csum_size = sizeof(struct ext4_dir_entry_tail);
2104
2105         if (!de) {
2106                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2107                                         blocksize - csum_size, fname, &de);
2108                 if (err)
2109                         return err;
2110         }
2111         BUFFER_TRACE(bh, "get_write_access");
2112         err = ext4_journal_get_write_access(handle, bh);
2113         if (err) {
2114                 ext4_std_error(dir->i_sb, err);
2115                 return err;
2116         }
2117
2118         /* By now the buffer is marked for journaling */
2119         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2120
2121         /*
2122          * XXX shouldn't update any times until successful
2123          * completion of syscall, but too many callers depend
2124          * on this.
2125          *
2126          * XXX similarly, too many callers depend on
2127          * ext4_new_inode() setting the times, but error
2128          * recovery deletes the inode, so the worst that can
2129          * happen is that the times are slightly out of date
2130          * and/or different from the directory change time.
2131          */
2132         dir->i_mtime = dir->i_ctime = current_time(dir);
2133         ext4_update_dx_flag(dir);
2134         inode_inc_iversion(dir);
2135         err2 = ext4_mark_inode_dirty(handle, dir);
2136         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2137         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2138         if (err)
2139                 ext4_std_error(dir->i_sb, err);
2140         return err ? err : err2;
2141 }
2142
2143 /*
2144  * This converts a one block unindexed directory to a 3 block indexed
2145  * directory, and adds the dentry to the indexed directory.
2146  */
2147 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2148                             struct inode *dir,
2149                             struct inode *inode, struct buffer_head *bh)
2150 {
2151         struct buffer_head *bh2;
2152         struct dx_root  *root;
2153         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2154         struct dx_entry *entries;
2155         struct ext4_dir_entry_2 *de, *de2;
2156         char            *data2, *top;
2157         unsigned        len;
2158         int             retval;
2159         unsigned        blocksize;
2160         ext4_lblk_t  block;
2161         struct fake_dirent *fde;
2162         int csum_size = 0;
2163
2164         if (ext4_has_metadata_csum(inode->i_sb))
2165                 csum_size = sizeof(struct ext4_dir_entry_tail);
2166
2167         blocksize =  dir->i_sb->s_blocksize;
2168         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2169         BUFFER_TRACE(bh, "get_write_access");
2170         retval = ext4_journal_get_write_access(handle, bh);
2171         if (retval) {
2172                 ext4_std_error(dir->i_sb, retval);
2173                 brelse(bh);
2174                 return retval;
2175         }
2176         root = (struct dx_root *) bh->b_data;
2177
2178         /* The 0th block becomes the root, move the dirents out */
2179         fde = &root->dotdot;
2180         de = (struct ext4_dir_entry_2 *)((char *)fde +
2181                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2182         if ((char *) de >= (((char *) root) + blocksize)) {
2183                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2184                 brelse(bh);
2185                 return -EFSCORRUPTED;
2186         }
2187         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2188
2189         /* Allocate new block for the 0th block's dirents */
2190         bh2 = ext4_append(handle, dir, &block);
2191         if (IS_ERR(bh2)) {
2192                 brelse(bh);
2193                 return PTR_ERR(bh2);
2194         }
2195         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2196         data2 = bh2->b_data;
2197
2198         memcpy(data2, de, len);
2199         memset(de, 0, len); /* wipe old data */
2200         de = (struct ext4_dir_entry_2 *) data2;
2201         top = data2 + len;
2202         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2203                 de = de2;
2204         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2205                                            (char *) de, blocksize);
2206
2207         if (csum_size)
2208                 ext4_initialize_dirent_tail(bh2, blocksize);
2209
2210         /* Initialize the root; the dot dirents already exist */
2211         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2212         de->rec_len = ext4_rec_len_to_disk(
2213                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2214         memset (&root->info, 0, sizeof(root->info));
2215         root->info.info_length = sizeof(root->info);
2216         if (ext4_hash_in_dirent(dir))
2217                 root->info.hash_version = DX_HASH_SIPHASH;
2218         else
2219                 root->info.hash_version =
2220                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2221
2222         entries = root->entries;
2223         dx_set_block(entries, 1);
2224         dx_set_count(entries, 1);
2225         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2226
2227         /* Initialize as for dx_probe */
2228         fname->hinfo.hash_version = root->info.hash_version;
2229         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2230                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2231         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2232
2233         /* casefolded encrypted hashes are computed on fname setup */
2234         if (!ext4_hash_in_dirent(dir))
2235                 ext4fs_dirhash(dir, fname_name(fname),
2236                                 fname_len(fname), &fname->hinfo);
2237
2238         memset(frames, 0, sizeof(frames));
2239         frame = frames;
2240         frame->entries = entries;
2241         frame->at = entries;
2242         frame->bh = bh;
2243
2244         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2245         if (retval)
2246                 goto out_frames;
2247         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2248         if (retval)
2249                 goto out_frames;
2250
2251         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2252         if (IS_ERR(de)) {
2253                 retval = PTR_ERR(de);
2254                 goto out_frames;
2255         }
2256
2257         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2258 out_frames:
2259         /*
2260          * Even if the block split failed, we have to properly write
2261          * out all the changes we did so far. Otherwise we can end up
2262          * with corrupted filesystem.
2263          */
2264         if (retval)
2265                 ext4_mark_inode_dirty(handle, dir);
2266         dx_release(frames);
2267         brelse(bh2);
2268         return retval;
2269 }
2270
2271 /*
2272  *      ext4_add_entry()
2273  *
2274  * adds a file entry to the specified directory, using the same
2275  * semantics as ext4_find_entry(). It returns NULL if it failed.
2276  *
2277  * NOTE!! The inode part of 'de' is left at 0 - which means you
2278  * may not sleep between calling this and putting something into
2279  * the entry, as someone else might have used it while you slept.
2280  */
2281 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2282                           struct inode *inode)
2283 {
2284         struct inode *dir = d_inode(dentry->d_parent);
2285         struct buffer_head *bh = NULL;
2286         struct ext4_dir_entry_2 *de;
2287         struct super_block *sb;
2288         struct ext4_filename fname;
2289         int     retval;
2290         int     dx_fallback=0;
2291         unsigned blocksize;
2292         ext4_lblk_t block, blocks;
2293         int     csum_size = 0;
2294
2295         if (ext4_has_metadata_csum(inode->i_sb))
2296                 csum_size = sizeof(struct ext4_dir_entry_tail);
2297
2298         sb = dir->i_sb;
2299         blocksize = sb->s_blocksize;
2300         if (!dentry->d_name.len)
2301                 return -EINVAL;
2302
2303         if (fscrypt_is_nokey_name(dentry))
2304                 return -ENOKEY;
2305
2306 #ifdef CONFIG_UNICODE
2307         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2308             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2309                 return -EINVAL;
2310 #endif
2311
2312         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2313         if (retval)
2314                 return retval;
2315
2316         if (ext4_has_inline_data(dir)) {
2317                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2318                 if (retval < 0)
2319                         goto out;
2320                 if (retval == 1) {
2321                         retval = 0;
2322                         goto out;
2323                 }
2324         }
2325
2326         if (is_dx(dir)) {
2327                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2328                 if (!retval || (retval != ERR_BAD_DX_DIR))
2329                         goto out;
2330                 /* Can we just ignore htree data? */
2331                 if (ext4_has_metadata_csum(sb)) {
2332                         EXT4_ERROR_INODE(dir,
2333                                 "Directory has corrupted htree index.");
2334                         retval = -EFSCORRUPTED;
2335                         goto out;
2336                 }
2337                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2338                 dx_fallback++;
2339                 retval = ext4_mark_inode_dirty(handle, dir);
2340                 if (unlikely(retval))
2341                         goto out;
2342         }
2343         blocks = dir->i_size >> sb->s_blocksize_bits;
2344         for (block = 0; block < blocks; block++) {
2345                 bh = ext4_read_dirblock(dir, block, DIRENT);
2346                 if (bh == NULL) {
2347                         bh = ext4_bread(handle, dir, block,
2348                                         EXT4_GET_BLOCKS_CREATE);
2349                         goto add_to_new_block;
2350                 }
2351                 if (IS_ERR(bh)) {
2352                         retval = PTR_ERR(bh);
2353                         bh = NULL;
2354                         goto out;
2355                 }
2356                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2357                                            NULL, bh);
2358                 if (retval != -ENOSPC)
2359                         goto out;
2360
2361                 if (blocks == 1 && !dx_fallback &&
2362                     ext4_has_feature_dir_index(sb)) {
2363                         retval = make_indexed_dir(handle, &fname, dir,
2364                                                   inode, bh);
2365                         bh = NULL; /* make_indexed_dir releases bh */
2366                         goto out;
2367                 }
2368                 brelse(bh);
2369         }
2370         bh = ext4_append(handle, dir, &block);
2371 add_to_new_block:
2372         if (IS_ERR(bh)) {
2373                 retval = PTR_ERR(bh);
2374                 bh = NULL;
2375                 goto out;
2376         }
2377         de = (struct ext4_dir_entry_2 *) bh->b_data;
2378         de->inode = 0;
2379         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2380
2381         if (csum_size)
2382                 ext4_initialize_dirent_tail(bh, blocksize);
2383
2384         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2385 out:
2386         ext4_fname_free_filename(&fname);
2387         brelse(bh);
2388         if (retval == 0)
2389                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2390         return retval;
2391 }
2392
2393 /*
2394  * Returns 0 for success, or a negative error value
2395  */
2396 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2397                              struct inode *dir, struct inode *inode)
2398 {
2399         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2400         struct dx_entry *entries, *at;
2401         struct buffer_head *bh;
2402         struct super_block *sb = dir->i_sb;
2403         struct ext4_dir_entry_2 *de;
2404         int restart;
2405         int err;
2406
2407 again:
2408         restart = 0;
2409         frame = dx_probe(fname, dir, NULL, frames);
2410         if (IS_ERR(frame))
2411                 return PTR_ERR(frame);
2412         entries = frame->entries;
2413         at = frame->at;
2414         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2415         if (IS_ERR(bh)) {
2416                 err = PTR_ERR(bh);
2417                 bh = NULL;
2418                 goto cleanup;
2419         }
2420
2421         BUFFER_TRACE(bh, "get_write_access");
2422         err = ext4_journal_get_write_access(handle, bh);
2423         if (err)
2424                 goto journal_error;
2425
2426         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2427         if (err != -ENOSPC)
2428                 goto cleanup;
2429
2430         err = 0;
2431         /* Block full, should compress but for now just split */
2432         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2433                        dx_get_count(entries), dx_get_limit(entries)));
2434         /* Need to split index? */
2435         if (dx_get_count(entries) == dx_get_limit(entries)) {
2436                 ext4_lblk_t newblock;
2437                 int levels = frame - frames + 1;
2438                 unsigned int icount;
2439                 int add_level = 1;
2440                 struct dx_entry *entries2;
2441                 struct dx_node *node2;
2442                 struct buffer_head *bh2;
2443
2444                 while (frame > frames) {
2445                         if (dx_get_count((frame - 1)->entries) <
2446                             dx_get_limit((frame - 1)->entries)) {
2447                                 add_level = 0;
2448                                 break;
2449                         }
2450                         frame--; /* split higher index block */
2451                         at = frame->at;
2452                         entries = frame->entries;
2453                         restart = 1;
2454                 }
2455                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2456                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2457                                          "reach max htree level :%d",
2458                                          dir->i_ino, levels);
2459                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2460                                 ext4_warning(sb, "Large directory feature is "
2461                                                  "not enabled on this "
2462                                                  "filesystem");
2463                         }
2464                         err = -ENOSPC;
2465                         goto cleanup;
2466                 }
2467                 icount = dx_get_count(entries);
2468                 bh2 = ext4_append(handle, dir, &newblock);
2469                 if (IS_ERR(bh2)) {
2470                         err = PTR_ERR(bh2);
2471                         goto cleanup;
2472                 }
2473                 node2 = (struct dx_node *)(bh2->b_data);
2474                 entries2 = node2->entries;
2475                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2476                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2477                                                            sb->s_blocksize);
2478                 BUFFER_TRACE(frame->bh, "get_write_access");
2479                 err = ext4_journal_get_write_access(handle, frame->bh);
2480                 if (err)
2481                         goto journal_error;
2482                 if (!add_level) {
2483                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2484                         unsigned hash2 = dx_get_hash(entries + icount1);
2485                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2486                                        icount1, icount2));
2487
2488                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2489                         err = ext4_journal_get_write_access(handle,
2490                                                              (frame - 1)->bh);
2491                         if (err)
2492                                 goto journal_error;
2493
2494                         memcpy((char *) entries2, (char *) (entries + icount1),
2495                                icount2 * sizeof(struct dx_entry));
2496                         dx_set_count(entries, icount1);
2497                         dx_set_count(entries2, icount2);
2498                         dx_set_limit(entries2, dx_node_limit(dir));
2499
2500                         /* Which index block gets the new entry? */
2501                         if (at - entries >= icount1) {
2502                                 frame->at = at = at - entries - icount1 + entries2;
2503                                 frame->entries = entries = entries2;
2504                                 swap(frame->bh, bh2);
2505                         }
2506                         dx_insert_block((frame - 1), hash2, newblock);
2507                         dxtrace(dx_show_index("node", frame->entries));
2508                         dxtrace(dx_show_index("node",
2509                                ((struct dx_node *) bh2->b_data)->entries));
2510                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2511                         if (err)
2512                                 goto journal_error;
2513                         brelse (bh2);
2514                         err = ext4_handle_dirty_dx_node(handle, dir,
2515                                                    (frame - 1)->bh);
2516                         if (err)
2517                                 goto journal_error;
2518                         err = ext4_handle_dirty_dx_node(handle, dir,
2519                                                         frame->bh);
2520                         if (err)
2521                                 goto journal_error;
2522                 } else {
2523                         struct dx_root *dxroot;
2524                         memcpy((char *) entries2, (char *) entries,
2525                                icount * sizeof(struct dx_entry));
2526                         dx_set_limit(entries2, dx_node_limit(dir));
2527
2528                         /* Set up root */
2529                         dx_set_count(entries, 1);
2530                         dx_set_block(entries + 0, newblock);
2531                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2532                         dxroot->info.indirect_levels += 1;
2533                         dxtrace(printk(KERN_DEBUG
2534                                        "Creating %d level index...\n",
2535                                        dxroot->info.indirect_levels));
2536                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2537                         if (err)
2538                                 goto journal_error;
2539                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2540                         brelse(bh2);
2541                         restart = 1;
2542                         goto journal_error;
2543                 }
2544         }
2545         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2546         if (IS_ERR(de)) {
2547                 err = PTR_ERR(de);
2548                 goto cleanup;
2549         }
2550         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2551         goto cleanup;
2552
2553 journal_error:
2554         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2555 cleanup:
2556         brelse(bh);
2557         dx_release(frames);
2558         /* @restart is true means htree-path has been changed, we need to
2559          * repeat dx_probe() to find out valid htree-path
2560          */
2561         if (restart && err == 0)
2562                 goto again;
2563         return err;
2564 }
2565
2566 /*
2567  * ext4_generic_delete_entry deletes a directory entry by merging it
2568  * with the previous entry
2569  */
2570 int ext4_generic_delete_entry(struct inode *dir,
2571                               struct ext4_dir_entry_2 *de_del,
2572                               struct buffer_head *bh,
2573                               void *entry_buf,
2574                               int buf_size,
2575                               int csum_size)
2576 {
2577         struct ext4_dir_entry_2 *de, *pde;
2578         unsigned int blocksize = dir->i_sb->s_blocksize;
2579         int i;
2580
2581         i = 0;
2582         pde = NULL;
2583         de = (struct ext4_dir_entry_2 *)entry_buf;
2584         while (i < buf_size - csum_size) {
2585                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2586                                          entry_buf, buf_size, i))
2587                         return -EFSCORRUPTED;
2588                 if (de == de_del)  {
2589                         if (pde) {
2590                                 pde->rec_len = ext4_rec_len_to_disk(
2591                                         ext4_rec_len_from_disk(pde->rec_len,
2592                                                                blocksize) +
2593                                         ext4_rec_len_from_disk(de->rec_len,
2594                                                                blocksize),
2595                                         blocksize);
2596
2597                                 /* wipe entire dir_entry */
2598                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2599                                                                 blocksize));
2600                         } else {
2601                                 /* wipe dir_entry excluding the rec_len field */
2602                                 de->inode = 0;
2603                                 memset(&de->name_len, 0,
2604                                         ext4_rec_len_from_disk(de->rec_len,
2605                                                                 blocksize) -
2606                                         offsetof(struct ext4_dir_entry_2,
2607                                                                 name_len));
2608                         }
2609
2610                         inode_inc_iversion(dir);
2611                         return 0;
2612                 }
2613                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2614                 pde = de;
2615                 de = ext4_next_entry(de, blocksize);
2616         }
2617         return -ENOENT;
2618 }
2619
2620 static int ext4_delete_entry(handle_t *handle,
2621                              struct inode *dir,
2622                              struct ext4_dir_entry_2 *de_del,
2623                              struct buffer_head *bh)
2624 {
2625         int err, csum_size = 0;
2626
2627         if (ext4_has_inline_data(dir)) {
2628                 int has_inline_data = 1;
2629                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2630                                                &has_inline_data);
2631                 if (has_inline_data)
2632                         return err;
2633         }
2634
2635         if (ext4_has_metadata_csum(dir->i_sb))
2636                 csum_size = sizeof(struct ext4_dir_entry_tail);
2637
2638         BUFFER_TRACE(bh, "get_write_access");
2639         err = ext4_journal_get_write_access(handle, bh);
2640         if (unlikely(err))
2641                 goto out;
2642
2643         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2644                                         dir->i_sb->s_blocksize, csum_size);
2645         if (err)
2646                 goto out;
2647
2648         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2649         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2650         if (unlikely(err))
2651                 goto out;
2652
2653         return 0;
2654 out:
2655         if (err != -ENOENT)
2656                 ext4_std_error(dir->i_sb, err);
2657         return err;
2658 }
2659
2660 /*
2661  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2662  * since this indicates that nlinks count was previously 1 to avoid overflowing
2663  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2664  * that subdirectory link counts are not being maintained accurately.
2665  *
2666  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2667  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2668  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2669  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2670  */
2671 static void ext4_inc_count(struct inode *inode)
2672 {
2673         inc_nlink(inode);
2674         if (is_dx(inode) &&
2675             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2676                 set_nlink(inode, 1);
2677 }
2678
2679 /*
2680  * If a directory had nlink == 1, then we should let it be 1. This indicates
2681  * directory has >EXT4_LINK_MAX subdirs.
2682  */
2683 static void ext4_dec_count(struct inode *inode)
2684 {
2685         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2686                 drop_nlink(inode);
2687 }
2688
2689
2690 /*
2691  * Add non-directory inode to a directory. On success, the inode reference is
2692  * consumed by dentry is instantiation. This is also indicated by clearing of
2693  * *inodep pointer. On failure, the caller is responsible for dropping the
2694  * inode reference in the safe context.
2695  */
2696 static int ext4_add_nondir(handle_t *handle,
2697                 struct dentry *dentry, struct inode **inodep)
2698 {
2699         struct inode *dir = d_inode(dentry->d_parent);
2700         struct inode *inode = *inodep;
2701         int err = ext4_add_entry(handle, dentry, inode);
2702         if (!err) {
2703                 err = ext4_mark_inode_dirty(handle, inode);
2704                 if (IS_DIRSYNC(dir))
2705                         ext4_handle_sync(handle);
2706                 d_instantiate_new(dentry, inode);
2707                 *inodep = NULL;
2708                 return err;
2709         }
2710         drop_nlink(inode);
2711         ext4_orphan_add(handle, inode);
2712         unlock_new_inode(inode);
2713         return err;
2714 }
2715
2716 /*
2717  * By the time this is called, we already have created
2718  * the directory cache entry for the new file, but it
2719  * is so far negative - it has no inode.
2720  *
2721  * If the create succeeds, we fill in the inode information
2722  * with d_instantiate().
2723  */
2724 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2725                        struct dentry *dentry, umode_t mode, bool excl)
2726 {
2727         handle_t *handle;
2728         struct inode *inode;
2729         int err, credits, retries = 0;
2730
2731         err = dquot_initialize(dir);
2732         if (err)
2733                 return err;
2734
2735         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2736                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2737 retry:
2738         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2739                                             0, NULL, EXT4_HT_DIR, credits);
2740         handle = ext4_journal_current_handle();
2741         err = PTR_ERR(inode);
2742         if (!IS_ERR(inode)) {
2743                 inode->i_op = &ext4_file_inode_operations;
2744                 inode->i_fop = &ext4_file_operations;
2745                 ext4_set_aops(inode);
2746                 err = ext4_add_nondir(handle, dentry, &inode);
2747                 if (!err)
2748                         ext4_fc_track_create(handle, dentry);
2749         }
2750         if (handle)
2751                 ext4_journal_stop(handle);
2752         if (!IS_ERR_OR_NULL(inode))
2753                 iput(inode);
2754         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2755                 goto retry;
2756         return err;
2757 }
2758
2759 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2760                       struct dentry *dentry, umode_t mode, dev_t rdev)
2761 {
2762         handle_t *handle;
2763         struct inode *inode;
2764         int err, credits, retries = 0;
2765
2766         err = dquot_initialize(dir);
2767         if (err)
2768                 return err;
2769
2770         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2771                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2772 retry:
2773         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2774                                             0, NULL, EXT4_HT_DIR, credits);
2775         handle = ext4_journal_current_handle();
2776         err = PTR_ERR(inode);
2777         if (!IS_ERR(inode)) {
2778                 init_special_inode(inode, inode->i_mode, rdev);
2779                 inode->i_op = &ext4_special_inode_operations;
2780                 err = ext4_add_nondir(handle, dentry, &inode);
2781                 if (!err)
2782                         ext4_fc_track_create(handle, dentry);
2783         }
2784         if (handle)
2785                 ext4_journal_stop(handle);
2786         if (!IS_ERR_OR_NULL(inode))
2787                 iput(inode);
2788         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2789                 goto retry;
2790         return err;
2791 }
2792
2793 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2794                         struct dentry *dentry, umode_t mode)
2795 {
2796         handle_t *handle;
2797         struct inode *inode;
2798         int err, retries = 0;
2799
2800         err = dquot_initialize(dir);
2801         if (err)
2802                 return err;
2803
2804 retry:
2805         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2806                                             NULL, 0, NULL,
2807                                             EXT4_HT_DIR,
2808                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2809                           4 + EXT4_XATTR_TRANS_BLOCKS);
2810         handle = ext4_journal_current_handle();
2811         err = PTR_ERR(inode);
2812         if (!IS_ERR(inode)) {
2813                 inode->i_op = &ext4_file_inode_operations;
2814                 inode->i_fop = &ext4_file_operations;
2815                 ext4_set_aops(inode);
2816                 d_tmpfile(dentry, inode);
2817                 err = ext4_orphan_add(handle, inode);
2818                 if (err)
2819                         goto err_unlock_inode;
2820                 mark_inode_dirty(inode);
2821                 unlock_new_inode(inode);
2822         }
2823         if (handle)
2824                 ext4_journal_stop(handle);
2825         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2826                 goto retry;
2827         return err;
2828 err_unlock_inode:
2829         ext4_journal_stop(handle);
2830         unlock_new_inode(inode);
2831         return err;
2832 }
2833
2834 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2835                           struct ext4_dir_entry_2 *de,
2836                           int blocksize, int csum_size,
2837                           unsigned int parent_ino, int dotdot_real_len)
2838 {
2839         de->inode = cpu_to_le32(inode->i_ino);
2840         de->name_len = 1;
2841         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2842                                            blocksize);
2843         strcpy(de->name, ".");
2844         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2845
2846         de = ext4_next_entry(de, blocksize);
2847         de->inode = cpu_to_le32(parent_ino);
2848         de->name_len = 2;
2849         if (!dotdot_real_len)
2850                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2851                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2852                                         blocksize);
2853         else
2854                 de->rec_len = ext4_rec_len_to_disk(
2855                                         ext4_dir_rec_len(de->name_len, NULL),
2856                                         blocksize);
2857         strcpy(de->name, "..");
2858         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2859
2860         return ext4_next_entry(de, blocksize);
2861 }
2862
2863 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2864                              struct inode *inode)
2865 {
2866         struct buffer_head *dir_block = NULL;
2867         struct ext4_dir_entry_2 *de;
2868         ext4_lblk_t block = 0;
2869         unsigned int blocksize = dir->i_sb->s_blocksize;
2870         int csum_size = 0;
2871         int err;
2872
2873         if (ext4_has_metadata_csum(dir->i_sb))
2874                 csum_size = sizeof(struct ext4_dir_entry_tail);
2875
2876         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2877                 err = ext4_try_create_inline_dir(handle, dir, inode);
2878                 if (err < 0 && err != -ENOSPC)
2879                         goto out;
2880                 if (!err)
2881                         goto out;
2882         }
2883
2884         inode->i_size = 0;
2885         dir_block = ext4_append(handle, inode, &block);
2886         if (IS_ERR(dir_block))
2887                 return PTR_ERR(dir_block);
2888         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2889         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2890         set_nlink(inode, 2);
2891         if (csum_size)
2892                 ext4_initialize_dirent_tail(dir_block, blocksize);
2893
2894         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2895         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2896         if (err)
2897                 goto out;
2898         set_buffer_verified(dir_block);
2899 out:
2900         brelse(dir_block);
2901         return err;
2902 }
2903
2904 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2905                       struct dentry *dentry, umode_t mode)
2906 {
2907         handle_t *handle;
2908         struct inode *inode;
2909         int err, err2 = 0, credits, retries = 0;
2910
2911         if (EXT4_DIR_LINK_MAX(dir))
2912                 return -EMLINK;
2913
2914         err = dquot_initialize(dir);
2915         if (err)
2916                 return err;
2917
2918         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2919                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2920 retry:
2921         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2922                                             &dentry->d_name,
2923                                             0, NULL, EXT4_HT_DIR, credits);
2924         handle = ext4_journal_current_handle();
2925         err = PTR_ERR(inode);
2926         if (IS_ERR(inode))
2927                 goto out_stop;
2928
2929         inode->i_op = &ext4_dir_inode_operations;
2930         inode->i_fop = &ext4_dir_operations;
2931         err = ext4_init_new_dir(handle, dir, inode);
2932         if (err)
2933                 goto out_clear_inode;
2934         err = ext4_mark_inode_dirty(handle, inode);
2935         if (!err)
2936                 err = ext4_add_entry(handle, dentry, inode);
2937         if (err) {
2938 out_clear_inode:
2939                 clear_nlink(inode);
2940                 ext4_orphan_add(handle, inode);
2941                 unlock_new_inode(inode);
2942                 err2 = ext4_mark_inode_dirty(handle, inode);
2943                 if (unlikely(err2))
2944                         err = err2;
2945                 ext4_journal_stop(handle);
2946                 iput(inode);
2947                 goto out_retry;
2948         }
2949         ext4_inc_count(dir);
2950
2951         ext4_update_dx_flag(dir);
2952         err = ext4_mark_inode_dirty(handle, dir);
2953         if (err)
2954                 goto out_clear_inode;
2955         d_instantiate_new(dentry, inode);
2956         ext4_fc_track_create(handle, dentry);
2957         if (IS_DIRSYNC(dir))
2958                 ext4_handle_sync(handle);
2959
2960 out_stop:
2961         if (handle)
2962                 ext4_journal_stop(handle);
2963 out_retry:
2964         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2965                 goto retry;
2966         return err;
2967 }
2968
2969 /*
2970  * routine to check that the specified directory is empty (for rmdir)
2971  */
2972 bool ext4_empty_dir(struct inode *inode)
2973 {
2974         unsigned int offset;
2975         struct buffer_head *bh;
2976         struct ext4_dir_entry_2 *de;
2977         struct super_block *sb;
2978
2979         if (ext4_has_inline_data(inode)) {
2980                 int has_inline_data = 1;
2981                 int ret;
2982
2983                 ret = empty_inline_dir(inode, &has_inline_data);
2984                 if (has_inline_data)
2985                         return ret;
2986         }
2987
2988         sb = inode->i_sb;
2989         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
2990                                         ext4_dir_rec_len(2, NULL)) {
2991                 EXT4_ERROR_INODE(inode, "invalid size");
2992                 return true;
2993         }
2994         /* The first directory block must not be a hole,
2995          * so treat it as DIRENT_HTREE
2996          */
2997         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2998         if (IS_ERR(bh))
2999                 return true;
3000
3001         de = (struct ext4_dir_entry_2 *) bh->b_data;
3002         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3003                                  0) ||
3004             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3005                 ext4_warning_inode(inode, "directory missing '.'");
3006                 brelse(bh);
3007                 return true;
3008         }
3009         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3010         de = ext4_next_entry(de, sb->s_blocksize);
3011         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3012                                  offset) ||
3013             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3014                 ext4_warning_inode(inode, "directory missing '..'");
3015                 brelse(bh);
3016                 return true;
3017         }
3018         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3019         while (offset < inode->i_size) {
3020                 if (!(offset & (sb->s_blocksize - 1))) {
3021                         unsigned int lblock;
3022                         brelse(bh);
3023                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3024                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3025                         if (bh == NULL) {
3026                                 offset += sb->s_blocksize;
3027                                 continue;
3028                         }
3029                         if (IS_ERR(bh))
3030                                 return true;
3031                 }
3032                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3033                                         (offset & (sb->s_blocksize - 1)));
3034                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3035                                          bh->b_data, bh->b_size, offset)) {
3036                         offset = (offset | (sb->s_blocksize - 1)) + 1;
3037                         continue;
3038                 }
3039                 if (le32_to_cpu(de->inode)) {
3040                         brelse(bh);
3041                         return false;
3042                 }
3043                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3044         }
3045         brelse(bh);
3046         return true;
3047 }
3048
3049 /*
3050  * ext4_orphan_add() links an unlinked or truncated inode into a list of
3051  * such inodes, starting at the superblock, in case we crash before the
3052  * file is closed/deleted, or in case the inode truncate spans multiple
3053  * transactions and the last transaction is not recovered after a crash.
3054  *
3055  * At filesystem recovery time, we walk this list deleting unlinked
3056  * inodes and truncating linked inodes in ext4_orphan_cleanup().
3057  *
3058  * Orphan list manipulation functions must be called under i_mutex unless
3059  * we are just creating the inode or deleting it.
3060  */
3061 int ext4_orphan_add(handle_t *handle, struct inode *inode)
3062 {
3063         struct super_block *sb = inode->i_sb;
3064         struct ext4_sb_info *sbi = EXT4_SB(sb);
3065         struct ext4_iloc iloc;
3066         int err = 0, rc;
3067         bool dirty = false;
3068
3069         if (!sbi->s_journal || is_bad_inode(inode))
3070                 return 0;
3071
3072         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3073                      !inode_is_locked(inode));
3074         /*
3075          * Exit early if inode already is on orphan list. This is a big speedup
3076          * since we don't have to contend on the global s_orphan_lock.
3077          */
3078         if (!list_empty(&EXT4_I(inode)->i_orphan))
3079                 return 0;
3080
3081         /*
3082          * Orphan handling is only valid for files with data blocks
3083          * being truncated, or files being unlinked. Note that we either
3084          * hold i_mutex, or the inode can not be referenced from outside,
3085          * so i_nlink should not be bumped due to race
3086          */
3087         ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3088                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
3089
3090         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3091         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3092         if (err)
3093                 goto out;
3094
3095         err = ext4_reserve_inode_write(handle, inode, &iloc);
3096         if (err)
3097                 goto out;
3098
3099         mutex_lock(&sbi->s_orphan_lock);
3100         /*
3101          * Due to previous errors inode may be already a part of on-disk
3102          * orphan list. If so skip on-disk list modification.
3103          */
3104         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
3105             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
3106                 /* Insert this inode at the head of the on-disk orphan list */
3107                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
3108                 lock_buffer(sbi->s_sbh);
3109                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
3110                 ext4_superblock_csum_set(sb);
3111                 unlock_buffer(sbi->s_sbh);
3112                 dirty = true;
3113         }
3114         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
3115         mutex_unlock(&sbi->s_orphan_lock);
3116
3117         if (dirty) {
3118                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3119                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3120                 if (!err)
3121                         err = rc;
3122                 if (err) {
3123                         /*
3124                          * We have to remove inode from in-memory list if
3125                          * addition to on disk orphan list failed. Stray orphan
3126                          * list entries can cause panics at unmount time.
3127                          */
3128                         mutex_lock(&sbi->s_orphan_lock);
3129                         list_del_init(&EXT4_I(inode)->i_orphan);
3130                         mutex_unlock(&sbi->s_orphan_lock);
3131                 }
3132         } else
3133                 brelse(iloc.bh);
3134
3135         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3136         jbd_debug(4, "orphan inode %lu will point to %d\n",
3137                         inode->i_ino, NEXT_ORPHAN(inode));
3138 out:
3139         ext4_std_error(sb, err);
3140         return err;
3141 }
3142
3143 /*
3144  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3145  * of such inodes stored on disk, because it is finally being cleaned up.
3146  */
3147 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3148 {
3149         struct list_head *prev;
3150         struct ext4_inode_info *ei = EXT4_I(inode);
3151         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3152         __u32 ino_next;
3153         struct ext4_iloc iloc;
3154         int err = 0;
3155
3156         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3157                 return 0;
3158
3159         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3160                      !inode_is_locked(inode));
3161         /* Do this quick check before taking global s_orphan_lock. */
3162         if (list_empty(&ei->i_orphan))
3163                 return 0;
3164
3165         if (handle) {
3166                 /* Grab inode buffer early before taking global s_orphan_lock */
3167                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3168         }
3169
3170         mutex_lock(&sbi->s_orphan_lock);
3171         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3172
3173         prev = ei->i_orphan.prev;
3174         list_del_init(&ei->i_orphan);
3175
3176         /* If we're on an error path, we may not have a valid
3177          * transaction handle with which to update the orphan list on
3178          * disk, but we still need to remove the inode from the linked
3179          * list in memory. */
3180         if (!handle || err) {
3181                 mutex_unlock(&sbi->s_orphan_lock);
3182                 goto out_err;
3183         }
3184
3185         ino_next = NEXT_ORPHAN(inode);
3186         if (prev == &sbi->s_orphan) {
3187                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3188                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3189                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3190                 if (err) {
3191                         mutex_unlock(&sbi->s_orphan_lock);
3192                         goto out_brelse;
3193                 }
3194                 lock_buffer(sbi->s_sbh);
3195                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3196                 ext4_superblock_csum_set(inode->i_sb);
3197                 unlock_buffer(sbi->s_sbh);
3198                 mutex_unlock(&sbi->s_orphan_lock);
3199                 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3200         } else {
3201                 struct ext4_iloc iloc2;
3202                 struct inode *i_prev =
3203                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3204
3205                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3206                           i_prev->i_ino, ino_next);
3207                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3208                 if (err) {
3209                         mutex_unlock(&sbi->s_orphan_lock);
3210                         goto out_brelse;
3211                 }
3212                 NEXT_ORPHAN(i_prev) = ino_next;
3213                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3214                 mutex_unlock(&sbi->s_orphan_lock);
3215         }
3216         if (err)
3217                 goto out_brelse;
3218         NEXT_ORPHAN(inode) = 0;
3219         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3220 out_err:
3221         ext4_std_error(inode->i_sb, err);
3222         return err;
3223
3224 out_brelse:
3225         brelse(iloc.bh);
3226         goto out_err;
3227 }
3228
3229 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3230 {
3231         int retval;
3232         struct inode *inode;
3233         struct buffer_head *bh;
3234         struct ext4_dir_entry_2 *de;
3235         handle_t *handle = NULL;
3236
3237         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3238                 return -EIO;
3239
3240         /* Initialize quotas before so that eventual writes go in
3241          * separate transaction */
3242         retval = dquot_initialize(dir);
3243         if (retval)
3244                 return retval;
3245         retval = dquot_initialize(d_inode(dentry));
3246         if (retval)
3247                 return retval;
3248
3249         retval = -ENOENT;
3250         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3251         if (IS_ERR(bh))
3252                 return PTR_ERR(bh);
3253         if (!bh)
3254                 goto end_rmdir;
3255
3256         inode = d_inode(dentry);
3257
3258         retval = -EFSCORRUPTED;
3259         if (le32_to_cpu(de->inode) != inode->i_ino)
3260                 goto end_rmdir;
3261
3262         retval = -ENOTEMPTY;
3263         if (!ext4_empty_dir(inode))
3264                 goto end_rmdir;
3265
3266         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3267                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3268         if (IS_ERR(handle)) {
3269                 retval = PTR_ERR(handle);
3270                 handle = NULL;
3271                 goto end_rmdir;
3272         }
3273
3274         if (IS_DIRSYNC(dir))
3275                 ext4_handle_sync(handle);
3276
3277         retval = ext4_delete_entry(handle, dir, de, bh);
3278         if (retval)
3279                 goto end_rmdir;
3280         if (!EXT4_DIR_LINK_EMPTY(inode))
3281                 ext4_warning_inode(inode,
3282                              "empty directory '%.*s' has too many links (%u)",
3283                              dentry->d_name.len, dentry->d_name.name,
3284                              inode->i_nlink);
3285         inode_inc_iversion(inode);
3286         clear_nlink(inode);
3287         /* There's no need to set i_disksize: the fact that i_nlink is
3288          * zero will ensure that the right thing happens during any
3289          * recovery. */
3290         inode->i_size = 0;
3291         ext4_orphan_add(handle, inode);
3292         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3293         retval = ext4_mark_inode_dirty(handle, inode);
3294         if (retval)
3295                 goto end_rmdir;
3296         ext4_dec_count(dir);
3297         ext4_update_dx_flag(dir);
3298         ext4_fc_track_unlink(handle, dentry);
3299         retval = ext4_mark_inode_dirty(handle, dir);
3300
3301 #ifdef CONFIG_UNICODE
3302         /* VFS negative dentries are incompatible with Encoding and
3303          * Case-insensitiveness. Eventually we'll want avoid
3304          * invalidating the dentries here, alongside with returning the
3305          * negative dentries at ext4_lookup(), when it is better
3306          * supported by the VFS for the CI case.
3307          */
3308         if (IS_CASEFOLDED(dir))
3309                 d_invalidate(dentry);
3310 #endif
3311
3312 end_rmdir:
3313         brelse(bh);
3314         if (handle)
3315                 ext4_journal_stop(handle);
3316         return retval;
3317 }
3318
3319 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3320                   struct inode *inode)
3321 {
3322         int retval = -ENOENT;
3323         struct buffer_head *bh;
3324         struct ext4_dir_entry_2 *de;
3325         int skip_remove_dentry = 0;
3326
3327         bh = ext4_find_entry(dir, d_name, &de, NULL);
3328         if (IS_ERR(bh))
3329                 return PTR_ERR(bh);
3330
3331         if (!bh)
3332                 return -ENOENT;
3333
3334         if (le32_to_cpu(de->inode) != inode->i_ino) {
3335                 /*
3336                  * It's okay if we find dont find dentry which matches
3337                  * the inode. That's because it might have gotten
3338                  * renamed to a different inode number
3339                  */
3340                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3341                         skip_remove_dentry = 1;
3342                 else
3343                         goto out;
3344         }
3345
3346         if (IS_DIRSYNC(dir))
3347                 ext4_handle_sync(handle);
3348
3349         if (!skip_remove_dentry) {
3350                 retval = ext4_delete_entry(handle, dir, de, bh);
3351                 if (retval)
3352                         goto out;
3353                 dir->i_ctime = dir->i_mtime = current_time(dir);
3354                 ext4_update_dx_flag(dir);
3355                 retval = ext4_mark_inode_dirty(handle, dir);
3356                 if (retval)
3357                         goto out;
3358         } else {
3359                 retval = 0;
3360         }
3361         if (inode->i_nlink == 0)
3362                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3363                                    d_name->len, d_name->name);
3364         else
3365                 drop_nlink(inode);
3366         if (!inode->i_nlink)
3367                 ext4_orphan_add(handle, inode);
3368         inode->i_ctime = current_time(inode);
3369         retval = ext4_mark_inode_dirty(handle, inode);
3370
3371 out:
3372         brelse(bh);
3373         return retval;
3374 }
3375
3376 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3377 {
3378         handle_t *handle;
3379         int retval;
3380
3381         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3382                 return -EIO;
3383
3384         trace_ext4_unlink_enter(dir, dentry);
3385         /*
3386          * Initialize quotas before so that eventual writes go
3387          * in separate transaction
3388          */
3389         retval = dquot_initialize(dir);
3390         if (retval)
3391                 goto out_trace;
3392         retval = dquot_initialize(d_inode(dentry));
3393         if (retval)
3394                 goto out_trace;
3395
3396         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3397                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3398         if (IS_ERR(handle)) {
3399                 retval = PTR_ERR(handle);
3400                 goto out_trace;
3401         }
3402
3403         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3404         if (!retval)
3405                 ext4_fc_track_unlink(handle, dentry);
3406 #ifdef CONFIG_UNICODE
3407         /* VFS negative dentries are incompatible with Encoding and
3408          * Case-insensitiveness. Eventually we'll want avoid
3409          * invalidating the dentries here, alongside with returning the
3410          * negative dentries at ext4_lookup(), when it is  better
3411          * supported by the VFS for the CI case.
3412          */
3413         if (IS_CASEFOLDED(dir))
3414                 d_invalidate(dentry);
3415 #endif
3416         if (handle)
3417                 ext4_journal_stop(handle);
3418
3419 out_trace:
3420         trace_ext4_unlink_exit(dentry, retval);
3421         return retval;
3422 }
3423
3424 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3425                         struct dentry *dentry, const char *symname)
3426 {
3427         handle_t *handle;
3428         struct inode *inode;
3429         int err, len = strlen(symname);
3430         int credits;
3431         struct fscrypt_str disk_link;
3432
3433         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3434                 return -EIO;
3435
3436         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3437                                       &disk_link);
3438         if (err)
3439                 return err;
3440
3441         err = dquot_initialize(dir);
3442         if (err)
3443                 return err;
3444
3445         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3446                 /*
3447                  * For non-fast symlinks, we just allocate inode and put it on
3448                  * orphan list in the first transaction => we need bitmap,
3449                  * group descriptor, sb, inode block, quota blocks, and
3450                  * possibly selinux xattr blocks.
3451                  */
3452                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3453                           EXT4_XATTR_TRANS_BLOCKS;
3454         } else {
3455                 /*
3456                  * Fast symlink. We have to add entry to directory
3457                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3458                  * allocate new inode (bitmap, group descriptor, inode block,
3459                  * quota blocks, sb is already counted in previous macros).
3460                  */
3461                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3462                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3463         }
3464
3465         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3466                                             &dentry->d_name, 0, NULL,
3467                                             EXT4_HT_DIR, credits);
3468         handle = ext4_journal_current_handle();
3469         if (IS_ERR(inode)) {
3470                 if (handle)
3471                         ext4_journal_stop(handle);
3472                 return PTR_ERR(inode);
3473         }
3474
3475         if (IS_ENCRYPTED(inode)) {
3476                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3477                 if (err)
3478                         goto err_drop_inode;
3479                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3480         }
3481
3482         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3483                 if (!IS_ENCRYPTED(inode))
3484                         inode->i_op = &ext4_symlink_inode_operations;
3485                 inode_nohighmem(inode);
3486                 ext4_set_aops(inode);
3487                 /*
3488                  * We cannot call page_symlink() with transaction started
3489                  * because it calls into ext4_write_begin() which can wait
3490                  * for transaction commit if we are running out of space
3491                  * and thus we deadlock. So we have to stop transaction now
3492                  * and restart it when symlink contents is written.
3493                  *
3494                  * To keep fs consistent in case of crash, we have to put inode
3495                  * to orphan list in the mean time.
3496                  */
3497                 drop_nlink(inode);
3498                 err = ext4_orphan_add(handle, inode);
3499                 if (handle)
3500                         ext4_journal_stop(handle);
3501                 handle = NULL;
3502                 if (err)
3503                         goto err_drop_inode;
3504                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3505                 if (err)
3506                         goto err_drop_inode;
3507                 /*
3508                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3509                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3510                  */
3511                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3512                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3513                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3514                 if (IS_ERR(handle)) {
3515                         err = PTR_ERR(handle);
3516                         handle = NULL;
3517                         goto err_drop_inode;
3518                 }
3519                 set_nlink(inode, 1);
3520                 err = ext4_orphan_del(handle, inode);
3521                 if (err)
3522                         goto err_drop_inode;
3523         } else {
3524                 /* clear the extent format for fast symlink */
3525                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3526                 if (!IS_ENCRYPTED(inode)) {
3527                         inode->i_op = &ext4_fast_symlink_inode_operations;
3528                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3529                 }
3530                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3531                        disk_link.len);
3532                 inode->i_size = disk_link.len - 1;
3533         }
3534         EXT4_I(inode)->i_disksize = inode->i_size;
3535         err = ext4_add_nondir(handle, dentry, &inode);
3536         if (handle)
3537                 ext4_journal_stop(handle);
3538         if (inode)
3539                 iput(inode);
3540         goto out_free_encrypted_link;
3541
3542 err_drop_inode:
3543         if (handle)
3544                 ext4_journal_stop(handle);
3545         clear_nlink(inode);
3546         unlock_new_inode(inode);
3547         iput(inode);
3548 out_free_encrypted_link:
3549         if (disk_link.name != (unsigned char *)symname)
3550                 kfree(disk_link.name);
3551         return err;
3552 }
3553
3554 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3555 {
3556         handle_t *handle;
3557         int err, retries = 0;
3558 retry:
3559         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3560                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3561                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3562         if (IS_ERR(handle))
3563                 return PTR_ERR(handle);
3564
3565         if (IS_DIRSYNC(dir))
3566                 ext4_handle_sync(handle);
3567
3568         inode->i_ctime = current_time(inode);
3569         ext4_inc_count(inode);
3570         ihold(inode);
3571
3572         err = ext4_add_entry(handle, dentry, inode);
3573         if (!err) {
3574                 err = ext4_mark_inode_dirty(handle, inode);
3575                 /* this can happen only for tmpfile being
3576                  * linked the first time
3577                  */
3578                 if (inode->i_nlink == 1)
3579                         ext4_orphan_del(handle, inode);
3580                 d_instantiate(dentry, inode);
3581                 ext4_fc_track_link(handle, dentry);
3582         } else {
3583                 drop_nlink(inode);
3584                 iput(inode);
3585         }
3586         ext4_journal_stop(handle);
3587         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3588                 goto retry;
3589         return err;
3590 }
3591
3592 static int ext4_link(struct dentry *old_dentry,
3593                      struct inode *dir, struct dentry *dentry)
3594 {
3595         struct inode *inode = d_inode(old_dentry);
3596         int err;
3597
3598         if (inode->i_nlink >= EXT4_LINK_MAX)
3599                 return -EMLINK;
3600
3601         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3602         if (err)
3603                 return err;
3604
3605         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3606             (!projid_eq(EXT4_I(dir)->i_projid,
3607                         EXT4_I(old_dentry->d_inode)->i_projid)))
3608                 return -EXDEV;
3609
3610         err = dquot_initialize(dir);
3611         if (err)
3612                 return err;
3613         return __ext4_link(dir, inode, dentry);
3614 }
3615
3616 /*
3617  * Try to find buffer head where contains the parent block.
3618  * It should be the inode block if it is inlined or the 1st block
3619  * if it is a normal dir.
3620  */
3621 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3622                                         struct inode *inode,
3623                                         int *retval,
3624                                         struct ext4_dir_entry_2 **parent_de,
3625                                         int *inlined)
3626 {
3627         struct buffer_head *bh;
3628
3629         if (!ext4_has_inline_data(inode)) {
3630                 /* The first directory block must not be a hole, so
3631                  * treat it as DIRENT_HTREE
3632                  */
3633                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3634                 if (IS_ERR(bh)) {
3635                         *retval = PTR_ERR(bh);
3636                         return NULL;
3637                 }
3638                 *parent_de = ext4_next_entry(
3639                                         (struct ext4_dir_entry_2 *)bh->b_data,
3640                                         inode->i_sb->s_blocksize);
3641                 return bh;
3642         }
3643
3644         *inlined = 1;
3645         return ext4_get_first_inline_block(inode, parent_de, retval);
3646 }
3647
3648 struct ext4_renament {
3649         struct inode *dir;
3650         struct dentry *dentry;
3651         struct inode *inode;
3652         bool is_dir;
3653         int dir_nlink_delta;
3654
3655         /* entry for "dentry" */
3656         struct buffer_head *bh;
3657         struct ext4_dir_entry_2 *de;
3658         int inlined;
3659
3660         /* entry for ".." in inode if it's a directory */
3661         struct buffer_head *dir_bh;
3662         struct ext4_dir_entry_2 *parent_de;
3663         int dir_inlined;
3664 };
3665
3666 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3667 {
3668         int retval;
3669
3670         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3671                                               &retval, &ent->parent_de,
3672                                               &ent->dir_inlined);
3673         if (!ent->dir_bh)
3674                 return retval;
3675         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3676                 return -EFSCORRUPTED;
3677         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3678         return ext4_journal_get_write_access(handle, ent->dir_bh);
3679 }
3680
3681 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3682                                   unsigned dir_ino)
3683 {
3684         int retval;
3685
3686         ent->parent_de->inode = cpu_to_le32(dir_ino);
3687         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3688         if (!ent->dir_inlined) {
3689                 if (is_dx(ent->inode)) {
3690                         retval = ext4_handle_dirty_dx_node(handle,
3691                                                            ent->inode,
3692                                                            ent->dir_bh);
3693                 } else {
3694                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3695                                                             ent->dir_bh);
3696                 }
3697         } else {
3698                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3699         }
3700         if (retval) {
3701                 ext4_std_error(ent->dir->i_sb, retval);
3702                 return retval;
3703         }
3704         return 0;
3705 }
3706
3707 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3708                        unsigned ino, unsigned file_type)
3709 {
3710         int retval, retval2;
3711
3712         BUFFER_TRACE(ent->bh, "get write access");
3713         retval = ext4_journal_get_write_access(handle, ent->bh);
3714         if (retval)
3715                 return retval;
3716         ent->de->inode = cpu_to_le32(ino);
3717         if (ext4_has_feature_filetype(ent->dir->i_sb))
3718                 ent->de->file_type = file_type;
3719         inode_inc_iversion(ent->dir);
3720         ent->dir->i_ctime = ent->dir->i_mtime =
3721                 current_time(ent->dir);
3722         retval = ext4_mark_inode_dirty(handle, ent->dir);
3723         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3724         if (!ent->inlined) {
3725                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3726                 if (unlikely(retval2)) {
3727                         ext4_std_error(ent->dir->i_sb, retval2);
3728                         return retval2;
3729                 }
3730         }
3731         return retval;
3732 }
3733
3734 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3735                           unsigned ino, unsigned file_type)
3736 {
3737         struct ext4_renament old = *ent;
3738         int retval = 0;
3739
3740         /*
3741          * old->de could have moved from under us during make indexed dir,
3742          * so the old->de may no longer valid and need to find it again
3743          * before reset old inode info.
3744          */
3745         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3746         if (IS_ERR(old.bh))
3747                 retval = PTR_ERR(old.bh);
3748         if (!old.bh)
3749                 retval = -ENOENT;
3750         if (retval) {
3751                 ext4_std_error(old.dir->i_sb, retval);
3752                 return;
3753         }
3754
3755         ext4_setent(handle, &old, ino, file_type);
3756         brelse(old.bh);
3757 }
3758
3759 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3760                                   const struct qstr *d_name)
3761 {
3762         int retval = -ENOENT;
3763         struct buffer_head *bh;
3764         struct ext4_dir_entry_2 *de;
3765
3766         bh = ext4_find_entry(dir, d_name, &de, NULL);
3767         if (IS_ERR(bh))
3768                 return PTR_ERR(bh);
3769         if (bh) {
3770                 retval = ext4_delete_entry(handle, dir, de, bh);
3771                 brelse(bh);
3772         }
3773         return retval;
3774 }
3775
3776 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3777                                int force_reread)
3778 {
3779         int retval;
3780         /*
3781          * ent->de could have moved from under us during htree split, so make
3782          * sure that we are deleting the right entry.  We might also be pointing
3783          * to a stale entry in the unused part of ent->bh so just checking inum
3784          * and the name isn't enough.
3785          */
3786         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3787             ent->de->name_len != ent->dentry->d_name.len ||
3788             strncmp(ent->de->name, ent->dentry->d_name.name,
3789                     ent->de->name_len) ||
3790             force_reread) {
3791                 retval = ext4_find_delete_entry(handle, ent->dir,
3792                                                 &ent->dentry->d_name);
3793         } else {
3794                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3795                 if (retval == -ENOENT) {
3796                         retval = ext4_find_delete_entry(handle, ent->dir,
3797                                                         &ent->dentry->d_name);
3798                 }
3799         }
3800
3801         if (retval) {
3802                 ext4_warning_inode(ent->dir,
3803                                    "Deleting old file: nlink %d, error=%d",
3804                                    ent->dir->i_nlink, retval);
3805         }
3806 }
3807
3808 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3809 {
3810         if (ent->dir_nlink_delta) {
3811                 if (ent->dir_nlink_delta == -1)
3812                         ext4_dec_count(ent->dir);
3813                 else
3814                         ext4_inc_count(ent->dir);
3815                 ext4_mark_inode_dirty(handle, ent->dir);
3816         }
3817 }
3818
3819 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3820                                               struct ext4_renament *ent,
3821                                               int credits, handle_t **h)
3822 {
3823         struct inode *wh;
3824         handle_t *handle;
3825         int retries = 0;
3826
3827         /*
3828          * for inode block, sb block, group summaries,
3829          * and inode bitmap
3830          */
3831         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3832                     EXT4_XATTR_TRANS_BLOCKS + 4);
3833 retry:
3834         wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3835                                          S_IFCHR | WHITEOUT_MODE,
3836                                          &ent->dentry->d_name, 0, NULL,
3837                                          EXT4_HT_DIR, credits);
3838
3839         handle = ext4_journal_current_handle();
3840         if (IS_ERR(wh)) {
3841                 if (handle)
3842                         ext4_journal_stop(handle);
3843                 if (PTR_ERR(wh) == -ENOSPC &&
3844                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3845                         goto retry;
3846         } else {
3847                 *h = handle;
3848                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3849                 wh->i_op = &ext4_special_inode_operations;
3850         }
3851         return wh;
3852 }
3853
3854 /*
3855  * Anybody can rename anything with this: the permission checks are left to the
3856  * higher-level routines.
3857  *
3858  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3859  * while new_{dentry,inode) refers to the destination dentry/inode
3860  * This comes from rename(const char *oldpath, const char *newpath)
3861  */
3862 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3863                        struct dentry *old_dentry, struct inode *new_dir,
3864                        struct dentry *new_dentry, unsigned int flags)
3865 {
3866         handle_t *handle = NULL;
3867         struct ext4_renament old = {
3868                 .dir = old_dir,
3869                 .dentry = old_dentry,
3870                 .inode = d_inode(old_dentry),
3871         };
3872         struct ext4_renament new = {
3873                 .dir = new_dir,
3874                 .dentry = new_dentry,
3875                 .inode = d_inode(new_dentry),
3876         };
3877         int force_reread;
3878         int retval;
3879         struct inode *whiteout = NULL;
3880         int credits;
3881         u8 old_file_type;
3882
3883         if (new.inode && new.inode->i_nlink == 0) {
3884                 EXT4_ERROR_INODE(new.inode,
3885                                  "target of rename is already freed");
3886                 return -EFSCORRUPTED;
3887         }
3888
3889         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3890             (!projid_eq(EXT4_I(new_dir)->i_projid,
3891                         EXT4_I(old_dentry->d_inode)->i_projid)))
3892                 return -EXDEV;
3893
3894         retval = dquot_initialize(old.dir);
3895         if (retval)
3896                 return retval;
3897         retval = dquot_initialize(new.dir);
3898         if (retval)
3899                 return retval;
3900
3901         /* Initialize quotas before so that eventual writes go
3902          * in separate transaction */
3903         if (new.inode) {
3904                 retval = dquot_initialize(new.inode);
3905                 if (retval)
3906                         return retval;
3907         }
3908
3909         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3910         if (IS_ERR(old.bh))
3911                 return PTR_ERR(old.bh);
3912         /*
3913          *  Check for inode number is _not_ due to possible IO errors.
3914          *  We might rmdir the source, keep it as pwd of some process
3915          *  and merrily kill the link to whatever was created under the
3916          *  same name. Goodbye sticky bit ;-<
3917          */
3918         retval = -ENOENT;
3919         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3920                 goto release_bh;
3921
3922         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3923                                  &new.de, &new.inlined);
3924         if (IS_ERR(new.bh)) {
3925                 retval = PTR_ERR(new.bh);
3926                 new.bh = NULL;
3927                 goto release_bh;
3928         }
3929         if (new.bh) {
3930                 if (!new.inode) {
3931                         brelse(new.bh);
3932                         new.bh = NULL;
3933                 }
3934         }
3935         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3936                 ext4_alloc_da_blocks(old.inode);
3937
3938         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3939                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3940         if (!(flags & RENAME_WHITEOUT)) {
3941                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3942                 if (IS_ERR(handle)) {
3943                         retval = PTR_ERR(handle);
3944                         goto release_bh;
3945                 }
3946         } else {
3947                 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3948                 if (IS_ERR(whiteout)) {
3949                         retval = PTR_ERR(whiteout);
3950                         goto release_bh;
3951                 }
3952         }
3953
3954         old_file_type = old.de->file_type;
3955         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3956                 ext4_handle_sync(handle);
3957
3958         if (S_ISDIR(old.inode->i_mode)) {
3959                 if (new.inode) {
3960                         retval = -ENOTEMPTY;
3961                         if (!ext4_empty_dir(new.inode))
3962                                 goto end_rename;
3963                 } else {
3964                         retval = -EMLINK;
3965                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3966                                 goto end_rename;
3967                 }
3968                 retval = ext4_rename_dir_prepare(handle, &old);
3969                 if (retval)
3970                         goto end_rename;
3971         }
3972         /*
3973          * If we're renaming a file within an inline_data dir and adding or
3974          * setting the new dirent causes a conversion from inline_data to
3975          * extents/blockmap, we need to force the dirent delete code to
3976          * re-read the directory, or else we end up trying to delete a dirent
3977          * from what is now the extent tree root (or a block map).
3978          */
3979         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3980                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3981
3982         if (whiteout) {
3983                 /*
3984                  * Do this before adding a new entry, so the old entry is sure
3985                  * to be still pointing to the valid old entry.
3986                  */
3987                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3988                                      EXT4_FT_CHRDEV);
3989                 if (retval)
3990                         goto end_rename;
3991                 retval = ext4_mark_inode_dirty(handle, whiteout);
3992                 if (unlikely(retval))
3993                         goto end_rename;
3994
3995         }
3996         if (!new.bh) {
3997                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3998                 if (retval)
3999                         goto end_rename;
4000         } else {
4001                 retval = ext4_setent(handle, &new,
4002                                      old.inode->i_ino, old_file_type);
4003                 if (retval)
4004                         goto end_rename;
4005         }
4006         if (force_reread)
4007                 force_reread = !ext4_test_inode_flag(new.dir,
4008                                                      EXT4_INODE_INLINE_DATA);
4009
4010         /*
4011          * Like most other Unix systems, set the ctime for inodes on a
4012          * rename.
4013          */
4014         old.inode->i_ctime = current_time(old.inode);
4015         retval = ext4_mark_inode_dirty(handle, old.inode);
4016         if (unlikely(retval))
4017                 goto end_rename;
4018
4019         if (!whiteout) {
4020                 /*
4021                  * ok, that's it
4022                  */
4023                 ext4_rename_delete(handle, &old, force_reread);
4024         }
4025
4026         if (new.inode) {
4027                 ext4_dec_count(new.inode);
4028                 new.inode->i_ctime = current_time(new.inode);
4029         }
4030         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
4031         ext4_update_dx_flag(old.dir);
4032         if (old.dir_bh) {
4033                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4034                 if (retval)
4035                         goto end_rename;
4036
4037                 ext4_dec_count(old.dir);
4038                 if (new.inode) {
4039                         /* checked ext4_empty_dir above, can't have another
4040                          * parent, ext4_dec_count() won't work for many-linked
4041                          * dirs */
4042                         clear_nlink(new.inode);
4043                 } else {
4044                         ext4_inc_count(new.dir);
4045                         ext4_update_dx_flag(new.dir);
4046                         retval = ext4_mark_inode_dirty(handle, new.dir);
4047                         if (unlikely(retval))
4048                                 goto end_rename;
4049                 }
4050         }
4051         retval = ext4_mark_inode_dirty(handle, old.dir);
4052         if (unlikely(retval))
4053                 goto end_rename;
4054
4055         if (S_ISDIR(old.inode->i_mode)) {
4056                 /*
4057                  * We disable fast commits here that's because the
4058                  * replay code is not yet capable of changing dot dot
4059                  * dirents in directories.
4060                  */
4061                 ext4_fc_mark_ineligible(old.inode->i_sb,
4062                         EXT4_FC_REASON_RENAME_DIR);
4063         } else {
4064                 if (new.inode)
4065                         ext4_fc_track_unlink(handle, new.dentry);
4066                 __ext4_fc_track_link(handle, old.inode, new.dentry);
4067                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4068                 if (whiteout)
4069                         __ext4_fc_track_create(handle, whiteout, old.dentry);
4070         }
4071
4072         if (new.inode) {
4073                 retval = ext4_mark_inode_dirty(handle, new.inode);
4074                 if (unlikely(retval))
4075                         goto end_rename;
4076                 if (!new.inode->i_nlink)
4077                         ext4_orphan_add(handle, new.inode);
4078         }
4079         retval = 0;
4080
4081 end_rename:
4082         if (whiteout) {
4083                 if (retval) {
4084                         ext4_resetent(handle, &old,
4085                                       old.inode->i_ino, old_file_type);
4086                         drop_nlink(whiteout);
4087                         ext4_orphan_add(handle, whiteout);
4088                 }
4089                 unlock_new_inode(whiteout);
4090                 ext4_journal_stop(handle);
4091                 iput(whiteout);
4092         } else {
4093                 ext4_journal_stop(handle);
4094         }
4095 release_bh:
4096         brelse(old.dir_bh);
4097         brelse(old.bh);
4098         brelse(new.bh);
4099         return retval;
4100 }
4101
4102 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4103                              struct inode *new_dir, struct dentry *new_dentry)
4104 {
4105         handle_t *handle = NULL;
4106         struct ext4_renament old = {
4107                 .dir = old_dir,
4108                 .dentry = old_dentry,
4109                 .inode = d_inode(old_dentry),
4110         };
4111         struct ext4_renament new = {
4112                 .dir = new_dir,
4113                 .dentry = new_dentry,
4114                 .inode = d_inode(new_dentry),
4115         };
4116         u8 new_file_type;
4117         int retval;
4118         struct timespec64 ctime;
4119
4120         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4121              !projid_eq(EXT4_I(new_dir)->i_projid,
4122                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
4123             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4124              !projid_eq(EXT4_I(old_dir)->i_projid,
4125                         EXT4_I(new_dentry->d_inode)->i_projid)))
4126                 return -EXDEV;
4127
4128         retval = dquot_initialize(old.dir);
4129         if (retval)
4130                 return retval;
4131         retval = dquot_initialize(new.dir);
4132         if (retval)
4133                 return retval;
4134
4135         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4136                                  &old.de, &old.inlined);
4137         if (IS_ERR(old.bh))
4138                 return PTR_ERR(old.bh);
4139         /*
4140          *  Check for inode number is _not_ due to possible IO errors.
4141          *  We might rmdir the source, keep it as pwd of some process
4142          *  and merrily kill the link to whatever was created under the
4143          *  same name. Goodbye sticky bit ;-<
4144          */
4145         retval = -ENOENT;
4146         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4147                 goto end_rename;
4148
4149         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4150                                  &new.de, &new.inlined);
4151         if (IS_ERR(new.bh)) {
4152                 retval = PTR_ERR(new.bh);
4153                 new.bh = NULL;
4154                 goto end_rename;
4155         }
4156
4157         /* RENAME_EXCHANGE case: old *and* new must both exist */
4158         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4159                 goto end_rename;
4160
4161         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4162                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4163                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4164         if (IS_ERR(handle)) {
4165                 retval = PTR_ERR(handle);
4166                 handle = NULL;
4167                 goto end_rename;
4168         }
4169
4170         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4171                 ext4_handle_sync(handle);
4172
4173         if (S_ISDIR(old.inode->i_mode)) {
4174                 old.is_dir = true;
4175                 retval = ext4_rename_dir_prepare(handle, &old);
4176                 if (retval)
4177                         goto end_rename;
4178         }
4179         if (S_ISDIR(new.inode->i_mode)) {
4180                 new.is_dir = true;
4181                 retval = ext4_rename_dir_prepare(handle, &new);
4182                 if (retval)
4183                         goto end_rename;
4184         }
4185
4186         /*
4187          * Other than the special case of overwriting a directory, parents'
4188          * nlink only needs to be modified if this is a cross directory rename.
4189          */
4190         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4191                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4192                 new.dir_nlink_delta = -old.dir_nlink_delta;
4193                 retval = -EMLINK;
4194                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4195                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4196                         goto end_rename;
4197         }
4198
4199         new_file_type = new.de->file_type;
4200         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4201         if (retval)
4202                 goto end_rename;
4203
4204         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4205         if (retval)
4206                 goto end_rename;
4207
4208         /*
4209          * Like most other Unix systems, set the ctime for inodes on a
4210          * rename.
4211          */
4212         ctime = current_time(old.inode);
4213         old.inode->i_ctime = ctime;
4214         new.inode->i_ctime = ctime;
4215         retval = ext4_mark_inode_dirty(handle, old.inode);
4216         if (unlikely(retval))
4217                 goto end_rename;
4218         retval = ext4_mark_inode_dirty(handle, new.inode);
4219         if (unlikely(retval))
4220                 goto end_rename;
4221         ext4_fc_mark_ineligible(new.inode->i_sb,
4222                                 EXT4_FC_REASON_CROSS_RENAME);
4223         if (old.dir_bh) {
4224                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4225                 if (retval)
4226                         goto end_rename;
4227         }
4228         if (new.dir_bh) {
4229                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4230                 if (retval)
4231                         goto end_rename;
4232         }
4233         ext4_update_dir_count(handle, &old);
4234         ext4_update_dir_count(handle, &new);
4235         retval = 0;
4236
4237 end_rename:
4238         brelse(old.dir_bh);
4239         brelse(new.dir_bh);
4240         brelse(old.bh);
4241         brelse(new.bh);
4242         if (handle)
4243                 ext4_journal_stop(handle);
4244         return retval;
4245 }
4246
4247 static int ext4_rename2(struct user_namespace *mnt_userns,
4248                         struct inode *old_dir, struct dentry *old_dentry,
4249                         struct inode *new_dir, struct dentry *new_dentry,
4250                         unsigned int flags)
4251 {
4252         int err;
4253
4254         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4255                 return -EIO;
4256
4257         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4258                 return -EINVAL;
4259
4260         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4261                                      flags);
4262         if (err)
4263                 return err;
4264
4265         if (flags & RENAME_EXCHANGE) {
4266                 return ext4_cross_rename(old_dir, old_dentry,
4267                                          new_dir, new_dentry);
4268         }
4269
4270         return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4271 }
4272
4273 /*
4274  * directories can handle most operations...
4275  */
4276 const struct inode_operations ext4_dir_inode_operations = {
4277         .create         = ext4_create,
4278         .lookup         = ext4_lookup,
4279         .link           = ext4_link,
4280         .unlink         = ext4_unlink,
4281         .symlink        = ext4_symlink,
4282         .mkdir          = ext4_mkdir,
4283         .rmdir          = ext4_rmdir,
4284         .mknod          = ext4_mknod,
4285         .tmpfile        = ext4_tmpfile,
4286         .rename         = ext4_rename2,
4287         .setattr        = ext4_setattr,
4288         .getattr        = ext4_getattr,
4289         .listxattr      = ext4_listxattr,
4290         .get_acl        = ext4_get_acl,
4291         .set_acl        = ext4_set_acl,
4292         .fiemap         = ext4_fiemap,
4293         .fileattr_get   = ext4_fileattr_get,
4294         .fileattr_set   = ext4_fileattr_set,
4295 };
4296
4297 const struct inode_operations ext4_special_inode_operations = {
4298         .setattr        = ext4_setattr,
4299         .getattr        = ext4_getattr,
4300         .listxattr      = ext4_listxattr,
4301         .get_acl        = ext4_get_acl,
4302         .set_acl        = ext4_set_acl,
4303 };