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         bh = ext4_bread(NULL, inode, block, 0);
 113         if (IS_ERR(bh)) {
 114                 __ext4_warning(inode->i_sb, func, line,
 115                                "inode #%lu: lblock %lu: comm %s: "
 116                                "error %ld reading directory block",
 117                                inode->i_ino, (unsigned long)block,
 118                                current->comm, PTR_ERR(bh));
 119
 120                 return bh;
 121         }
 122         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 123                 ext4_error_inode(inode, func, line, block,
 124                                  "Directory hole found for htree %s block",
 125                                  (type == INDEX) ? "index" : "leaf");
 126                 return ERR_PTR(-EFSCORRUPTED);
 127         }
 128         if (!bh)
 129                 return NULL;
 130         dirent = (struct ext4_dir_entry *) bh->b_data;
 131         /* Determine whether or not we have an index block */
 132         if (is_dx(inode)) {
 133                 if (block == 0)
 134                         is_dx_block = 1;
 135                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 136                                                 inode->i_sb->s_blocksize) ==
 137                          inode->i_sb->s_blocksize)
 138                         is_dx_block = 1;
 139         }
 140         if (!is_dx_block && type == INDEX) {
 141                 ext4_error_inode(inode, func, line, block,
 142                        "directory leaf block found instead of index block");
 143                 brelse(bh);
 144                 return ERR_PTR(-EFSCORRUPTED);
 145         }
 146         if (!ext4_has_metadata_csum(inode->i_sb) ||
 147             buffer_verified(bh))
 148                 return bh;
 149
 150         /*
 151          * An empty leaf block can get mistaken for a index block; for
 152          * this reason, we can only check the index checksum when the
 153          * caller is sure it should be an index block.
 154          */
 155         if (is_dx_block && type == INDEX) {
 156                 if (ext4_dx_csum_verify(inode, dirent))
 157                         set_buffer_verified(bh);
 158                 else {
 159                         ext4_error_inode(inode, func, line, block,
 160                                          "Directory index failed checksum");
 161                         brelse(bh);
 162                         return ERR_PTR(-EFSBADCRC);
 163                 }
 164         }
 165         if (!is_dx_block) {
 166                 if (ext4_dirblock_csum_verify(inode, bh))
 167                         set_buffer_verified(bh);
 168                 else {
 169                         ext4_error_inode(inode, func, line, block,
 170                                          "Directory block failed checksum");
 171                         brelse(bh);
 172                         return ERR_PTR(-EFSBADCRC);
 173                 }
 174         }
 175         return bh;
 176 }
 177
 178 #ifndef assert
 179 #define assert(test) J_ASSERT(test)
 180 #endif
 181
 182 #ifdef DX_DEBUG
 183 #define dxtrace(command) command
 184 #else
 185 #define dxtrace(command)
 186 #endif
 187
 188 struct fake_dirent
 189 {
 190         __le32 inode;
 191         __le16 rec_len;
 192         u8 name_len;
 193         u8 file_type;
 194 };
 195
 196 struct dx_countlimit
 197 {
 198         __le16 limit;
 199         __le16 count;
 200 };
 201
 202 struct dx_entry
 203 {
 204         __le32 hash;
 205         __le32 block;
 206 };
 207
 208 /*
 209  * dx_root_info is laid out so that if it should somehow get overlaid by a
 210  * dirent the two low bits of the hash version will be zero.  Therefore, the
 211  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 212  */
 213
 214 struct dx_root
 215 {
 216         struct fake_dirent dot;
 217         char dot_name[4];
 218         struct fake_dirent dotdot;
 219         char dotdot_name[4];
 220         struct dx_root_info
 221         {
 222                 __le32 reserved_zero;
 223                 u8 hash_version;
 224                 u8 info_length; /* 8 */
 225                 u8 indirect_levels;
 226                 u8 unused_flags;
 227         }
 228         info;
 229         struct dx_entry entries[0];
 230 };
 231
 232 struct dx_node
 233 {
 234         struct fake_dirent fake;
 235         struct dx_entry entries[0];
 236 };
 237
 238
 239 struct dx_frame
 240 {
 241         struct buffer_head *bh;
 242         struct dx_entry *entries;
 243         struct dx_entry *at;
 244 };
 245
 246 struct dx_map_entry
 247 {
 248         u32 hash;
 249         u16 offs;
 250         u16 size;
 251 };
 252
 253 /*
 254  * This goes at the end of each htree block.
 255  */
 256 struct dx_tail {
 257         u32 dt_reserved;
 258         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 259 };
 260
 261 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 262 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 263 static inline unsigned dx_get_hash(struct dx_entry *entry);
 264 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 265 static unsigned dx_get_count(struct dx_entry *entries);
 266 static unsigned dx_get_limit(struct dx_entry *entries);
 267 static void dx_set_count(struct dx_entry *entries, unsigned value);
 268 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 269 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 270 static unsigned dx_node_limit(struct inode *dir);
 271 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 272                                  struct inode *dir,
 273                                  struct dx_hash_info *hinfo,
 274                                  struct dx_frame *frame);
 275 static void dx_release(struct dx_frame *frames);
 276 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 277                        unsigned blocksize, struct dx_hash_info *hinfo,
 278                        struct dx_map_entry map[]);
 279 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 280 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 281                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 282 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 283 static void dx_insert_block(struct dx_frame *frame,
 284                                         u32 hash, ext4_lblk_t block);
 285 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 286                                  struct dx_frame *frame,
 287                                  struct dx_frame *frames,
 288                                  __u32 *start_hash);
 289 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 290                 struct ext4_filename *fname,
 291                 struct ext4_dir_entry_2 **res_dir);
 292 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 293                              struct inode *dir, struct inode *inode);
 294
 295 /* checksumming functions */
 296 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 297                                  unsigned int blocksize)
 298 {
 299         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 300
 301         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 302         t->det_rec_len = ext4_rec_len_to_disk(
 303                         sizeof(struct ext4_dir_entry_tail), blocksize);
 304         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 305 }
 306
 307 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 308 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 309                                                    struct buffer_head *bh)
 310 {
 311         struct ext4_dir_entry_tail *t;
 312
 313 #ifdef PARANOID
 314         struct ext4_dir_entry *d, *top;
 315
 316         d = (struct ext4_dir_entry *)bh->b_data;
 317         top = (struct ext4_dir_entry *)(bh->b_data +
 318                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 319                  sizeof(struct ext4_dir_entry_tail)));
 320         while (d < top && d->rec_len)
 321                 d = (struct ext4_dir_entry *)(((void *)d) +
 322                     le16_to_cpu(d->rec_len));
 323
 324         if (d != top)
 325                 return NULL;
 326
 327         t = (struct ext4_dir_entry_tail *)d;
 328 #else
 329         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 330 #endif
 331
 332         if (t->det_reserved_zero1 ||
 333             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 334             t->det_reserved_zero2 ||
 335             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 336                 return NULL;
 337
 338         return t;
 339 }
 340
 341 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 342 {
 343         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 344         struct ext4_inode_info *ei = EXT4_I(inode);
 345         __u32 csum;
 346
 347         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 348         return cpu_to_le32(csum);
 349 }
 350
 351 #define warn_no_space_for_csum(inode)                                   \
 352         __warn_no_space_for_csum((inode), __func__, __LINE__)
 353
 354 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 355                                      unsigned int line)
 356 {
 357         __ext4_warning_inode(inode, func, line,
 358                 "No space for directory leaf checksum. Please run e2fsck -D.");
 359 }
 360
 361 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 362 {
 363         struct ext4_dir_entry_tail *t;
 364
 365         if (!ext4_has_metadata_csum(inode->i_sb))
 366                 return 1;
 367
 368         t = get_dirent_tail(inode, bh);
 369         if (!t) {
 370                 warn_no_space_for_csum(inode);
 371                 return 0;
 372         }
 373
 374         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 375                                                   (char *)t - bh->b_data))
 376                 return 0;
 377
 378         return 1;
 379 }
 380
 381 static void ext4_dirblock_csum_set(struct inode *inode,
 382                                  struct buffer_head *bh)
 383 {
 384         struct ext4_dir_entry_tail *t;
 385
 386         if (!ext4_has_metadata_csum(inode->i_sb))
 387                 return;
 388
 389         t = get_dirent_tail(inode, bh);
 390         if (!t) {
 391                 warn_no_space_for_csum(inode);
 392                 return;
 393         }
 394
 395         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 396                                              (char *)t - bh->b_data);
 397 }
 398
 399 int ext4_handle_dirty_dirblock(handle_t *handle,
 400                                struct inode *inode,
 401                                struct buffer_head *bh)
 402 {
 403         ext4_dirblock_csum_set(inode, bh);
 404         return ext4_handle_dirty_metadata(handle, inode, bh);
 405 }
 406
 407 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 408                                                struct ext4_dir_entry *dirent,
 409                                                int *offset)
 410 {
 411         struct ext4_dir_entry *dp;
 412         struct dx_root_info *root;
 413         int count_offset;
 414
 415         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 416                 count_offset = 8;
 417         else if (le16_to_cpu(dirent->rec_len) == 12) {
 418                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 419                 if (le16_to_cpu(dp->rec_len) !=
 420                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 421                         return NULL;
 422                 root = (struct dx_root_info *)(((void *)dp + 12));
 423                 if (root->reserved_zero ||
 424                     root->info_length != sizeof(struct dx_root_info))
 425                         return NULL;
 426                 count_offset = 32;
 427         } else
 428                 return NULL;
 429
 430         if (offset)
 431                 *offset = count_offset;
 432         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 433 }
 434
 435 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 436                            int count_offset, int count, struct dx_tail *t)
 437 {
 438         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 439         struct ext4_inode_info *ei = EXT4_I(inode);
 440         __u32 csum;
 441         int size;
 442         __u32 dummy_csum = 0;
 443         int offset = offsetof(struct dx_tail, dt_checksum);
 444
 445         size = count_offset + (count * sizeof(struct dx_entry));
 446         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 447         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 448         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 449
 450         return cpu_to_le32(csum);
 451 }
 452
 453 static int ext4_dx_csum_verify(struct inode *inode,
 454                                struct ext4_dir_entry *dirent)
 455 {
 456         struct dx_countlimit *c;
 457         struct dx_tail *t;
 458         int count_offset, limit, count;
 459
 460         if (!ext4_has_metadata_csum(inode->i_sb))
 461                 return 1;
 462
 463         c = get_dx_countlimit(inode, dirent, &count_offset);
 464         if (!c) {
 465                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 466                 return 0;
 467         }
 468         limit = le16_to_cpu(c->limit);
 469         count = le16_to_cpu(c->count);
 470         if (count_offset + (limit * sizeof(struct dx_entry)) >
 471             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 472                 warn_no_space_for_csum(inode);
 473                 return 0;
 474         }
 475         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 476
 477         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 478                                             count, t))
 479                 return 0;
 480         return 1;
 481 }
 482
 483 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 484 {
 485         struct dx_countlimit *c;
 486         struct dx_tail *t;
 487         int count_offset, limit, count;
 488
 489         if (!ext4_has_metadata_csum(inode->i_sb))
 490                 return;
 491
 492         c = get_dx_countlimit(inode, dirent, &count_offset);
 493         if (!c) {
 494                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 495                 return;
 496         }
 497         limit = le16_to_cpu(c->limit);
 498         count = le16_to_cpu(c->count);
 499         if (count_offset + (limit * sizeof(struct dx_entry)) >
 500             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 501                 warn_no_space_for_csum(inode);
 502                 return;
 503         }
 504         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 505
 506         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 507 }
 508
 509 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 510                                             struct inode *inode,
 511                                             struct buffer_head *bh)
 512 {
 513         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 514         return ext4_handle_dirty_metadata(handle, inode, bh);
 515 }
 516
 517 /*
 518  * p is at least 6 bytes before the end of page
 519  */
 520 static inline struct ext4_dir_entry_2 *
 521 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 522 {
 523         return (struct ext4_dir_entry_2 *)((char *)p +
 524                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 525 }
 526
 527 /*
 528  * Future: use high four bits of block for coalesce-on-delete flags
 529  * Mask them off for now.
 530  */
 531
 532 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 533 {
 534         return le32_to_cpu(entry->block) & 0x0fffffff;
 535 }
 536
 537 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 538 {
 539         entry->block = cpu_to_le32(value);
 540 }
 541
 542 static inline unsigned dx_get_hash(struct dx_entry *entry)
 543 {
 544         return le32_to_cpu(entry->hash);
 545 }
 546
 547 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 548 {
 549         entry->hash = cpu_to_le32(value);
 550 }
 551
 552 static inline unsigned dx_get_count(struct dx_entry *entries)
 553 {
 554         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 555 }
 556
 557 static inline unsigned dx_get_limit(struct dx_entry *entries)
 558 {
 559         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 560 }
 561
 562 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 563 {
 564         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 565 }
 566
 567 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 568 {
 569         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 570 }
 571
 572 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 573 {
 574         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 575                 EXT4_DIR_REC_LEN(2) - infosize;
 576
 577         if (ext4_has_metadata_csum(dir->i_sb))
 578                 entry_space -= sizeof(struct dx_tail);
 579         return entry_space / sizeof(struct dx_entry);
 580 }
 581
 582 static inline unsigned dx_node_limit(struct inode *dir)
 583 {
 584         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 585
 586         if (ext4_has_metadata_csum(dir->i_sb))
 587                 entry_space -= sizeof(struct dx_tail);
 588         return entry_space / sizeof(struct dx_entry);
 589 }
 590
 591 /*
 592  * Debug
 593  */
 594 #ifdef DX_DEBUG
 595 static void dx_show_index(char * label, struct dx_entry *entries)
 596 {
 597         int i, n = dx_get_count (entries);
 598         printk(KERN_DEBUG "%s index", label);
 599         for (i = 0; i < n; i++) {
 600                 printk(KERN_CONT " %x->%lu",
 601                        i ? dx_get_hash(entries + i) : 0,
 602                        (unsigned long)dx_get_block(entries + i));
 603         }
 604         printk(KERN_CONT "\n");
 605 }
 606
 607 struct stats
 608 {
 609         unsigned names;
 610         unsigned space;
 611         unsigned bcount;
 612 };
 613
 614 static struct stats dx_show_leaf(struct inode *dir,
 615                                 struct dx_hash_info *hinfo,
 616                                 struct ext4_dir_entry_2 *de,
 617                                 int size, int show_names)
 618 {
 619         unsigned names = 0, space = 0;
 620         char *base = (char *) de;
 621         struct dx_hash_info h = *hinfo;
 622
 623         printk("names: ");
 624         while ((char *) de < base + size)
 625         {
 626                 if (de->inode)
 627                 {
 628                         if (show_names)
 629                         {
 630 #ifdef CONFIG_FS_ENCRYPTION
 631                                 int len;
 632                                 char *name;
 633                                 struct fscrypt_str fname_crypto_str =
 634                                         FSTR_INIT(NULL, 0);
 635                                 int res = 0;
 636
 637                                 name  = de->name;
 638                                 len = de->name_len;
 639                                 if (IS_ENCRYPTED(dir))
 640                                         res = fscrypt_get_encryption_info(dir);
 641                                 if (res) {
 642                                         printk(KERN_WARNING "Error setting up"
 643                                                " fname crypto: %d\n", res);
 644                                 }
 645                                 if (!fscrypt_has_encryption_key(dir)) {
 646                                         /* Directory is not encrypted */
 647                                         ext4fs_dirhash(dir, de->name,
 648                                                 de->name_len, &h);
 649                                         printk("%*.s:(U)%x.%u ", len,
 650                                                name, h.hash,
 651                                                (unsigned) ((char *) de
 652                                                            - base));
 653                                 } else {
 654                                         struct fscrypt_str de_name =
 655                                                 FSTR_INIT(name, len);
 656
 657                                         /* Directory is encrypted */
 658                                         res = fscrypt_fname_alloc_buffer(
 659                                                 dir, len,
 660                                                 &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                                         ext4fs_dirhash(dir, de->name,
 681                                                        de->name_len, &h);
 682                                         printk("%*.s:(E)%x.%u ", len, name,
 683                                                h.hash, (unsigned) ((char *) de
 684                                                                    - base));
 685                                         fscrypt_fname_free_buffer(
 686                                                         &fname_crypto_str);
 687                                 }
 688 #else
 689                                 int len = de->name_len;
 690                                 char *name = de->name;
 691                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 692                                 printk("%*.s:%x.%u ", len, name, h.hash,
 693                                        (unsigned) ((char *) de - base));
 694 #endif
 695                         }
 696                         space += EXT4_DIR_REC_LEN(de->name_len);
 697                         names++;
 698                 }
 699                 de = ext4_next_entry(de, size);
 700         }
 701         printk(KERN_CONT "(%i)\n", names);
 702         return (struct stats) { names, space, 1 };
 703 }
 704
 705 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 706                              struct dx_entry *entries, int levels)
 707 {
 708         unsigned blocksize = dir->i_sb->s_blocksize;
 709         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 710         unsigned bcount = 0;
 711         struct buffer_head *bh;
 712         printk("%i indexed blocks...\n", count);
 713         for (i = 0; i < count; i++, entries++)
 714         {
 715                 ext4_lblk_t block = dx_get_block(entries);
 716                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 717                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 718                 struct stats stats;
 719                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 720                 bh = ext4_bread(NULL,dir, block, 0);
 721                 if (!bh || IS_ERR(bh))
 722                         continue;
 723                 stats = levels?
 724                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 725                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 726                         bh->b_data, blocksize, 0);
 727                 names += stats.names;
 728                 space += stats.space;
 729                 bcount += stats.bcount;
 730                 brelse(bh);
 731         }
 732         if (bcount)
 733                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 734                        levels ? "" : "   ", names, space/bcount,
 735                        (space/bcount)*100/blocksize);
 736         return (struct stats) { names, space, bcount};
 737 }
 738 #endif /* DX_DEBUG */
 739
 740 /*
 741  * Probe for a directory leaf block to search.
 742  *
 743  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 744  * error in the directory index, and the caller should fall back to
 745  * searching the directory normally.  The callers of dx_probe **MUST**
 746  * check for this error code, and make sure it never gets reflected
 747  * back to userspace.
 748  */
 749 static struct dx_frame *
 750 dx_probe(struct ext4_filename *fname, struct inode *dir,
 751          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 752 {
 753         unsigned count, indirect;
 754         struct dx_entry *at, *entries, *p, *q, *m;
 755         struct dx_root *root;
 756         struct dx_frame *frame = frame_in;
 757         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 758         u32 hash;
 759
 760         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 761         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 762         if (IS_ERR(frame->bh))
 763                 return (struct dx_frame *) frame->bh;
 764
 765         root = (struct dx_root *) frame->bh->b_data;
 766         if (root->info.hash_version != DX_HASH_TEA &&
 767             root->info.hash_version != DX_HASH_HALF_MD4 &&
 768             root->info.hash_version != DX_HASH_LEGACY) {
 769                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 770                                    root->info.hash_version);
 771                 goto fail;
 772         }
 773         if (fname)
 774                 hinfo = &fname->hinfo;
 775         hinfo->hash_version = root->info.hash_version;
 776         if (hinfo->hash_version <= DX_HASH_TEA)
 777                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 778         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 779         if (fname && fname_name(fname))
 780                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 781         hash = hinfo->hash;
 782
 783         if (root->info.unused_flags & 1) {
 784                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 785                                    root->info.unused_flags);
 786                 goto fail;
 787         }
 788
 789         indirect = root->info.indirect_levels;
 790         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 791                 ext4_warning(dir->i_sb,
 792                              "Directory (ino: %lu) htree depth %#06x exceed"
 793                              "supported value", dir->i_ino,
 794                              ext4_dir_htree_level(dir->i_sb));
 795                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 796                         ext4_warning(dir->i_sb, "Enable large directory "
 797                                                 "feature to access it");
 798                 }
 799                 goto fail;
 800         }
 801
 802         entries = (struct dx_entry *)(((char *)&root->info) +
 803                                       root->info.info_length);
 804
 805         if (dx_get_limit(entries) != dx_root_limit(dir,
 806                                                    root->info.info_length)) {
 807                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 808                                    dx_get_limit(entries),
 809                                    dx_root_limit(dir, root->info.info_length));
 810                 goto fail;
 811         }
 812
 813         dxtrace(printk("Look up %x", hash));
 814         while (1) {
 815                 count = dx_get_count(entries);
 816                 if (!count || count > dx_get_limit(entries)) {
 817                         ext4_warning_inode(dir,
 818                                            "dx entry: count %u beyond limit %u",
 819                                            count, dx_get_limit(entries));
 820                         goto fail;
 821                 }
 822
 823                 p = entries + 1;
 824                 q = entries + count - 1;
 825                 while (p <= q) {
 826                         m = p + (q - p) / 2;
 827                         dxtrace(printk(KERN_CONT "."));
 828                         if (dx_get_hash(m) > hash)
 829                                 q = m - 1;
 830                         else
 831                                 p = m + 1;
 832                 }
 833
 834                 if (0) { // linear search cross check
 835                         unsigned n = count - 1;
 836                         at = entries;
 837                         while (n--)
 838                         {
 839                                 dxtrace(printk(KERN_CONT ","));
 840                                 if (dx_get_hash(++at) > hash)
 841                                 {
 842                                         at--;
 843                                         break;
 844                                 }
 845                         }
 846                         assert (at == p - 1);
 847                 }
 848
 849                 at = p - 1;
 850                 dxtrace(printk(KERN_CONT " %x->%u\n",
 851                                at == entries ? 0 : dx_get_hash(at),
 852                                dx_get_block(at)));
 853                 frame->entries = entries;
 854                 frame->at = at;
 855                 if (!indirect--)
 856                         return frame;
 857                 frame++;
 858                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 859                 if (IS_ERR(frame->bh)) {
 860                         ret_err = (struct dx_frame *) frame->bh;
 861                         frame->bh = NULL;
 862                         goto fail;
 863                 }
 864                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 865
 866                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 867                         ext4_warning_inode(dir,
 868                                 "dx entry: limit %u != node limit %u",
 869                                 dx_get_limit(entries), dx_node_limit(dir));
 870                         goto fail;
 871                 }
 872         }
 873 fail:
 874         while (frame >= frame_in) {
 875                 brelse(frame->bh);
 876                 frame--;
 877         }
 878
 879         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 880                 ext4_warning_inode(dir,
 881                         "Corrupt directory, running e2fsck is recommended");
 882         return ret_err;
 883 }
 884
 885 static void dx_release(struct dx_frame *frames)
 886 {
 887         struct dx_root_info *info;
 888         int i;
 889         unsigned int indirect_levels;
 890
 891         if (frames[0].bh == NULL)
 892                 return;
 893
 894         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 895         /* save local copy, "info" may be freed after brelse() */
 896         indirect_levels = info->indirect_levels;
 897         for (i = 0; i <= indirect_levels; i++) {
 898                 if (frames[i].bh == NULL)
 899                         break;
 900                 brelse(frames[i].bh);
 901                 frames[i].bh = NULL;
 902         }
 903 }
 904
 905 /*
 906  * This function increments the frame pointer to search the next leaf
 907  * block, and reads in the necessary intervening nodes if the search
 908  * should be necessary.  Whether or not the search is necessary is
 909  * controlled by the hash parameter.  If the hash value is even, then
 910  * the search is only continued if the next block starts with that
 911  * hash value.  This is used if we are searching for a specific file.
 912  *
 913  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 914  *
 915  * This function returns 1 if the caller should continue to search,
 916  * or 0 if it should not.  If there is an error reading one of the
 917  * index blocks, it will a negative error code.
 918  *
 919  * If start_hash is non-null, it will be filled in with the starting
 920  * hash of the next page.
 921  */
 922 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 923                                  struct dx_frame *frame,
 924                                  struct dx_frame *frames,
 925                                  __u32 *start_hash)
 926 {
 927         struct dx_frame *p;
 928         struct buffer_head *bh;
 929         int num_frames = 0;
 930         __u32 bhash;
 931
 932         p = frame;
 933         /*
 934          * Find the next leaf page by incrementing the frame pointer.
 935          * If we run out of entries in the interior node, loop around and
 936          * increment pointer in the parent node.  When we break out of
 937          * this loop, num_frames indicates the number of interior
 938          * nodes need to be read.
 939          */
 940         while (1) {
 941                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 942                         break;
 943                 if (p == frames)
 944                         return 0;
 945                 num_frames++;
 946                 p--;
 947         }
 948
 949         /*
 950          * If the hash is 1, then continue only if the next page has a
 951          * continuation hash of any value.  This is used for readdir
 952          * handling.  Otherwise, check to see if the hash matches the
 953          * desired contiuation hash.  If it doesn't, return since
 954          * there's no point to read in the successive index pages.
 955          */
 956         bhash = dx_get_hash(p->at);
 957         if (start_hash)
 958                 *start_hash = bhash;
 959         if ((hash & 1) == 0) {
 960                 if ((bhash & ~1) != hash)
 961                         return 0;
 962         }
 963         /*
 964          * If the hash is HASH_NB_ALWAYS, we always go to the next
 965          * block so no check is necessary
 966          */
 967         while (num_frames--) {
 968                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 969                 if (IS_ERR(bh))
 970                         return PTR_ERR(bh);
 971                 p++;
 972                 brelse(p->bh);
 973                 p->bh = bh;
 974                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 975         }
 976         return 1;
 977 }
 978
 979
 980 /*
 981  * This function fills a red-black tree with information from a
 982  * directory block.  It returns the number directory entries loaded
 983  * into the tree.  If there is an error it is returned in err.
 984  */
 985 static int htree_dirblock_to_tree(struct file *dir_file,
 986                                   struct inode *dir, ext4_lblk_t block,
 987                                   struct dx_hash_info *hinfo,
 988                                   __u32 start_hash, __u32 start_minor_hash)
 989 {
 990         struct buffer_head *bh;
 991         struct ext4_dir_entry_2 *de, *top;
 992         int err = 0, count = 0;
 993         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 994
 995         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 996                                                         (unsigned long)block));
 997         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
 998         if (IS_ERR(bh))
 999                 return PTR_ERR(bh);
1000
1001         de = (struct ext4_dir_entry_2 *) bh->b_data;
1002         top = (struct ext4_dir_entry_2 *) ((char *) de +
1003                                            dir->i_sb->s_blocksize -
1004                                            EXT4_DIR_REC_LEN(0));
1005 #ifdef CONFIG_FS_ENCRYPTION
1006         /* Check if the directory is encrypted */
1007         if (IS_ENCRYPTED(dir)) {
1008                 err = fscrypt_get_encryption_info(dir);
1009                 if (err < 0) {
1010                         brelse(bh);
1011                         return err;
1012                 }
1013                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1014                                                      &fname_crypto_str);
1015                 if (err < 0) {
1016                         brelse(bh);
1017                         return err;
1018                 }
1019         }
1020 #endif
1021         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1022                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1023                                 bh->b_data, bh->b_size,
1024                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1025                                          + ((char *)de - bh->b_data))) {
1026                         /* silently ignore the rest of the block */
1027                         break;
1028                 }
1029                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1030                 if ((hinfo->hash < start_hash) ||
1031                     ((hinfo->hash == start_hash) &&
1032                      (hinfo->minor_hash < start_minor_hash)))
1033                         continue;
1034                 if (de->inode == 0)
1035                         continue;
1036                 if (!IS_ENCRYPTED(dir)) {
1037                         tmp_str.name = de->name;
1038                         tmp_str.len = de->name_len;
1039                         err = ext4_htree_store_dirent(dir_file,
1040                                    hinfo->hash, hinfo->minor_hash, de,
1041                                    &tmp_str);
1042                 } else {
1043                         int save_len = fname_crypto_str.len;
1044                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1045                                                                 de->name_len);
1046
1047                         /* Directory is encrypted */
1048                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1049                                         hinfo->minor_hash, &de_name,
1050                                         &fname_crypto_str);
1051                         if (err) {
1052                                 count = err;
1053                                 goto errout;
1054                         }
1055                         err = ext4_htree_store_dirent(dir_file,
1056                                    hinfo->hash, hinfo->minor_hash, de,
1057                                         &fname_crypto_str);
1058                         fname_crypto_str.len = save_len;
1059                 }
1060                 if (err != 0) {
1061                         count = err;
1062                         goto errout;
1063                 }
1064                 count++;
1065         }
1066 errout:
1067         brelse(bh);
1068 #ifdef CONFIG_FS_ENCRYPTION
1069         fscrypt_fname_free_buffer(&fname_crypto_str);
1070 #endif
1071         return count;
1072 }
1073
1074
1075 /*
1076  * This function fills a red-black tree with information from a
1077  * directory.  We start scanning the directory in hash order, starting
1078  * at start_hash and start_minor_hash.
1079  *
1080  * This function returns the number of entries inserted into the tree,
1081  * or a negative error code.
1082  */
1083 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1084                          __u32 start_minor_hash, __u32 *next_hash)
1085 {
1086         struct dx_hash_info hinfo;
1087         struct ext4_dir_entry_2 *de;
1088         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1089         struct inode *dir;
1090         ext4_lblk_t block;
1091         int count = 0;
1092         int ret, err;
1093         __u32 hashval;
1094         struct fscrypt_str tmp_str;
1095
1096         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1097                        start_hash, start_minor_hash));
1098         dir = file_inode(dir_file);
1099         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1100                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1101                 if (hinfo.hash_version <= DX_HASH_TEA)
1102                         hinfo.hash_version +=
1103                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1104                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1105                 if (ext4_has_inline_data(dir)) {
1106                         int has_inline_data = 1;
1107                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1108                                                        &hinfo, start_hash,
1109                                                        start_minor_hash,
1110                                                        &has_inline_data);
1111                         if (has_inline_data) {
1112                                 *next_hash = ~0;
1113                                 return count;
1114                         }
1115                 }
1116                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1117                                                start_hash, start_minor_hash);
1118                 *next_hash = ~0;
1119                 return count;
1120         }
1121         hinfo.hash = start_hash;
1122         hinfo.minor_hash = 0;
1123         frame = dx_probe(NULL, dir, &hinfo, frames);
1124         if (IS_ERR(frame))
1125                 return PTR_ERR(frame);
1126
1127         /* Add '.' and '..' from the htree header */
1128         if (!start_hash && !start_minor_hash) {
1129                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1130                 tmp_str.name = de->name;
1131                 tmp_str.len = de->name_len;
1132                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1133                                               de, &tmp_str);
1134                 if (err != 0)
1135                         goto errout;
1136                 count++;
1137         }
1138         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1139                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1140                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1141                 tmp_str.name = de->name;
1142                 tmp_str.len = de->name_len;
1143                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1144                                               de, &tmp_str);
1145                 if (err != 0)
1146                         goto errout;
1147                 count++;
1148         }
1149
1150         while (1) {
1151                 if (fatal_signal_pending(current)) {
1152                         err = -ERESTARTSYS;
1153                         goto errout;
1154                 }
1155                 cond_resched();
1156                 block = dx_get_block(frame->at);
1157                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1158                                              start_hash, start_minor_hash);
1159                 if (ret < 0) {
1160                         err = ret;
1161                         goto errout;
1162                 }
1163                 count += ret;
1164                 hashval = ~0;
1165                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1166                                             frame, frames, &hashval);
1167                 *next_hash = hashval;
1168                 if (ret < 0) {
1169                         err = ret;
1170                         goto errout;
1171                 }
1172                 /*
1173                  * Stop if:  (a) there are no more entries, or
1174                  * (b) we have inserted at least one entry and the
1175                  * next hash value is not a continuation
1176                  */
1177                 if ((ret == 0) ||
1178                     (count && ((hashval & 1) == 0)))
1179                         break;
1180         }
1181         dx_release(frames);
1182         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1183                        "next hash: %x\n", count, *next_hash));
1184         return count;
1185 errout:
1186         dx_release(frames);
1187         return (err);
1188 }
1189
1190 static inline int search_dirblock(struct buffer_head *bh,
1191                                   struct inode *dir,
1192                                   struct ext4_filename *fname,
1193                                   unsigned int offset,
1194                                   struct ext4_dir_entry_2 **res_dir)
1195 {
1196         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1197                                fname, offset, res_dir);
1198 }
1199
1200 /*
1201  * Directory block splitting, compacting
1202  */
1203
1204 /*
1205  * Create map of hash values, offsets, and sizes, stored at end of block.
1206  * Returns number of entries mapped.
1207  */
1208 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1209                        unsigned blocksize, struct dx_hash_info *hinfo,
1210                        struct dx_map_entry *map_tail)
1211 {
1212         int count = 0;
1213         char *base = (char *) de;
1214         struct dx_hash_info h = *hinfo;
1215
1216         while ((char *) de < base + blocksize) {
1217                 if (de->name_len && de->inode) {
1218                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1219                         map_tail--;
1220                         map_tail->hash = h.hash;
1221                         map_tail->offs = ((char *) de - base)>>2;
1222                         map_tail->size = le16_to_cpu(de->rec_len);
1223                         count++;
1224                         cond_resched();
1225                 }
1226                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1227                 de = ext4_next_entry(de, blocksize);
1228         }
1229         return count;
1230 }
1231
1232 /* Sort map by hash value */
1233 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1234 {
1235         struct dx_map_entry *p, *q, *top = map + count - 1;
1236         int more;
1237         /* Combsort until bubble sort doesn't suck */
1238         while (count > 2) {
1239                 count = count*10/13;
1240                 if (count - 9 < 2) /* 9, 10 -> 11 */
1241                         count = 11;
1242                 for (p = top, q = p - count; q >= map; p--, q--)
1243                         if (p->hash < q->hash)
1244                                 swap(*p, *q);
1245         }
1246         /* Garden variety bubble sort */
1247         do {
1248                 more = 0;
1249                 q = top;
1250                 while (q-- > map) {
1251                         if (q[1].hash >= q[0].hash)
1252                                 continue;
1253                         swap(*(q+1), *q);
1254                         more = 1;
1255                 }
1256         } while(more);
1257 }
1258
1259 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1260 {
1261         struct dx_entry *entries = frame->entries;
1262         struct dx_entry *old = frame->at, *new = old + 1;
1263         int count = dx_get_count(entries);
1264
1265         assert(count < dx_get_limit(entries));
1266         assert(old < entries + count);
1267         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1268         dx_set_hash(new, hash);
1269         dx_set_block(new, block);
1270         dx_set_count(entries, count + 1);
1271 }
1272
1273 #ifdef CONFIG_UNICODE
1274 /*
1275  * Test whether a case-insensitive directory entry matches the filename
1276  * being searched for.  If quick is set, assume the name being looked up
1277  * is already in the casefolded form.
1278  *
1279  * Returns: 0 if the directory entry matches, more than 0 if it
1280  * doesn't match or less than zero on error.
1281  */
1282 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1283                     const struct qstr *entry, bool quick)
1284 {
1285         const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1286         const struct unicode_map *um = sbi->s_encoding;
1287         int ret;
1288
1289         if (quick)
1290                 ret = utf8_strncasecmp_folded(um, name, entry);
1291         else
1292                 ret = utf8_strncasecmp(um, name, entry);
1293
1294         if (ret < 0) {
1295                 /* Handle invalid character sequence as either an error
1296                  * or as an opaque byte sequence.
1297                  */
1298                 if (ext4_has_strict_mode(sbi))
1299                         return -EINVAL;
1300
1301                 if (name->len != entry->len)
1302                         return 1;
1303
1304                 return !!memcmp(name->name, entry->name, name->len);
1305         }
1306
1307         return ret;
1308 }
1309
1310 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1311                                   struct fscrypt_str *cf_name)
1312 {
1313         int len;
1314
1315         if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1316                 cf_name->name = NULL;
1317                 return;
1318         }
1319
1320         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1321         if (!cf_name->name)
1322                 return;
1323
1324         len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1325                             iname, cf_name->name,
1326                             EXT4_NAME_LEN);
1327         if (len <= 0) {
1328                 kfree(cf_name->name);
1329                 cf_name->name = NULL;
1330                 return;
1331         }
1332         cf_name->len = (unsigned) len;
1333
1334 }
1335 #endif
1336
1337 /*
1338  * Test whether a directory entry matches the filename being searched for.
1339  *
1340  * Return: %true if the directory entry matches, otherwise %false.
1341  */
1342 static inline bool ext4_match(const struct inode *parent,
1343                               const struct ext4_filename *fname,
1344                               const struct ext4_dir_entry_2 *de)
1345 {
1346         struct fscrypt_name f;
1347 #ifdef CONFIG_UNICODE
1348         const struct qstr entry = {.name = de->name, .len = de->name_len};
1349 #endif
1350
1351         if (!de->inode)
1352                 return false;
1353
1354         f.usr_fname = fname->usr_fname;
1355         f.disk_name = fname->disk_name;
1356 #ifdef CONFIG_FS_ENCRYPTION
1357         f.crypto_buf = fname->crypto_buf;
1358 #endif
1359
1360 #ifdef CONFIG_UNICODE
1361         if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1362                 if (fname->cf_name.name) {
1363                         struct qstr cf = {.name = fname->cf_name.name,
1364                                           .len = fname->cf_name.len};
1365                         return !ext4_ci_compare(parent, &cf, &entry, true);
1366                 }
1367                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1368                                         false);
1369         }
1370 #endif
1371
1372         return fscrypt_match_name(&f, de->name, de->name_len);
1373 }
1374
1375 /*
1376  * Returns 0 if not found, -1 on failure, and 1 on success
1377  */
1378 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1379                     struct inode *dir, struct ext4_filename *fname,
1380                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1381 {
1382         struct ext4_dir_entry_2 * de;
1383         char * dlimit;
1384         int de_len;
1385
1386         de = (struct ext4_dir_entry_2 *)search_buf;
1387         dlimit = search_buf + buf_size;
1388         while ((char *) de < dlimit) {
1389                 /* this code is executed quadratically often */
1390                 /* do minimal checking `by hand' */
1391                 if ((char *) de + de->name_len <= dlimit &&
1392                     ext4_match(dir, fname, de)) {
1393                         /* found a match - just to be sure, do
1394                          * a full check */
1395                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1396                                                  bh->b_size, offset))
1397                                 return -1;
1398                         *res_dir = de;
1399                         return 1;
1400                 }
1401                 /* prevent looping on a bad block */
1402                 de_len = ext4_rec_len_from_disk(de->rec_len,
1403                                                 dir->i_sb->s_blocksize);
1404                 if (de_len <= 0)
1405                         return -1;
1406                 offset += de_len;
1407                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1408         }
1409         return 0;
1410 }
1411
1412 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1413                                struct ext4_dir_entry *de)
1414 {
1415         struct super_block *sb = dir->i_sb;
1416
1417         if (!is_dx(dir))
1418                 return 0;
1419         if (block == 0)
1420                 return 1;
1421         if (de->inode == 0 &&
1422             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1423                         sb->s_blocksize)
1424                 return 1;
1425         return 0;
1426 }
1427
1428 /*
1429  *      __ext4_find_entry()
1430  *
1431  * finds an entry in the specified directory with the wanted name. It
1432  * returns the cache buffer in which the entry was found, and the entry
1433  * itself (as a parameter - res_dir). It does NOT read the inode of the
1434  * entry - you'll have to do that yourself if you want to.
1435  *
1436  * The returned buffer_head has ->b_count elevated.  The caller is expected
1437  * to brelse() it when appropriate.
1438  */
1439 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1440                                              struct ext4_filename *fname,
1441                                              struct ext4_dir_entry_2 **res_dir,
1442                                              int *inlined)
1443 {
1444         struct super_block *sb;
1445         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1446         struct buffer_head *bh, *ret = NULL;
1447         ext4_lblk_t start, block;
1448         const u8 *name = fname->usr_fname->name;
1449         size_t ra_max = 0;      /* Number of bh's in the readahead
1450                                    buffer, bh_use[] */
1451         size_t ra_ptr = 0;      /* Current index into readahead
1452                                    buffer */
1453         ext4_lblk_t  nblocks;
1454         int i, namelen, retval;
1455
1456         *res_dir = NULL;
1457         sb = dir->i_sb;
1458         namelen = fname->usr_fname->len;
1459         if (namelen > EXT4_NAME_LEN)
1460                 return NULL;
1461
1462         if (ext4_has_inline_data(dir)) {
1463                 int has_inline_data = 1;
1464                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1465                                              &has_inline_data);
1466                 if (has_inline_data) {
1467                         if (inlined)
1468                                 *inlined = 1;
1469                         goto cleanup_and_exit;
1470                 }
1471         }
1472
1473         if ((namelen <= 2) && (name[0] == '.') &&
1474             (name[1] == '.' || name[1] == '\0')) {
1475                 /*
1476                  * "." or ".." will only be in the first block
1477                  * NFS may look up ".."; "." should be handled by the VFS
1478                  */
1479                 block = start = 0;
1480                 nblocks = 1;
1481                 goto restart;
1482         }
1483         if (is_dx(dir)) {
1484                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1485                 /*
1486                  * On success, or if the error was file not found,
1487                  * return.  Otherwise, fall back to doing a search the
1488                  * old fashioned way.
1489                  */
1490                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1491                         goto cleanup_and_exit;
1492                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1493                                "falling back\n"));
1494                 ret = NULL;
1495         }
1496         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1497         if (!nblocks) {
1498                 ret = NULL;
1499                 goto cleanup_and_exit;
1500         }
1501         start = EXT4_I(dir)->i_dir_start_lookup;
1502         if (start >= nblocks)
1503                 start = 0;
1504         block = start;
1505 restart:
1506         do {
1507                 /*
1508                  * We deal with the read-ahead logic here.
1509                  */
1510                 if (ra_ptr >= ra_max) {
1511                         /* Refill the readahead buffer */
1512                         ra_ptr = 0;
1513                         if (block < start)
1514                                 ra_max = start - block;
1515                         else
1516                                 ra_max = nblocks - block;
1517                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1518                         retval = ext4_bread_batch(dir, block, ra_max,
1519                                                   false /* wait */, bh_use);
1520                         if (retval) {
1521                                 ret = ERR_PTR(retval);
1522                                 ra_max = 0;
1523                                 goto cleanup_and_exit;
1524                         }
1525                 }
1526                 if ((bh = bh_use[ra_ptr++]) == NULL)
1527                         goto next;
1528                 wait_on_buffer(bh);
1529                 if (!buffer_uptodate(bh)) {
1530                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1531                                          (unsigned long) block);
1532                         brelse(bh);
1533                         ret = ERR_PTR(-EIO);
1534                         goto cleanup_and_exit;
1535                 }
1536                 if (!buffer_verified(bh) &&
1537                     !is_dx_internal_node(dir, block,
1538                                          (struct ext4_dir_entry *)bh->b_data) &&
1539                     !ext4_dirblock_csum_verify(dir, bh)) {
1540                         EXT4_ERROR_INODE(dir, "checksumming directory "
1541                                          "block %lu", (unsigned long)block);
1542                         brelse(bh);
1543                         ret = ERR_PTR(-EFSBADCRC);
1544                         goto cleanup_and_exit;
1545                 }
1546                 set_buffer_verified(bh);
1547                 i = search_dirblock(bh, dir, fname,
1548                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1549                 if (i == 1) {
1550                         EXT4_I(dir)->i_dir_start_lookup = block;
1551                         ret = bh;
1552                         goto cleanup_and_exit;
1553                 } else {
1554                         brelse(bh);
1555                         if (i < 0)
1556                                 goto cleanup_and_exit;
1557                 }
1558         next:
1559                 if (++block >= nblocks)
1560                         block = 0;
1561         } while (block != start);
1562
1563         /*
1564          * If the directory has grown while we were searching, then
1565          * search the last part of the directory before giving up.
1566          */
1567         block = nblocks;
1568         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1569         if (block < nblocks) {
1570                 start = 0;
1571                 goto restart;
1572         }
1573
1574 cleanup_and_exit:
1575         /* Clean up the read-ahead blocks */
1576         for (; ra_ptr < ra_max; ra_ptr++)
1577                 brelse(bh_use[ra_ptr]);
1578         return ret;
1579 }
1580
1581 static struct buffer_head *ext4_find_entry(struct inode *dir,
1582                                            const struct qstr *d_name,
1583                                            struct ext4_dir_entry_2 **res_dir,
1584                                            int *inlined)
1585 {
1586         int err;
1587         struct ext4_filename fname;
1588         struct buffer_head *bh;
1589
1590         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1591         if (err == -ENOENT)
1592                 return NULL;
1593         if (err)
1594                 return ERR_PTR(err);
1595
1596         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1597
1598         ext4_fname_free_filename(&fname);
1599         return bh;
1600 }
1601
1602 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1603                                              struct dentry *dentry,
1604                                              struct ext4_dir_entry_2 **res_dir)
1605 {
1606         int err;
1607         struct ext4_filename fname;
1608         struct buffer_head *bh;
1609
1610         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1611         if (err == -ENOENT)
1612                 return NULL;
1613         if (err)
1614                 return ERR_PTR(err);
1615
1616         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1617
1618         ext4_fname_free_filename(&fname);
1619         return bh;
1620 }
1621
1622 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1623                         struct ext4_filename *fname,
1624                         struct ext4_dir_entry_2 **res_dir)
1625 {
1626         struct super_block * sb = dir->i_sb;
1627         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1628         struct buffer_head *bh;
1629         ext4_lblk_t block;
1630         int retval;
1631
1632 #ifdef CONFIG_FS_ENCRYPTION
1633         *res_dir = NULL;
1634 #endif
1635         frame = dx_probe(fname, dir, NULL, frames);
1636         if (IS_ERR(frame))
1637                 return (struct buffer_head *) frame;
1638         do {
1639                 block = dx_get_block(frame->at);
1640                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1641                 if (IS_ERR(bh))
1642                         goto errout;
1643
1644                 retval = search_dirblock(bh, dir, fname,
1645                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1646                                          res_dir);
1647                 if (retval == 1)
1648                         goto success;
1649                 brelse(bh);
1650                 if (retval == -1) {
1651                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1652                         goto errout;
1653                 }
1654
1655                 /* Check to see if we should continue to search */
1656                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1657                                                frames, NULL);
1658                 if (retval < 0) {
1659                         ext4_warning_inode(dir,
1660                                 "error %d reading directory index block",
1661                                 retval);
1662                         bh = ERR_PTR(retval);
1663                         goto errout;
1664                 }
1665         } while (retval == 1);
1666
1667         bh = NULL;
1668 errout:
1669         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1670 success:
1671         dx_release(frames);
1672         return bh;
1673 }
1674
1675 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1676 {
1677         struct inode *inode;
1678         struct ext4_dir_entry_2 *de;
1679         struct buffer_head *bh;
1680
1681         if (dentry->d_name.len > EXT4_NAME_LEN)
1682                 return ERR_PTR(-ENAMETOOLONG);
1683
1684         bh = ext4_lookup_entry(dir, dentry, &de);
1685         if (IS_ERR(bh))
1686                 return ERR_CAST(bh);
1687         inode = NULL;
1688         if (bh) {
1689                 __u32 ino = le32_to_cpu(de->inode);
1690                 brelse(bh);
1691                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1692                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1693                         return ERR_PTR(-EFSCORRUPTED);
1694                 }
1695                 if (unlikely(ino == dir->i_ino)) {
1696                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1697                                          dentry);
1698                         return ERR_PTR(-EFSCORRUPTED);
1699                 }
1700                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1701                 if (inode == ERR_PTR(-ESTALE)) {
1702                         EXT4_ERROR_INODE(dir,
1703                                          "deleted inode referenced: %u",
1704                                          ino);
1705                         return ERR_PTR(-EFSCORRUPTED);
1706                 }
1707                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1708                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1709                     !fscrypt_has_permitted_context(dir, inode)) {
1710                         ext4_warning(inode->i_sb,
1711                                      "Inconsistent encryption contexts: %lu/%lu",
1712                                      dir->i_ino, inode->i_ino);
1713                         iput(inode);
1714                         return ERR_PTR(-EPERM);
1715                 }
1716         }
1717
1718 #ifdef CONFIG_UNICODE
1719         if (!inode && IS_CASEFOLDED(dir)) {
1720                 /* Eventually we want to call d_add_ci(dentry, NULL)
1721                  * for negative dentries in the encoding case as
1722                  * well.  For now, prevent the negative dentry
1723                  * from being cached.
1724                  */
1725                 return NULL;
1726         }
1727 #endif
1728         return d_splice_alias(inode, dentry);
1729 }
1730
1731
1732 struct dentry *ext4_get_parent(struct dentry *child)
1733 {
1734         __u32 ino;
1735         static const struct qstr dotdot = QSTR_INIT("..", 2);
1736         struct ext4_dir_entry_2 * de;
1737         struct buffer_head *bh;
1738
1739         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1740         if (IS_ERR(bh))
1741                 return ERR_CAST(bh);
1742         if (!bh)
1743                 return ERR_PTR(-ENOENT);
1744         ino = le32_to_cpu(de->inode);
1745         brelse(bh);
1746
1747         if (!ext4_valid_inum(child->d_sb, ino)) {
1748                 EXT4_ERROR_INODE(d_inode(child),
1749                                  "bad parent inode number: %u", ino);
1750                 return ERR_PTR(-EFSCORRUPTED);
1751         }
1752
1753         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1754 }
1755
1756 /*
1757  * Move count entries from end of map between two memory locations.
1758  * Returns pointer to last entry moved.
1759  */
1760 static struct ext4_dir_entry_2 *
1761 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1762                 unsigned blocksize)
1763 {
1764         unsigned rec_len = 0;
1765
1766         while (count--) {
1767                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1768                                                 (from + (map->offs<<2));
1769                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1770                 memcpy (to, de, rec_len);
1771                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1772                                 ext4_rec_len_to_disk(rec_len, blocksize);
1773                 de->inode = 0;
1774                 map++;
1775                 to += rec_len;
1776         }
1777         return (struct ext4_dir_entry_2 *) (to - rec_len);
1778 }
1779
1780 /*
1781  * Compact each dir entry in the range to the minimal rec_len.
1782  * Returns pointer to last entry in range.
1783  */
1784 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1785 {
1786         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1787         unsigned rec_len = 0;
1788
1789         prev = to = de;
1790         while ((char*)de < base + blocksize) {
1791                 next = ext4_next_entry(de, blocksize);
1792                 if (de->inode && de->name_len) {
1793                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1794                         if (de > to)
1795                                 memmove(to, de, rec_len);
1796                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1797                         prev = to;
1798                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1799                 }
1800                 de = next;
1801         }
1802         return prev;
1803 }
1804
1805 /*
1806  * Split a full leaf block to make room for a new dir entry.
1807  * Allocate a new block, and move entries so that they are approx. equally full.
1808  * Returns pointer to de in block into which the new entry will be inserted.
1809  */
1810 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1811                         struct buffer_head **bh,struct dx_frame *frame,
1812                         struct dx_hash_info *hinfo)
1813 {
1814         unsigned blocksize = dir->i_sb->s_blocksize;
1815         unsigned count, continued;
1816         struct buffer_head *bh2;
1817         ext4_lblk_t newblock;
1818         u32 hash2;
1819         struct dx_map_entry *map;
1820         char *data1 = (*bh)->b_data, *data2;
1821         unsigned split, move, size;
1822         struct ext4_dir_entry_2 *de = NULL, *de2;
1823         int     csum_size = 0;
1824         int     err = 0, i;
1825
1826         if (ext4_has_metadata_csum(dir->i_sb))
1827                 csum_size = sizeof(struct ext4_dir_entry_tail);
1828
1829         bh2 = ext4_append(handle, dir, &newblock);
1830         if (IS_ERR(bh2)) {
1831                 brelse(*bh);
1832                 *bh = NULL;
1833                 return (struct ext4_dir_entry_2 *) bh2;
1834         }
1835
1836         BUFFER_TRACE(*bh, "get_write_access");
1837         err = ext4_journal_get_write_access(handle, *bh);
1838         if (err)
1839                 goto journal_error;
1840
1841         BUFFER_TRACE(frame->bh, "get_write_access");
1842         err = ext4_journal_get_write_access(handle, frame->bh);
1843         if (err)
1844                 goto journal_error;
1845
1846         data2 = bh2->b_data;
1847
1848         /* create map in the end of data2 block */
1849         map = (struct dx_map_entry *) (data2 + blocksize);
1850         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1851                              blocksize, hinfo, map);
1852         map -= count;
1853         dx_sort_map(map, count);
1854         /* Split the existing block in the middle, size-wise */
1855         size = 0;
1856         move = 0;
1857         for (i = count-1; i >= 0; i--) {
1858                 /* is more than half of this entry in 2nd half of the block? */
1859                 if (size + map[i].size/2 > blocksize/2)
1860                         break;
1861                 size += map[i].size;
1862                 move++;
1863         }
1864         /* map index at which we will split */
1865         split = count - move;
1866         hash2 = map[split].hash;
1867         continued = hash2 == map[split - 1].hash;
1868         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1869                         (unsigned long)dx_get_block(frame->at),
1870                                         hash2, split, count-split));
1871
1872         /* Fancy dance to stay within two buffers */
1873         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1874                               blocksize);
1875         de = dx_pack_dirents(data1, blocksize);
1876         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1877                                            (char *) de,
1878                                            blocksize);
1879         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1880                                             (char *) de2,
1881                                             blocksize);
1882         if (csum_size) {
1883                 ext4_initialize_dirent_tail(*bh, blocksize);
1884                 ext4_initialize_dirent_tail(bh2, blocksize);
1885         }
1886
1887         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1888                         blocksize, 1));
1889         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1890                         blocksize, 1));
1891
1892         /* Which block gets the new entry? */
1893         if (hinfo->hash >= hash2) {
1894                 swap(*bh, bh2);
1895                 de = de2;
1896         }
1897         dx_insert_block(frame, hash2 + continued, newblock);
1898         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1899         if (err)
1900                 goto journal_error;
1901         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1902         if (err)
1903                 goto journal_error;
1904         brelse(bh2);
1905         dxtrace(dx_show_index("frame", frame->entries));
1906         return de;
1907
1908 journal_error:
1909         brelse(*bh);
1910         brelse(bh2);
1911         *bh = NULL;
1912         ext4_std_error(dir->i_sb, err);
1913         return ERR_PTR(err);
1914 }
1915
1916 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1917                       struct buffer_head *bh,
1918                       void *buf, int buf_size,
1919                       struct ext4_filename *fname,
1920                       struct ext4_dir_entry_2 **dest_de)
1921 {
1922         struct ext4_dir_entry_2 *de;
1923         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1924         int nlen, rlen;
1925         unsigned int offset = 0;
1926         char *top;
1927
1928         de = (struct ext4_dir_entry_2 *)buf;
1929         top = buf + buf_size - reclen;
1930         while ((char *) de <= top) {
1931                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1932                                          buf, buf_size, offset))
1933                         return -EFSCORRUPTED;
1934                 if (ext4_match(dir, fname, de))
1935                         return -EEXIST;
1936                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1937                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1938                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1939                         break;
1940                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1941                 offset += rlen;
1942         }
1943         if ((char *) de > top)
1944                 return -ENOSPC;
1945
1946         *dest_de = de;
1947         return 0;
1948 }
1949
1950 void ext4_insert_dentry(struct inode *inode,
1951                         struct ext4_dir_entry_2 *de,
1952                         int buf_size,
1953                         struct ext4_filename *fname)
1954 {
1955
1956         int nlen, rlen;
1957
1958         nlen = EXT4_DIR_REC_LEN(de->name_len);
1959         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1960         if (de->inode) {
1961                 struct ext4_dir_entry_2 *de1 =
1962                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1963                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1964                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1965                 de = de1;
1966         }
1967         de->file_type = EXT4_FT_UNKNOWN;
1968         de->inode = cpu_to_le32(inode->i_ino);
1969         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1970         de->name_len = fname_len(fname);
1971         memcpy(de->name, fname_name(fname), fname_len(fname));
1972 }
1973
1974 /*
1975  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1976  * it points to a directory entry which is guaranteed to be large
1977  * enough for new directory entry.  If de is NULL, then
1978  * add_dirent_to_buf will attempt search the directory block for
1979  * space.  It will return -ENOSPC if no space is available, and -EIO
1980  * and -EEXIST if directory entry already exists.
1981  */
1982 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1983                              struct inode *dir,
1984                              struct inode *inode, struct ext4_dir_entry_2 *de,
1985                              struct buffer_head *bh)
1986 {
1987         unsigned int    blocksize = dir->i_sb->s_blocksize;
1988         int             csum_size = 0;
1989         int             err;
1990
1991         if (ext4_has_metadata_csum(inode->i_sb))
1992                 csum_size = sizeof(struct ext4_dir_entry_tail);
1993
1994         if (!de) {
1995                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1996                                         blocksize - csum_size, fname, &de);
1997                 if (err)
1998                         return err;
1999         }
2000         BUFFER_TRACE(bh, "get_write_access");
2001         err = ext4_journal_get_write_access(handle, bh);
2002         if (err) {
2003                 ext4_std_error(dir->i_sb, err);
2004                 return err;
2005         }
2006
2007         /* By now the buffer is marked for journaling */
2008         ext4_insert_dentry(inode, de, blocksize, fname);
2009
2010         /*
2011          * XXX shouldn't update any times until successful
2012          * completion of syscall, but too many callers depend
2013          * on this.
2014          *
2015          * XXX similarly, too many callers depend on
2016          * ext4_new_inode() setting the times, but error
2017          * recovery deletes the inode, so the worst that can
2018          * happen is that the times are slightly out of date
2019          * and/or different from the directory change time.
2020          */
2021         dir->i_mtime = dir->i_ctime = current_time(dir);
2022         ext4_update_dx_flag(dir);
2023         inode_inc_iversion(dir);
2024         ext4_mark_inode_dirty(handle, dir);
2025         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2026         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2027         if (err)
2028                 ext4_std_error(dir->i_sb, err);
2029         return 0;
2030 }
2031
2032 /*
2033  * This converts a one block unindexed directory to a 3 block indexed
2034  * directory, and adds the dentry to the indexed directory.
2035  */
2036 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2037                             struct inode *dir,
2038                             struct inode *inode, struct buffer_head *bh)
2039 {
2040         struct buffer_head *bh2;
2041         struct dx_root  *root;
2042         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2043         struct dx_entry *entries;
2044         struct ext4_dir_entry_2 *de, *de2;
2045         char            *data2, *top;
2046         unsigned        len;
2047         int             retval;
2048         unsigned        blocksize;
2049         ext4_lblk_t  block;
2050         struct fake_dirent *fde;
2051         int csum_size = 0;
2052
2053         if (ext4_has_metadata_csum(inode->i_sb))
2054                 csum_size = sizeof(struct ext4_dir_entry_tail);
2055
2056         blocksize =  dir->i_sb->s_blocksize;
2057         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2058         BUFFER_TRACE(bh, "get_write_access");
2059         retval = ext4_journal_get_write_access(handle, bh);
2060         if (retval) {
2061                 ext4_std_error(dir->i_sb, retval);
2062                 brelse(bh);
2063                 return retval;
2064         }
2065         root = (struct dx_root *) bh->b_data;
2066
2067         /* The 0th block becomes the root, move the dirents out */
2068         fde = &root->dotdot;
2069         de = (struct ext4_dir_entry_2 *)((char *)fde +
2070                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2071         if ((char *) de >= (((char *) root) + blocksize)) {
2072                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2073                 brelse(bh);
2074                 return -EFSCORRUPTED;
2075         }
2076         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2077
2078         /* Allocate new block for the 0th block's dirents */
2079         bh2 = ext4_append(handle, dir, &block);
2080         if (IS_ERR(bh2)) {
2081                 brelse(bh);
2082                 return PTR_ERR(bh2);
2083         }
2084         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2085         data2 = bh2->b_data;
2086
2087         memcpy(data2, de, len);
2088         de = (struct ext4_dir_entry_2 *) data2;
2089         top = data2 + len;
2090         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2091                 de = de2;
2092         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2093                                            (char *) de, blocksize);
2094
2095         if (csum_size)
2096                 ext4_initialize_dirent_tail(bh2, blocksize);
2097
2098         /* Initialize the root; the dot dirents already exist */
2099         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2100         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2101                                            blocksize);
2102         memset (&root->info, 0, sizeof(root->info));
2103         root->info.info_length = sizeof(root->info);
2104         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2105         entries = root->entries;
2106         dx_set_block(entries, 1);
2107         dx_set_count(entries, 1);
2108         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2109
2110         /* Initialize as for dx_probe */
2111         fname->hinfo.hash_version = root->info.hash_version;
2112         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2113                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2114         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2115         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2116
2117         memset(frames, 0, sizeof(frames));
2118         frame = frames;
2119         frame->entries = entries;
2120         frame->at = entries;
2121         frame->bh = bh;
2122
2123         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2124         if (retval)
2125                 goto out_frames;
2126         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2127         if (retval)
2128                 goto out_frames;
2129
2130         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2131         if (IS_ERR(de)) {
2132                 retval = PTR_ERR(de);
2133                 goto out_frames;
2134         }
2135
2136         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2137 out_frames:
2138         /*
2139          * Even if the block split failed, we have to properly write
2140          * out all the changes we did so far. Otherwise we can end up
2141          * with corrupted filesystem.
2142          */
2143         if (retval)
2144                 ext4_mark_inode_dirty(handle, dir);
2145         dx_release(frames);
2146         brelse(bh2);
2147         return retval;
2148 }
2149
2150 /*
2151  *      ext4_add_entry()
2152  *
2153  * adds a file entry to the specified directory, using the same
2154  * semantics as ext4_find_entry(). It returns NULL if it failed.
2155  *
2156  * NOTE!! The inode part of 'de' is left at 0 - which means you
2157  * may not sleep between calling this and putting something into
2158  * the entry, as someone else might have used it while you slept.
2159  */
2160 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2161                           struct inode *inode)
2162 {
2163         struct inode *dir = d_inode(dentry->d_parent);
2164         struct buffer_head *bh = NULL;
2165         struct ext4_dir_entry_2 *de;
2166         struct super_block *sb;
2167         struct ext4_sb_info *sbi;
2168         struct ext4_filename fname;
2169         int     retval;
2170         int     dx_fallback=0;
2171         unsigned blocksize;
2172         ext4_lblk_t block, blocks;
2173         int     csum_size = 0;
2174
2175         if (ext4_has_metadata_csum(inode->i_sb))
2176                 csum_size = sizeof(struct ext4_dir_entry_tail);
2177
2178         sb = dir->i_sb;
2179         sbi = EXT4_SB(sb);
2180         blocksize = sb->s_blocksize;
2181         if (!dentry->d_name.len)
2182                 return -EINVAL;
2183
2184 #ifdef CONFIG_UNICODE
2185         if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2186             sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2187                 return -EINVAL;
2188 #endif
2189
2190         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2191         if (retval)
2192                 return retval;
2193
2194         if (ext4_has_inline_data(dir)) {
2195                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2196                 if (retval < 0)
2197                         goto out;
2198                 if (retval == 1) {
2199                         retval = 0;
2200                         goto out;
2201                 }
2202         }
2203
2204         if (is_dx(dir)) {
2205                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2206                 if (!retval || (retval != ERR_BAD_DX_DIR))
2207                         goto out;
2208                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2209                 dx_fallback++;
2210                 ext4_mark_inode_dirty(handle, dir);
2211         }
2212         blocks = dir->i_size >> sb->s_blocksize_bits;
2213         for (block = 0; block < blocks; block++) {
2214                 bh = ext4_read_dirblock(dir, block, DIRENT);
2215                 if (bh == NULL) {
2216                         bh = ext4_bread(handle, dir, block,
2217                                         EXT4_GET_BLOCKS_CREATE);
2218                         goto add_to_new_block;
2219                 }
2220                 if (IS_ERR(bh)) {
2221                         retval = PTR_ERR(bh);
2222                         bh = NULL;
2223                         goto out;
2224                 }
2225                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2226                                            NULL, bh);
2227                 if (retval != -ENOSPC)
2228                         goto out;
2229
2230                 if (blocks == 1 && !dx_fallback &&
2231                     ext4_has_feature_dir_index(sb)) {
2232                         retval = make_indexed_dir(handle, &fname, dir,
2233                                                   inode, bh);
2234                         bh = NULL; /* make_indexed_dir releases bh */
2235                         goto out;
2236                 }
2237                 brelse(bh);
2238         }
2239         bh = ext4_append(handle, dir, &block);
2240 add_to_new_block:
2241         if (IS_ERR(bh)) {
2242                 retval = PTR_ERR(bh);
2243                 bh = NULL;
2244                 goto out;
2245         }
2246         de = (struct ext4_dir_entry_2 *) bh->b_data;
2247         de->inode = 0;
2248         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2249
2250         if (csum_size)
2251                 ext4_initialize_dirent_tail(bh, blocksize);
2252
2253         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2254 out:
2255         ext4_fname_free_filename(&fname);
2256         brelse(bh);
2257         if (retval == 0)
2258                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2259         return retval;
2260 }
2261
2262 /*
2263  * Returns 0 for success, or a negative error value
2264  */
2265 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2266                              struct inode *dir, struct inode *inode)
2267 {
2268         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2269         struct dx_entry *entries, *at;
2270         struct buffer_head *bh;
2271         struct super_block *sb = dir->i_sb;
2272         struct ext4_dir_entry_2 *de;
2273         int restart;
2274         int err;
2275
2276 again:
2277         restart = 0;
2278         frame = dx_probe(fname, dir, NULL, frames);
2279         if (IS_ERR(frame))
2280                 return PTR_ERR(frame);
2281         entries = frame->entries;
2282         at = frame->at;
2283         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2284         if (IS_ERR(bh)) {
2285                 err = PTR_ERR(bh);
2286                 bh = NULL;
2287                 goto cleanup;
2288         }
2289
2290         BUFFER_TRACE(bh, "get_write_access");
2291         err = ext4_journal_get_write_access(handle, bh);
2292         if (err)
2293                 goto journal_error;
2294
2295         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2296         if (err != -ENOSPC)
2297                 goto cleanup;
2298
2299         err = 0;
2300         /* Block full, should compress but for now just split */
2301         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2302                        dx_get_count(entries), dx_get_limit(entries)));
2303         /* Need to split index? */
2304         if (dx_get_count(entries) == dx_get_limit(entries)) {
2305                 ext4_lblk_t newblock;
2306                 int levels = frame - frames + 1;
2307                 unsigned int icount;
2308                 int add_level = 1;
2309                 struct dx_entry *entries2;
2310                 struct dx_node *node2;
2311                 struct buffer_head *bh2;
2312
2313                 while (frame > frames) {
2314                         if (dx_get_count((frame - 1)->entries) <
2315                             dx_get_limit((frame - 1)->entries)) {
2316                                 add_level = 0;
2317                                 break;
2318                         }
2319                         frame--; /* split higher index block */
2320                         at = frame->at;
2321                         entries = frame->entries;
2322                         restart = 1;
2323                 }
2324                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2325                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2326                                          "reach max htree level :%d",
2327                                          dir->i_ino, levels);
2328                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2329                                 ext4_warning(sb, "Large directory feature is "
2330                                                  "not enabled on this "
2331                                                  "filesystem");
2332                         }
2333                         err = -ENOSPC;
2334                         goto cleanup;
2335                 }
2336                 icount = dx_get_count(entries);
2337                 bh2 = ext4_append(handle, dir, &newblock);
2338                 if (IS_ERR(bh2)) {
2339                         err = PTR_ERR(bh2);
2340                         goto cleanup;
2341                 }
2342                 node2 = (struct dx_node *)(bh2->b_data);
2343                 entries2 = node2->entries;
2344                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2345                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2346                                                            sb->s_blocksize);
2347                 BUFFER_TRACE(frame->bh, "get_write_access");
2348                 err = ext4_journal_get_write_access(handle, frame->bh);
2349                 if (err)
2350                         goto journal_error;
2351                 if (!add_level) {
2352                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2353                         unsigned hash2 = dx_get_hash(entries + icount1);
2354                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2355                                        icount1, icount2));
2356
2357                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2358                         err = ext4_journal_get_write_access(handle,
2359                                                              (frame - 1)->bh);
2360                         if (err)
2361                                 goto journal_error;
2362
2363                         memcpy((char *) entries2, (char *) (entries + icount1),
2364                                icount2 * sizeof(struct dx_entry));
2365                         dx_set_count(entries, icount1);
2366                         dx_set_count(entries2, icount2);
2367                         dx_set_limit(entries2, dx_node_limit(dir));
2368
2369                         /* Which index block gets the new entry? */
2370                         if (at - entries >= icount1) {
2371                                 frame->at = at = at - entries - icount1 + entries2;
2372                                 frame->entries = entries = entries2;
2373                                 swap(frame->bh, bh2);
2374                         }
2375                         dx_insert_block((frame - 1), hash2, newblock);
2376                         dxtrace(dx_show_index("node", frame->entries));
2377                         dxtrace(dx_show_index("node",
2378                                ((struct dx_node *) bh2->b_data)->entries));
2379                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2380                         if (err)
2381                                 goto journal_error;
2382                         brelse (bh2);
2383                         err = ext4_handle_dirty_dx_node(handle, dir,
2384                                                    (frame - 1)->bh);
2385                         if (err)
2386                                 goto journal_error;
2387                         if (restart) {
2388                                 err = ext4_handle_dirty_dx_node(handle, dir,
2389                                                            frame->bh);
2390                                 goto journal_error;
2391                         }
2392                 } else {
2393                         struct dx_root *dxroot;
2394                         memcpy((char *) entries2, (char *) entries,
2395                                icount * sizeof(struct dx_entry));
2396                         dx_set_limit(entries2, dx_node_limit(dir));
2397
2398                         /* Set up root */
2399                         dx_set_count(entries, 1);
2400                         dx_set_block(entries + 0, newblock);
2401                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2402                         dxroot->info.indirect_levels += 1;
2403                         dxtrace(printk(KERN_DEBUG
2404                                        "Creating %d level index...\n",
2405                                        dxroot->info.indirect_levels));
2406                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2407                         if (err)
2408                                 goto journal_error;
2409                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2410                         brelse(bh2);
2411                         restart = 1;
2412                         goto journal_error;
2413                 }
2414         }
2415         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2416         if (IS_ERR(de)) {
2417                 err = PTR_ERR(de);
2418                 goto cleanup;
2419         }
2420         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2421         goto cleanup;
2422
2423 journal_error:
2424         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2425 cleanup:
2426         brelse(bh);
2427         dx_release(frames);
2428         /* @restart is true means htree-path has been changed, we need to
2429          * repeat dx_probe() to find out valid htree-path
2430          */
2431         if (restart && err == 0)
2432                 goto again;
2433         return err;
2434 }
2435
2436 /*
2437  * ext4_generic_delete_entry deletes a directory entry by merging it
2438  * with the previous entry
2439  */
2440 int ext4_generic_delete_entry(handle_t *handle,
2441                               struct inode *dir,
2442                               struct ext4_dir_entry_2 *de_del,
2443                               struct buffer_head *bh,
2444                               void *entry_buf,
2445                               int buf_size,
2446                               int csum_size)
2447 {
2448         struct ext4_dir_entry_2 *de, *pde;
2449         unsigned int blocksize = dir->i_sb->s_blocksize;
2450         int i;
2451
2452         i = 0;
2453         pde = NULL;
2454         de = (struct ext4_dir_entry_2 *)entry_buf;
2455         while (i < buf_size - csum_size) {
2456                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2457                                          bh->b_data, bh->b_size, i))
2458                         return -EFSCORRUPTED;
2459                 if (de == de_del)  {
2460                         if (pde)
2461                                 pde->rec_len = ext4_rec_len_to_disk(
2462                                         ext4_rec_len_from_disk(pde->rec_len,
2463                                                                blocksize) +
2464                                         ext4_rec_len_from_disk(de->rec_len,
2465                                                                blocksize),
2466                                         blocksize);
2467                         else
2468                                 de->inode = 0;
2469                         inode_inc_iversion(dir);
2470                         return 0;
2471                 }
2472                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2473                 pde = de;
2474                 de = ext4_next_entry(de, blocksize);
2475         }
2476         return -ENOENT;
2477 }
2478
2479 static int ext4_delete_entry(handle_t *handle,
2480                              struct inode *dir,
2481                              struct ext4_dir_entry_2 *de_del,
2482                              struct buffer_head *bh)
2483 {
2484         int err, csum_size = 0;
2485
2486         if (ext4_has_inline_data(dir)) {
2487                 int has_inline_data = 1;
2488                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2489                                                &has_inline_data);
2490                 if (has_inline_data)
2491                         return err;
2492         }
2493
2494         if (ext4_has_metadata_csum(dir->i_sb))
2495                 csum_size = sizeof(struct ext4_dir_entry_tail);
2496
2497         BUFFER_TRACE(bh, "get_write_access");
2498         err = ext4_journal_get_write_access(handle, bh);
2499         if (unlikely(err))
2500                 goto out;
2501
2502         err = ext4_generic_delete_entry(handle, dir, de_del,
2503                                         bh, bh->b_data,
2504                                         dir->i_sb->s_blocksize, csum_size);
2505         if (err)
2506                 goto out;
2507
2508         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2509         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2510         if (unlikely(err))
2511                 goto out;
2512
2513         return 0;
2514 out:
2515         if (err != -ENOENT)
2516                 ext4_std_error(dir->i_sb, err);
2517         return err;
2518 }
2519
2520 /*
2521  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2522  * since this indicates that nlinks count was previously 1 to avoid overflowing
2523  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2524  * that subdirectory link counts are not being maintained accurately.
2525  *
2526  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2527  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2528  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2529  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2530  */
2531 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2532 {
2533         inc_nlink(inode);
2534         if (is_dx(inode) &&
2535             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2536                 set_nlink(inode, 1);
2537 }
2538
2539 /*
2540  * If a directory had nlink == 1, then we should let it be 1. This indicates
2541  * directory has >EXT4_LINK_MAX subdirs.
2542  */
2543 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2544 {
2545         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2546                 drop_nlink(inode);
2547 }
2548
2549
2550 /*
2551  * Add non-directory inode to a directory. On success, the inode reference is
2552  * consumed by dentry is instantiation. This is also indicated by clearing of
2553  * *inodep pointer. On failure, the caller is responsible for dropping the
2554  * inode reference in the safe context.
2555  */
2556 static int ext4_add_nondir(handle_t *handle,
2557                 struct dentry *dentry, struct inode **inodep)
2558 {
2559         struct inode *dir = d_inode(dentry->d_parent);
2560         struct inode *inode = *inodep;
2561         int err = ext4_add_entry(handle, dentry, inode);
2562         if (!err) {
2563                 ext4_mark_inode_dirty(handle, inode);
2564                 if (IS_DIRSYNC(dir))
2565                         ext4_handle_sync(handle);
2566                 d_instantiate_new(dentry, inode);
2567                 *inodep = NULL;
2568                 return 0;
2569         }
2570         drop_nlink(inode);
2571         ext4_orphan_add(handle, inode);
2572         unlock_new_inode(inode);
2573         return err;
2574 }
2575
2576 /*
2577  * By the time this is called, we already have created
2578  * the directory cache entry for the new file, but it
2579  * is so far negative - it has no inode.
2580  *
2581  * If the create succeeds, we fill in the inode information
2582  * with d_instantiate().
2583  */
2584 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2585                        bool excl)
2586 {
2587         handle_t *handle;
2588         struct inode *inode;
2589         int err, credits, retries = 0;
2590
2591         err = dquot_initialize(dir);
2592         if (err)
2593                 return err;
2594
2595         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2596                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2597 retry:
2598         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2599                                             NULL, EXT4_HT_DIR, credits);
2600         handle = ext4_journal_current_handle();
2601         err = PTR_ERR(inode);
2602         if (!IS_ERR(inode)) {
2603                 inode->i_op = &ext4_file_inode_operations;
2604                 inode->i_fop = &ext4_file_operations;
2605                 ext4_set_aops(inode);
2606                 err = ext4_add_nondir(handle, dentry, &inode);
2607         }
2608         if (handle)
2609                 ext4_journal_stop(handle);
2610         if (!IS_ERR_OR_NULL(inode))
2611                 iput(inode);
2612         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2613                 goto retry;
2614         return err;
2615 }
2616
2617 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2618                       umode_t mode, dev_t rdev)
2619 {
2620         handle_t *handle;
2621         struct inode *inode;
2622         int err, credits, retries = 0;
2623
2624         err = dquot_initialize(dir);
2625         if (err)
2626                 return err;
2627
2628         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2629                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2630 retry:
2631         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2632                                             NULL, EXT4_HT_DIR, credits);
2633         handle = ext4_journal_current_handle();
2634         err = PTR_ERR(inode);
2635         if (!IS_ERR(inode)) {
2636                 init_special_inode(inode, inode->i_mode, rdev);
2637                 inode->i_op = &ext4_special_inode_operations;
2638                 err = ext4_add_nondir(handle, dentry, &inode);
2639         }
2640         if (handle)
2641                 ext4_journal_stop(handle);
2642         if (!IS_ERR_OR_NULL(inode))
2643                 iput(inode);
2644         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2645                 goto retry;
2646         return err;
2647 }
2648
2649 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2650 {
2651         handle_t *handle;
2652         struct inode *inode;
2653         int err, retries = 0;
2654
2655         err = dquot_initialize(dir);
2656         if (err)
2657                 return err;
2658
2659 retry:
2660         inode = ext4_new_inode_start_handle(dir, mode,
2661                                             NULL, 0, NULL,
2662                                             EXT4_HT_DIR,
2663                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2664                           4 + EXT4_XATTR_TRANS_BLOCKS);
2665         handle = ext4_journal_current_handle();
2666         err = PTR_ERR(inode);
2667         if (!IS_ERR(inode)) {
2668                 inode->i_op = &ext4_file_inode_operations;
2669                 inode->i_fop = &ext4_file_operations;
2670                 ext4_set_aops(inode);
2671                 d_tmpfile(dentry, inode);
2672                 err = ext4_orphan_add(handle, inode);
2673                 if (err)
2674                         goto err_unlock_inode;
2675                 mark_inode_dirty(inode);
2676                 unlock_new_inode(inode);
2677         }
2678         if (handle)
2679                 ext4_journal_stop(handle);
2680         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2681                 goto retry;
2682         return err;
2683 err_unlock_inode:
2684         ext4_journal_stop(handle);
2685         unlock_new_inode(inode);
2686         return err;
2687 }
2688
2689 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2690                           struct ext4_dir_entry_2 *de,
2691                           int blocksize, int csum_size,
2692                           unsigned int parent_ino, int dotdot_real_len)
2693 {
2694         de->inode = cpu_to_le32(inode->i_ino);
2695         de->name_len = 1;
2696         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2697                                            blocksize);
2698         strcpy(de->name, ".");
2699         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2700
2701         de = ext4_next_entry(de, blocksize);
2702         de->inode = cpu_to_le32(parent_ino);
2703         de->name_len = 2;
2704         if (!dotdot_real_len)
2705                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2706                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2707                                         blocksize);
2708         else
2709                 de->rec_len = ext4_rec_len_to_disk(
2710                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2711         strcpy(de->name, "..");
2712         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2713
2714         return ext4_next_entry(de, blocksize);
2715 }
2716
2717 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2718                              struct inode *inode)
2719 {
2720         struct buffer_head *dir_block = NULL;
2721         struct ext4_dir_entry_2 *de;
2722         ext4_lblk_t block = 0;
2723         unsigned int blocksize = dir->i_sb->s_blocksize;
2724         int csum_size = 0;
2725         int err;
2726
2727         if (ext4_has_metadata_csum(dir->i_sb))
2728                 csum_size = sizeof(struct ext4_dir_entry_tail);
2729
2730         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2731                 err = ext4_try_create_inline_dir(handle, dir, inode);
2732                 if (err < 0 && err != -ENOSPC)
2733                         goto out;
2734                 if (!err)
2735                         goto out;
2736         }
2737
2738         inode->i_size = 0;
2739         dir_block = ext4_append(handle, inode, &block);
2740         if (IS_ERR(dir_block))
2741                 return PTR_ERR(dir_block);
2742         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2743         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2744         set_nlink(inode, 2);
2745         if (csum_size)
2746                 ext4_initialize_dirent_tail(dir_block, blocksize);
2747
2748         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2749         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2750         if (err)
2751                 goto out;
2752         set_buffer_verified(dir_block);
2753 out:
2754         brelse(dir_block);
2755         return err;
2756 }
2757
2758 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2759 {
2760         handle_t *handle;
2761         struct inode *inode;
2762         int err, credits, retries = 0;
2763
2764         if (EXT4_DIR_LINK_MAX(dir))
2765                 return -EMLINK;
2766
2767         err = dquot_initialize(dir);
2768         if (err)
2769                 return err;
2770
2771         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2772                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2773 retry:
2774         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2775                                             &dentry->d_name,
2776                                             0, NULL, EXT4_HT_DIR, credits);
2777         handle = ext4_journal_current_handle();
2778         err = PTR_ERR(inode);
2779         if (IS_ERR(inode))
2780                 goto out_stop;
2781
2782         inode->i_op = &ext4_dir_inode_operations;
2783         inode->i_fop = &ext4_dir_operations;
2784         err = ext4_init_new_dir(handle, dir, inode);
2785         if (err)
2786                 goto out_clear_inode;
2787         err = ext4_mark_inode_dirty(handle, inode);
2788         if (!err)
2789                 err = ext4_add_entry(handle, dentry, inode);
2790         if (err) {
2791 out_clear_inode:
2792                 clear_nlink(inode);
2793                 ext4_orphan_add(handle, inode);
2794                 unlock_new_inode(inode);
2795                 ext4_mark_inode_dirty(handle, inode);
2796                 ext4_journal_stop(handle);
2797                 iput(inode);
2798                 goto out_retry;
2799         }
2800         ext4_inc_count(handle, dir);
2801         ext4_update_dx_flag(dir);
2802         err = ext4_mark_inode_dirty(handle, dir);
2803         if (err)
2804                 goto out_clear_inode;
2805         d_instantiate_new(dentry, inode);
2806         if (IS_DIRSYNC(dir))
2807                 ext4_handle_sync(handle);
2808
2809 out_stop:
2810         if (handle)
2811                 ext4_journal_stop(handle);
2812 out_retry:
2813         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2814                 goto retry;
2815         return err;
2816 }
2817
2818 /*
2819  * routine to check that the specified directory is empty (for rmdir)
2820  */
2821 bool ext4_empty_dir(struct inode *inode)
2822 {
2823         unsigned int offset;
2824         struct buffer_head *bh;
2825         struct ext4_dir_entry_2 *de, *de1;
2826         struct super_block *sb;
2827
2828         if (ext4_has_inline_data(inode)) {
2829                 int has_inline_data = 1;
2830                 int ret;
2831
2832                 ret = empty_inline_dir(inode, &has_inline_data);
2833                 if (has_inline_data)
2834                         return ret;
2835         }
2836
2837         sb = inode->i_sb;
2838         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2839                 EXT4_ERROR_INODE(inode, "invalid size");
2840                 return true;
2841         }
2842         /* The first directory block must not be a hole,
2843          * so treat it as DIRENT_HTREE
2844          */
2845         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2846         if (IS_ERR(bh))
2847                 return true;
2848
2849         de = (struct ext4_dir_entry_2 *) bh->b_data;
2850         de1 = ext4_next_entry(de, sb->s_blocksize);
2851         if (le32_to_cpu(de->inode) != inode->i_ino ||
2852                         le32_to_cpu(de1->inode) == 0 ||
2853                         strcmp(".", de->name) || strcmp("..", de1->name)) {
2854                 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2855                 brelse(bh);
2856                 return true;
2857         }
2858         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2859                  ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2860         de = ext4_next_entry(de1, sb->s_blocksize);
2861         while (offset < inode->i_size) {
2862                 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2863                         unsigned int lblock;
2864                         brelse(bh);
2865                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2866                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2867                         if (bh == NULL) {
2868                                 offset += sb->s_blocksize;
2869                                 continue;
2870                         }
2871                         if (IS_ERR(bh))
2872                                 return true;
2873                         de = (struct ext4_dir_entry_2 *) bh->b_data;
2874                 }
2875                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2876                                          bh->b_data, bh->b_size, offset)) {
2877                         de = (struct ext4_dir_entry_2 *)(bh->b_data +
2878                                                          sb->s_blocksize);
2879                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2880                         continue;
2881                 }
2882                 if (le32_to_cpu(de->inode)) {
2883                         brelse(bh);
2884                         return false;
2885                 }
2886                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2887                 de = ext4_next_entry(de, sb->s_blocksize);
2888         }
2889         brelse(bh);
2890         return true;
2891 }
2892
2893 /*
2894  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2895  * such inodes, starting at the superblock, in case we crash before the
2896  * file is closed/deleted, or in case the inode truncate spans multiple
2897  * transactions and the last transaction is not recovered after a crash.
2898  *
2899  * At filesystem recovery time, we walk this list deleting unlinked
2900  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2901  *
2902  * Orphan list manipulation functions must be called under i_mutex unless
2903  * we are just creating the inode or deleting it.
2904  */
2905 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2906 {
2907         struct super_block *sb = inode->i_sb;
2908         struct ext4_sb_info *sbi = EXT4_SB(sb);
2909         struct ext4_iloc iloc;
2910         int err = 0, rc;
2911         bool dirty = false;
2912
2913         if (!sbi->s_journal || is_bad_inode(inode))
2914                 return 0;
2915
2916         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2917                      !inode_is_locked(inode));
2918         /*
2919          * Exit early if inode already is on orphan list. This is a big speedup
2920          * since we don't have to contend on the global s_orphan_lock.
2921          */
2922         if (!list_empty(&EXT4_I(inode)->i_orphan))
2923                 return 0;
2924
2925         /*
2926          * Orphan handling is only valid for files with data blocks
2927          * being truncated, or files being unlinked. Note that we either
2928          * hold i_mutex, or the inode can not be referenced from outside,
2929          * so i_nlink should not be bumped due to race
2930          */
2931         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2932                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2933
2934         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2935         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2936         if (err)
2937                 goto out;
2938
2939         err = ext4_reserve_inode_write(handle, inode, &iloc);
2940         if (err)
2941                 goto out;
2942
2943         mutex_lock(&sbi->s_orphan_lock);
2944         /*
2945          * Due to previous errors inode may be already a part of on-disk
2946          * orphan list. If so skip on-disk list modification.
2947          */
2948         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2949             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2950                 /* Insert this inode at the head of the on-disk orphan list */
2951                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2952                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2953                 dirty = true;
2954         }
2955         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2956         mutex_unlock(&sbi->s_orphan_lock);
2957
2958         if (dirty) {
2959                 err = ext4_handle_dirty_super(handle, sb);
2960                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2961                 if (!err)
2962                         err = rc;
2963                 if (err) {
2964                         /*
2965                          * We have to remove inode from in-memory list if
2966                          * addition to on disk orphan list failed. Stray orphan
2967                          * list entries can cause panics at unmount time.
2968                          */
2969                         mutex_lock(&sbi->s_orphan_lock);
2970                         list_del_init(&EXT4_I(inode)->i_orphan);
2971                         mutex_unlock(&sbi->s_orphan_lock);
2972                 }
2973         } else
2974                 brelse(iloc.bh);
2975
2976         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2977         jbd_debug(4, "orphan inode %lu will point to %d\n",
2978                         inode->i_ino, NEXT_ORPHAN(inode));
2979 out:
2980         ext4_std_error(sb, err);
2981         return err;
2982 }
2983
2984 /*
2985  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2986  * of such inodes stored on disk, because it is finally being cleaned up.
2987  */
2988 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2989 {
2990         struct list_head *prev;
2991         struct ext4_inode_info *ei = EXT4_I(inode);
2992         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2993         __u32 ino_next;
2994         struct ext4_iloc iloc;
2995         int err = 0;
2996
2997         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2998                 return 0;
2999
3000         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3001                      !inode_is_locked(inode));
3002         /* Do this quick check before taking global s_orphan_lock. */
3003         if (list_empty(&ei->i_orphan))
3004                 return 0;
3005
3006         if (handle) {
3007                 /* Grab inode buffer early before taking global s_orphan_lock */
3008                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3009         }
3010
3011         mutex_lock(&sbi->s_orphan_lock);
3012         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3013
3014         prev = ei->i_orphan.prev;
3015         list_del_init(&ei->i_orphan);
3016
3017         /* If we're on an error path, we may not have a valid
3018          * transaction handle with which to update the orphan list on
3019          * disk, but we still need to remove the inode from the linked
3020          * list in memory. */
3021         if (!handle || err) {
3022                 mutex_unlock(&sbi->s_orphan_lock);
3023                 goto out_err;
3024         }
3025
3026         ino_next = NEXT_ORPHAN(inode);
3027         if (prev == &sbi->s_orphan) {
3028                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3029                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3030                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3031                 if (err) {
3032                         mutex_unlock(&sbi->s_orphan_lock);
3033                         goto out_brelse;
3034                 }
3035                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3036                 mutex_unlock(&sbi->s_orphan_lock);
3037                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3038         } else {
3039                 struct ext4_iloc iloc2;
3040                 struct inode *i_prev =
3041                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3042
3043                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3044                           i_prev->i_ino, ino_next);
3045                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3046                 if (err) {
3047                         mutex_unlock(&sbi->s_orphan_lock);
3048                         goto out_brelse;
3049                 }
3050                 NEXT_ORPHAN(i_prev) = ino_next;
3051                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3052                 mutex_unlock(&sbi->s_orphan_lock);
3053         }
3054         if (err)
3055                 goto out_brelse;
3056         NEXT_ORPHAN(inode) = 0;
3057         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3058 out_err:
3059         ext4_std_error(inode->i_sb, err);
3060         return err;
3061
3062 out_brelse:
3063         brelse(iloc.bh);
3064         goto out_err;
3065 }
3066
3067 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3068 {
3069         int retval;
3070         struct inode *inode;
3071         struct buffer_head *bh;
3072         struct ext4_dir_entry_2 *de;
3073         handle_t *handle = NULL;
3074
3075         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3076                 return -EIO;
3077
3078         /* Initialize quotas before so that eventual writes go in
3079          * separate transaction */
3080         retval = dquot_initialize(dir);
3081         if (retval)
3082                 return retval;
3083         retval = dquot_initialize(d_inode(dentry));
3084         if (retval)
3085                 return retval;
3086
3087         retval = -ENOENT;
3088         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3089         if (IS_ERR(bh))
3090                 return PTR_ERR(bh);
3091         if (!bh)
3092                 goto end_rmdir;
3093
3094         inode = d_inode(dentry);
3095
3096         retval = -EFSCORRUPTED;
3097         if (le32_to_cpu(de->inode) != inode->i_ino)
3098                 goto end_rmdir;
3099
3100         retval = -ENOTEMPTY;
3101         if (!ext4_empty_dir(inode))
3102                 goto end_rmdir;
3103
3104         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3105                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3106         if (IS_ERR(handle)) {
3107                 retval = PTR_ERR(handle);
3108                 handle = NULL;
3109                 goto end_rmdir;
3110         }
3111
3112         if (IS_DIRSYNC(dir))
3113                 ext4_handle_sync(handle);
3114
3115         retval = ext4_delete_entry(handle, dir, de, bh);
3116         if (retval)
3117                 goto end_rmdir;
3118         if (!EXT4_DIR_LINK_EMPTY(inode))
3119                 ext4_warning_inode(inode,
3120                              "empty directory '%.*s' has too many links (%u)",
3121                              dentry->d_name.len, dentry->d_name.name,
3122                              inode->i_nlink);
3123         inode_inc_iversion(inode);
3124         clear_nlink(inode);
3125         /* There's no need to set i_disksize: the fact that i_nlink is
3126          * zero will ensure that the right thing happens during any
3127          * recovery. */
3128         inode->i_size = 0;
3129         ext4_orphan_add(handle, inode);
3130         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3131         ext4_mark_inode_dirty(handle, inode);
3132         ext4_dec_count(handle, dir);
3133         ext4_update_dx_flag(dir);
3134         ext4_mark_inode_dirty(handle, dir);
3135
3136 #ifdef CONFIG_UNICODE
3137         /* VFS negative dentries are incompatible with Encoding and
3138          * Case-insensitiveness. Eventually we'll want avoid
3139          * invalidating the dentries here, alongside with returning the
3140          * negative dentries at ext4_lookup(), when it is better
3141          * supported by the VFS for the CI case.
3142          */
3143         if (IS_CASEFOLDED(dir))
3144                 d_invalidate(dentry);
3145 #endif
3146
3147 end_rmdir:
3148         brelse(bh);
3149         if (handle)
3150                 ext4_journal_stop(handle);
3151         return retval;
3152 }
3153
3154 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3155 {
3156         int retval;
3157         struct inode *inode;
3158         struct buffer_head *bh;
3159         struct ext4_dir_entry_2 *de;
3160         handle_t *handle = NULL;
3161
3162         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3163                 return -EIO;
3164
3165         trace_ext4_unlink_enter(dir, dentry);
3166         /* Initialize quotas before so that eventual writes go
3167          * in separate transaction */
3168         retval = dquot_initialize(dir);
3169         if (retval)
3170                 return retval;
3171         retval = dquot_initialize(d_inode(dentry));
3172         if (retval)
3173                 return retval;
3174
3175         retval = -ENOENT;
3176         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3177         if (IS_ERR(bh))
3178                 return PTR_ERR(bh);
3179         if (!bh)
3180                 goto end_unlink;
3181
3182         inode = d_inode(dentry);
3183
3184         retval = -EFSCORRUPTED;
3185         if (le32_to_cpu(de->inode) != inode->i_ino)
3186                 goto end_unlink;
3187
3188         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3189                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3190         if (IS_ERR(handle)) {
3191                 retval = PTR_ERR(handle);
3192                 handle = NULL;
3193                 goto end_unlink;
3194         }
3195
3196         if (IS_DIRSYNC(dir))
3197                 ext4_handle_sync(handle);
3198
3199         retval = ext4_delete_entry(handle, dir, de, bh);
3200         if (retval)
3201                 goto end_unlink;
3202         dir->i_ctime = dir->i_mtime = current_time(dir);
3203         ext4_update_dx_flag(dir);
3204         ext4_mark_inode_dirty(handle, dir);
3205         if (inode->i_nlink == 0)
3206                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3207                                    dentry->d_name.len, dentry->d_name.name);
3208         else
3209                 drop_nlink(inode);
3210         if (!inode->i_nlink)
3211                 ext4_orphan_add(handle, inode);
3212         inode->i_ctime = current_time(inode);
3213         ext4_mark_inode_dirty(handle, inode);
3214
3215 #ifdef CONFIG_UNICODE
3216         /* VFS negative dentries are incompatible with Encoding and
3217          * Case-insensitiveness. Eventually we'll want avoid
3218          * invalidating the dentries here, alongside with returning the
3219          * negative dentries at ext4_lookup(), when it is  better
3220          * supported by the VFS for the CI case.
3221          */
3222         if (IS_CASEFOLDED(dir))
3223                 d_invalidate(dentry);
3224 #endif
3225
3226 end_unlink:
3227         brelse(bh);
3228         if (handle)
3229                 ext4_journal_stop(handle);
3230         trace_ext4_unlink_exit(dentry, retval);
3231         return retval;
3232 }
3233
3234 static int ext4_symlink(struct inode *dir,
3235                         struct dentry *dentry, const char *symname)
3236 {
3237         handle_t *handle;
3238         struct inode *inode;
3239         int err, len = strlen(symname);
3240         int credits;
3241         struct fscrypt_str disk_link;
3242
3243         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3244                 return -EIO;
3245
3246         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3247                                       &disk_link);
3248         if (err)
3249                 return err;
3250
3251         err = dquot_initialize(dir);
3252         if (err)
3253                 return err;
3254
3255         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3256                 /*
3257                  * For non-fast symlinks, we just allocate inode and put it on
3258                  * orphan list in the first transaction => we need bitmap,
3259                  * group descriptor, sb, inode block, quota blocks, and
3260                  * possibly selinux xattr blocks.
3261                  */
3262                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3263                           EXT4_XATTR_TRANS_BLOCKS;
3264         } else {
3265                 /*
3266                  * Fast symlink. We have to add entry to directory
3267                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3268                  * allocate new inode (bitmap, group descriptor, inode block,
3269                  * quota blocks, sb is already counted in previous macros).
3270                  */
3271                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3272                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3273         }
3274
3275         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3276                                             &dentry->d_name, 0, NULL,
3277                                             EXT4_HT_DIR, credits);
3278         handle = ext4_journal_current_handle();
3279         if (IS_ERR(inode)) {
3280                 if (handle)
3281                         ext4_journal_stop(handle);
3282                 return PTR_ERR(inode);
3283         }
3284
3285         if (IS_ENCRYPTED(inode)) {
3286                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3287                 if (err)
3288                         goto err_drop_inode;
3289                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3290         }
3291
3292         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3293                 if (!IS_ENCRYPTED(inode))
3294                         inode->i_op = &ext4_symlink_inode_operations;
3295                 inode_nohighmem(inode);
3296                 ext4_set_aops(inode);
3297                 /*
3298                  * We cannot call page_symlink() with transaction started
3299                  * because it calls into ext4_write_begin() which can wait
3300                  * for transaction commit if we are running out of space
3301                  * and thus we deadlock. So we have to stop transaction now
3302                  * and restart it when symlink contents is written.
3303                  *
3304                  * To keep fs consistent in case of crash, we have to put inode
3305                  * to orphan list in the mean time.
3306                  */
3307                 drop_nlink(inode);
3308                 err = ext4_orphan_add(handle, inode);
3309                 ext4_journal_stop(handle);
3310                 handle = NULL;
3311                 if (err)
3312                         goto err_drop_inode;
3313                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3314                 if (err)
3315                         goto err_drop_inode;
3316                 /*
3317                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3318                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3319                  */
3320                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3321                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3322                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3323                 if (IS_ERR(handle)) {
3324                         err = PTR_ERR(handle);
3325                         handle = NULL;
3326                         goto err_drop_inode;
3327                 }
3328                 set_nlink(inode, 1);
3329                 err = ext4_orphan_del(handle, inode);
3330                 if (err)
3331                         goto err_drop_inode;
3332         } else {
3333                 /* clear the extent format for fast symlink */
3334                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3335                 if (!IS_ENCRYPTED(inode)) {
3336                         inode->i_op = &ext4_fast_symlink_inode_operations;
3337                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3338                 }
3339                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3340                        disk_link.len);
3341                 inode->i_size = disk_link.len - 1;
3342         }
3343         EXT4_I(inode)->i_disksize = inode->i_size;
3344         err = ext4_add_nondir(handle, dentry, &inode);
3345         if (handle)
3346                 ext4_journal_stop(handle);
3347         if (inode)
3348                 iput(inode);
3349         goto out_free_encrypted_link;
3350
3351 err_drop_inode:
3352         if (handle)
3353                 ext4_journal_stop(handle);
3354         clear_nlink(inode);
3355         unlock_new_inode(inode);
3356         iput(inode);
3357 out_free_encrypted_link:
3358         if (disk_link.name != (unsigned char *)symname)
3359                 kfree(disk_link.name);
3360         return err;
3361 }
3362
3363 static int ext4_link(struct dentry *old_dentry,
3364                      struct inode *dir, struct dentry *dentry)
3365 {
3366         handle_t *handle;
3367         struct inode *inode = d_inode(old_dentry);
3368         int err, retries = 0;
3369
3370         if (inode->i_nlink >= EXT4_LINK_MAX)
3371                 return -EMLINK;
3372
3373         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3374         if (err)
3375                 return err;
3376
3377         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3378             (!projid_eq(EXT4_I(dir)->i_projid,
3379                         EXT4_I(old_dentry->d_inode)->i_projid)))
3380                 return -EXDEV;
3381
3382         err = dquot_initialize(dir);
3383         if (err)
3384                 return err;
3385
3386 retry:
3387         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3388                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3389                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3390         if (IS_ERR(handle))
3391                 return PTR_ERR(handle);
3392
3393         if (IS_DIRSYNC(dir))
3394                 ext4_handle_sync(handle);
3395
3396         inode->i_ctime = current_time(inode);
3397         ext4_inc_count(handle, inode);
3398         ihold(inode);
3399
3400         err = ext4_add_entry(handle, dentry, inode);
3401         if (!err) {
3402                 ext4_mark_inode_dirty(handle, inode);
3403                 /* this can happen only for tmpfile being
3404                  * linked the first time
3405                  */
3406                 if (inode->i_nlink == 1)
3407                         ext4_orphan_del(handle, inode);
3408                 d_instantiate(dentry, inode);
3409         } else {
3410                 drop_nlink(inode);
3411                 iput(inode);
3412         }
3413         ext4_journal_stop(handle);
3414         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3415                 goto retry;
3416         return err;
3417 }
3418
3419
3420 /*
3421  * Try to find buffer head where contains the parent block.
3422  * It should be the inode block if it is inlined or the 1st block
3423  * if it is a normal dir.
3424  */
3425 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3426                                         struct inode *inode,
3427                                         int *retval,
3428                                         struct ext4_dir_entry_2 **parent_de,
3429                                         int *inlined)
3430 {
3431         struct buffer_head *bh;
3432
3433         if (!ext4_has_inline_data(inode)) {
3434                 /* The first directory block must not be a hole, so
3435                  * treat it as DIRENT_HTREE
3436                  */
3437                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3438                 if (IS_ERR(bh)) {
3439                         *retval = PTR_ERR(bh);
3440                         return NULL;
3441                 }
3442                 *parent_de = ext4_next_entry(
3443                                         (struct ext4_dir_entry_2 *)bh->b_data,
3444                                         inode->i_sb->s_blocksize);
3445                 return bh;
3446         }
3447
3448         *inlined = 1;
3449         return ext4_get_first_inline_block(inode, parent_de, retval);
3450 }
3451
3452 struct ext4_renament {
3453         struct inode *dir;
3454         struct dentry *dentry;
3455         struct inode *inode;
3456         bool is_dir;
3457         int dir_nlink_delta;
3458
3459         /* entry for "dentry" */
3460         struct buffer_head *bh;
3461         struct ext4_dir_entry_2 *de;
3462         int inlined;
3463
3464         /* entry for ".." in inode if it's a directory */
3465         struct buffer_head *dir_bh;
3466         struct ext4_dir_entry_2 *parent_de;
3467         int dir_inlined;
3468 };
3469
3470 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3471 {
3472         int retval;
3473
3474         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3475                                               &retval, &ent->parent_de,
3476                                               &ent->dir_inlined);
3477         if (!ent->dir_bh)
3478                 return retval;
3479         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3480                 return -EFSCORRUPTED;
3481         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3482         return ext4_journal_get_write_access(handle, ent->dir_bh);
3483 }
3484
3485 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3486                                   unsigned dir_ino)
3487 {
3488         int retval;
3489
3490         ent->parent_de->inode = cpu_to_le32(dir_ino);
3491         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3492         if (!ent->dir_inlined) {
3493                 if (is_dx(ent->inode)) {
3494                         retval = ext4_handle_dirty_dx_node(handle,
3495                                                            ent->inode,
3496                                                            ent->dir_bh);
3497                 } else {
3498                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3499                                                             ent->dir_bh);
3500                 }
3501         } else {
3502                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3503         }
3504         if (retval) {
3505                 ext4_std_error(ent->dir->i_sb, retval);
3506                 return retval;
3507         }
3508         return 0;
3509 }
3510
3511 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3512                        unsigned ino, unsigned file_type)
3513 {
3514         int retval;
3515
3516         BUFFER_TRACE(ent->bh, "get write access");
3517         retval = ext4_journal_get_write_access(handle, ent->bh);
3518         if (retval)
3519                 return retval;
3520         ent->de->inode = cpu_to_le32(ino);
3521         if (ext4_has_feature_filetype(ent->dir->i_sb))
3522                 ent->de->file_type = file_type;
3523         inode_inc_iversion(ent->dir);
3524         ent->dir->i_ctime = ent->dir->i_mtime =
3525                 current_time(ent->dir);
3526         ext4_mark_inode_dirty(handle, ent->dir);
3527         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3528         if (!ent->inlined) {
3529                 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3530                 if (unlikely(retval)) {
3531                         ext4_std_error(ent->dir->i_sb, retval);
3532                         return retval;
3533                 }
3534         }
3535         brelse(ent->bh);
3536         ent->bh = NULL;
3537
3538         return 0;
3539 }
3540
3541 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3542                                   const struct qstr *d_name)
3543 {
3544         int retval = -ENOENT;
3545         struct buffer_head *bh;
3546         struct ext4_dir_entry_2 *de;
3547
3548         bh = ext4_find_entry(dir, d_name, &de, NULL);
3549         if (IS_ERR(bh))
3550                 return PTR_ERR(bh);
3551         if (bh) {
3552                 retval = ext4_delete_entry(handle, dir, de, bh);
3553                 brelse(bh);
3554         }
3555         return retval;
3556 }
3557
3558 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3559                                int force_reread)
3560 {
3561         int retval;
3562         /*
3563          * ent->de could have moved from under us during htree split, so make
3564          * sure that we are deleting the right entry.  We might also be pointing
3565          * to a stale entry in the unused part of ent->bh so just checking inum
3566          * and the name isn't enough.
3567          */
3568         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3569             ent->de->name_len != ent->dentry->d_name.len ||
3570             strncmp(ent->de->name, ent->dentry->d_name.name,
3571                     ent->de->name_len) ||
3572             force_reread) {
3573                 retval = ext4_find_delete_entry(handle, ent->dir,
3574                                                 &ent->dentry->d_name);
3575         } else {
3576                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3577                 if (retval == -ENOENT) {
3578                         retval = ext4_find_delete_entry(handle, ent->dir,
3579                                                         &ent->dentry->d_name);
3580                 }
3581         }
3582
3583         if (retval) {
3584                 ext4_warning_inode(ent->dir,
3585                                    "Deleting old file: nlink %d, error=%d",
3586                                    ent->dir->i_nlink, retval);
3587         }
3588 }
3589
3590 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3591 {
3592         if (ent->dir_nlink_delta) {
3593                 if (ent->dir_nlink_delta == -1)
3594                         ext4_dec_count(handle, ent->dir);
3595                 else
3596                         ext4_inc_count(handle, ent->dir);
3597                 ext4_mark_inode_dirty(handle, ent->dir);
3598         }
3599 }
3600
3601 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3602                                               int credits, handle_t **h)
3603 {
3604         struct inode *wh;
3605         handle_t *handle;
3606         int retries = 0;
3607
3608         /*
3609          * for inode block, sb block, group summaries,
3610          * and inode bitmap
3611          */
3612         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3613                     EXT4_XATTR_TRANS_BLOCKS + 4);
3614 retry:
3615         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3616                                          &ent->dentry->d_name, 0, NULL,
3617                                          EXT4_HT_DIR, credits);
3618
3619         handle = ext4_journal_current_handle();
3620         if (IS_ERR(wh)) {
3621                 if (handle)
3622                         ext4_journal_stop(handle);
3623                 if (PTR_ERR(wh) == -ENOSPC &&
3624                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3625                         goto retry;
3626         } else {
3627                 *h = handle;
3628                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3629                 wh->i_op = &ext4_special_inode_operations;
3630         }
3631         return wh;
3632 }
3633
3634 /*
3635  * Anybody can rename anything with this: the permission checks are left to the
3636  * higher-level routines.
3637  *
3638  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3639  * while new_{dentry,inode) refers to the destination dentry/inode
3640  * This comes from rename(const char *oldpath, const char *newpath)
3641  */
3642 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3643                        struct inode *new_dir, struct dentry *new_dentry,
3644                        unsigned int flags)
3645 {
3646         handle_t *handle = NULL;
3647         struct ext4_renament old = {
3648                 .dir = old_dir,
3649                 .dentry = old_dentry,
3650                 .inode = d_inode(old_dentry),
3651         };
3652         struct ext4_renament new = {
3653                 .dir = new_dir,
3654                 .dentry = new_dentry,
3655                 .inode = d_inode(new_dentry),
3656         };
3657         int force_reread;
3658         int retval;
3659         struct inode *whiteout = NULL;
3660         int credits;
3661         u8 old_file_type;
3662
3663         if (new.inode && new.inode->i_nlink == 0) {
3664                 EXT4_ERROR_INODE(new.inode,
3665                                  "target of rename is already freed");
3666                 return -EFSCORRUPTED;
3667         }
3668
3669         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3670             (!projid_eq(EXT4_I(new_dir)->i_projid,
3671                         EXT4_I(old_dentry->d_inode)->i_projid)))
3672                 return -EXDEV;
3673
3674         retval = dquot_initialize(old.dir);
3675         if (retval)
3676                 return retval;
3677         retval = dquot_initialize(new.dir);
3678         if (retval)
3679                 return retval;
3680
3681         /* Initialize quotas before so that eventual writes go
3682          * in separate transaction */
3683         if (new.inode) {
3684                 retval = dquot_initialize(new.inode);
3685                 if (retval)
3686                         return retval;
3687         }
3688
3689         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3690         if (IS_ERR(old.bh))
3691                 return PTR_ERR(old.bh);
3692         /*
3693          *  Check for inode number is _not_ due to possible IO errors.
3694          *  We might rmdir the source, keep it as pwd of some process
3695          *  and merrily kill the link to whatever was created under the
3696          *  same name. Goodbye sticky bit ;-<
3697          */
3698         retval = -ENOENT;
3699         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3700                 goto end_rename;
3701
3702         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3703                                  &new.de, &new.inlined);
3704         if (IS_ERR(new.bh)) {
3705                 retval = PTR_ERR(new.bh);
3706                 new.bh = NULL;
3707                 goto end_rename;
3708         }
3709         if (new.bh) {
3710                 if (!new.inode) {
3711                         brelse(new.bh);
3712                         new.bh = NULL;
3713                 }
3714         }
3715         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3716                 ext4_alloc_da_blocks(old.inode);
3717
3718         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3719                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3720         if (!(flags & RENAME_WHITEOUT)) {
3721                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3722                 if (IS_ERR(handle)) {
3723                         retval = PTR_ERR(handle);
3724                         handle = NULL;
3725                         goto end_rename;
3726                 }
3727         } else {
3728                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3729                 if (IS_ERR(whiteout)) {
3730                         retval = PTR_ERR(whiteout);
3731                         whiteout = NULL;
3732                         goto end_rename;
3733                 }
3734         }
3735
3736         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3737                 ext4_handle_sync(handle);
3738
3739         if (S_ISDIR(old.inode->i_mode)) {
3740                 if (new.inode) {
3741                         retval = -ENOTEMPTY;
3742                         if (!ext4_empty_dir(new.inode))
3743                                 goto end_rename;
3744                 } else {
3745                         retval = -EMLINK;
3746                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3747                                 goto end_rename;
3748                 }
3749                 retval = ext4_rename_dir_prepare(handle, &old);
3750                 if (retval)
3751                         goto end_rename;
3752         }
3753         /*
3754          * If we're renaming a file within an inline_data dir and adding or
3755          * setting the new dirent causes a conversion from inline_data to
3756          * extents/blockmap, we need to force the dirent delete code to
3757          * re-read the directory, or else we end up trying to delete a dirent
3758          * from what is now the extent tree root (or a block map).
3759          */
3760         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3761                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3762
3763         old_file_type = old.de->file_type;
3764         if (whiteout) {
3765                 /*
3766                  * Do this before adding a new entry, so the old entry is sure
3767                  * to be still pointing to the valid old entry.
3768                  */
3769                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3770                                      EXT4_FT_CHRDEV);
3771                 if (retval)
3772                         goto end_rename;
3773                 ext4_mark_inode_dirty(handle, whiteout);
3774         }
3775         if (!new.bh) {
3776                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3777                 if (retval)
3778                         goto end_rename;
3779         } else {
3780                 retval = ext4_setent(handle, &new,
3781                                      old.inode->i_ino, old_file_type);
3782                 if (retval)
3783                         goto end_rename;
3784         }
3785         if (force_reread)
3786                 force_reread = !ext4_test_inode_flag(new.dir,
3787                                                      EXT4_INODE_INLINE_DATA);
3788
3789         /*
3790          * Like most other Unix systems, set the ctime for inodes on a
3791          * rename.
3792          */
3793         old.inode->i_ctime = current_time(old.inode);
3794         ext4_mark_inode_dirty(handle, old.inode);
3795
3796         if (!whiteout) {
3797                 /*
3798                  * ok, that's it
3799                  */
3800                 ext4_rename_delete(handle, &old, force_reread);
3801         }
3802
3803         if (new.inode) {
3804                 ext4_dec_count(handle, new.inode);
3805                 new.inode->i_ctime = current_time(new.inode);
3806         }
3807         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3808         ext4_update_dx_flag(old.dir);
3809         if (old.dir_bh) {
3810                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3811                 if (retval)
3812                         goto end_rename;
3813
3814                 ext4_dec_count(handle, old.dir);
3815                 if (new.inode) {
3816                         /* checked ext4_empty_dir above, can't have another
3817                          * parent, ext4_dec_count() won't work for many-linked
3818                          * dirs */
3819                         clear_nlink(new.inode);
3820                 } else {
3821                         ext4_inc_count(handle, new.dir);
3822                         ext4_update_dx_flag(new.dir);
3823                         ext4_mark_inode_dirty(handle, new.dir);
3824                 }
3825         }
3826         ext4_mark_inode_dirty(handle, old.dir);
3827         if (new.inode) {
3828                 ext4_mark_inode_dirty(handle, new.inode);
3829                 if (!new.inode->i_nlink)
3830                         ext4_orphan_add(handle, new.inode);
3831         }
3832         retval = 0;
3833
3834 end_rename:
3835         brelse(old.dir_bh);
3836         brelse(old.bh);
3837         brelse(new.bh);
3838         if (whiteout) {
3839                 if (retval)
3840                         drop_nlink(whiteout);
3841                 unlock_new_inode(whiteout);
3842                 iput(whiteout);
3843         }
3844         if (handle)
3845                 ext4_journal_stop(handle);
3846         return retval;
3847 }
3848
3849 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3850                              struct inode *new_dir, struct dentry *new_dentry)
3851 {
3852         handle_t *handle = NULL;
3853         struct ext4_renament old = {
3854                 .dir = old_dir,
3855                 .dentry = old_dentry,
3856                 .inode = d_inode(old_dentry),
3857         };
3858         struct ext4_renament new = {
3859                 .dir = new_dir,
3860                 .dentry = new_dentry,
3861                 .inode = d_inode(new_dentry),
3862         };
3863         u8 new_file_type;
3864         int retval;
3865         struct timespec64 ctime;
3866
3867         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3868              !projid_eq(EXT4_I(new_dir)->i_projid,
3869                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3870             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3871              !projid_eq(EXT4_I(old_dir)->i_projid,
3872                         EXT4_I(new_dentry->d_inode)->i_projid)))
3873                 return -EXDEV;
3874
3875         retval = dquot_initialize(old.dir);
3876         if (retval)
3877                 return retval;
3878         retval = dquot_initialize(new.dir);
3879         if (retval)
3880                 return retval;
3881
3882         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3883                                  &old.de, &old.inlined);
3884         if (IS_ERR(old.bh))
3885                 return PTR_ERR(old.bh);
3886         /*
3887          *  Check for inode number is _not_ due to possible IO errors.
3888          *  We might rmdir the source, keep it as pwd of some process
3889          *  and merrily kill the link to whatever was created under the
3890          *  same name. Goodbye sticky bit ;-<
3891          */
3892         retval = -ENOENT;
3893         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3894                 goto end_rename;
3895
3896         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3897                                  &new.de, &new.inlined);
3898         if (IS_ERR(new.bh)) {
3899                 retval = PTR_ERR(new.bh);
3900                 new.bh = NULL;
3901                 goto end_rename;
3902         }
3903
3904         /* RENAME_EXCHANGE case: old *and* new must both exist */
3905         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3906                 goto end_rename;
3907
3908         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3909                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3910                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3911         if (IS_ERR(handle)) {
3912                 retval = PTR_ERR(handle);
3913                 handle = NULL;
3914                 goto end_rename;
3915         }
3916
3917         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3918                 ext4_handle_sync(handle);
3919
3920         if (S_ISDIR(old.inode->i_mode)) {
3921                 old.is_dir = true;
3922                 retval = ext4_rename_dir_prepare(handle, &old);
3923                 if (retval)
3924                         goto end_rename;
3925         }
3926         if (S_ISDIR(new.inode->i_mode)) {
3927                 new.is_dir = true;
3928                 retval = ext4_rename_dir_prepare(handle, &new);
3929                 if (retval)
3930                         goto end_rename;
3931         }
3932
3933         /*
3934          * Other than the special case of overwriting a directory, parents'
3935          * nlink only needs to be modified if this is a cross directory rename.
3936          */
3937         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3938                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3939                 new.dir_nlink_delta = -old.dir_nlink_delta;
3940                 retval = -EMLINK;
3941                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3942                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3943                         goto end_rename;
3944         }
3945
3946         new_file_type = new.de->file_type;
3947         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3948         if (retval)
3949                 goto end_rename;
3950
3951         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3952         if (retval)
3953                 goto end_rename;
3954
3955         /*
3956          * Like most other Unix systems, set the ctime for inodes on a
3957          * rename.
3958          */
3959         ctime = current_time(old.inode);
3960         old.inode->i_ctime = ctime;
3961         new.inode->i_ctime = ctime;
3962         ext4_mark_inode_dirty(handle, old.inode);
3963         ext4_mark_inode_dirty(handle, new.inode);
3964
3965         if (old.dir_bh) {
3966                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3967                 if (retval)
3968                         goto end_rename;
3969         }
3970         if (new.dir_bh) {
3971                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3972                 if (retval)
3973                         goto end_rename;
3974         }
3975         ext4_update_dir_count(handle, &old);
3976         ext4_update_dir_count(handle, &new);
3977         retval = 0;
3978
3979 end_rename:
3980         brelse(old.dir_bh);
3981         brelse(new.dir_bh);
3982         brelse(old.bh);
3983         brelse(new.bh);
3984         if (handle)
3985                 ext4_journal_stop(handle);
3986         return retval;
3987 }
3988
3989 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3990                         struct inode *new_dir, struct dentry *new_dentry,
3991                         unsigned int flags)
3992 {
3993         int err;
3994
3995         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3996                 return -EIO;
3997
3998         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3999                 return -EINVAL;
4000
4001         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4002                                      flags);
4003         if (err)
4004                 return err;
4005
4006         if (flags & RENAME_EXCHANGE) {
4007                 return ext4_cross_rename(old_dir, old_dentry,
4008                                          new_dir, new_dentry);
4009         }
4010
4011         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4012 }
4013
4014 /*
4015  * directories can handle most operations...
4016  */
4017 const struct inode_operations ext4_dir_inode_operations = {
4018         .create         = ext4_create,
4019         .lookup         = ext4_lookup,
4020         .link           = ext4_link,
4021         .unlink         = ext4_unlink,
4022         .symlink        = ext4_symlink,
4023         .mkdir          = ext4_mkdir,
4024         .rmdir          = ext4_rmdir,
4025         .mknod          = ext4_mknod,
4026         .tmpfile        = ext4_tmpfile,
4027         .rename         = ext4_rename2,
4028         .setattr        = ext4_setattr,
4029         .getattr        = ext4_getattr,
4030         .listxattr      = ext4_listxattr,
4031         .get_acl        = ext4_get_acl,
4032         .set_acl        = ext4_set_acl,
4033         .fiemap         = ext4_fiemap,
4034 };
4035
4036 const struct inode_operations ext4_special_inode_operations = {
4037         .setattr        = ext4_setattr,
4038         .getattr        = ext4_getattr,
4039         .listxattr      = ext4_listxattr,
4040         .get_acl        = ext4_get_acl,
4041         .set_acl        = ext4_set_acl,
4042 };