fs/ext4/namei.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
ext4_append
__ext4_read_dirblock
ext4_initialize_dirent_tail
get_dirent_tail
ext4_dirblock_csum
__warn_no_space_for_csum
ext4_dirblock_csum_verify
ext4_dirblock_csum_set
ext4_handle_dirty_dirblock
get_dx_countlimit
ext4_dx_csum
ext4_dx_csum_verify
ext4_dx_csum_set
ext4_handle_dirty_dx_node
ext4_next_entry
dx_get_block
dx_set_block
dx_get_hash
dx_set_hash
dx_get_count
dx_get_limit
dx_set_count
dx_set_limit
dx_root_limit
dx_node_limit
dx_show_index
dx_show_leaf
dx_show_entries
dx_probe
dx_release
ext4_htree_next_block
htree_dirblock_to_tree
ext4_htree_fill_tree
search_dirblock
dx_make_map
dx_sort_map
dx_insert_block
ext4_ci_compare
ext4_fname_setup_ci_filename
ext4_match
ext4_search_dir
is_dx_internal_node
__ext4_find_entry
ext4_find_entry
ext4_lookup_entry
ext4_dx_find_entry
ext4_lookup
ext4_get_parent
dx_move_dirents
dx_pack_dirents
do_split
ext4_find_dest_de
ext4_insert_dentry
add_dirent_to_buf
make_indexed_dir
ext4_add_entry
ext4_dx_add_entry
ext4_generic_delete_entry
ext4_delete_entry
ext4_inc_count
ext4_dec_count
ext4_add_nondir
ext4_create
ext4_mknod
ext4_tmpfile
ext4_init_dot_dotdot
ext4_init_new_dir
ext4_mkdir
ext4_empty_dir
ext4_orphan_add
ext4_orphan_del
ext4_rmdir
ext4_unlink
ext4_symlink
ext4_link
ext4_get_first_dir_block
ext4_rename_dir_prepare
ext4_rename_dir_finish
ext4_setent
ext4_find_delete_entry
ext4_rename_delete
ext4_update_dir_count
ext4_whiteout_for_rename
ext4_rename
ext4_cross_rename
ext4_rename2
   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                 cond_resched();
1511                 if (ra_ptr >= ra_max) {
1512                         /* Refill the readahead buffer */
1513                         ra_ptr = 0;
1514                         if (block < start)
1515                                 ra_max = start - block;
1516                         else
1517                                 ra_max = nblocks - block;
1518                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1519                         retval = ext4_bread_batch(dir, block, ra_max,
1520                                                   false /* wait */, bh_use);
1521                         if (retval) {
1522                                 ret = ERR_PTR(retval);
1523                                 ra_max = 0;
1524                                 goto cleanup_and_exit;
1525                         }
1526                 }
1527                 if ((bh = bh_use[ra_ptr++]) == NULL)
1528                         goto next;
1529                 wait_on_buffer(bh);
1530                 if (!buffer_uptodate(bh)) {
1531                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1532                                          (unsigned long) block);
1533                         brelse(bh);
1534                         ret = ERR_PTR(-EIO);
1535                         goto cleanup_and_exit;
1536                 }
1537                 if (!buffer_verified(bh) &&
1538                     !is_dx_internal_node(dir, block,
1539                                          (struct ext4_dir_entry *)bh->b_data) &&
1540                     !ext4_dirblock_csum_verify(dir, bh)) {
1541                         EXT4_ERROR_INODE(dir, "checksumming directory "
1542                                          "block %lu", (unsigned long)block);
1543                         brelse(bh);
1544                         ret = ERR_PTR(-EFSBADCRC);
1545                         goto cleanup_and_exit;
1546                 }
1547                 set_buffer_verified(bh);
1548                 i = search_dirblock(bh, dir, fname,
1549                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1550                 if (i == 1) {
1551                         EXT4_I(dir)->i_dir_start_lookup = block;
1552                         ret = bh;
1553                         goto cleanup_and_exit;
1554                 } else {
1555                         brelse(bh);
1556                         if (i < 0)
1557                                 goto cleanup_and_exit;
1558                 }
1559         next:
1560                 if (++block >= nblocks)
1561                         block = 0;
1562         } while (block != start);
1563 
1564         /*
1565          * If the directory has grown while we were searching, then
1566          * search the last part of the directory before giving up.
1567          */
1568         block = nblocks;
1569         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1570         if (block < nblocks) {
1571                 start = 0;
1572                 goto restart;
1573         }
1574 
1575 cleanup_and_exit:
1576         /* Clean up the read-ahead blocks */
1577         for (; ra_ptr < ra_max; ra_ptr++)
1578                 brelse(bh_use[ra_ptr]);
1579         return ret;
1580 }
1581 
1582 static struct buffer_head *ext4_find_entry(struct inode *dir,
1583                                            const struct qstr *d_name,
1584                                            struct ext4_dir_entry_2 **res_dir,
1585                                            int *inlined)
1586 {
1587         int err;
1588         struct ext4_filename fname;
1589         struct buffer_head *bh;
1590 
1591         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1592         if (err == -ENOENT)
1593                 return NULL;
1594         if (err)
1595                 return ERR_PTR(err);
1596 
1597         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1598 
1599         ext4_fname_free_filename(&fname);
1600         return bh;
1601 }
1602 
1603 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1604                                              struct dentry *dentry,
1605                                              struct ext4_dir_entry_2 **res_dir)
1606 {
1607         int err;
1608         struct ext4_filename fname;
1609         struct buffer_head *bh;
1610 
1611         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1612         if (err == -ENOENT)
1613                 return NULL;
1614         if (err)
1615                 return ERR_PTR(err);
1616 
1617         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1618 
1619         ext4_fname_free_filename(&fname);
1620         return bh;
1621 }
1622 
1623 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1624                         struct ext4_filename *fname,
1625                         struct ext4_dir_entry_2 **res_dir)
1626 {
1627         struct super_block * sb = dir->i_sb;
1628         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1629         struct buffer_head *bh;
1630         ext4_lblk_t block;
1631         int retval;
1632 
1633 #ifdef CONFIG_FS_ENCRYPTION
1634         *res_dir = NULL;
1635 #endif
1636         frame = dx_probe(fname, dir, NULL, frames);
1637         if (IS_ERR(frame))
1638                 return (struct buffer_head *) frame;
1639         do {
1640                 block = dx_get_block(frame->at);
1641                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1642                 if (IS_ERR(bh))
1643                         goto errout;
1644 
1645                 retval = search_dirblock(bh, dir, fname,
1646                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1647                                          res_dir);
1648                 if (retval == 1)
1649                         goto success;
1650                 brelse(bh);
1651                 if (retval == -1) {
1652                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1653                         goto errout;
1654                 }
1655 
1656                 /* Check to see if we should continue to search */
1657                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1658                                                frames, NULL);
1659                 if (retval < 0) {
1660                         ext4_warning_inode(dir,
1661                                 "error %d reading directory index block",
1662                                 retval);
1663                         bh = ERR_PTR(retval);
1664                         goto errout;
1665                 }
1666         } while (retval == 1);
1667 
1668         bh = NULL;
1669 errout:
1670         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1671 success:
1672         dx_release(frames);
1673         return bh;
1674 }
1675 
1676 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1677 {
1678         struct inode *inode;
1679         struct ext4_dir_entry_2 *de;
1680         struct buffer_head *bh;
1681 
1682         if (dentry->d_name.len > EXT4_NAME_LEN)
1683                 return ERR_PTR(-ENAMETOOLONG);
1684 
1685         bh = ext4_lookup_entry(dir, dentry, &de);
1686         if (IS_ERR(bh))
1687                 return ERR_CAST(bh);
1688         inode = NULL;
1689         if (bh) {
1690                 __u32 ino = le32_to_cpu(de->inode);
1691                 brelse(bh);
1692                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1693                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1694                         return ERR_PTR(-EFSCORRUPTED);
1695                 }
1696                 if (unlikely(ino == dir->i_ino)) {
1697                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1698                                          dentry);
1699                         return ERR_PTR(-EFSCORRUPTED);
1700                 }
1701                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1702                 if (inode == ERR_PTR(-ESTALE)) {
1703                         EXT4_ERROR_INODE(dir,
1704                                          "deleted inode referenced: %u",
1705                                          ino);
1706                         return ERR_PTR(-EFSCORRUPTED);
1707                 }
1708                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1709                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1710                     !fscrypt_has_permitted_context(dir, inode)) {
1711                         ext4_warning(inode->i_sb,
1712                                      "Inconsistent encryption contexts: %lu/%lu",
1713                                      dir->i_ino, inode->i_ino);
1714                         iput(inode);
1715                         return ERR_PTR(-EPERM);
1716                 }
1717         }
1718 
1719 #ifdef CONFIG_UNICODE
1720         if (!inode && IS_CASEFOLDED(dir)) {
1721                 /* Eventually we want to call d_add_ci(dentry, NULL)
1722                  * for negative dentries in the encoding case as
1723                  * well.  For now, prevent the negative dentry
1724                  * from being cached.
1725                  */
1726                 return NULL;
1727         }
1728 #endif
1729         return d_splice_alias(inode, dentry);
1730 }
1731 
1732 
1733 struct dentry *ext4_get_parent(struct dentry *child)
1734 {
1735         __u32 ino;
1736         static const struct qstr dotdot = QSTR_INIT("..", 2);
1737         struct ext4_dir_entry_2 * de;
1738         struct buffer_head *bh;
1739 
1740         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1741         if (IS_ERR(bh))
1742                 return ERR_CAST(bh);
1743         if (!bh)
1744                 return ERR_PTR(-ENOENT);
1745         ino = le32_to_cpu(de->inode);
1746         brelse(bh);
1747 
1748         if (!ext4_valid_inum(child->d_sb, ino)) {
1749                 EXT4_ERROR_INODE(d_inode(child),
1750                                  "bad parent inode number: %u", ino);
1751                 return ERR_PTR(-EFSCORRUPTED);
1752         }
1753 
1754         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1755 }
1756 
1757 /*
1758  * Move count entries from end of map between two memory locations.
1759  * Returns pointer to last entry moved.
1760  */
1761 static struct ext4_dir_entry_2 *
1762 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1763                 unsigned blocksize)
1764 {
1765         unsigned rec_len = 0;
1766 
1767         while (count--) {
1768                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1769                                                 (from + (map->offs<<2));
1770                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1771                 memcpy (to, de, rec_len);
1772                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1773                                 ext4_rec_len_to_disk(rec_len, blocksize);
1774                 de->inode = 0;
1775                 map++;
1776                 to += rec_len;
1777         }
1778         return (struct ext4_dir_entry_2 *) (to - rec_len);
1779 }
1780 
1781 /*
1782  * Compact each dir entry in the range to the minimal rec_len.
1783  * Returns pointer to last entry in range.
1784  */
1785 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1786 {
1787         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1788         unsigned rec_len = 0;
1789 
1790         prev = to = de;
1791         while ((char*)de < base + blocksize) {
1792                 next = ext4_next_entry(de, blocksize);
1793                 if (de->inode && de->name_len) {
1794                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1795                         if (de > to)
1796                                 memmove(to, de, rec_len);
1797                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1798                         prev = to;
1799                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1800                 }
1801                 de = next;
1802         }
1803         return prev;
1804 }
1805 
1806 /*
1807  * Split a full leaf block to make room for a new dir entry.
1808  * Allocate a new block, and move entries so that they are approx. equally full.
1809  * Returns pointer to de in block into which the new entry will be inserted.
1810  */
1811 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1812                         struct buffer_head **bh,struct dx_frame *frame,
1813                         struct dx_hash_info *hinfo)
1814 {
1815         unsigned blocksize = dir->i_sb->s_blocksize;
1816         unsigned count, continued;
1817         struct buffer_head *bh2;
1818         ext4_lblk_t newblock;
1819         u32 hash2;
1820         struct dx_map_entry *map;
1821         char *data1 = (*bh)->b_data, *data2;
1822         unsigned split, move, size;
1823         struct ext4_dir_entry_2 *de = NULL, *de2;
1824         int     csum_size = 0;
1825         int     err = 0, i;
1826 
1827         if (ext4_has_metadata_csum(dir->i_sb))
1828                 csum_size = sizeof(struct ext4_dir_entry_tail);
1829 
1830         bh2 = ext4_append(handle, dir, &newblock);
1831         if (IS_ERR(bh2)) {
1832                 brelse(*bh);
1833                 *bh = NULL;
1834                 return (struct ext4_dir_entry_2 *) bh2;
1835         }
1836 
1837         BUFFER_TRACE(*bh, "get_write_access");
1838         err = ext4_journal_get_write_access(handle, *bh);
1839         if (err)
1840                 goto journal_error;
1841 
1842         BUFFER_TRACE(frame->bh, "get_write_access");
1843         err = ext4_journal_get_write_access(handle, frame->bh);
1844         if (err)
1845                 goto journal_error;
1846 
1847         data2 = bh2->b_data;
1848 
1849         /* create map in the end of data2 block */
1850         map = (struct dx_map_entry *) (data2 + blocksize);
1851         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1852                              blocksize, hinfo, map);
1853         map -= count;
1854         dx_sort_map(map, count);
1855         /* Split the existing block in the middle, size-wise */
1856         size = 0;
1857         move = 0;
1858         for (i = count-1; i >= 0; i--) {
1859                 /* is more than half of this entry in 2nd half of the block? */
1860                 if (size + map[i].size/2 > blocksize/2)
1861                         break;
1862                 size += map[i].size;
1863                 move++;
1864         }
1865         /* map index at which we will split */
1866         split = count - move;
1867         hash2 = map[split].hash;
1868         continued = hash2 == map[split - 1].hash;
1869         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1870                         (unsigned long)dx_get_block(frame->at),
1871                                         hash2, split, count-split));
1872 
1873         /* Fancy dance to stay within two buffers */
1874         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1875                               blocksize);
1876         de = dx_pack_dirents(data1, blocksize);
1877         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1878                                            (char *) de,
1879                                            blocksize);
1880         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1881                                             (char *) de2,
1882                                             blocksize);
1883         if (csum_size) {
1884                 ext4_initialize_dirent_tail(*bh, blocksize);
1885                 ext4_initialize_dirent_tail(bh2, blocksize);
1886         }
1887 
1888         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1889                         blocksize, 1));
1890         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1891                         blocksize, 1));
1892 
1893         /* Which block gets the new entry? */
1894         if (hinfo->hash >= hash2) {
1895                 swap(*bh, bh2);
1896                 de = de2;
1897         }
1898         dx_insert_block(frame, hash2 + continued, newblock);
1899         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1900         if (err)
1901                 goto journal_error;
1902         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1903         if (err)
1904                 goto journal_error;
1905         brelse(bh2);
1906         dxtrace(dx_show_index("frame", frame->entries));
1907         return de;
1908 
1909 journal_error:
1910         brelse(*bh);
1911         brelse(bh2);
1912         *bh = NULL;
1913         ext4_std_error(dir->i_sb, err);
1914         return ERR_PTR(err);
1915 }
1916 
1917 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1918                       struct buffer_head *bh,
1919                       void *buf, int buf_size,
1920                       struct ext4_filename *fname,
1921                       struct ext4_dir_entry_2 **dest_de)
1922 {
1923         struct ext4_dir_entry_2 *de;
1924         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1925         int nlen, rlen;
1926         unsigned int offset = 0;
1927         char *top;
1928 
1929         de = (struct ext4_dir_entry_2 *)buf;
1930         top = buf + buf_size - reclen;
1931         while ((char *) de <= top) {
1932                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1933                                          buf, buf_size, offset))
1934                         return -EFSCORRUPTED;
1935                 if (ext4_match(dir, fname, de))
1936                         return -EEXIST;
1937                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1938                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1939                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1940                         break;
1941                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1942                 offset += rlen;
1943         }
1944         if ((char *) de > top)
1945                 return -ENOSPC;
1946 
1947         *dest_de = de;
1948         return 0;
1949 }
1950 
1951 void ext4_insert_dentry(struct inode *inode,
1952                         struct ext4_dir_entry_2 *de,
1953                         int buf_size,
1954                         struct ext4_filename *fname)
1955 {
1956 
1957         int nlen, rlen;
1958 
1959         nlen = EXT4_DIR_REC_LEN(de->name_len);
1960         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1961         if (de->inode) {
1962                 struct ext4_dir_entry_2 *de1 =
1963                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1964                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1965                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1966                 de = de1;
1967         }
1968         de->file_type = EXT4_FT_UNKNOWN;
1969         de->inode = cpu_to_le32(inode->i_ino);
1970         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1971         de->name_len = fname_len(fname);
1972         memcpy(de->name, fname_name(fname), fname_len(fname));
1973 }
1974 
1975 /*
1976  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1977  * it points to a directory entry which is guaranteed to be large
1978  * enough for new directory entry.  If de is NULL, then
1979  * add_dirent_to_buf will attempt search the directory block for
1980  * space.  It will return -ENOSPC if no space is available, and -EIO
1981  * and -EEXIST if directory entry already exists.
1982  */
1983 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1984                              struct inode *dir,
1985                              struct inode *inode, struct ext4_dir_entry_2 *de,
1986                              struct buffer_head *bh)
1987 {
1988         unsigned int    blocksize = dir->i_sb->s_blocksize;
1989         int             csum_size = 0;
1990         int             err;
1991 
1992         if (ext4_has_metadata_csum(inode->i_sb))
1993                 csum_size = sizeof(struct ext4_dir_entry_tail);
1994 
1995         if (!de) {
1996                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1997                                         blocksize - csum_size, fname, &de);
1998                 if (err)
1999                         return err;
2000         }
2001         BUFFER_TRACE(bh, "get_write_access");
2002         err = ext4_journal_get_write_access(handle, bh);
2003         if (err) {
2004                 ext4_std_error(dir->i_sb, err);
2005                 return err;
2006         }
2007 
2008         /* By now the buffer is marked for journaling */
2009         ext4_insert_dentry(inode, de, blocksize, fname);
2010 
2011         /*
2012          * XXX shouldn't update any times until successful
2013          * completion of syscall, but too many callers depend
2014          * on this.
2015          *
2016          * XXX similarly, too many callers depend on
2017          * ext4_new_inode() setting the times, but error
2018          * recovery deletes the inode, so the worst that can
2019          * happen is that the times are slightly out of date
2020          * and/or different from the directory change time.
2021          */
2022         dir->i_mtime = dir->i_ctime = current_time(dir);
2023         ext4_update_dx_flag(dir);
2024         inode_inc_iversion(dir);
2025         ext4_mark_inode_dirty(handle, dir);
2026         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2027         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2028         if (err)
2029                 ext4_std_error(dir->i_sb, err);
2030         return 0;
2031 }
2032 
2033 /*
2034  * This converts a one block unindexed directory to a 3 block indexed
2035  * directory, and adds the dentry to the indexed directory.
2036  */
2037 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2038                             struct inode *dir,
2039                             struct inode *inode, struct buffer_head *bh)
2040 {
2041         struct buffer_head *bh2;
2042         struct dx_root  *root;
2043         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2044         struct dx_entry *entries;
2045         struct ext4_dir_entry_2 *de, *de2;
2046         char            *data2, *top;
2047         unsigned        len;
2048         int             retval;
2049         unsigned        blocksize;
2050         ext4_lblk_t  block;
2051         struct fake_dirent *fde;
2052         int csum_size = 0;
2053 
2054         if (ext4_has_metadata_csum(inode->i_sb))
2055                 csum_size = sizeof(struct ext4_dir_entry_tail);
2056 
2057         blocksize =  dir->i_sb->s_blocksize;
2058         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2059         BUFFER_TRACE(bh, "get_write_access");
2060         retval = ext4_journal_get_write_access(handle, bh);
2061         if (retval) {
2062                 ext4_std_error(dir->i_sb, retval);
2063                 brelse(bh);
2064                 return retval;
2065         }
2066         root = (struct dx_root *) bh->b_data;
2067 
2068         /* The 0th block becomes the root, move the dirents out */
2069         fde = &root->dotdot;
2070         de = (struct ext4_dir_entry_2 *)((char *)fde +
2071                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2072         if ((char *) de >= (((char *) root) + blocksize)) {
2073                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2074                 brelse(bh);
2075                 return -EFSCORRUPTED;
2076         }
2077         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2078 
2079         /* Allocate new block for the 0th block's dirents */
2080         bh2 = ext4_append(handle, dir, &block);
2081         if (IS_ERR(bh2)) {
2082                 brelse(bh);
2083                 return PTR_ERR(bh2);
2084         }
2085         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2086         data2 = bh2->b_data;
2087 
2088         memcpy(data2, de, len);
2089         de = (struct ext4_dir_entry_2 *) data2;
2090         top = data2 + len;
2091         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2092                 de = de2;
2093         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2094                                            (char *) de, blocksize);
2095 
2096         if (csum_size)
2097                 ext4_initialize_dirent_tail(bh2, blocksize);
2098 
2099         /* Initialize the root; the dot dirents already exist */
2100         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2101         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2102                                            blocksize);
2103         memset (&root->info, 0, sizeof(root->info));
2104         root->info.info_length = sizeof(root->info);
2105         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2106         entries = root->entries;
2107         dx_set_block(entries, 1);
2108         dx_set_count(entries, 1);
2109         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2110 
2111         /* Initialize as for dx_probe */
2112         fname->hinfo.hash_version = root->info.hash_version;
2113         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2114                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2115         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2116         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2117 
2118         memset(frames, 0, sizeof(frames));
2119         frame = frames;
2120         frame->entries = entries;
2121         frame->at = entries;
2122         frame->bh = bh;
2123 
2124         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2125         if (retval)
2126                 goto out_frames;        
2127         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2128         if (retval)
2129                 goto out_frames;        
2130 
2131         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2132         if (IS_ERR(de)) {
2133                 retval = PTR_ERR(de);
2134                 goto out_frames;
2135         }
2136 
2137         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2138 out_frames:
2139         /*
2140          * Even if the block split failed, we have to properly write
2141          * out all the changes we did so far. Otherwise we can end up
2142          * with corrupted filesystem.
2143          */
2144         if (retval)
2145                 ext4_mark_inode_dirty(handle, dir);
2146         dx_release(frames);
2147         brelse(bh2);
2148         return retval;
2149 }
2150 
2151 /*
2152  *      ext4_add_entry()
2153  *
2154  * adds a file entry to the specified directory, using the same
2155  * semantics as ext4_find_entry(). It returns NULL if it failed.
2156  *
2157  * NOTE!! The inode part of 'de' is left at 0 - which means you
2158  * may not sleep between calling this and putting something into
2159  * the entry, as someone else might have used it while you slept.
2160  */
2161 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2162                           struct inode *inode)
2163 {
2164         struct inode *dir = d_inode(dentry->d_parent);
2165         struct buffer_head *bh = NULL;
2166         struct ext4_dir_entry_2 *de;
2167         struct super_block *sb;
2168         struct ext4_sb_info *sbi;
2169         struct ext4_filename fname;
2170         int     retval;
2171         int     dx_fallback=0;
2172         unsigned blocksize;
2173         ext4_lblk_t block, blocks;
2174         int     csum_size = 0;
2175 
2176         if (ext4_has_metadata_csum(inode->i_sb))
2177                 csum_size = sizeof(struct ext4_dir_entry_tail);
2178 
2179         sb = dir->i_sb;
2180         sbi = EXT4_SB(sb);
2181         blocksize = sb->s_blocksize;
2182         if (!dentry->d_name.len)
2183                 return -EINVAL;
2184 
2185 #ifdef CONFIG_UNICODE
2186         if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2187             sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2188                 return -EINVAL;
2189 #endif
2190 
2191         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2192         if (retval)
2193                 return retval;
2194 
2195         if (ext4_has_inline_data(dir)) {
2196                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2197                 if (retval < 0)
2198                         goto out;
2199                 if (retval == 1) {
2200                         retval = 0;
2201                         goto out;
2202                 }
2203         }
2204 
2205         if (is_dx(dir)) {
2206                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2207                 if (!retval || (retval != ERR_BAD_DX_DIR))
2208                         goto out;
2209                 /* Can we just ignore htree data? */
2210                 if (ext4_has_metadata_csum(sb)) {
2211                         EXT4_ERROR_INODE(dir,
2212                                 "Directory has corrupted htree index.");
2213                         retval = -EFSCORRUPTED;
2214                         goto out;
2215                 }
2216                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2217                 dx_fallback++;
2218                 ext4_mark_inode_dirty(handle, dir);
2219         }
2220         blocks = dir->i_size >> sb->s_blocksize_bits;
2221         for (block = 0; block < blocks; block++) {
2222                 bh = ext4_read_dirblock(dir, block, DIRENT);
2223                 if (bh == NULL) {
2224                         bh = ext4_bread(handle, dir, block,
2225                                         EXT4_GET_BLOCKS_CREATE);
2226                         goto add_to_new_block;
2227                 }
2228                 if (IS_ERR(bh)) {
2229                         retval = PTR_ERR(bh);
2230                         bh = NULL;
2231                         goto out;
2232                 }
2233                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2234                                            NULL, bh);
2235                 if (retval != -ENOSPC)
2236                         goto out;
2237 
2238                 if (blocks == 1 && !dx_fallback &&
2239                     ext4_has_feature_dir_index(sb)) {
2240                         retval = make_indexed_dir(handle, &fname, dir,
2241                                                   inode, bh);
2242                         bh = NULL; /* make_indexed_dir releases bh */
2243                         goto out;
2244                 }
2245                 brelse(bh);
2246         }
2247         bh = ext4_append(handle, dir, &block);
2248 add_to_new_block:
2249         if (IS_ERR(bh)) {
2250                 retval = PTR_ERR(bh);
2251                 bh = NULL;
2252                 goto out;
2253         }
2254         de = (struct ext4_dir_entry_2 *) bh->b_data;
2255         de->inode = 0;
2256         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2257 
2258         if (csum_size)
2259                 ext4_initialize_dirent_tail(bh, blocksize);
2260 
2261         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2262 out:
2263         ext4_fname_free_filename(&fname);
2264         brelse(bh);
2265         if (retval == 0)
2266                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2267         return retval;
2268 }
2269 
2270 /*
2271  * Returns 0 for success, or a negative error value
2272  */
2273 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2274                              struct inode *dir, struct inode *inode)
2275 {
2276         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2277         struct dx_entry *entries, *at;
2278         struct buffer_head *bh;
2279         struct super_block *sb = dir->i_sb;
2280         struct ext4_dir_entry_2 *de;
2281         int restart;
2282         int err;
2283 
2284 again:
2285         restart = 0;
2286         frame = dx_probe(fname, dir, NULL, frames);
2287         if (IS_ERR(frame))
2288                 return PTR_ERR(frame);
2289         entries = frame->entries;
2290         at = frame->at;
2291         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2292         if (IS_ERR(bh)) {
2293                 err = PTR_ERR(bh);
2294                 bh = NULL;
2295                 goto cleanup;
2296         }
2297 
2298         BUFFER_TRACE(bh, "get_write_access");
2299         err = ext4_journal_get_write_access(handle, bh);
2300         if (err)
2301                 goto journal_error;
2302 
2303         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2304         if (err != -ENOSPC)
2305                 goto cleanup;
2306 
2307         err = 0;
2308         /* Block full, should compress but for now just split */
2309         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2310                        dx_get_count(entries), dx_get_limit(entries)));
2311         /* Need to split index? */
2312         if (dx_get_count(entries) == dx_get_limit(entries)) {
2313                 ext4_lblk_t newblock;
2314                 int levels = frame - frames + 1;
2315                 unsigned int icount;
2316                 int add_level = 1;
2317                 struct dx_entry *entries2;
2318                 struct dx_node *node2;
2319                 struct buffer_head *bh2;
2320 
2321                 while (frame > frames) {
2322                         if (dx_get_count((frame - 1)->entries) <
2323                             dx_get_limit((frame - 1)->entries)) {
2324                                 add_level = 0;
2325                                 break;
2326                         }
2327                         frame--; /* split higher index block */
2328                         at = frame->at;
2329                         entries = frame->entries;
2330                         restart = 1;
2331                 }
2332                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2333                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2334                                          "reach max htree level :%d",
2335                                          dir->i_ino, levels);
2336                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2337                                 ext4_warning(sb, "Large directory feature is "
2338                                                  "not enabled on this "
2339                                                  "filesystem");
2340                         }
2341                         err = -ENOSPC;
2342                         goto cleanup;
2343                 }
2344                 icount = dx_get_count(entries);
2345                 bh2 = ext4_append(handle, dir, &newblock);
2346                 if (IS_ERR(bh2)) {
2347                         err = PTR_ERR(bh2);
2348                         goto cleanup;
2349                 }
2350                 node2 = (struct dx_node *)(bh2->b_data);
2351                 entries2 = node2->entries;
2352                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2353                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2354                                                            sb->s_blocksize);
2355                 BUFFER_TRACE(frame->bh, "get_write_access");
2356                 err = ext4_journal_get_write_access(handle, frame->bh);
2357                 if (err)
2358                         goto journal_error;
2359                 if (!add_level) {
2360                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2361                         unsigned hash2 = dx_get_hash(entries + icount1);
2362                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2363                                        icount1, icount2));
2364 
2365                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2366                         err = ext4_journal_get_write_access(handle,
2367                                                              (frame - 1)->bh);
2368                         if (err)
2369                                 goto journal_error;
2370 
2371                         memcpy((char *) entries2, (char *) (entries + icount1),
2372                                icount2 * sizeof(struct dx_entry));
2373                         dx_set_count(entries, icount1);
2374                         dx_set_count(entries2, icount2);
2375                         dx_set_limit(entries2, dx_node_limit(dir));
2376 
2377                         /* Which index block gets the new entry? */
2378                         if (at - entries >= icount1) {
2379                                 frame->at = at = at - entries - icount1 + entries2;
2380                                 frame->entries = entries = entries2;
2381                                 swap(frame->bh, bh2);
2382                         }
2383                         dx_insert_block((frame - 1), hash2, newblock);
2384                         dxtrace(dx_show_index("node", frame->entries));
2385                         dxtrace(dx_show_index("node",
2386                                ((struct dx_node *) bh2->b_data)->entries));
2387                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2388                         if (err)
2389                                 goto journal_error;
2390                         brelse (bh2);
2391                         err = ext4_handle_dirty_dx_node(handle, dir,
2392                                                    (frame - 1)->bh);
2393                         if (err)
2394                                 goto journal_error;
2395                         if (restart) {
2396                                 err = ext4_handle_dirty_dx_node(handle, dir,
2397                                                            frame->bh);
2398                                 goto journal_error;
2399                         }
2400                 } else {
2401                         struct dx_root *dxroot;
2402                         memcpy((char *) entries2, (char *) entries,
2403                                icount * sizeof(struct dx_entry));
2404                         dx_set_limit(entries2, dx_node_limit(dir));
2405 
2406                         /* Set up root */
2407                         dx_set_count(entries, 1);
2408                         dx_set_block(entries + 0, newblock);
2409                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2410                         dxroot->info.indirect_levels += 1;
2411                         dxtrace(printk(KERN_DEBUG
2412                                        "Creating %d level index...\n",
2413                                        dxroot->info.indirect_levels));
2414                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2415                         if (err)
2416                                 goto journal_error;
2417                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2418                         brelse(bh2);
2419                         restart = 1;
2420                         goto journal_error;
2421                 }
2422         }
2423         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2424         if (IS_ERR(de)) {
2425                 err = PTR_ERR(de);
2426                 goto cleanup;
2427         }
2428         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2429         goto cleanup;
2430 
2431 journal_error:
2432         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2433 cleanup:
2434         brelse(bh);
2435         dx_release(frames);
2436         /* @restart is true means htree-path has been changed, we need to
2437          * repeat dx_probe() to find out valid htree-path
2438          */
2439         if (restart && err == 0)
2440                 goto again;
2441         return err;
2442 }
2443 
2444 /*
2445  * ext4_generic_delete_entry deletes a directory entry by merging it
2446  * with the previous entry
2447  */
2448 int ext4_generic_delete_entry(handle_t *handle,
2449                               struct inode *dir,
2450                               struct ext4_dir_entry_2 *de_del,
2451                               struct buffer_head *bh,
2452                               void *entry_buf,
2453                               int buf_size,
2454                               int csum_size)
2455 {
2456         struct ext4_dir_entry_2 *de, *pde;
2457         unsigned int blocksize = dir->i_sb->s_blocksize;
2458         int i;
2459 
2460         i = 0;
2461         pde = NULL;
2462         de = (struct ext4_dir_entry_2 *)entry_buf;
2463         while (i < buf_size - csum_size) {
2464                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2465                                          bh->b_data, bh->b_size, i))
2466                         return -EFSCORRUPTED;
2467                 if (de == de_del)  {
2468                         if (pde)
2469                                 pde->rec_len = ext4_rec_len_to_disk(
2470                                         ext4_rec_len_from_disk(pde->rec_len,
2471                                                                blocksize) +
2472                                         ext4_rec_len_from_disk(de->rec_len,
2473                                                                blocksize),
2474                                         blocksize);
2475                         else
2476                                 de->inode = 0;
2477                         inode_inc_iversion(dir);
2478                         return 0;
2479                 }
2480                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2481                 pde = de;
2482                 de = ext4_next_entry(de, blocksize);
2483         }
2484         return -ENOENT;
2485 }
2486 
2487 static int ext4_delete_entry(handle_t *handle,
2488                              struct inode *dir,
2489                              struct ext4_dir_entry_2 *de_del,
2490                              struct buffer_head *bh)
2491 {
2492         int err, csum_size = 0;
2493 
2494         if (ext4_has_inline_data(dir)) {
2495                 int has_inline_data = 1;
2496                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2497                                                &has_inline_data);
2498                 if (has_inline_data)
2499                         return err;
2500         }
2501 
2502         if (ext4_has_metadata_csum(dir->i_sb))
2503                 csum_size = sizeof(struct ext4_dir_entry_tail);
2504 
2505         BUFFER_TRACE(bh, "get_write_access");
2506         err = ext4_journal_get_write_access(handle, bh);
2507         if (unlikely(err))
2508                 goto out;
2509 
2510         err = ext4_generic_delete_entry(handle, dir, de_del,
2511                                         bh, bh->b_data,
2512                                         dir->i_sb->s_blocksize, csum_size);
2513         if (err)
2514                 goto out;
2515 
2516         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2517         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2518         if (unlikely(err))
2519                 goto out;
2520 
2521         return 0;
2522 out:
2523         if (err != -ENOENT)
2524                 ext4_std_error(dir->i_sb, err);
2525         return err;
2526 }
2527 
2528 /*
2529  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2530  * since this indicates that nlinks count was previously 1 to avoid overflowing
2531  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2532  * that subdirectory link counts are not being maintained accurately.
2533  *
2534  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2535  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2536  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2537  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2538  */
2539 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2540 {
2541         inc_nlink(inode);
2542         if (is_dx(inode) &&
2543             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2544                 set_nlink(inode, 1);
2545 }
2546 
2547 /*
2548  * If a directory had nlink == 1, then we should let it be 1. This indicates
2549  * directory has >EXT4_LINK_MAX subdirs.
2550  */
2551 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2552 {
2553         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2554                 drop_nlink(inode);
2555 }
2556 
2557 
2558 static int ext4_add_nondir(handle_t *handle,
2559                 struct dentry *dentry, struct inode *inode)
2560 {
2561         int err = ext4_add_entry(handle, dentry, inode);
2562         if (!err) {
2563                 ext4_mark_inode_dirty(handle, inode);
2564                 d_instantiate_new(dentry, inode);
2565                 return 0;
2566         }
2567         drop_nlink(inode);
2568         unlock_new_inode(inode);
2569         iput(inode);
2570         return err;
2571 }
2572 
2573 /*
2574  * By the time this is called, we already have created
2575  * the directory cache entry for the new file, but it
2576  * is so far negative - it has no inode.
2577  *
2578  * If the create succeeds, we fill in the inode information
2579  * with d_instantiate().
2580  */
2581 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2582                        bool excl)
2583 {
2584         handle_t *handle;
2585         struct inode *inode;
2586         int err, credits, retries = 0;
2587 
2588         err = dquot_initialize(dir);
2589         if (err)
2590                 return err;
2591 
2592         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2593                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2594 retry:
2595         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2596                                             NULL, EXT4_HT_DIR, credits);
2597         handle = ext4_journal_current_handle();
2598         err = PTR_ERR(inode);
2599         if (!IS_ERR(inode)) {
2600                 inode->i_op = &ext4_file_inode_operations;
2601                 inode->i_fop = &ext4_file_operations;
2602                 ext4_set_aops(inode);
2603                 err = ext4_add_nondir(handle, dentry, inode);
2604                 if (!err && IS_DIRSYNC(dir))
2605                         ext4_handle_sync(handle);
2606         }
2607         if (handle)
2608                 ext4_journal_stop(handle);
2609         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2610                 goto retry;
2611         return err;
2612 }
2613 
2614 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2615                       umode_t mode, dev_t rdev)
2616 {
2617         handle_t *handle;
2618         struct inode *inode;
2619         int err, credits, retries = 0;
2620 
2621         err = dquot_initialize(dir);
2622         if (err)
2623                 return err;
2624 
2625         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2626                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2627 retry:
2628         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2629                                             NULL, EXT4_HT_DIR, credits);
2630         handle = ext4_journal_current_handle();
2631         err = PTR_ERR(inode);
2632         if (!IS_ERR(inode)) {
2633                 init_special_inode(inode, inode->i_mode, rdev);
2634                 inode->i_op = &ext4_special_inode_operations;
2635                 err = ext4_add_nondir(handle, dentry, inode);
2636                 if (!err && IS_DIRSYNC(dir))
2637                         ext4_handle_sync(handle);
2638         }
2639         if (handle)
2640                 ext4_journal_stop(handle);
2641         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2642                 goto retry;
2643         return err;
2644 }
2645 
2646 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2647 {
2648         handle_t *handle;
2649         struct inode *inode;
2650         int err, retries = 0;
2651 
2652         err = dquot_initialize(dir);
2653         if (err)
2654                 return err;
2655 
2656 retry:
2657         inode = ext4_new_inode_start_handle(dir, mode,
2658                                             NULL, 0, NULL,
2659                                             EXT4_HT_DIR,
2660                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2661                           4 + EXT4_XATTR_TRANS_BLOCKS);
2662         handle = ext4_journal_current_handle();
2663         err = PTR_ERR(inode);
2664         if (!IS_ERR(inode)) {
2665                 inode->i_op = &ext4_file_inode_operations;
2666                 inode->i_fop = &ext4_file_operations;
2667                 ext4_set_aops(inode);
2668                 d_tmpfile(dentry, inode);
2669                 err = ext4_orphan_add(handle, inode);
2670                 if (err)
2671                         goto err_unlock_inode;
2672                 mark_inode_dirty(inode);
2673                 unlock_new_inode(inode);
2674         }
2675         if (handle)
2676                 ext4_journal_stop(handle);
2677         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2678                 goto retry;
2679         return err;
2680 err_unlock_inode:
2681         ext4_journal_stop(handle);
2682         unlock_new_inode(inode);
2683         return err;
2684 }
2685 
2686 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2687                           struct ext4_dir_entry_2 *de,
2688                           int blocksize, int csum_size,
2689                           unsigned int parent_ino, int dotdot_real_len)
2690 {
2691         de->inode = cpu_to_le32(inode->i_ino);
2692         de->name_len = 1;
2693         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2694                                            blocksize);
2695         strcpy(de->name, ".");
2696         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2697 
2698         de = ext4_next_entry(de, blocksize);
2699         de->inode = cpu_to_le32(parent_ino);
2700         de->name_len = 2;
2701         if (!dotdot_real_len)
2702                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2703                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2704                                         blocksize);
2705         else
2706                 de->rec_len = ext4_rec_len_to_disk(
2707                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2708         strcpy(de->name, "..");
2709         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2710 
2711         return ext4_next_entry(de, blocksize);
2712 }
2713 
2714 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2715                              struct inode *inode)
2716 {
2717         struct buffer_head *dir_block = NULL;
2718         struct ext4_dir_entry_2 *de;
2719         ext4_lblk_t block = 0;
2720         unsigned int blocksize = dir->i_sb->s_blocksize;
2721         int csum_size = 0;
2722         int err;
2723 
2724         if (ext4_has_metadata_csum(dir->i_sb))
2725                 csum_size = sizeof(struct ext4_dir_entry_tail);
2726 
2727         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2728                 err = ext4_try_create_inline_dir(handle, dir, inode);
2729                 if (err < 0 && err != -ENOSPC)
2730                         goto out;
2731                 if (!err)
2732                         goto out;
2733         }
2734 
2735         inode->i_size = 0;
2736         dir_block = ext4_append(handle, inode, &block);
2737         if (IS_ERR(dir_block))
2738                 return PTR_ERR(dir_block);
2739         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2740         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2741         set_nlink(inode, 2);
2742         if (csum_size)
2743                 ext4_initialize_dirent_tail(dir_block, blocksize);
2744 
2745         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2746         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2747         if (err)
2748                 goto out;
2749         set_buffer_verified(dir_block);
2750 out:
2751         brelse(dir_block);
2752         return err;
2753 }
2754 
2755 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2756 {
2757         handle_t *handle;
2758         struct inode *inode;
2759         int err, credits, retries = 0;
2760 
2761         if (EXT4_DIR_LINK_MAX(dir))
2762                 return -EMLINK;
2763 
2764         err = dquot_initialize(dir);
2765         if (err)
2766                 return err;
2767 
2768         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2769                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2770 retry:
2771         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2772                                             &dentry->d_name,
2773                                             0, NULL, EXT4_HT_DIR, credits);
2774         handle = ext4_journal_current_handle();
2775         err = PTR_ERR(inode);
2776         if (IS_ERR(inode))
2777                 goto out_stop;
2778 
2779         inode->i_op = &ext4_dir_inode_operations;
2780         inode->i_fop = &ext4_dir_operations;
2781         err = ext4_init_new_dir(handle, dir, inode);
2782         if (err)
2783                 goto out_clear_inode;
2784         err = ext4_mark_inode_dirty(handle, inode);
2785         if (!err)
2786                 err = ext4_add_entry(handle, dentry, inode);
2787         if (err) {
2788 out_clear_inode:
2789                 clear_nlink(inode);
2790                 unlock_new_inode(inode);
2791                 ext4_mark_inode_dirty(handle, inode);
2792                 iput(inode);
2793                 goto out_stop;
2794         }
2795         ext4_inc_count(handle, dir);
2796         ext4_update_dx_flag(dir);
2797         err = ext4_mark_inode_dirty(handle, dir);
2798         if (err)
2799                 goto out_clear_inode;
2800         d_instantiate_new(dentry, inode);
2801         if (IS_DIRSYNC(dir))
2802                 ext4_handle_sync(handle);
2803 
2804 out_stop:
2805         if (handle)
2806                 ext4_journal_stop(handle);
2807         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2808                 goto retry;
2809         return err;
2810 }
2811 
2812 /*
2813  * routine to check that the specified directory is empty (for rmdir)
2814  */
2815 bool ext4_empty_dir(struct inode *inode)
2816 {
2817         unsigned int offset;
2818         struct buffer_head *bh;
2819         struct ext4_dir_entry_2 *de;
2820         struct super_block *sb;
2821 
2822         if (ext4_has_inline_data(inode)) {
2823                 int has_inline_data = 1;
2824                 int ret;
2825 
2826                 ret = empty_inline_dir(inode, &has_inline_data);
2827                 if (has_inline_data)
2828                         return ret;
2829         }
2830 
2831         sb = inode->i_sb;
2832         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2833                 EXT4_ERROR_INODE(inode, "invalid size");
2834                 return true;
2835         }
2836         /* The first directory block must not be a hole,
2837          * so treat it as DIRENT_HTREE
2838          */
2839         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2840         if (IS_ERR(bh))
2841                 return true;
2842 
2843         de = (struct ext4_dir_entry_2 *) bh->b_data;
2844         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2845                                  0) ||
2846             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2847                 ext4_warning_inode(inode, "directory missing '.'");
2848                 brelse(bh);
2849                 return true;
2850         }
2851         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2852         de = ext4_next_entry(de, sb->s_blocksize);
2853         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2854                                  offset) ||
2855             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2856                 ext4_warning_inode(inode, "directory missing '..'");
2857                 brelse(bh);
2858                 return true;
2859         }
2860         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2861         while (offset < inode->i_size) {
2862                 if (!(offset & (sb->s_blocksize - 1))) {
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                 }
2874                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2875                                         (offset & (sb->s_blocksize - 1)));
2876                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2877                                          bh->b_data, bh->b_size, offset)) {
2878                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2879                         continue;
2880                 }
2881                 if (le32_to_cpu(de->inode)) {
2882                         brelse(bh);
2883                         return false;
2884                 }
2885                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2886         }
2887         brelse(bh);
2888         return true;
2889 }
2890 
2891 /*
2892  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2893  * such inodes, starting at the superblock, in case we crash before the
2894  * file is closed/deleted, or in case the inode truncate spans multiple
2895  * transactions and the last transaction is not recovered after a crash.
2896  *
2897  * At filesystem recovery time, we walk this list deleting unlinked
2898  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2899  *
2900  * Orphan list manipulation functions must be called under i_mutex unless
2901  * we are just creating the inode or deleting it.
2902  */
2903 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2904 {
2905         struct super_block *sb = inode->i_sb;
2906         struct ext4_sb_info *sbi = EXT4_SB(sb);
2907         struct ext4_iloc iloc;
2908         int err = 0, rc;
2909         bool dirty = false;
2910 
2911         if (!sbi->s_journal || is_bad_inode(inode))
2912                 return 0;
2913 
2914         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2915                      !inode_is_locked(inode));
2916         /*
2917          * Exit early if inode already is on orphan list. This is a big speedup
2918          * since we don't have to contend on the global s_orphan_lock.
2919          */
2920         if (!list_empty(&EXT4_I(inode)->i_orphan))
2921                 return 0;
2922 
2923         /*
2924          * Orphan handling is only valid for files with data blocks
2925          * being truncated, or files being unlinked. Note that we either
2926          * hold i_mutex, or the inode can not be referenced from outside,
2927          * so i_nlink should not be bumped due to race
2928          */
2929         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2930                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2931 
2932         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2933         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2934         if (err)
2935                 goto out;
2936 
2937         err = ext4_reserve_inode_write(handle, inode, &iloc);
2938         if (err)
2939                 goto out;
2940 
2941         mutex_lock(&sbi->s_orphan_lock);
2942         /*
2943          * Due to previous errors inode may be already a part of on-disk
2944          * orphan list. If so skip on-disk list modification.
2945          */
2946         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2947             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2948                 /* Insert this inode at the head of the on-disk orphan list */
2949                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2950                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2951                 dirty = true;
2952         }
2953         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2954         mutex_unlock(&sbi->s_orphan_lock);
2955 
2956         if (dirty) {
2957                 err = ext4_handle_dirty_super(handle, sb);
2958                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2959                 if (!err)
2960                         err = rc;
2961                 if (err) {
2962                         /*
2963                          * We have to remove inode from in-memory list if
2964                          * addition to on disk orphan list failed. Stray orphan
2965                          * list entries can cause panics at unmount time.
2966                          */
2967                         mutex_lock(&sbi->s_orphan_lock);
2968                         list_del_init(&EXT4_I(inode)->i_orphan);
2969                         mutex_unlock(&sbi->s_orphan_lock);
2970                 }
2971         } else
2972                 brelse(iloc.bh);
2973 
2974         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2975         jbd_debug(4, "orphan inode %lu will point to %d\n",
2976                         inode->i_ino, NEXT_ORPHAN(inode));
2977 out:
2978         ext4_std_error(sb, err);
2979         return err;
2980 }
2981 
2982 /*
2983  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2984  * of such inodes stored on disk, because it is finally being cleaned up.
2985  */
2986 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2987 {
2988         struct list_head *prev;
2989         struct ext4_inode_info *ei = EXT4_I(inode);
2990         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2991         __u32 ino_next;
2992         struct ext4_iloc iloc;
2993         int err = 0;
2994 
2995         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2996                 return 0;
2997 
2998         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2999                      !inode_is_locked(inode));
3000         /* Do this quick check before taking global s_orphan_lock. */
3001         if (list_empty(&ei->i_orphan))
3002                 return 0;
3003 
3004         if (handle) {
3005                 /* Grab inode buffer early before taking global s_orphan_lock */
3006                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3007         }
3008 
3009         mutex_lock(&sbi->s_orphan_lock);
3010         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3011 
3012         prev = ei->i_orphan.prev;
3013         list_del_init(&ei->i_orphan);
3014 
3015         /* If we're on an error path, we may not have a valid
3016          * transaction handle with which to update the orphan list on
3017          * disk, but we still need to remove the inode from the linked
3018          * list in memory. */
3019         if (!handle || err) {
3020                 mutex_unlock(&sbi->s_orphan_lock);
3021                 goto out_err;
3022         }
3023 
3024         ino_next = NEXT_ORPHAN(inode);
3025         if (prev == &sbi->s_orphan) {
3026                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3027                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3028                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3029                 if (err) {
3030                         mutex_unlock(&sbi->s_orphan_lock);
3031                         goto out_brelse;
3032                 }
3033                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3034                 mutex_unlock(&sbi->s_orphan_lock);
3035                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3036         } else {
3037                 struct ext4_iloc iloc2;
3038                 struct inode *i_prev =
3039                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3040 
3041                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3042                           i_prev->i_ino, ino_next);
3043                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3044                 if (err) {
3045                         mutex_unlock(&sbi->s_orphan_lock);
3046                         goto out_brelse;
3047                 }
3048                 NEXT_ORPHAN(i_prev) = ino_next;
3049                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3050                 mutex_unlock(&sbi->s_orphan_lock);
3051         }
3052         if (err)
3053                 goto out_brelse;
3054         NEXT_ORPHAN(inode) = 0;
3055         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3056 out_err:
3057         ext4_std_error(inode->i_sb, err);
3058         return err;
3059 
3060 out_brelse:
3061         brelse(iloc.bh);
3062         goto out_err;
3063 }
3064 
3065 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3066 {
3067         int retval;
3068         struct inode *inode;
3069         struct buffer_head *bh;
3070         struct ext4_dir_entry_2 *de;
3071         handle_t *handle = NULL;
3072 
3073         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3074                 return -EIO;
3075 
3076         /* Initialize quotas before so that eventual writes go in
3077          * separate transaction */
3078         retval = dquot_initialize(dir);
3079         if (retval)
3080                 return retval;
3081         retval = dquot_initialize(d_inode(dentry));
3082         if (retval)
3083                 return retval;
3084 
3085         retval = -ENOENT;
3086         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3087         if (IS_ERR(bh))
3088                 return PTR_ERR(bh);
3089         if (!bh)
3090                 goto end_rmdir;
3091 
3092         inode = d_inode(dentry);
3093 
3094         retval = -EFSCORRUPTED;
3095         if (le32_to_cpu(de->inode) != inode->i_ino)
3096                 goto end_rmdir;
3097 
3098         retval = -ENOTEMPTY;
3099         if (!ext4_empty_dir(inode))
3100                 goto end_rmdir;
3101 
3102         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3103                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3104         if (IS_ERR(handle)) {
3105                 retval = PTR_ERR(handle);
3106                 handle = NULL;
3107                 goto end_rmdir;
3108         }
3109 
3110         if (IS_DIRSYNC(dir))
3111                 ext4_handle_sync(handle);
3112 
3113         retval = ext4_delete_entry(handle, dir, de, bh);
3114         if (retval)
3115                 goto end_rmdir;
3116         if (!EXT4_DIR_LINK_EMPTY(inode))
3117                 ext4_warning_inode(inode,
3118                              "empty directory '%.*s' has too many links (%u)",
3119                              dentry->d_name.len, dentry->d_name.name,
3120                              inode->i_nlink);
3121         inode_inc_iversion(inode);
3122         clear_nlink(inode);
3123         /* There's no need to set i_disksize: the fact that i_nlink is
3124          * zero will ensure that the right thing happens during any
3125          * recovery. */
3126         inode->i_size = 0;
3127         ext4_orphan_add(handle, inode);
3128         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3129         ext4_mark_inode_dirty(handle, inode);
3130         ext4_dec_count(handle, dir);
3131         ext4_update_dx_flag(dir);
3132         ext4_mark_inode_dirty(handle, dir);
3133 
3134 #ifdef CONFIG_UNICODE
3135         /* VFS negative dentries are incompatible with Encoding and
3136          * Case-insensitiveness. Eventually we'll want avoid
3137          * invalidating the dentries here, alongside with returning the
3138          * negative dentries at ext4_lookup(), when it is better
3139          * supported by the VFS for the CI case.
3140          */
3141         if (IS_CASEFOLDED(dir))
3142                 d_invalidate(dentry);
3143 #endif
3144 
3145 end_rmdir:
3146         brelse(bh);
3147         if (handle)
3148                 ext4_journal_stop(handle);
3149         return retval;
3150 }
3151 
3152 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3153 {
3154         int retval;
3155         struct inode *inode;
3156         struct buffer_head *bh;
3157         struct ext4_dir_entry_2 *de;
3158         handle_t *handle = NULL;
3159 
3160         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3161                 return -EIO;
3162 
3163         trace_ext4_unlink_enter(dir, dentry);
3164         /* Initialize quotas before so that eventual writes go
3165          * in separate transaction */
3166         retval = dquot_initialize(dir);
3167         if (retval)
3168                 return retval;
3169         retval = dquot_initialize(d_inode(dentry));
3170         if (retval)
3171                 return retval;
3172 
3173         retval = -ENOENT;
3174         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3175         if (IS_ERR(bh))
3176                 return PTR_ERR(bh);
3177         if (!bh)
3178                 goto end_unlink;
3179 
3180         inode = d_inode(dentry);
3181 
3182         retval = -EFSCORRUPTED;
3183         if (le32_to_cpu(de->inode) != inode->i_ino)
3184                 goto end_unlink;
3185 
3186         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3187                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3188         if (IS_ERR(handle)) {
3189                 retval = PTR_ERR(handle);
3190                 handle = NULL;
3191                 goto end_unlink;
3192         }
3193 
3194         if (IS_DIRSYNC(dir))
3195                 ext4_handle_sync(handle);
3196 
3197         retval = ext4_delete_entry(handle, dir, de, bh);
3198         if (retval)
3199                 goto end_unlink;
3200         dir->i_ctime = dir->i_mtime = current_time(dir);
3201         ext4_update_dx_flag(dir);
3202         ext4_mark_inode_dirty(handle, dir);
3203         if (inode->i_nlink == 0)
3204                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3205                                    dentry->d_name.len, dentry->d_name.name);
3206         else
3207                 drop_nlink(inode);
3208         if (!inode->i_nlink)
3209                 ext4_orphan_add(handle, inode);
3210         inode->i_ctime = current_time(inode);
3211         ext4_mark_inode_dirty(handle, inode);
3212 
3213 #ifdef CONFIG_UNICODE
3214         /* VFS negative dentries are incompatible with Encoding and
3215          * Case-insensitiveness. Eventually we'll want avoid
3216          * invalidating the dentries here, alongside with returning the
3217          * negative dentries at ext4_lookup(), when it is  better
3218          * supported by the VFS for the CI case.
3219          */
3220         if (IS_CASEFOLDED(dir))
3221                 d_invalidate(dentry);
3222 #endif
3223 
3224 end_unlink:
3225         brelse(bh);
3226         if (handle)
3227                 ext4_journal_stop(handle);
3228         trace_ext4_unlink_exit(dentry, retval);
3229         return retval;
3230 }
3231 
3232 static int ext4_symlink(struct inode *dir,
3233                         struct dentry *dentry, const char *symname)
3234 {
3235         handle_t *handle;
3236         struct inode *inode;
3237         int err, len = strlen(symname);
3238         int credits;
3239         struct fscrypt_str disk_link;
3240 
3241         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3242                 return -EIO;
3243 
3244         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3245                                       &disk_link);
3246         if (err)
3247                 return err;
3248 
3249         err = dquot_initialize(dir);
3250         if (err)
3251                 return err;
3252 
3253         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3254                 /*
3255                  * For non-fast symlinks, we just allocate inode and put it on
3256                  * orphan list in the first transaction => we need bitmap,
3257                  * group descriptor, sb, inode block, quota blocks, and
3258                  * possibly selinux xattr blocks.
3259                  */
3260                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3261                           EXT4_XATTR_TRANS_BLOCKS;
3262         } else {
3263                 /*
3264                  * Fast symlink. We have to add entry to directory
3265                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3266                  * allocate new inode (bitmap, group descriptor, inode block,
3267                  * quota blocks, sb is already counted in previous macros).
3268                  */
3269                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3270                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3271         }
3272 
3273         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3274                                             &dentry->d_name, 0, NULL,
3275                                             EXT4_HT_DIR, credits);
3276         handle = ext4_journal_current_handle();
3277         if (IS_ERR(inode)) {
3278                 if (handle)
3279                         ext4_journal_stop(handle);
3280                 return PTR_ERR(inode);
3281         }
3282 
3283         if (IS_ENCRYPTED(inode)) {
3284                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3285                 if (err)
3286                         goto err_drop_inode;
3287                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3288         }
3289 
3290         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3291                 if (!IS_ENCRYPTED(inode))
3292                         inode->i_op = &ext4_symlink_inode_operations;
3293                 inode_nohighmem(inode);
3294                 ext4_set_aops(inode);
3295                 /*
3296                  * We cannot call page_symlink() with transaction started
3297                  * because it calls into ext4_write_begin() which can wait
3298                  * for transaction commit if we are running out of space
3299                  * and thus we deadlock. So we have to stop transaction now
3300                  * and restart it when symlink contents is written.
3301                  * 
3302                  * To keep fs consistent in case of crash, we have to put inode
3303                  * to orphan list in the mean time.
3304                  */
3305                 drop_nlink(inode);
3306                 err = ext4_orphan_add(handle, inode);
3307                 ext4_journal_stop(handle);
3308                 handle = NULL;
3309                 if (err)
3310                         goto err_drop_inode;
3311                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3312                 if (err)
3313                         goto err_drop_inode;
3314                 /*
3315                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3316                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3317                  */
3318                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3319                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3320                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3321                 if (IS_ERR(handle)) {
3322                         err = PTR_ERR(handle);
3323                         handle = NULL;
3324                         goto err_drop_inode;
3325                 }
3326                 set_nlink(inode, 1);
3327                 err = ext4_orphan_del(handle, inode);
3328                 if (err)
3329                         goto err_drop_inode;
3330         } else {
3331                 /* clear the extent format for fast symlink */
3332                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3333                 if (!IS_ENCRYPTED(inode)) {
3334                         inode->i_op = &ext4_fast_symlink_inode_operations;
3335                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3336                 }
3337                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3338                        disk_link.len);
3339                 inode->i_size = disk_link.len - 1;
3340         }
3341         EXT4_I(inode)->i_disksize = inode->i_size;
3342         err = ext4_add_nondir(handle, dentry, inode);
3343         if (!err && IS_DIRSYNC(dir))
3344                 ext4_handle_sync(handle);
3345 
3346         if (handle)
3347                 ext4_journal_stop(handle);
3348         goto out_free_encrypted_link;
3349 
3350 err_drop_inode:
3351         if (handle)
3352                 ext4_journal_stop(handle);
3353         clear_nlink(inode);
3354         unlock_new_inode(inode);
3355         iput(inode);
3356 out_free_encrypted_link:
3357         if (disk_link.name != (unsigned char *)symname)
3358                 kfree(disk_link.name);
3359         return err;
3360 }
3361 
3362 static int ext4_link(struct dentry *old_dentry,
3363                      struct inode *dir, struct dentry *dentry)
3364 {
3365         handle_t *handle;
3366         struct inode *inode = d_inode(old_dentry);
3367         int err, retries = 0;
3368 
3369         if (inode->i_nlink >= EXT4_LINK_MAX)
3370                 return -EMLINK;
3371 
3372         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3373         if (err)
3374                 return err;
3375 
3376         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3377             (!projid_eq(EXT4_I(dir)->i_projid,
3378                         EXT4_I(old_dentry->d_inode)->i_projid)))
3379                 return -EXDEV;
3380 
3381         err = dquot_initialize(dir);
3382         if (err)
3383                 return err;
3384 
3385 retry:
3386         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3387                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3388                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3389         if (IS_ERR(handle))
3390                 return PTR_ERR(handle);
3391 
3392         if (IS_DIRSYNC(dir))
3393                 ext4_handle_sync(handle);
3394 
3395         inode->i_ctime = current_time(inode);
3396         ext4_inc_count(handle, inode);
3397         ihold(inode);
3398 
3399         err = ext4_add_entry(handle, dentry, inode);
3400         if (!err) {
3401                 ext4_mark_inode_dirty(handle, inode);
3402                 /* this can happen only for tmpfile being
3403                  * linked the first time
3404                  */
3405                 if (inode->i_nlink == 1)
3406                         ext4_orphan_del(handle, inode);
3407                 d_instantiate(dentry, inode);
3408         } else {
3409                 drop_nlink(inode);
3410                 iput(inode);
3411         }
3412         ext4_journal_stop(handle);
3413         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3414                 goto retry;
3415         return err;
3416 }
3417 
3418 
3419 /*
3420  * Try to find buffer head where contains the parent block.
3421  * It should be the inode block if it is inlined or the 1st block
3422  * if it is a normal dir.
3423  */
3424 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3425                                         struct inode *inode,
3426                                         int *retval,
3427                                         struct ext4_dir_entry_2 **parent_de,
3428                                         int *inlined)
3429 {
3430         struct buffer_head *bh;
3431 
3432         if (!ext4_has_inline_data(inode)) {
3433                 /* The first directory block must not be a hole, so
3434                  * treat it as DIRENT_HTREE
3435                  */
3436                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3437                 if (IS_ERR(bh)) {
3438                         *retval = PTR_ERR(bh);
3439                         return NULL;
3440                 }
3441                 *parent_de = ext4_next_entry(
3442                                         (struct ext4_dir_entry_2 *)bh->b_data,
3443                                         inode->i_sb->s_blocksize);
3444                 return bh;
3445         }
3446 
3447         *inlined = 1;
3448         return ext4_get_first_inline_block(inode, parent_de, retval);
3449 }
3450 
3451 struct ext4_renament {
3452         struct inode *dir;
3453         struct dentry *dentry;
3454         struct inode *inode;
3455         bool is_dir;
3456         int dir_nlink_delta;
3457 
3458         /* entry for "dentry" */
3459         struct buffer_head *bh;
3460         struct ext4_dir_entry_2 *de;
3461         int inlined;
3462 
3463         /* entry for ".." in inode if it's a directory */
3464         struct buffer_head *dir_bh;
3465         struct ext4_dir_entry_2 *parent_de;
3466         int dir_inlined;
3467 };
3468 
3469 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3470 {
3471         int retval;
3472 
3473         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3474                                               &retval, &ent->parent_de,
3475                                               &ent->dir_inlined);
3476         if (!ent->dir_bh)
3477                 return retval;
3478         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3479                 return -EFSCORRUPTED;
3480         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3481         return ext4_journal_get_write_access(handle, ent->dir_bh);
3482 }
3483 
3484 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3485                                   unsigned dir_ino)
3486 {
3487         int retval;
3488 
3489         ent->parent_de->inode = cpu_to_le32(dir_ino);
3490         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3491         if (!ent->dir_inlined) {
3492                 if (is_dx(ent->inode)) {
3493                         retval = ext4_handle_dirty_dx_node(handle,
3494                                                            ent->inode,
3495                                                            ent->dir_bh);
3496                 } else {
3497                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3498                                                             ent->dir_bh);
3499                 }
3500         } else {
3501                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3502         }
3503         if (retval) {
3504                 ext4_std_error(ent->dir->i_sb, retval);
3505                 return retval;
3506         }
3507         return 0;
3508 }
3509 
3510 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3511                        unsigned ino, unsigned file_type)
3512 {
3513         int retval;
3514 
3515         BUFFER_TRACE(ent->bh, "get write access");
3516         retval = ext4_journal_get_write_access(handle, ent->bh);
3517         if (retval)
3518                 return retval;
3519         ent->de->inode = cpu_to_le32(ino);
3520         if (ext4_has_feature_filetype(ent->dir->i_sb))
3521                 ent->de->file_type = file_type;
3522         inode_inc_iversion(ent->dir);
3523         ent->dir->i_ctime = ent->dir->i_mtime =
3524                 current_time(ent->dir);
3525         ext4_mark_inode_dirty(handle, ent->dir);
3526         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3527         if (!ent->inlined) {
3528                 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3529                 if (unlikely(retval)) {
3530                         ext4_std_error(ent->dir->i_sb, retval);
3531                         return retval;
3532                 }
3533         }
3534         brelse(ent->bh);
3535         ent->bh = NULL;
3536 
3537         return 0;
3538 }
3539 
3540 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3541                                   const struct qstr *d_name)
3542 {
3543         int retval = -ENOENT;
3544         struct buffer_head *bh;
3545         struct ext4_dir_entry_2 *de;
3546 
3547         bh = ext4_find_entry(dir, d_name, &de, NULL);
3548         if (IS_ERR(bh))
3549                 return PTR_ERR(bh);
3550         if (bh) {
3551                 retval = ext4_delete_entry(handle, dir, de, bh);
3552                 brelse(bh);
3553         }
3554         return retval;
3555 }
3556 
3557 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3558                                int force_reread)
3559 {
3560         int retval;
3561         /*
3562          * ent->de could have moved from under us during htree split, so make
3563          * sure that we are deleting the right entry.  We might also be pointing
3564          * to a stale entry in the unused part of ent->bh so just checking inum
3565          * and the name isn't enough.
3566          */
3567         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3568             ent->de->name_len != ent->dentry->d_name.len ||
3569             strncmp(ent->de->name, ent->dentry->d_name.name,
3570                     ent->de->name_len) ||
3571             force_reread) {
3572                 retval = ext4_find_delete_entry(handle, ent->dir,
3573                                                 &ent->dentry->d_name);
3574         } else {
3575                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3576                 if (retval == -ENOENT) {
3577                         retval = ext4_find_delete_entry(handle, ent->dir,
3578                                                         &ent->dentry->d_name);
3579                 }
3580         }
3581 
3582         if (retval) {
3583                 ext4_warning_inode(ent->dir,
3584                                    "Deleting old file: nlink %d, error=%d",
3585                                    ent->dir->i_nlink, retval);
3586         }
3587 }
3588 
3589 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3590 {
3591         if (ent->dir_nlink_delta) {
3592                 if (ent->dir_nlink_delta == -1)
3593                         ext4_dec_count(handle, ent->dir);
3594                 else
3595                         ext4_inc_count(handle, ent->dir);
3596                 ext4_mark_inode_dirty(handle, ent->dir);
3597         }
3598 }
3599 
3600 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3601                                               int credits, handle_t **h)
3602 {
3603         struct inode *wh;
3604         handle_t *handle;
3605         int retries = 0;
3606 
3607         /*
3608          * for inode block, sb block, group summaries,
3609          * and inode bitmap
3610          */
3611         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3612                     EXT4_XATTR_TRANS_BLOCKS + 4);
3613 retry:
3614         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3615                                          &ent->dentry->d_name, 0, NULL,
3616                                          EXT4_HT_DIR, credits);
3617 
3618         handle = ext4_journal_current_handle();
3619         if (IS_ERR(wh)) {
3620                 if (handle)
3621                         ext4_journal_stop(handle);
3622                 if (PTR_ERR(wh) == -ENOSPC &&
3623                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3624                         goto retry;
3625         } else {
3626                 *h = handle;
3627                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3628                 wh->i_op = &ext4_special_inode_operations;
3629         }
3630         return wh;
3631 }
3632 
3633 /*
3634  * Anybody can rename anything with this: the permission checks are left to the
3635  * higher-level routines.
3636  *
3637  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3638  * while new_{dentry,inode) refers to the destination dentry/inode
3639  * This comes from rename(const char *oldpath, const char *newpath)
3640  */
3641 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3642                        struct inode *new_dir, struct dentry *new_dentry,
3643                        unsigned int flags)
3644 {
3645         handle_t *handle = NULL;
3646         struct ext4_renament old = {
3647                 .dir = old_dir,
3648                 .dentry = old_dentry,
3649                 .inode = d_inode(old_dentry),
3650         };
3651         struct ext4_renament new = {
3652                 .dir = new_dir,
3653                 .dentry = new_dentry,
3654                 .inode = d_inode(new_dentry),
3655         };
3656         int force_reread;
3657         int retval;
3658         struct inode *whiteout = NULL;
3659         int credits;
3660         u8 old_file_type;
3661 
3662         if (new.inode && new.inode->i_nlink == 0) {
3663                 EXT4_ERROR_INODE(new.inode,
3664                                  "target of rename is already freed");
3665                 return -EFSCORRUPTED;
3666         }
3667 
3668         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3669             (!projid_eq(EXT4_I(new_dir)->i_projid,
3670                         EXT4_I(old_dentry->d_inode)->i_projid)))
3671                 return -EXDEV;
3672 
3673         retval = dquot_initialize(old.dir);
3674         if (retval)
3675                 return retval;
3676         retval = dquot_initialize(new.dir);
3677         if (retval)
3678                 return retval;
3679 
3680         /* Initialize quotas before so that eventual writes go
3681          * in separate transaction */
3682         if (new.inode) {
3683                 retval = dquot_initialize(new.inode);
3684                 if (retval)
3685                         return retval;
3686         }
3687 
3688         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3689         if (IS_ERR(old.bh))
3690                 return PTR_ERR(old.bh);
3691         /*
3692          *  Check for inode number is _not_ due to possible IO errors.
3693          *  We might rmdir the source, keep it as pwd of some process
3694          *  and merrily kill the link to whatever was created under the
3695          *  same name. Goodbye sticky bit ;-<
3696          */
3697         retval = -ENOENT;
3698         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3699                 goto end_rename;
3700 
3701         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3702                                  &new.de, &new.inlined);
3703         if (IS_ERR(new.bh)) {
3704                 retval = PTR_ERR(new.bh);
3705                 new.bh = NULL;
3706                 goto end_rename;
3707         }
3708         if (new.bh) {
3709                 if (!new.inode) {
3710                         brelse(new.bh);
3711                         new.bh = NULL;
3712                 }
3713         }
3714         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3715                 ext4_alloc_da_blocks(old.inode);
3716 
3717         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3718                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3719         if (!(flags & RENAME_WHITEOUT)) {
3720                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3721                 if (IS_ERR(handle)) {
3722                         retval = PTR_ERR(handle);
3723                         handle = NULL;
3724                         goto end_rename;
3725                 }
3726         } else {
3727                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3728                 if (IS_ERR(whiteout)) {
3729                         retval = PTR_ERR(whiteout);
3730                         whiteout = NULL;
3731                         goto end_rename;
3732                 }
3733         }
3734 
3735         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3736                 ext4_handle_sync(handle);
3737 
3738         if (S_ISDIR(old.inode->i_mode)) {
3739                 if (new.inode) {
3740                         retval = -ENOTEMPTY;
3741                         if (!ext4_empty_dir(new.inode))
3742                                 goto end_rename;
3743                 } else {
3744                         retval = -EMLINK;
3745                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3746                                 goto end_rename;
3747                 }
3748                 retval = ext4_rename_dir_prepare(handle, &old);
3749                 if (retval)
3750                         goto end_rename;
3751         }
3752         /*
3753          * If we're renaming a file within an inline_data dir and adding or
3754          * setting the new dirent causes a conversion from inline_data to
3755          * extents/blockmap, we need to force the dirent delete code to
3756          * re-read the directory, or else we end up trying to delete a dirent
3757          * from what is now the extent tree root (or a block map).
3758          */
3759         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3760                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3761 
3762         old_file_type = old.de->file_type;
3763         if (whiteout) {
3764                 /*
3765                  * Do this before adding a new entry, so the old entry is sure
3766                  * to be still pointing to the valid old entry.
3767                  */
3768                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3769                                      EXT4_FT_CHRDEV);
3770                 if (retval)
3771                         goto end_rename;
3772                 ext4_mark_inode_dirty(handle, whiteout);
3773         }
3774         if (!new.bh) {
3775                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3776                 if (retval)
3777                         goto end_rename;
3778         } else {
3779                 retval = ext4_setent(handle, &new,
3780                                      old.inode->i_ino, old_file_type);
3781                 if (retval)
3782                         goto end_rename;
3783         }
3784         if (force_reread)
3785                 force_reread = !ext4_test_inode_flag(new.dir,
3786                                                      EXT4_INODE_INLINE_DATA);
3787 
3788         /*
3789          * Like most other Unix systems, set the ctime for inodes on a
3790          * rename.
3791          */
3792         old.inode->i_ctime = current_time(old.inode);
3793         ext4_mark_inode_dirty(handle, old.inode);
3794 
3795         if (!whiteout) {
3796                 /*
3797                  * ok, that's it
3798                  */
3799                 ext4_rename_delete(handle, &old, force_reread);
3800         }
3801 
3802         if (new.inode) {
3803                 ext4_dec_count(handle, new.inode);
3804                 new.inode->i_ctime = current_time(new.inode);
3805         }
3806         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3807         ext4_update_dx_flag(old.dir);
3808         if (old.dir_bh) {
3809                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3810                 if (retval)
3811                         goto end_rename;
3812 
3813                 ext4_dec_count(handle, old.dir);
3814                 if (new.inode) {
3815                         /* checked ext4_empty_dir above, can't have another
3816                          * parent, ext4_dec_count() won't work for many-linked
3817                          * dirs */
3818                         clear_nlink(new.inode);
3819                 } else {
3820                         ext4_inc_count(handle, new.dir);
3821                         ext4_update_dx_flag(new.dir);
3822                         ext4_mark_inode_dirty(handle, new.dir);
3823                 }
3824         }
3825         ext4_mark_inode_dirty(handle, old.dir);
3826         if (new.inode) {
3827                 ext4_mark_inode_dirty(handle, new.inode);
3828                 if (!new.inode->i_nlink)
3829                         ext4_orphan_add(handle, new.inode);
3830         }
3831         retval = 0;
3832 
3833 end_rename:
3834         brelse(old.dir_bh);
3835         brelse(old.bh);
3836         brelse(new.bh);
3837         if (whiteout) {
3838                 if (retval)
3839                         drop_nlink(whiteout);
3840                 unlock_new_inode(whiteout);
3841                 iput(whiteout);
3842         }
3843         if (handle)
3844                 ext4_journal_stop(handle);
3845         return retval;
3846 }
3847 
3848 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3849                              struct inode *new_dir, struct dentry *new_dentry)
3850 {
3851         handle_t *handle = NULL;
3852         struct ext4_renament old = {
3853                 .dir = old_dir,
3854                 .dentry = old_dentry,
3855                 .inode = d_inode(old_dentry),
3856         };
3857         struct ext4_renament new = {
3858                 .dir = new_dir,
3859                 .dentry = new_dentry,
3860                 .inode = d_inode(new_dentry),
3861         };
3862         u8 new_file_type;
3863         int retval;
3864         struct timespec64 ctime;
3865 
3866         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3867              !projid_eq(EXT4_I(new_dir)->i_projid,
3868                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3869             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3870              !projid_eq(EXT4_I(old_dir)->i_projid,
3871                         EXT4_I(new_dentry->d_inode)->i_projid)))
3872                 return -EXDEV;
3873 
3874         retval = dquot_initialize(old.dir);
3875         if (retval)
3876                 return retval;
3877         retval = dquot_initialize(new.dir);
3878         if (retval)
3879                 return retval;
3880 
3881         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3882                                  &old.de, &old.inlined);
3883         if (IS_ERR(old.bh))
3884                 return PTR_ERR(old.bh);
3885         /*
3886          *  Check for inode number is _not_ due to possible IO errors.
3887          *  We might rmdir the source, keep it as pwd of some process
3888          *  and merrily kill the link to whatever was created under the
3889          *  same name. Goodbye sticky bit ;-<
3890          */
3891         retval = -ENOENT;
3892         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3893                 goto end_rename;
3894 
3895         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3896                                  &new.de, &new.inlined);
3897         if (IS_ERR(new.bh)) {
3898                 retval = PTR_ERR(new.bh);
3899                 new.bh = NULL;
3900                 goto end_rename;
3901         }
3902 
3903         /* RENAME_EXCHANGE case: old *and* new must both exist */
3904         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3905                 goto end_rename;
3906 
3907         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3908                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3909                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3910         if (IS_ERR(handle)) {
3911                 retval = PTR_ERR(handle);
3912                 handle = NULL;
3913                 goto end_rename;
3914         }
3915 
3916         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3917                 ext4_handle_sync(handle);
3918 
3919         if (S_ISDIR(old.inode->i_mode)) {
3920                 old.is_dir = true;
3921                 retval = ext4_rename_dir_prepare(handle, &old);
3922                 if (retval)
3923                         goto end_rename;
3924         }
3925         if (S_ISDIR(new.inode->i_mode)) {
3926                 new.is_dir = true;
3927                 retval = ext4_rename_dir_prepare(handle, &new);
3928                 if (retval)
3929                         goto end_rename;
3930         }
3931 
3932         /*
3933          * Other than the special case of overwriting a directory, parents'
3934          * nlink only needs to be modified if this is a cross directory rename.
3935          */
3936         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3937                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3938                 new.dir_nlink_delta = -old.dir_nlink_delta;
3939                 retval = -EMLINK;
3940                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3941                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3942                         goto end_rename;
3943         }
3944 
3945         new_file_type = new.de->file_type;
3946         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3947         if (retval)
3948                 goto end_rename;
3949 
3950         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3951         if (retval)
3952                 goto end_rename;
3953 
3954         /*
3955          * Like most other Unix systems, set the ctime for inodes on a
3956          * rename.
3957          */
3958         ctime = current_time(old.inode);
3959         old.inode->i_ctime = ctime;
3960         new.inode->i_ctime = ctime;
3961         ext4_mark_inode_dirty(handle, old.inode);
3962         ext4_mark_inode_dirty(handle, new.inode);
3963 
3964         if (old.dir_bh) {
3965                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3966                 if (retval)
3967                         goto end_rename;
3968         }
3969         if (new.dir_bh) {
3970                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3971                 if (retval)
3972                         goto end_rename;
3973         }
3974         ext4_update_dir_count(handle, &old);
3975         ext4_update_dir_count(handle, &new);
3976         retval = 0;
3977 
3978 end_rename:
3979         brelse(old.dir_bh);
3980         brelse(new.dir_bh);
3981         brelse(old.bh);
3982         brelse(new.bh);
3983         if (handle)
3984                 ext4_journal_stop(handle);
3985         return retval;
3986 }
3987 
3988 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3989                         struct inode *new_dir, struct dentry *new_dentry,
3990                         unsigned int flags)
3991 {
3992         int err;
3993 
3994         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3995                 return -EIO;
3996 
3997         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3998                 return -EINVAL;
3999 
4000         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4001                                      flags);
4002         if (err)
4003                 return err;
4004 
4005         if (flags & RENAME_EXCHANGE) {
4006                 return ext4_cross_rename(old_dir, old_dentry,
4007                                          new_dir, new_dentry);
4008         }
4009 
4010         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4011 }
4012 
4013 /*
4014  * directories can handle most operations...
4015  */
4016 const struct inode_operations ext4_dir_inode_operations = {
4017         .create         = ext4_create,
4018         .lookup         = ext4_lookup,
4019         .link           = ext4_link,
4020         .unlink         = ext4_unlink,
4021         .symlink        = ext4_symlink,
4022         .mkdir          = ext4_mkdir,
4023         .rmdir          = ext4_rmdir,
4024         .mknod          = ext4_mknod,
4025         .tmpfile        = ext4_tmpfile,
4026         .rename         = ext4_rename2,
4027         .setattr        = ext4_setattr,
4028         .getattr        = ext4_getattr,
4029         .listxattr      = ext4_listxattr,
4030         .get_acl        = ext4_get_acl,
4031         .set_acl        = ext4_set_acl,
4032         .fiemap         = ext4_fiemap,
4033 };
4034 
4035 const struct inode_operations ext4_special_inode_operations = {
4036         .setattr        = ext4_setattr,
4037         .getattr        = ext4_getattr,
4038         .listxattr      = ext4_listxattr,
4039         .get_acl        = ext4_get_acl,
4040         .set_acl        = ext4_set_acl,
4041 };
/* [<][>][^][v][top][bottom][index][help] */
root/fs/ext4/namei.c

DEFINITIONS
