fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include "ext4.h"
  39 #include "ext4_jbd2.h"
  40
  41 #include "xattr.h"
  42 #include "acl.h"
  43
  44 #include <trace/events/ext4.h>
  45 /*
  46  * define how far ahead to read directories while searching them.
  47  */
  48 #define NAMEI_RA_CHUNKS  2
  49 #define NAMEI_RA_BLOCKS  4
  50 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  51
  52 static struct buffer_head *ext4_append(handle_t *handle,
  53                                         struct inode *inode,
  54                                         ext4_lblk_t *block)
  55 {
  56         struct buffer_head *bh;
  57         int err;
  58
  59         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  60                      ((inode->i_size >> 10) >=
  61                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  62                 return ERR_PTR(-ENOSPC);
  63
  64         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  65
  66         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  67         if (IS_ERR(bh))
  68                 return bh;
  69         inode->i_size += inode->i_sb->s_blocksize;
  70         EXT4_I(inode)->i_disksize = inode->i_size;
  71         BUFFER_TRACE(bh, "get_write_access");
  72         err = ext4_journal_get_write_access(handle, bh);
  73         if (err) {
  74                 brelse(bh);
  75                 ext4_std_error(inode->i_sb, err);
  76                 return ERR_PTR(err);
  77         }
  78         return bh;
  79 }
  80
  81 static int ext4_dx_csum_verify(struct inode *inode,
  82                                struct ext4_dir_entry *dirent);
  83
  84 /*
  85  * Hints to ext4_read_dirblock regarding whether we expect a directory
  86  * block being read to be an index block, or a block containing
  87  * directory entries (and if the latter, whether it was found via a
  88  * logical block in an htree index block).  This is used to control
  89  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  90  * directory block read from the storage device.  EITHER will means
  91  * the caller doesn't know what kind of directory block will be read,
  92  * so no specific verification will be done.
  93  */
  94 typedef enum {
  95         EITHER, INDEX, DIRENT, DIRENT_HTREE
  96 } dirblock_type_t;
  97
  98 #define ext4_read_dirblock(inode, block, type) \
  99         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 100
 101 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 102                                                 ext4_lblk_t block,
 103                                                 dirblock_type_t type,
 104                                                 const char *func,
 105                                                 unsigned int line)
 106 {
 107         struct buffer_head *bh;
 108         struct ext4_dir_entry *dirent;
 109         int is_dx_block = 0;
 110
 111         bh = ext4_bread(NULL, inode, block, 0);
 112         if (IS_ERR(bh)) {
 113                 __ext4_warning(inode->i_sb, func, line,
 114                                "inode #%lu: lblock %lu: comm %s: "
 115                                "error %ld reading directory block",
 116                                inode->i_ino, (unsigned long)block,
 117                                current->comm, PTR_ERR(bh));
 118
 119                 return bh;
 120         }
 121         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 122                 ext4_error_inode(inode, func, line, block,
 123                                  "Directory hole found for htree %s block",
 124                                  (type == INDEX) ? "index" : "leaf");
 125                 return ERR_PTR(-EFSCORRUPTED);
 126         }
 127         if (!bh)
 128                 return NULL;
 129         dirent = (struct ext4_dir_entry *) bh->b_data;
 130         /* Determine whether or not we have an index block */
 131         if (is_dx(inode)) {
 132                 if (block == 0)
 133                         is_dx_block = 1;
 134                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 135                                                 inode->i_sb->s_blocksize) ==
 136                          inode->i_sb->s_blocksize)
 137                         is_dx_block = 1;
 138         }
 139         if (!is_dx_block && type == INDEX) {
 140                 ext4_error_inode(inode, func, line, block,
 141                        "directory leaf block found instead of index block");
 142                 brelse(bh);
 143                 return ERR_PTR(-EFSCORRUPTED);
 144         }
 145         if (!ext4_has_metadata_csum(inode->i_sb) ||
 146             buffer_verified(bh))
 147                 return bh;
 148
 149         /*
 150          * An empty leaf block can get mistaken for a index block; for
 151          * this reason, we can only check the index checksum when the
 152          * caller is sure it should be an index block.
 153          */
 154         if (is_dx_block && type == INDEX) {
 155                 if (ext4_dx_csum_verify(inode, dirent))
 156                         set_buffer_verified(bh);
 157                 else {
 158                         ext4_error_inode(inode, func, line, block,
 159                                          "Directory index failed checksum");
 160                         brelse(bh);
 161                         return ERR_PTR(-EFSBADCRC);
 162                 }
 163         }
 164         if (!is_dx_block) {
 165                 if (ext4_dirent_csum_verify(inode, dirent))
 166                         set_buffer_verified(bh);
 167                 else {
 168                         ext4_error_inode(inode, func, line, block,
 169                                          "Directory block failed checksum");
 170                         brelse(bh);
 171                         return ERR_PTR(-EFSBADCRC);
 172                 }
 173         }
 174         return bh;
 175 }
 176
 177 #ifndef assert
 178 #define assert(test) J_ASSERT(test)
 179 #endif
 180
 181 #ifdef DX_DEBUG
 182 #define dxtrace(command) command
 183 #else
 184 #define dxtrace(command)
 185 #endif
 186
 187 struct fake_dirent
 188 {
 189         __le32 inode;
 190         __le16 rec_len;
 191         u8 name_len;
 192         u8 file_type;
 193 };
 194
 195 struct dx_countlimit
 196 {
 197         __le16 limit;
 198         __le16 count;
 199 };
 200
 201 struct dx_entry
 202 {
 203         __le32 hash;
 204         __le32 block;
 205 };
 206
 207 /*
 208  * dx_root_info is laid out so that if it should somehow get overlaid by a
 209  * dirent the two low bits of the hash version will be zero.  Therefore, the
 210  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 211  */
 212
 213 struct dx_root
 214 {
 215         struct fake_dirent dot;
 216         char dot_name[4];
 217         struct fake_dirent dotdot;
 218         char dotdot_name[4];
 219         struct dx_root_info
 220         {
 221                 __le32 reserved_zero;
 222                 u8 hash_version;
 223                 u8 info_length; /* 8 */
 224                 u8 indirect_levels;
 225                 u8 unused_flags;
 226         }
 227         info;
 228         struct dx_entry entries[0];
 229 };
 230
 231 struct dx_node
 232 {
 233         struct fake_dirent fake;
 234         struct dx_entry entries[0];
 235 };
 236
 237
 238 struct dx_frame
 239 {
 240         struct buffer_head *bh;
 241         struct dx_entry *entries;
 242         struct dx_entry *at;
 243 };
 244
 245 struct dx_map_entry
 246 {
 247         u32 hash;
 248         u16 offs;
 249         u16 size;
 250 };
 251
 252 /*
 253  * This goes at the end of each htree block.
 254  */
 255 struct dx_tail {
 256         u32 dt_reserved;
 257         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 258 };
 259
 260 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 261 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 262 static inline unsigned dx_get_hash(struct dx_entry *entry);
 263 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 264 static unsigned dx_get_count(struct dx_entry *entries);
 265 static unsigned dx_get_limit(struct dx_entry *entries);
 266 static void dx_set_count(struct dx_entry *entries, unsigned value);
 267 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 268 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 269 static unsigned dx_node_limit(struct inode *dir);
 270 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 271                                  struct inode *dir,
 272                                  struct dx_hash_info *hinfo,
 273                                  struct dx_frame *frame);
 274 static void dx_release(struct dx_frame *frames);
 275 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 276                        unsigned blocksize, struct dx_hash_info *hinfo,
 277                        struct dx_map_entry map[]);
 278 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 279 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 280                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 281 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 282 static void dx_insert_block(struct dx_frame *frame,
 283                                         u32 hash, ext4_lblk_t block);
 284 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 285                                  struct dx_frame *frame,
 286                                  struct dx_frame *frames,
 287                                  __u32 *start_hash);
 288 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 289                 struct ext4_filename *fname,
 290                 struct ext4_dir_entry_2 **res_dir);
 291 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 292                              struct inode *dir, struct inode *inode);
 293
 294 /* checksumming functions */
 295 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 296                             unsigned int blocksize)
 297 {
 298         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 299         t->det_rec_len = ext4_rec_len_to_disk(
 300                         sizeof(struct ext4_dir_entry_tail), blocksize);
 301         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 302 }
 303
 304 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 305 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 306                                                    struct ext4_dir_entry *de)
 307 {
 308         struct ext4_dir_entry_tail *t;
 309
 310 #ifdef PARANOID
 311         struct ext4_dir_entry *d, *top;
 312
 313         d = de;
 314         top = (struct ext4_dir_entry *)(((void *)de) +
 315                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 316                 sizeof(struct ext4_dir_entry_tail)));
 317         while (d < top && d->rec_len)
 318                 d = (struct ext4_dir_entry *)(((void *)d) +
 319                     le16_to_cpu(d->rec_len));
 320
 321         if (d != top)
 322                 return NULL;
 323
 324         t = (struct ext4_dir_entry_tail *)d;
 325 #else
 326         t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 327 #endif
 328
 329         if (t->det_reserved_zero1 ||
 330             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 331             t->det_reserved_zero2 ||
 332             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 333                 return NULL;
 334
 335         return t;
 336 }
 337
 338 static __le32 ext4_dirent_csum(struct inode *inode,
 339                                struct ext4_dir_entry *dirent, int size)
 340 {
 341         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 342         struct ext4_inode_info *ei = EXT4_I(inode);
 343         __u32 csum;
 344
 345         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 346         return cpu_to_le32(csum);
 347 }
 348
 349 #define warn_no_space_for_csum(inode)                                   \
 350         __warn_no_space_for_csum((inode), __func__, __LINE__)
 351
 352 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 353                                      unsigned int line)
 354 {
 355         __ext4_warning_inode(inode, func, line,
 356                 "No space for directory leaf checksum. Please run e2fsck -D.");
 357 }
 358
 359 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 360 {
 361         struct ext4_dir_entry_tail *t;
 362
 363         if (!ext4_has_metadata_csum(inode->i_sb))
 364                 return 1;
 365
 366         t = get_dirent_tail(inode, dirent);
 367         if (!t) {
 368                 warn_no_space_for_csum(inode);
 369                 return 0;
 370         }
 371
 372         if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 373                                                 (void *)t - (void *)dirent))
 374                 return 0;
 375
 376         return 1;
 377 }
 378
 379 static void ext4_dirent_csum_set(struct inode *inode,
 380                                  struct ext4_dir_entry *dirent)
 381 {
 382         struct ext4_dir_entry_tail *t;
 383
 384         if (!ext4_has_metadata_csum(inode->i_sb))
 385                 return;
 386
 387         t = get_dirent_tail(inode, dirent);
 388         if (!t) {
 389                 warn_no_space_for_csum(inode);
 390                 return;
 391         }
 392
 393         t->det_checksum = ext4_dirent_csum(inode, dirent,
 394                                            (void *)t - (void *)dirent);
 395 }
 396
 397 int ext4_handle_dirty_dirent_node(handle_t *handle,
 398                                   struct inode *inode,
 399                                   struct buffer_head *bh)
 400 {
 401         ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 402         return ext4_handle_dirty_metadata(handle, inode, bh);
 403 }
 404
 405 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 406                                                struct ext4_dir_entry *dirent,
 407                                                int *offset)
 408 {
 409         struct ext4_dir_entry *dp;
 410         struct dx_root_info *root;
 411         int count_offset;
 412
 413         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 414                 count_offset = 8;
 415         else if (le16_to_cpu(dirent->rec_len) == 12) {
 416                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 417                 if (le16_to_cpu(dp->rec_len) !=
 418                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 419                         return NULL;
 420                 root = (struct dx_root_info *)(((void *)dp + 12));
 421                 if (root->reserved_zero ||
 422                     root->info_length != sizeof(struct dx_root_info))
 423                         return NULL;
 424                 count_offset = 32;
 425         } else
 426                 return NULL;
 427
 428         if (offset)
 429                 *offset = count_offset;
 430         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 431 }
 432
 433 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 434                            int count_offset, int count, struct dx_tail *t)
 435 {
 436         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 437         struct ext4_inode_info *ei = EXT4_I(inode);
 438         __u32 csum;
 439         int size;
 440         __u32 dummy_csum = 0;
 441         int offset = offsetof(struct dx_tail, dt_checksum);
 442
 443         size = count_offset + (count * sizeof(struct dx_entry));
 444         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 445         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 446         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 447
 448         return cpu_to_le32(csum);
 449 }
 450
 451 static int ext4_dx_csum_verify(struct inode *inode,
 452                                struct ext4_dir_entry *dirent)
 453 {
 454         struct dx_countlimit *c;
 455         struct dx_tail *t;
 456         int count_offset, limit, count;
 457
 458         if (!ext4_has_metadata_csum(inode->i_sb))
 459                 return 1;
 460
 461         c = get_dx_countlimit(inode, dirent, &count_offset);
 462         if (!c) {
 463                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 464                 return 0;
 465         }
 466         limit = le16_to_cpu(c->limit);
 467         count = le16_to_cpu(c->count);
 468         if (count_offset + (limit * sizeof(struct dx_entry)) >
 469             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 470                 warn_no_space_for_csum(inode);
 471                 return 0;
 472         }
 473         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 474
 475         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 476                                             count, t))
 477                 return 0;
 478         return 1;
 479 }
 480
 481 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 482 {
 483         struct dx_countlimit *c;
 484         struct dx_tail *t;
 485         int count_offset, limit, count;
 486
 487         if (!ext4_has_metadata_csum(inode->i_sb))
 488                 return;
 489
 490         c = get_dx_countlimit(inode, dirent, &count_offset);
 491         if (!c) {
 492                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 493                 return;
 494         }
 495         limit = le16_to_cpu(c->limit);
 496         count = le16_to_cpu(c->count);
 497         if (count_offset + (limit * sizeof(struct dx_entry)) >
 498             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 499                 warn_no_space_for_csum(inode);
 500                 return;
 501         }
 502         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 503
 504         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 505 }
 506
 507 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 508                                             struct inode *inode,
 509                                             struct buffer_head *bh)
 510 {
 511         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 512         return ext4_handle_dirty_metadata(handle, inode, bh);
 513 }
 514
 515 /*
 516  * p is at least 6 bytes before the end of page
 517  */
 518 static inline struct ext4_dir_entry_2 *
 519 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 520 {
 521         return (struct ext4_dir_entry_2 *)((char *)p +
 522                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 523 }
 524
 525 /*
 526  * Future: use high four bits of block for coalesce-on-delete flags
 527  * Mask them off for now.
 528  */
 529
 530 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 531 {
 532         return le32_to_cpu(entry->block) & 0x0fffffff;
 533 }
 534
 535 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 536 {
 537         entry->block = cpu_to_le32(value);
 538 }
 539
 540 static inline unsigned dx_get_hash(struct dx_entry *entry)
 541 {
 542         return le32_to_cpu(entry->hash);
 543 }
 544
 545 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 546 {
 547         entry->hash = cpu_to_le32(value);
 548 }
 549
 550 static inline unsigned dx_get_count(struct dx_entry *entries)
 551 {
 552         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 553 }
 554
 555 static inline unsigned dx_get_limit(struct dx_entry *entries)
 556 {
 557         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 558 }
 559
 560 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 561 {
 562         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 563 }
 564
 565 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 566 {
 567         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 568 }
 569
 570 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 571 {
 572         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 573                 EXT4_DIR_REC_LEN(2) - infosize;
 574
 575         if (ext4_has_metadata_csum(dir->i_sb))
 576                 entry_space -= sizeof(struct dx_tail);
 577         return entry_space / sizeof(struct dx_entry);
 578 }
 579
 580 static inline unsigned dx_node_limit(struct inode *dir)
 581 {
 582         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 583
 584         if (ext4_has_metadata_csum(dir->i_sb))
 585                 entry_space -= sizeof(struct dx_tail);
 586         return entry_space / sizeof(struct dx_entry);
 587 }
 588
 589 /*
 590  * Debug
 591  */
 592 #ifdef DX_DEBUG
 593 static void dx_show_index(char * label, struct dx_entry *entries)
 594 {
 595         int i, n = dx_get_count (entries);
 596         printk(KERN_DEBUG "%s index", label);
 597         for (i = 0; i < n; i++) {
 598                 printk(KERN_CONT " %x->%lu",
 599                        i ? dx_get_hash(entries + i) : 0,
 600                        (unsigned long)dx_get_block(entries + i));
 601         }
 602         printk(KERN_CONT "\n");
 603 }
 604
 605 struct stats
 606 {
 607         unsigned names;
 608         unsigned space;
 609         unsigned bcount;
 610 };
 611
 612 static struct stats dx_show_leaf(struct inode *dir,
 613                                 struct dx_hash_info *hinfo,
 614                                 struct ext4_dir_entry_2 *de,
 615                                 int size, int show_names)
 616 {
 617         unsigned names = 0, space = 0;
 618         char *base = (char *) de;
 619         struct dx_hash_info h = *hinfo;
 620
 621         printk("names: ");
 622         while ((char *) de < base + size)
 623         {
 624                 if (de->inode)
 625                 {
 626                         if (show_names)
 627                         {
 628 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 629                                 int len;
 630                                 char *name;
 631                                 struct fscrypt_str fname_crypto_str =
 632                                         FSTR_INIT(NULL, 0);
 633                                 int res = 0;
 634
 635                                 name  = de->name;
 636                                 len = de->name_len;
 637                                 if (ext4_encrypted_inode(dir))
 638                                         res = fscrypt_get_encryption_info(dir);
 639                                 if (res) {
 640                                         printk(KERN_WARNING "Error setting up"
 641                                                " fname crypto: %d\n", res);
 642                                 }
 643                                 if (!fscrypt_has_encryption_key(dir)) {
 644                                         /* Directory is not encrypted */
 645                                         ext4fs_dirhash(de->name,
 646                                                 de->name_len, &h);
 647                                         printk("%*.s:(U)%x.%u ", len,
 648                                                name, h.hash,
 649                                                (unsigned) ((char *) de
 650                                                            - base));
 651                                 } else {
 652                                         struct fscrypt_str de_name =
 653                                                 FSTR_INIT(name, len);
 654
 655                                         /* Directory is encrypted */
 656                                         res = fscrypt_fname_alloc_buffer(
 657                                                 dir, len,
 658                                                 &fname_crypto_str);
 659                                         if (res)
 660                                                 printk(KERN_WARNING "Error "
 661                                                         "allocating crypto "
 662                                                         "buffer--skipping "
 663                                                         "crypto\n");
 664                                         res = fscrypt_fname_disk_to_usr(dir,
 665                                                 0, 0, &de_name,
 666                                                 &fname_crypto_str);
 667                                         if (res) {
 668                                                 printk(KERN_WARNING "Error "
 669                                                         "converting filename "
 670                                                         "from disk to usr"
 671                                                         "\n");
 672                                                 name = "??";
 673                                                 len = 2;
 674                                         } else {
 675                                                 name = fname_crypto_str.name;
 676                                                 len = fname_crypto_str.len;
 677                                         }
 678                                         ext4fs_dirhash(de->name, de->name_len,
 679                                                        &h);
 680                                         printk("%*.s:(E)%x.%u ", len, name,
 681                                                h.hash, (unsigned) ((char *) de
 682                                                                    - base));
 683                                         fscrypt_fname_free_buffer(
 684                                                         &fname_crypto_str);
 685                                 }
 686 #else
 687                                 int len = de->name_len;
 688                                 char *name = de->name;
 689                                 ext4fs_dirhash(de->name, de->name_len, &h);
 690                                 printk("%*.s:%x.%u ", len, name, h.hash,
 691                                        (unsigned) ((char *) de - base));
 692 #endif
 693                         }
 694                         space += EXT4_DIR_REC_LEN(de->name_len);
 695                         names++;
 696                 }
 697                 de = ext4_next_entry(de, size);
 698         }
 699         printk(KERN_CONT "(%i)\n", names);
 700         return (struct stats) { names, space, 1 };
 701 }
 702
 703 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 704                              struct dx_entry *entries, int levels)
 705 {
 706         unsigned blocksize = dir->i_sb->s_blocksize;
 707         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 708         unsigned bcount = 0;
 709         struct buffer_head *bh;
 710         printk("%i indexed blocks...\n", count);
 711         for (i = 0; i < count; i++, entries++)
 712         {
 713                 ext4_lblk_t block = dx_get_block(entries);
 714                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 715                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 716                 struct stats stats;
 717                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 718                 bh = ext4_bread(NULL,dir, block, 0);
 719                 if (!bh || IS_ERR(bh))
 720                         continue;
 721                 stats = levels?
 722                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 723                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 724                         bh->b_data, blocksize, 0);
 725                 names += stats.names;
 726                 space += stats.space;
 727                 bcount += stats.bcount;
 728                 brelse(bh);
 729         }
 730         if (bcount)
 731                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 732                        levels ? "" : "   ", names, space/bcount,
 733                        (space/bcount)*100/blocksize);
 734         return (struct stats) { names, space, bcount};
 735 }
 736 #endif /* DX_DEBUG */
 737
 738 /*
 739  * Probe for a directory leaf block to search.
 740  *
 741  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 742  * error in the directory index, and the caller should fall back to
 743  * searching the directory normally.  The callers of dx_probe **MUST**
 744  * check for this error code, and make sure it never gets reflected
 745  * back to userspace.
 746  */
 747 static struct dx_frame *
 748 dx_probe(struct ext4_filename *fname, struct inode *dir,
 749          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 750 {
 751         unsigned count, indirect;
 752         struct dx_entry *at, *entries, *p, *q, *m;
 753         struct dx_root *root;
 754         struct dx_frame *frame = frame_in;
 755         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 756         u32 hash;
 757
 758         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 759         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 760         if (IS_ERR(frame->bh))
 761                 return (struct dx_frame *) frame->bh;
 762
 763         root = (struct dx_root *) frame->bh->b_data;
 764         if (root->info.hash_version != DX_HASH_TEA &&
 765             root->info.hash_version != DX_HASH_HALF_MD4 &&
 766             root->info.hash_version != DX_HASH_LEGACY) {
 767                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 768                                    root->info.hash_version);
 769                 goto fail;
 770         }
 771         if (fname)
 772                 hinfo = &fname->hinfo;
 773         hinfo->hash_version = root->info.hash_version;
 774         if (hinfo->hash_version <= DX_HASH_TEA)
 775                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 776         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 777         if (fname && fname_name(fname))
 778                 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
 779         hash = hinfo->hash;
 780
 781         if (root->info.unused_flags & 1) {
 782                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 783                                    root->info.unused_flags);
 784                 goto fail;
 785         }
 786
 787         indirect = root->info.indirect_levels;
 788         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 789                 ext4_warning(dir->i_sb,
 790                              "Directory (ino: %lu) htree depth %#06x exceed"
 791                              "supported value", dir->i_ino,
 792                              ext4_dir_htree_level(dir->i_sb));
 793                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 794                         ext4_warning(dir->i_sb, "Enable large directory "
 795                                                 "feature to access it");
 796                 }
 797                 goto fail;
 798         }
 799
 800         entries = (struct dx_entry *)(((char *)&root->info) +
 801                                       root->info.info_length);
 802
 803         if (dx_get_limit(entries) != dx_root_limit(dir,
 804                                                    root->info.info_length)) {
 805                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 806                                    dx_get_limit(entries),
 807                                    dx_root_limit(dir, root->info.info_length));
 808                 goto fail;
 809         }
 810
 811         dxtrace(printk("Look up %x", hash));
 812         while (1) {
 813                 count = dx_get_count(entries);
 814                 if (!count || count > dx_get_limit(entries)) {
 815                         ext4_warning_inode(dir,
 816                                            "dx entry: count %u beyond limit %u",
 817                                            count, dx_get_limit(entries));
 818                         goto fail;
 819                 }
 820
 821                 p = entries + 1;
 822                 q = entries + count - 1;
 823                 while (p <= q) {
 824                         m = p + (q - p) / 2;
 825                         dxtrace(printk(KERN_CONT "."));
 826                         if (dx_get_hash(m) > hash)
 827                                 q = m - 1;
 828                         else
 829                                 p = m + 1;
 830                 }
 831
 832                 if (0) { // linear search cross check
 833                         unsigned n = count - 1;
 834                         at = entries;
 835                         while (n--)
 836                         {
 837                                 dxtrace(printk(KERN_CONT ","));
 838                                 if (dx_get_hash(++at) > hash)
 839                                 {
 840                                         at--;
 841                                         break;
 842                                 }
 843                         }
 844                         assert (at == p - 1);
 845                 }
 846
 847                 at = p - 1;
 848                 dxtrace(printk(KERN_CONT " %x->%u\n",
 849                                at == entries ? 0 : dx_get_hash(at),
 850                                dx_get_block(at)));
 851                 frame->entries = entries;
 852                 frame->at = at;
 853                 if (!indirect--)
 854                         return frame;
 855                 frame++;
 856                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 857                 if (IS_ERR(frame->bh)) {
 858                         ret_err = (struct dx_frame *) frame->bh;
 859                         frame->bh = NULL;
 860                         goto fail;
 861                 }
 862                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 863
 864                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 865                         ext4_warning_inode(dir,
 866                                 "dx entry: limit %u != node limit %u",
 867                                 dx_get_limit(entries), dx_node_limit(dir));
 868                         goto fail;
 869                 }
 870         }
 871 fail:
 872         while (frame >= frame_in) {
 873                 brelse(frame->bh);
 874                 frame--;
 875         }
 876
 877         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 878                 ext4_warning_inode(dir,
 879                         "Corrupt directory, running e2fsck is recommended");
 880         return ret_err;
 881 }
 882
 883 static void dx_release(struct dx_frame *frames)
 884 {
 885         struct dx_root_info *info;
 886         int i;
 887         unsigned int indirect_levels;
 888
 889         if (frames[0].bh == NULL)
 890                 return;
 891
 892         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 893         /* save local copy, "info" may be freed after brelse() */
 894         indirect_levels = info->indirect_levels;
 895         for (i = 0; i <= indirect_levels; i++) {
 896                 if (frames[i].bh == NULL)
 897                         break;
 898                 brelse(frames[i].bh);
 899                 frames[i].bh = NULL;
 900         }
 901 }
 902
 903 /*
 904  * This function increments the frame pointer to search the next leaf
 905  * block, and reads in the necessary intervening nodes if the search
 906  * should be necessary.  Whether or not the search is necessary is
 907  * controlled by the hash parameter.  If the hash value is even, then
 908  * the search is only continued if the next block starts with that
 909  * hash value.  This is used if we are searching for a specific file.
 910  *
 911  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 912  *
 913  * This function returns 1 if the caller should continue to search,
 914  * or 0 if it should not.  If there is an error reading one of the
 915  * index blocks, it will a negative error code.
 916  *
 917  * If start_hash is non-null, it will be filled in with the starting
 918  * hash of the next page.
 919  */
 920 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 921                                  struct dx_frame *frame,
 922                                  struct dx_frame *frames,
 923                                  __u32 *start_hash)
 924 {
 925         struct dx_frame *p;
 926         struct buffer_head *bh;
 927         int num_frames = 0;
 928         __u32 bhash;
 929
 930         p = frame;
 931         /*
 932          * Find the next leaf page by incrementing the frame pointer.
 933          * If we run out of entries in the interior node, loop around and
 934          * increment pointer in the parent node.  When we break out of
 935          * this loop, num_frames indicates the number of interior
 936          * nodes need to be read.
 937          */
 938         while (1) {
 939                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 940                         break;
 941                 if (p == frames)
 942                         return 0;
 943                 num_frames++;
 944                 p--;
 945         }
 946
 947         /*
 948          * If the hash is 1, then continue only if the next page has a
 949          * continuation hash of any value.  This is used for readdir
 950          * handling.  Otherwise, check to see if the hash matches the
 951          * desired contiuation hash.  If it doesn't, return since
 952          * there's no point to read in the successive index pages.
 953          */
 954         bhash = dx_get_hash(p->at);
 955         if (start_hash)
 956                 *start_hash = bhash;
 957         if ((hash & 1) == 0) {
 958                 if ((bhash & ~1) != hash)
 959                         return 0;
 960         }
 961         /*
 962          * If the hash is HASH_NB_ALWAYS, we always go to the next
 963          * block so no check is necessary
 964          */
 965         while (num_frames--) {
 966                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 967                 if (IS_ERR(bh))
 968                         return PTR_ERR(bh);
 969                 p++;
 970                 brelse(p->bh);
 971                 p->bh = bh;
 972                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 973         }
 974         return 1;
 975 }
 976
 977
 978 /*
 979  * This function fills a red-black tree with information from a
 980  * directory block.  It returns the number directory entries loaded
 981  * into the tree.  If there is an error it is returned in err.
 982  */
 983 static int htree_dirblock_to_tree(struct file *dir_file,
 984                                   struct inode *dir, ext4_lblk_t block,
 985                                   struct dx_hash_info *hinfo,
 986                                   __u32 start_hash, __u32 start_minor_hash)
 987 {
 988         struct buffer_head *bh;
 989         struct ext4_dir_entry_2 *de, *top;
 990         int err = 0, count = 0;
 991         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 992
 993         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 994                                                         (unsigned long)block));
 995         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
 996         if (IS_ERR(bh))
 997                 return PTR_ERR(bh);
 998
 999         de = (struct ext4_dir_entry_2 *) bh->b_data;
1000         top = (struct ext4_dir_entry_2 *) ((char *) de +
1001                                            dir->i_sb->s_blocksize -
1002                                            EXT4_DIR_REC_LEN(0));
1003 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1004         /* Check if the directory is encrypted */
1005         if (ext4_encrypted_inode(dir)) {
1006                 err = fscrypt_get_encryption_info(dir);
1007                 if (err < 0) {
1008                         brelse(bh);
1009                         return err;
1010                 }
1011                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1012                                                      &fname_crypto_str);
1013                 if (err < 0) {
1014                         brelse(bh);
1015                         return err;
1016                 }
1017         }
1018 #endif
1019         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1020                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1021                                 bh->b_data, bh->b_size,
1022                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1023                                          + ((char *)de - bh->b_data))) {
1024                         /* silently ignore the rest of the block */
1025                         break;
1026                 }
1027                 ext4fs_dirhash(de->name, de->name_len, hinfo);
1028                 if ((hinfo->hash < start_hash) ||
1029                     ((hinfo->hash == start_hash) &&
1030                      (hinfo->minor_hash < start_minor_hash)))
1031                         continue;
1032                 if (de->inode == 0)
1033                         continue;
1034                 if (!ext4_encrypted_inode(dir)) {
1035                         tmp_str.name = de->name;
1036                         tmp_str.len = de->name_len;
1037                         err = ext4_htree_store_dirent(dir_file,
1038                                    hinfo->hash, hinfo->minor_hash, de,
1039                                    &tmp_str);
1040                 } else {
1041                         int save_len = fname_crypto_str.len;
1042                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1043                                                                 de->name_len);
1044
1045                         /* Directory is encrypted */
1046                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1047                                         hinfo->minor_hash, &de_name,
1048                                         &fname_crypto_str);
1049                         if (err) {
1050                                 count = err;
1051                                 goto errout;
1052                         }
1053                         err = ext4_htree_store_dirent(dir_file,
1054                                    hinfo->hash, hinfo->minor_hash, de,
1055                                         &fname_crypto_str);
1056                         fname_crypto_str.len = save_len;
1057                 }
1058                 if (err != 0) {
1059                         count = err;
1060                         goto errout;
1061                 }
1062                 count++;
1063         }
1064 errout:
1065         brelse(bh);
1066 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1067         fscrypt_fname_free_buffer(&fname_crypto_str);
1068 #endif
1069         return count;
1070 }
1071
1072
1073 /*
1074  * This function fills a red-black tree with information from a
1075  * directory.  We start scanning the directory in hash order, starting
1076  * at start_hash and start_minor_hash.
1077  *
1078  * This function returns the number of entries inserted into the tree,
1079  * or a negative error code.
1080  */
1081 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1082                          __u32 start_minor_hash, __u32 *next_hash)
1083 {
1084         struct dx_hash_info hinfo;
1085         struct ext4_dir_entry_2 *de;
1086         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1087         struct inode *dir;
1088         ext4_lblk_t block;
1089         int count = 0;
1090         int ret, err;
1091         __u32 hashval;
1092         struct fscrypt_str tmp_str;
1093
1094         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1095                        start_hash, start_minor_hash));
1096         dir = file_inode(dir_file);
1097         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1098                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1099                 if (hinfo.hash_version <= DX_HASH_TEA)
1100                         hinfo.hash_version +=
1101                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1102                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1103                 if (ext4_has_inline_data(dir)) {
1104                         int has_inline_data = 1;
1105                         count = htree_inlinedir_to_tree(dir_file, dir, 0,
1106                                                         &hinfo, start_hash,
1107                                                         start_minor_hash,
1108                                                         &has_inline_data);
1109                         if (has_inline_data) {
1110                                 *next_hash = ~0;
1111                                 return count;
1112                         }
1113                 }
1114                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1115                                                start_hash, start_minor_hash);
1116                 *next_hash = ~0;
1117                 return count;
1118         }
1119         hinfo.hash = start_hash;
1120         hinfo.minor_hash = 0;
1121         frame = dx_probe(NULL, dir, &hinfo, frames);
1122         if (IS_ERR(frame))
1123                 return PTR_ERR(frame);
1124
1125         /* Add '.' and '..' from the htree header */
1126         if (!start_hash && !start_minor_hash) {
1127                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1128                 tmp_str.name = de->name;
1129                 tmp_str.len = de->name_len;
1130                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1131                                               de, &tmp_str);
1132                 if (err != 0)
1133                         goto errout;
1134                 count++;
1135         }
1136         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1137                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1138                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1139                 tmp_str.name = de->name;
1140                 tmp_str.len = de->name_len;
1141                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1142                                               de, &tmp_str);
1143                 if (err != 0)
1144                         goto errout;
1145                 count++;
1146         }
1147
1148         while (1) {
1149                 if (fatal_signal_pending(current)) {
1150                         err = -ERESTARTSYS;
1151                         goto errout;
1152                 }
1153                 cond_resched();
1154                 block = dx_get_block(frame->at);
1155                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1156                                              start_hash, start_minor_hash);
1157                 if (ret < 0) {
1158                         err = ret;
1159                         goto errout;
1160                 }
1161                 count += ret;
1162                 hashval = ~0;
1163                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1164                                             frame, frames, &hashval);
1165                 *next_hash = hashval;
1166                 if (ret < 0) {
1167                         err = ret;
1168                         goto errout;
1169                 }
1170                 /*
1171                  * Stop if:  (a) there are no more entries, or
1172                  * (b) we have inserted at least one entry and the
1173                  * next hash value is not a continuation
1174                  */
1175                 if ((ret == 0) ||
1176                     (count && ((hashval & 1) == 0)))
1177                         break;
1178         }
1179         dx_release(frames);
1180         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1181                        "next hash: %x\n", count, *next_hash));
1182         return count;
1183 errout:
1184         dx_release(frames);
1185         return (err);
1186 }
1187
1188 static inline int search_dirblock(struct buffer_head *bh,
1189                                   struct inode *dir,
1190                                   struct ext4_filename *fname,
1191                                   unsigned int offset,
1192                                   struct ext4_dir_entry_2 **res_dir)
1193 {
1194         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1195                                fname, offset, res_dir);
1196 }
1197
1198 /*
1199  * Directory block splitting, compacting
1200  */
1201
1202 /*
1203  * Create map of hash values, offsets, and sizes, stored at end of block.
1204  * Returns number of entries mapped.
1205  */
1206 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1207                        unsigned blocksize, struct dx_hash_info *hinfo,
1208                        struct dx_map_entry *map_tail)
1209 {
1210         int count = 0;
1211         char *base = (char *) de;
1212         struct dx_hash_info h = *hinfo;
1213
1214         while ((char *) de < base + blocksize) {
1215                 if (de->name_len && de->inode) {
1216                         ext4fs_dirhash(de->name, de->name_len, &h);
1217                         map_tail--;
1218                         map_tail->hash = h.hash;
1219                         map_tail->offs = ((char *) de - base)>>2;
1220                         map_tail->size = le16_to_cpu(de->rec_len);
1221                         count++;
1222                         cond_resched();
1223                 }
1224                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1225                 de = ext4_next_entry(de, blocksize);
1226         }
1227         return count;
1228 }
1229
1230 /* Sort map by hash value */
1231 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1232 {
1233         struct dx_map_entry *p, *q, *top = map + count - 1;
1234         int more;
1235         /* Combsort until bubble sort doesn't suck */
1236         while (count > 2) {
1237                 count = count*10/13;
1238                 if (count - 9 < 2) /* 9, 10 -> 11 */
1239                         count = 11;
1240                 for (p = top, q = p - count; q >= map; p--, q--)
1241                         if (p->hash < q->hash)
1242                                 swap(*p, *q);
1243         }
1244         /* Garden variety bubble sort */
1245         do {
1246                 more = 0;
1247                 q = top;
1248                 while (q-- > map) {
1249                         if (q[1].hash >= q[0].hash)
1250                                 continue;
1251                         swap(*(q+1), *q);
1252                         more = 1;
1253                 }
1254         } while(more);
1255 }
1256
1257 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1258 {
1259         struct dx_entry *entries = frame->entries;
1260         struct dx_entry *old = frame->at, *new = old + 1;
1261         int count = dx_get_count(entries);
1262
1263         assert(count < dx_get_limit(entries));
1264         assert(old < entries + count);
1265         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1266         dx_set_hash(new, hash);
1267         dx_set_block(new, block);
1268         dx_set_count(entries, count + 1);
1269 }
1270
1271 /*
1272  * Test whether a directory entry matches the filename being searched for.
1273  *
1274  * Return: %true if the directory entry matches, otherwise %false.
1275  */
1276 static inline bool ext4_match(const struct ext4_filename *fname,
1277                               const struct ext4_dir_entry_2 *de)
1278 {
1279         struct fscrypt_name f;
1280
1281         if (!de->inode)
1282                 return false;
1283
1284         f.usr_fname = fname->usr_fname;
1285         f.disk_name = fname->disk_name;
1286 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1287         f.crypto_buf = fname->crypto_buf;
1288 #endif
1289         return fscrypt_match_name(&f, de->name, de->name_len);
1290 }
1291
1292 /*
1293  * Returns 0 if not found, -1 on failure, and 1 on success
1294  */
1295 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1296                     struct inode *dir, struct ext4_filename *fname,
1297                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1298 {
1299         struct ext4_dir_entry_2 * de;
1300         char * dlimit;
1301         int de_len;
1302
1303         de = (struct ext4_dir_entry_2 *)search_buf;
1304         dlimit = search_buf + buf_size;
1305         while ((char *) de < dlimit) {
1306                 /* this code is executed quadratically often */
1307                 /* do minimal checking `by hand' */
1308                 if ((char *) de + de->name_len <= dlimit &&
1309                     ext4_match(fname, de)) {
1310                         /* found a match - just to be sure, do
1311                          * a full check */
1312                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1313                                                  buf_size, offset))
1314                                 return -1;
1315                         *res_dir = de;
1316                         return 1;
1317                 }
1318                 /* prevent looping on a bad block */
1319                 de_len = ext4_rec_len_from_disk(de->rec_len,
1320                                                 dir->i_sb->s_blocksize);
1321                 if (de_len <= 0)
1322                         return -1;
1323                 offset += de_len;
1324                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1325         }
1326         return 0;
1327 }
1328
1329 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1330                                struct ext4_dir_entry *de)
1331 {
1332         struct super_block *sb = dir->i_sb;
1333
1334         if (!is_dx(dir))
1335                 return 0;
1336         if (block == 0)
1337                 return 1;
1338         if (de->inode == 0 &&
1339             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1340                         sb->s_blocksize)
1341                 return 1;
1342         return 0;
1343 }
1344
1345 /*
1346  *      __ext4_find_entry()
1347  *
1348  * finds an entry in the specified directory with the wanted name. It
1349  * returns the cache buffer in which the entry was found, and the entry
1350  * itself (as a parameter - res_dir). It does NOT read the inode of the
1351  * entry - you'll have to do that yourself if you want to.
1352  *
1353  * The returned buffer_head has ->b_count elevated.  The caller is expected
1354  * to brelse() it when appropriate.
1355  */
1356 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1357                                              struct ext4_filename *fname,
1358                                              struct ext4_dir_entry_2 **res_dir,
1359                                              int *inlined)
1360 {
1361         struct super_block *sb;
1362         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1363         struct buffer_head *bh, *ret = NULL;
1364         ext4_lblk_t start, block;
1365         const u8 *name = fname->usr_fname->name;
1366         size_t ra_max = 0;      /* Number of bh's in the readahead
1367                                    buffer, bh_use[] */
1368         size_t ra_ptr = 0;      /* Current index into readahead
1369                                    buffer */
1370         ext4_lblk_t  nblocks;
1371         int i, namelen, retval;
1372
1373         *res_dir = NULL;
1374         sb = dir->i_sb;
1375         namelen = fname->usr_fname->len;
1376         if (namelen > EXT4_NAME_LEN)
1377                 return NULL;
1378
1379         if (ext4_has_inline_data(dir)) {
1380                 int has_inline_data = 1;
1381                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1382                                              &has_inline_data);
1383                 if (has_inline_data) {
1384                         if (inlined)
1385                                 *inlined = 1;
1386                         goto cleanup_and_exit;
1387                 }
1388         }
1389
1390         if ((namelen <= 2) && (name[0] == '.') &&
1391             (name[1] == '.' || name[1] == '\0')) {
1392                 /*
1393                  * "." or ".." will only be in the first block
1394                  * NFS may look up ".."; "." should be handled by the VFS
1395                  */
1396                 block = start = 0;
1397                 nblocks = 1;
1398                 goto restart;
1399         }
1400         if (is_dx(dir)) {
1401                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1402                 /*
1403                  * On success, or if the error was file not found,
1404                  * return.  Otherwise, fall back to doing a search the
1405                  * old fashioned way.
1406                  */
1407                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1408                         goto cleanup_and_exit;
1409                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1410                                "falling back\n"));
1411                 ret = NULL;
1412         }
1413         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1414         if (!nblocks) {
1415                 ret = NULL;
1416                 goto cleanup_and_exit;
1417         }
1418         start = EXT4_I(dir)->i_dir_start_lookup;
1419         if (start >= nblocks)
1420                 start = 0;
1421         block = start;
1422 restart:
1423         do {
1424                 /*
1425                  * We deal with the read-ahead logic here.
1426                  */
1427                 cond_resched();
1428                 if (ra_ptr >= ra_max) {
1429                         /* Refill the readahead buffer */
1430                         ra_ptr = 0;
1431                         if (block < start)
1432                                 ra_max = start - block;
1433                         else
1434                                 ra_max = nblocks - block;
1435                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1436                         retval = ext4_bread_batch(dir, block, ra_max,
1437                                                   false /* wait */, bh_use);
1438                         if (retval) {
1439                                 ret = ERR_PTR(retval);
1440                                 ra_max = 0;
1441                                 goto cleanup_and_exit;
1442                         }
1443                 }
1444                 if ((bh = bh_use[ra_ptr++]) == NULL)
1445                         goto next;
1446                 wait_on_buffer(bh);
1447                 if (!buffer_uptodate(bh)) {
1448                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1449                                          (unsigned long) block);
1450                         brelse(bh);
1451                         ret = ERR_PTR(-EIO);
1452                         goto cleanup_and_exit;
1453                 }
1454                 if (!buffer_verified(bh) &&
1455                     !is_dx_internal_node(dir, block,
1456                                          (struct ext4_dir_entry *)bh->b_data) &&
1457                     !ext4_dirent_csum_verify(dir,
1458                                 (struct ext4_dir_entry *)bh->b_data)) {
1459                         EXT4_ERROR_INODE(dir, "checksumming directory "
1460                                          "block %lu", (unsigned long)block);
1461                         brelse(bh);
1462                         ret = ERR_PTR(-EFSBADCRC);
1463                         goto cleanup_and_exit;
1464                 }
1465                 set_buffer_verified(bh);
1466                 i = search_dirblock(bh, dir, fname,
1467                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1468                 if (i == 1) {
1469                         EXT4_I(dir)->i_dir_start_lookup = block;
1470                         ret = bh;
1471                         goto cleanup_and_exit;
1472                 } else {
1473                         brelse(bh);
1474                         if (i < 0)
1475                                 goto cleanup_and_exit;
1476                 }
1477         next:
1478                 if (++block >= nblocks)
1479                         block = 0;
1480         } while (block != start);
1481
1482         /*
1483          * If the directory has grown while we were searching, then
1484          * search the last part of the directory before giving up.
1485          */
1486         block = nblocks;
1487         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1488         if (block < nblocks) {
1489                 start = 0;
1490                 goto restart;
1491         }
1492
1493 cleanup_and_exit:
1494         /* Clean up the read-ahead blocks */
1495         for (; ra_ptr < ra_max; ra_ptr++)
1496                 brelse(bh_use[ra_ptr]);
1497         return ret;
1498 }
1499
1500 static struct buffer_head *ext4_find_entry(struct inode *dir,
1501                                            const struct qstr *d_name,
1502                                            struct ext4_dir_entry_2 **res_dir,
1503                                            int *inlined)
1504 {
1505         int err;
1506         struct ext4_filename fname;
1507         struct buffer_head *bh;
1508
1509         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1510         if (err == -ENOENT)
1511                 return NULL;
1512         if (err)
1513                 return ERR_PTR(err);
1514
1515         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1516
1517         ext4_fname_free_filename(&fname);
1518         return bh;
1519 }
1520
1521 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1522                                              struct dentry *dentry,
1523                                              struct ext4_dir_entry_2 **res_dir)
1524 {
1525         int err;
1526         struct ext4_filename fname;
1527         struct buffer_head *bh;
1528
1529         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1530         if (err == -ENOENT)
1531                 return NULL;
1532         if (err)
1533                 return ERR_PTR(err);
1534
1535         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1536
1537         ext4_fname_free_filename(&fname);
1538         return bh;
1539 }
1540
1541 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1542                         struct ext4_filename *fname,
1543                         struct ext4_dir_entry_2 **res_dir)
1544 {
1545         struct super_block * sb = dir->i_sb;
1546         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1547         struct buffer_head *bh;
1548         ext4_lblk_t block;
1549         int retval;
1550
1551 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1552         *res_dir = NULL;
1553 #endif
1554         frame = dx_probe(fname, dir, NULL, frames);
1555         if (IS_ERR(frame))
1556                 return (struct buffer_head *) frame;
1557         do {
1558                 block = dx_get_block(frame->at);
1559                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1560                 if (IS_ERR(bh))
1561                         goto errout;
1562
1563                 retval = search_dirblock(bh, dir, fname,
1564                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1565                                          res_dir);
1566                 if (retval == 1)
1567                         goto success;
1568                 brelse(bh);
1569                 if (retval == -1) {
1570                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1571                         goto errout;
1572                 }
1573
1574                 /* Check to see if we should continue to search */
1575                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1576                                                frames, NULL);
1577                 if (retval < 0) {
1578                         ext4_warning_inode(dir,
1579                                 "error %d reading directory index block",
1580                                 retval);
1581                         bh = ERR_PTR(retval);
1582                         goto errout;
1583                 }
1584         } while (retval == 1);
1585
1586         bh = NULL;
1587 errout:
1588         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1589 success:
1590         dx_release(frames);
1591         return bh;
1592 }
1593
1594 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1595 {
1596         struct inode *inode;
1597         struct ext4_dir_entry_2 *de;
1598         struct buffer_head *bh;
1599
1600         if (dentry->d_name.len > EXT4_NAME_LEN)
1601                 return ERR_PTR(-ENAMETOOLONG);
1602
1603         bh = ext4_lookup_entry(dir, dentry, &de);
1604         if (IS_ERR(bh))
1605                 return (struct dentry *) bh;
1606         inode = NULL;
1607         if (bh) {
1608                 __u32 ino = le32_to_cpu(de->inode);
1609                 brelse(bh);
1610                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1611                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1612                         return ERR_PTR(-EFSCORRUPTED);
1613                 }
1614                 if (unlikely(ino == dir->i_ino)) {
1615                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1616                                          dentry);
1617                         return ERR_PTR(-EFSCORRUPTED);
1618                 }
1619                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1620                 if (inode == ERR_PTR(-ESTALE)) {
1621                         EXT4_ERROR_INODE(dir,
1622                                          "deleted inode referenced: %u",
1623                                          ino);
1624                         return ERR_PTR(-EFSCORRUPTED);
1625                 }
1626                 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1627                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1628                     !fscrypt_has_permitted_context(dir, inode)) {
1629                         ext4_warning(inode->i_sb,
1630                                      "Inconsistent encryption contexts: %lu/%lu",
1631                                      dir->i_ino, inode->i_ino);
1632                         iput(inode);
1633                         return ERR_PTR(-EPERM);
1634                 }
1635         }
1636         return d_splice_alias(inode, dentry);
1637 }
1638
1639
1640 struct dentry *ext4_get_parent(struct dentry *child)
1641 {
1642         __u32 ino;
1643         static const struct qstr dotdot = QSTR_INIT("..", 2);
1644         struct ext4_dir_entry_2 * de;
1645         struct buffer_head *bh;
1646
1647         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1648         if (IS_ERR(bh))
1649                 return (struct dentry *) bh;
1650         if (!bh)
1651                 return ERR_PTR(-ENOENT);
1652         ino = le32_to_cpu(de->inode);
1653         brelse(bh);
1654
1655         if (!ext4_valid_inum(child->d_sb, ino)) {
1656                 EXT4_ERROR_INODE(d_inode(child),
1657                                  "bad parent inode number: %u", ino);
1658                 return ERR_PTR(-EFSCORRUPTED);
1659         }
1660
1661         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1662 }
1663
1664 /*
1665  * Move count entries from end of map between two memory locations.
1666  * Returns pointer to last entry moved.
1667  */
1668 static struct ext4_dir_entry_2 *
1669 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1670                 unsigned blocksize)
1671 {
1672         unsigned rec_len = 0;
1673
1674         while (count--) {
1675                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1676                                                 (from + (map->offs<<2));
1677                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1678                 memcpy (to, de, rec_len);
1679                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1680                                 ext4_rec_len_to_disk(rec_len, blocksize);
1681                 de->inode = 0;
1682                 map++;
1683                 to += rec_len;
1684         }
1685         return (struct ext4_dir_entry_2 *) (to - rec_len);
1686 }
1687
1688 /*
1689  * Compact each dir entry in the range to the minimal rec_len.
1690  * Returns pointer to last entry in range.
1691  */
1692 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1693 {
1694         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1695         unsigned rec_len = 0;
1696
1697         prev = to = de;
1698         while ((char*)de < base + blocksize) {
1699                 next = ext4_next_entry(de, blocksize);
1700                 if (de->inode && de->name_len) {
1701                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1702                         if (de > to)
1703                                 memmove(to, de, rec_len);
1704                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1705                         prev = to;
1706                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1707                 }
1708                 de = next;
1709         }
1710         return prev;
1711 }
1712
1713 /*
1714  * Split a full leaf block to make room for a new dir entry.
1715  * Allocate a new block, and move entries so that they are approx. equally full.
1716  * Returns pointer to de in block into which the new entry will be inserted.
1717  */
1718 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1719                         struct buffer_head **bh,struct dx_frame *frame,
1720                         struct dx_hash_info *hinfo)
1721 {
1722         unsigned blocksize = dir->i_sb->s_blocksize;
1723         unsigned count, continued;
1724         struct buffer_head *bh2;
1725         ext4_lblk_t newblock;
1726         u32 hash2;
1727         struct dx_map_entry *map;
1728         char *data1 = (*bh)->b_data, *data2;
1729         unsigned split, move, size;
1730         struct ext4_dir_entry_2 *de = NULL, *de2;
1731         struct ext4_dir_entry_tail *t;
1732         int     csum_size = 0;
1733         int     err = 0, i;
1734
1735         if (ext4_has_metadata_csum(dir->i_sb))
1736                 csum_size = sizeof(struct ext4_dir_entry_tail);
1737
1738         bh2 = ext4_append(handle, dir, &newblock);
1739         if (IS_ERR(bh2)) {
1740                 brelse(*bh);
1741                 *bh = NULL;
1742                 return (struct ext4_dir_entry_2 *) bh2;
1743         }
1744
1745         BUFFER_TRACE(*bh, "get_write_access");
1746         err = ext4_journal_get_write_access(handle, *bh);
1747         if (err)
1748                 goto journal_error;
1749
1750         BUFFER_TRACE(frame->bh, "get_write_access");
1751         err = ext4_journal_get_write_access(handle, frame->bh);
1752         if (err)
1753                 goto journal_error;
1754
1755         data2 = bh2->b_data;
1756
1757         /* create map in the end of data2 block */
1758         map = (struct dx_map_entry *) (data2 + blocksize);
1759         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1760                              blocksize, hinfo, map);
1761         map -= count;
1762         dx_sort_map(map, count);
1763         /* Ensure that neither split block is over half full */
1764         size = 0;
1765         move = 0;
1766         for (i = count-1; i >= 0; i--) {
1767                 /* is more than half of this entry in 2nd half of the block? */
1768                 if (size + map[i].size/2 > blocksize/2)
1769                         break;
1770                 size += map[i].size;
1771                 move++;
1772         }
1773         /*
1774          * map index at which we will split
1775          *
1776          * If the sum of active entries didn't exceed half the block size, just
1777          * split it in half by count; each resulting block will have at least
1778          * half the space free.
1779          */
1780         if (i > 0)
1781                 split = count - move;
1782         else
1783                 split = count/2;
1784
1785         hash2 = map[split].hash;
1786         continued = hash2 == map[split - 1].hash;
1787         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1788                         (unsigned long)dx_get_block(frame->at),
1789                                         hash2, split, count-split));
1790
1791         /* Fancy dance to stay within two buffers */
1792         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1793                               blocksize);
1794         de = dx_pack_dirents(data1, blocksize);
1795         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1796                                            (char *) de,
1797                                            blocksize);
1798         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1799                                             (char *) de2,
1800                                             blocksize);
1801         if (csum_size) {
1802                 t = EXT4_DIRENT_TAIL(data2, blocksize);
1803                 initialize_dirent_tail(t, blocksize);
1804
1805                 t = EXT4_DIRENT_TAIL(data1, blocksize);
1806                 initialize_dirent_tail(t, blocksize);
1807         }
1808
1809         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1810                         blocksize, 1));
1811         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1812                         blocksize, 1));
1813
1814         /* Which block gets the new entry? */
1815         if (hinfo->hash >= hash2) {
1816                 swap(*bh, bh2);
1817                 de = de2;
1818         }
1819         dx_insert_block(frame, hash2 + continued, newblock);
1820         err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1821         if (err)
1822                 goto journal_error;
1823         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1824         if (err)
1825                 goto journal_error;
1826         brelse(bh2);
1827         dxtrace(dx_show_index("frame", frame->entries));
1828         return de;
1829
1830 journal_error:
1831         brelse(*bh);
1832         brelse(bh2);
1833         *bh = NULL;
1834         ext4_std_error(dir->i_sb, err);
1835         return ERR_PTR(err);
1836 }
1837
1838 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1839                       struct buffer_head *bh,
1840                       void *buf, int buf_size,
1841                       struct ext4_filename *fname,
1842                       struct ext4_dir_entry_2 **dest_de)
1843 {
1844         struct ext4_dir_entry_2 *de;
1845         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1846         int nlen, rlen;
1847         unsigned int offset = 0;
1848         char *top;
1849
1850         de = (struct ext4_dir_entry_2 *)buf;
1851         top = buf + buf_size - reclen;
1852         while ((char *) de <= top) {
1853                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1854                                          buf, buf_size, offset))
1855                         return -EFSCORRUPTED;
1856                 if (ext4_match(fname, de))
1857                         return -EEXIST;
1858                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1859                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1860                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1861                         break;
1862                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1863                 offset += rlen;
1864         }
1865         if ((char *) de > top)
1866                 return -ENOSPC;
1867
1868         *dest_de = de;
1869         return 0;
1870 }
1871
1872 void ext4_insert_dentry(struct inode *inode,
1873                         struct ext4_dir_entry_2 *de,
1874                         int buf_size,
1875                         struct ext4_filename *fname)
1876 {
1877
1878         int nlen, rlen;
1879
1880         nlen = EXT4_DIR_REC_LEN(de->name_len);
1881         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1882         if (de->inode) {
1883                 struct ext4_dir_entry_2 *de1 =
1884                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1885                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1886                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1887                 de = de1;
1888         }
1889         de->file_type = EXT4_FT_UNKNOWN;
1890         de->inode = cpu_to_le32(inode->i_ino);
1891         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1892         de->name_len = fname_len(fname);
1893         memcpy(de->name, fname_name(fname), fname_len(fname));
1894 }
1895
1896 /*
1897  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1898  * it points to a directory entry which is guaranteed to be large
1899  * enough for new directory entry.  If de is NULL, then
1900  * add_dirent_to_buf will attempt search the directory block for
1901  * space.  It will return -ENOSPC if no space is available, and -EIO
1902  * and -EEXIST if directory entry already exists.
1903  */
1904 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1905                              struct inode *dir,
1906                              struct inode *inode, struct ext4_dir_entry_2 *de,
1907                              struct buffer_head *bh)
1908 {
1909         unsigned int    blocksize = dir->i_sb->s_blocksize;
1910         int             csum_size = 0;
1911         int             err;
1912
1913         if (ext4_has_metadata_csum(inode->i_sb))
1914                 csum_size = sizeof(struct ext4_dir_entry_tail);
1915
1916         if (!de) {
1917                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1918                                         blocksize - csum_size, fname, &de);
1919                 if (err)
1920                         return err;
1921         }
1922         BUFFER_TRACE(bh, "get_write_access");
1923         err = ext4_journal_get_write_access(handle, bh);
1924         if (err) {
1925                 ext4_std_error(dir->i_sb, err);
1926                 return err;
1927         }
1928
1929         /* By now the buffer is marked for journaling */
1930         ext4_insert_dentry(inode, de, blocksize, fname);
1931
1932         /*
1933          * XXX shouldn't update any times until successful
1934          * completion of syscall, but too many callers depend
1935          * on this.
1936          *
1937          * XXX similarly, too many callers depend on
1938          * ext4_new_inode() setting the times, but error
1939          * recovery deletes the inode, so the worst that can
1940          * happen is that the times are slightly out of date
1941          * and/or different from the directory change time.
1942          */
1943         dir->i_mtime = dir->i_ctime = current_time(dir);
1944         ext4_update_dx_flag(dir);
1945         inode_inc_iversion(dir);
1946         ext4_mark_inode_dirty(handle, dir);
1947         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1948         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1949         if (err)
1950                 ext4_std_error(dir->i_sb, err);
1951         return 0;
1952 }
1953
1954 /*
1955  * This converts a one block unindexed directory to a 3 block indexed
1956  * directory, and adds the dentry to the indexed directory.
1957  */
1958 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1959                             struct inode *dir,
1960                             struct inode *inode, struct buffer_head *bh)
1961 {
1962         struct buffer_head *bh2;
1963         struct dx_root  *root;
1964         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1965         struct dx_entry *entries;
1966         struct ext4_dir_entry_2 *de, *de2;
1967         struct ext4_dir_entry_tail *t;
1968         char            *data1, *top;
1969         unsigned        len;
1970         int             retval;
1971         unsigned        blocksize;
1972         ext4_lblk_t  block;
1973         struct fake_dirent *fde;
1974         int csum_size = 0;
1975
1976         if (ext4_has_metadata_csum(inode->i_sb))
1977                 csum_size = sizeof(struct ext4_dir_entry_tail);
1978
1979         blocksize =  dir->i_sb->s_blocksize;
1980         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1981         BUFFER_TRACE(bh, "get_write_access");
1982         retval = ext4_journal_get_write_access(handle, bh);
1983         if (retval) {
1984                 ext4_std_error(dir->i_sb, retval);
1985                 brelse(bh);
1986                 return retval;
1987         }
1988         root = (struct dx_root *) bh->b_data;
1989
1990         /* The 0th block becomes the root, move the dirents out */
1991         fde = &root->dotdot;
1992         de = (struct ext4_dir_entry_2 *)((char *)fde +
1993                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1994         if ((char *) de >= (((char *) root) + blocksize)) {
1995                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1996                 brelse(bh);
1997                 return -EFSCORRUPTED;
1998         }
1999         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2000
2001         /* Allocate new block for the 0th block's dirents */
2002         bh2 = ext4_append(handle, dir, &block);
2003         if (IS_ERR(bh2)) {
2004                 brelse(bh);
2005                 return PTR_ERR(bh2);
2006         }
2007         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2008         data1 = bh2->b_data;
2009
2010         memcpy (data1, de, len);
2011         de = (struct ext4_dir_entry_2 *) data1;
2012         top = data1 + len;
2013         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2014                 de = de2;
2015         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2016                                            (char *) de,
2017                                            blocksize);
2018
2019         if (csum_size) {
2020                 t = EXT4_DIRENT_TAIL(data1, blocksize);
2021                 initialize_dirent_tail(t, blocksize);
2022         }
2023
2024         /* Initialize the root; the dot dirents already exist */
2025         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2026         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2027                                            blocksize);
2028         memset (&root->info, 0, sizeof(root->info));
2029         root->info.info_length = sizeof(root->info);
2030         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2031         entries = root->entries;
2032         dx_set_block(entries, 1);
2033         dx_set_count(entries, 1);
2034         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2035
2036         /* Initialize as for dx_probe */
2037         fname->hinfo.hash_version = root->info.hash_version;
2038         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2039                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2040         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2041         ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2042
2043         memset(frames, 0, sizeof(frames));
2044         frame = frames;
2045         frame->entries = entries;
2046         frame->at = entries;
2047         frame->bh = bh;
2048
2049         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2050         if (retval)
2051                 goto out_frames;
2052         retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2053         if (retval)
2054                 goto out_frames;
2055
2056         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2057         if (IS_ERR(de)) {
2058                 retval = PTR_ERR(de);
2059                 goto out_frames;
2060         }
2061
2062         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2063 out_frames:
2064         /*
2065          * Even if the block split failed, we have to properly write
2066          * out all the changes we did so far. Otherwise we can end up
2067          * with corrupted filesystem.
2068          */
2069         if (retval)
2070                 ext4_mark_inode_dirty(handle, dir);
2071         dx_release(frames);
2072         brelse(bh2);
2073         return retval;
2074 }
2075
2076 /*
2077  *      ext4_add_entry()
2078  *
2079  * adds a file entry to the specified directory, using the same
2080  * semantics as ext4_find_entry(). It returns NULL if it failed.
2081  *
2082  * NOTE!! The inode part of 'de' is left at 0 - which means you
2083  * may not sleep between calling this and putting something into
2084  * the entry, as someone else might have used it while you slept.
2085  */
2086 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2087                           struct inode *inode)
2088 {
2089         struct inode *dir = d_inode(dentry->d_parent);
2090         struct buffer_head *bh = NULL;
2091         struct ext4_dir_entry_2 *de;
2092         struct ext4_dir_entry_tail *t;
2093         struct super_block *sb;
2094         struct ext4_filename fname;
2095         int     retval;
2096         int     dx_fallback=0;
2097         unsigned blocksize;
2098         ext4_lblk_t block, blocks;
2099         int     csum_size = 0;
2100
2101         if (ext4_has_metadata_csum(inode->i_sb))
2102                 csum_size = sizeof(struct ext4_dir_entry_tail);
2103
2104         sb = dir->i_sb;
2105         blocksize = sb->s_blocksize;
2106         if (!dentry->d_name.len)
2107                 return -EINVAL;
2108
2109         if (fscrypt_is_nokey_name(dentry))
2110                 return -ENOKEY;
2111
2112         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2113         if (retval)
2114                 return retval;
2115
2116         if (ext4_has_inline_data(dir)) {
2117                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2118                 if (retval < 0)
2119                         goto out;
2120                 if (retval == 1) {
2121                         retval = 0;
2122                         goto out;
2123                 }
2124         }
2125
2126         if (is_dx(dir)) {
2127                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2128                 if (!retval || (retval != ERR_BAD_DX_DIR))
2129                         goto out;
2130                 /* Can we just ignore htree data? */
2131                 if (ext4_has_metadata_csum(sb)) {
2132                         EXT4_ERROR_INODE(dir,
2133                                 "Directory has corrupted htree index.");
2134                         retval = -EFSCORRUPTED;
2135                         goto out;
2136                 }
2137                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2138                 dx_fallback++;
2139                 ext4_mark_inode_dirty(handle, dir);
2140         }
2141         blocks = dir->i_size >> sb->s_blocksize_bits;
2142         for (block = 0; block < blocks; block++) {
2143                 bh = ext4_read_dirblock(dir, block, DIRENT);
2144                 if (bh == NULL) {
2145                         bh = ext4_bread(handle, dir, block,
2146                                         EXT4_GET_BLOCKS_CREATE);
2147                         goto add_to_new_block;
2148                 }
2149                 if (IS_ERR(bh)) {
2150                         retval = PTR_ERR(bh);
2151                         bh = NULL;
2152                         goto out;
2153                 }
2154                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2155                                            NULL, bh);
2156                 if (retval != -ENOSPC)
2157                         goto out;
2158
2159                 if (blocks == 1 && !dx_fallback &&
2160                     ext4_has_feature_dir_index(sb)) {
2161                         retval = make_indexed_dir(handle, &fname, dir,
2162                                                   inode, bh);
2163                         bh = NULL; /* make_indexed_dir releases bh */
2164                         goto out;
2165                 }
2166                 brelse(bh);
2167         }
2168         bh = ext4_append(handle, dir, &block);
2169 add_to_new_block:
2170         if (IS_ERR(bh)) {
2171                 retval = PTR_ERR(bh);
2172                 bh = NULL;
2173                 goto out;
2174         }
2175         de = (struct ext4_dir_entry_2 *) bh->b_data;
2176         de->inode = 0;
2177         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2178
2179         if (csum_size) {
2180                 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2181                 initialize_dirent_tail(t, blocksize);
2182         }
2183
2184         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2185 out:
2186         ext4_fname_free_filename(&fname);
2187         brelse(bh);
2188         if (retval == 0)
2189                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2190         return retval;
2191 }
2192
2193 /*
2194  * Returns 0 for success, or a negative error value
2195  */
2196 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2197                              struct inode *dir, struct inode *inode)
2198 {
2199         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2200         struct dx_entry *entries, *at;
2201         struct buffer_head *bh;
2202         struct super_block *sb = dir->i_sb;
2203         struct ext4_dir_entry_2 *de;
2204         int restart;
2205         int err;
2206
2207 again:
2208         restart = 0;
2209         frame = dx_probe(fname, dir, NULL, frames);
2210         if (IS_ERR(frame))
2211                 return PTR_ERR(frame);
2212         entries = frame->entries;
2213         at = frame->at;
2214         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2215         if (IS_ERR(bh)) {
2216                 err = PTR_ERR(bh);
2217                 bh = NULL;
2218                 goto cleanup;
2219         }
2220
2221         BUFFER_TRACE(bh, "get_write_access");
2222         err = ext4_journal_get_write_access(handle, bh);
2223         if (err)
2224                 goto journal_error;
2225
2226         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2227         if (err != -ENOSPC)
2228                 goto cleanup;
2229
2230         err = 0;
2231         /* Block full, should compress but for now just split */
2232         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2233                        dx_get_count(entries), dx_get_limit(entries)));
2234         /* Need to split index? */
2235         if (dx_get_count(entries) == dx_get_limit(entries)) {
2236                 ext4_lblk_t newblock;
2237                 int levels = frame - frames + 1;
2238                 unsigned int icount;
2239                 int add_level = 1;
2240                 struct dx_entry *entries2;
2241                 struct dx_node *node2;
2242                 struct buffer_head *bh2;
2243
2244                 while (frame > frames) {
2245                         if (dx_get_count((frame - 1)->entries) <
2246                             dx_get_limit((frame - 1)->entries)) {
2247                                 add_level = 0;
2248                                 break;
2249                         }
2250                         frame--; /* split higher index block */
2251                         at = frame->at;
2252                         entries = frame->entries;
2253                         restart = 1;
2254                 }
2255                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2256                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2257                                          "reach max htree level :%d",
2258                                          dir->i_ino, levels);
2259                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2260                                 ext4_warning(sb, "Large directory feature is "
2261                                                  "not enabled on this "
2262                                                  "filesystem");
2263                         }
2264                         err = -ENOSPC;
2265                         goto cleanup;
2266                 }
2267                 icount = dx_get_count(entries);
2268                 bh2 = ext4_append(handle, dir, &newblock);
2269                 if (IS_ERR(bh2)) {
2270                         err = PTR_ERR(bh2);
2271                         goto cleanup;
2272                 }
2273                 node2 = (struct dx_node *)(bh2->b_data);
2274                 entries2 = node2->entries;
2275                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2276                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2277                                                            sb->s_blocksize);
2278                 BUFFER_TRACE(frame->bh, "get_write_access");
2279                 err = ext4_journal_get_write_access(handle, frame->bh);
2280                 if (err)
2281                         goto journal_error;
2282                 if (!add_level) {
2283                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2284                         unsigned hash2 = dx_get_hash(entries + icount1);
2285                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2286                                        icount1, icount2));
2287
2288                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2289                         err = ext4_journal_get_write_access(handle,
2290                                                              (frame - 1)->bh);
2291                         if (err)
2292                                 goto journal_error;
2293
2294                         memcpy((char *) entries2, (char *) (entries + icount1),
2295                                icount2 * sizeof(struct dx_entry));
2296                         dx_set_count(entries, icount1);
2297                         dx_set_count(entries2, icount2);
2298                         dx_set_limit(entries2, dx_node_limit(dir));
2299
2300                         /* Which index block gets the new entry? */
2301                         if (at - entries >= icount1) {
2302                                 frame->at = at = at - entries - icount1 + entries2;
2303                                 frame->entries = entries = entries2;
2304                                 swap(frame->bh, bh2);
2305                         }
2306                         dx_insert_block((frame - 1), hash2, newblock);
2307                         dxtrace(dx_show_index("node", frame->entries));
2308                         dxtrace(dx_show_index("node",
2309                                ((struct dx_node *) bh2->b_data)->entries));
2310                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2311                         if (err)
2312                                 goto journal_error;
2313                         brelse (bh2);
2314                         err = ext4_handle_dirty_dx_node(handle, dir,
2315                                                    (frame - 1)->bh);
2316                         if (err)
2317                                 goto journal_error;
2318                         err = ext4_handle_dirty_dx_node(handle, dir,
2319                                                         frame->bh);
2320                         if (restart || err)
2321                                 goto journal_error;
2322                 } else {
2323                         struct dx_root *dxroot;
2324                         memcpy((char *) entries2, (char *) entries,
2325                                icount * sizeof(struct dx_entry));
2326                         dx_set_limit(entries2, dx_node_limit(dir));
2327
2328                         /* Set up root */
2329                         dx_set_count(entries, 1);
2330                         dx_set_block(entries + 0, newblock);
2331                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2332                         dxroot->info.indirect_levels += 1;
2333                         dxtrace(printk(KERN_DEBUG
2334                                        "Creating %d level index...\n",
2335                                        dxroot->info.indirect_levels));
2336                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2337                         if (err)
2338                                 goto journal_error;
2339                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2340                         brelse(bh2);
2341                         restart = 1;
2342                         goto journal_error;
2343                 }
2344         }
2345         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2346         if (IS_ERR(de)) {
2347                 err = PTR_ERR(de);
2348                 goto cleanup;
2349         }
2350         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2351         goto cleanup;
2352
2353 journal_error:
2354         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2355 cleanup:
2356         brelse(bh);
2357         dx_release(frames);
2358         /* @restart is true means htree-path has been changed, we need to
2359          * repeat dx_probe() to find out valid htree-path
2360          */
2361         if (restart && err == 0)
2362                 goto again;
2363         return err;
2364 }
2365
2366 /*
2367  * ext4_generic_delete_entry deletes a directory entry by merging it
2368  * with the previous entry
2369  */
2370 int ext4_generic_delete_entry(handle_t *handle,
2371                               struct inode *dir,
2372                               struct ext4_dir_entry_2 *de_del,
2373                               struct buffer_head *bh,
2374                               void *entry_buf,
2375                               int buf_size,
2376                               int csum_size)
2377 {
2378         struct ext4_dir_entry_2 *de, *pde;
2379         unsigned int blocksize = dir->i_sb->s_blocksize;
2380         int i;
2381
2382         i = 0;
2383         pde = NULL;
2384         de = (struct ext4_dir_entry_2 *)entry_buf;
2385         while (i < buf_size - csum_size) {
2386                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2387                                          entry_buf, buf_size, i))
2388                         return -EFSCORRUPTED;
2389                 if (de == de_del)  {
2390                         if (pde)
2391                                 pde->rec_len = ext4_rec_len_to_disk(
2392                                         ext4_rec_len_from_disk(pde->rec_len,
2393                                                                blocksize) +
2394                                         ext4_rec_len_from_disk(de->rec_len,
2395                                                                blocksize),
2396                                         blocksize);
2397                         else
2398                                 de->inode = 0;
2399                         inode_inc_iversion(dir);
2400                         return 0;
2401                 }
2402                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2403                 pde = de;
2404                 de = ext4_next_entry(de, blocksize);
2405         }
2406         return -ENOENT;
2407 }
2408
2409 static int ext4_delete_entry(handle_t *handle,
2410                              struct inode *dir,
2411                              struct ext4_dir_entry_2 *de_del,
2412                              struct buffer_head *bh)
2413 {
2414         int err, csum_size = 0;
2415
2416         if (ext4_has_inline_data(dir)) {
2417                 int has_inline_data = 1;
2418                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2419                                                &has_inline_data);
2420                 if (has_inline_data)
2421                         return err;
2422         }
2423
2424         if (ext4_has_metadata_csum(dir->i_sb))
2425                 csum_size = sizeof(struct ext4_dir_entry_tail);
2426
2427         BUFFER_TRACE(bh, "get_write_access");
2428         err = ext4_journal_get_write_access(handle, bh);
2429         if (unlikely(err))
2430                 goto out;
2431
2432         err = ext4_generic_delete_entry(handle, dir, de_del,
2433                                         bh, bh->b_data,
2434                                         dir->i_sb->s_blocksize, csum_size);
2435         if (err)
2436                 goto out;
2437
2438         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2439         err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2440         if (unlikely(err))
2441                 goto out;
2442
2443         return 0;
2444 out:
2445         if (err != -ENOENT)
2446                 ext4_std_error(dir->i_sb, err);
2447         return err;
2448 }
2449
2450 /*
2451  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2452  * since this indicates that nlinks count was previously 1 to avoid overflowing
2453  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2454  * that subdirectory link counts are not being maintained accurately.
2455  *
2456  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2457  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2458  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2459  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2460  */
2461 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2462 {
2463         inc_nlink(inode);
2464         if (is_dx(inode) &&
2465             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2466                 set_nlink(inode, 1);
2467 }
2468
2469 /*
2470  * If a directory had nlink == 1, then we should let it be 1. This indicates
2471  * directory has >EXT4_LINK_MAX subdirs.
2472  */
2473 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2474 {
2475         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2476                 drop_nlink(inode);
2477 }
2478
2479
2480 static int ext4_add_nondir(handle_t *handle,
2481                 struct dentry *dentry, struct inode *inode)
2482 {
2483         int err = ext4_add_entry(handle, dentry, inode);
2484         if (!err) {
2485                 ext4_mark_inode_dirty(handle, inode);
2486                 d_instantiate_new(dentry, inode);
2487                 return 0;
2488         }
2489         drop_nlink(inode);
2490         unlock_new_inode(inode);
2491         iput(inode);
2492         return err;
2493 }
2494
2495 /*
2496  * By the time this is called, we already have created
2497  * the directory cache entry for the new file, but it
2498  * is so far negative - it has no inode.
2499  *
2500  * If the create succeeds, we fill in the inode information
2501  * with d_instantiate().
2502  */
2503 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2504                        bool excl)
2505 {
2506         handle_t *handle;
2507         struct inode *inode;
2508         int err, credits, retries = 0;
2509
2510         err = dquot_initialize(dir);
2511         if (err)
2512                 return err;
2513
2514         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2515                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2516 retry:
2517         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2518                                             NULL, EXT4_HT_DIR, credits);
2519         handle = ext4_journal_current_handle();
2520         err = PTR_ERR(inode);
2521         if (!IS_ERR(inode)) {
2522                 inode->i_op = &ext4_file_inode_operations;
2523                 inode->i_fop = &ext4_file_operations;
2524                 ext4_set_aops(inode);
2525                 err = ext4_add_nondir(handle, dentry, inode);
2526                 if (!err && IS_DIRSYNC(dir))
2527                         ext4_handle_sync(handle);
2528         }
2529         if (handle)
2530                 ext4_journal_stop(handle);
2531         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2532                 goto retry;
2533         return err;
2534 }
2535
2536 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2537                       umode_t mode, dev_t rdev)
2538 {
2539         handle_t *handle;
2540         struct inode *inode;
2541         int err, credits, retries = 0;
2542
2543         err = dquot_initialize(dir);
2544         if (err)
2545                 return err;
2546
2547         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2548                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2549 retry:
2550         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2551                                             NULL, EXT4_HT_DIR, credits);
2552         handle = ext4_journal_current_handle();
2553         err = PTR_ERR(inode);
2554         if (!IS_ERR(inode)) {
2555                 init_special_inode(inode, inode->i_mode, rdev);
2556                 inode->i_op = &ext4_special_inode_operations;
2557                 err = ext4_add_nondir(handle, dentry, inode);
2558                 if (!err && IS_DIRSYNC(dir))
2559                         ext4_handle_sync(handle);
2560         }
2561         if (handle)
2562                 ext4_journal_stop(handle);
2563         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2564                 goto retry;
2565         return err;
2566 }
2567
2568 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2569 {
2570         handle_t *handle;
2571         struct inode *inode;
2572         int err, retries = 0;
2573
2574         err = dquot_initialize(dir);
2575         if (err)
2576                 return err;
2577
2578 retry:
2579         inode = ext4_new_inode_start_handle(dir, mode,
2580                                             NULL, 0, NULL,
2581                                             EXT4_HT_DIR,
2582                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2583                           4 + EXT4_XATTR_TRANS_BLOCKS);
2584         handle = ext4_journal_current_handle();
2585         err = PTR_ERR(inode);
2586         if (!IS_ERR(inode)) {
2587                 inode->i_op = &ext4_file_inode_operations;
2588                 inode->i_fop = &ext4_file_operations;
2589                 ext4_set_aops(inode);
2590                 d_tmpfile(dentry, inode);
2591                 err = ext4_orphan_add(handle, inode);
2592                 if (err)
2593                         goto err_unlock_inode;
2594                 mark_inode_dirty(inode);
2595                 unlock_new_inode(inode);
2596         }
2597         if (handle)
2598                 ext4_journal_stop(handle);
2599         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2600                 goto retry;
2601         return err;
2602 err_unlock_inode:
2603         ext4_journal_stop(handle);
2604         unlock_new_inode(inode);
2605         return err;
2606 }
2607
2608 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2609                           struct ext4_dir_entry_2 *de,
2610                           int blocksize, int csum_size,
2611                           unsigned int parent_ino, int dotdot_real_len)
2612 {
2613         de->inode = cpu_to_le32(inode->i_ino);
2614         de->name_len = 1;
2615         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2616                                            blocksize);
2617         strcpy(de->name, ".");
2618         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2619
2620         de = ext4_next_entry(de, blocksize);
2621         de->inode = cpu_to_le32(parent_ino);
2622         de->name_len = 2;
2623         if (!dotdot_real_len)
2624                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2625                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2626                                         blocksize);
2627         else
2628                 de->rec_len = ext4_rec_len_to_disk(
2629                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2630         strcpy(de->name, "..");
2631         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2632
2633         return ext4_next_entry(de, blocksize);
2634 }
2635
2636 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2637                              struct inode *inode)
2638 {
2639         struct buffer_head *dir_block = NULL;
2640         struct ext4_dir_entry_2 *de;
2641         struct ext4_dir_entry_tail *t;
2642         ext4_lblk_t block = 0;
2643         unsigned int blocksize = dir->i_sb->s_blocksize;
2644         int csum_size = 0;
2645         int err;
2646
2647         if (ext4_has_metadata_csum(dir->i_sb))
2648                 csum_size = sizeof(struct ext4_dir_entry_tail);
2649
2650         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2651                 err = ext4_try_create_inline_dir(handle, dir, inode);
2652                 if (err < 0 && err != -ENOSPC)
2653                         goto out;
2654                 if (!err)
2655                         goto out;
2656         }
2657
2658         inode->i_size = 0;
2659         dir_block = ext4_append(handle, inode, &block);
2660         if (IS_ERR(dir_block))
2661                 return PTR_ERR(dir_block);
2662         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2663         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2664         set_nlink(inode, 2);
2665         if (csum_size) {
2666                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2667                 initialize_dirent_tail(t, blocksize);
2668         }
2669
2670         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2671         err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2672         if (err)
2673                 goto out;
2674         set_buffer_verified(dir_block);
2675 out:
2676         brelse(dir_block);
2677         return err;
2678 }
2679
2680 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2681 {
2682         handle_t *handle;
2683         struct inode *inode;
2684         int err, credits, retries = 0;
2685
2686         if (EXT4_DIR_LINK_MAX(dir))
2687                 return -EMLINK;
2688
2689         err = dquot_initialize(dir);
2690         if (err)
2691                 return err;
2692
2693         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2694                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2695 retry:
2696         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2697                                             &dentry->d_name,
2698                                             0, NULL, EXT4_HT_DIR, credits);
2699         handle = ext4_journal_current_handle();
2700         err = PTR_ERR(inode);
2701         if (IS_ERR(inode))
2702                 goto out_stop;
2703
2704         inode->i_op = &ext4_dir_inode_operations;
2705         inode->i_fop = &ext4_dir_operations;
2706         err = ext4_init_new_dir(handle, dir, inode);
2707         if (err)
2708                 goto out_clear_inode;
2709         err = ext4_mark_inode_dirty(handle, inode);
2710         if (!err)
2711                 err = ext4_add_entry(handle, dentry, inode);
2712         if (err) {
2713 out_clear_inode:
2714                 clear_nlink(inode);
2715                 unlock_new_inode(inode);
2716                 ext4_mark_inode_dirty(handle, inode);
2717                 iput(inode);
2718                 goto out_stop;
2719         }
2720         ext4_inc_count(handle, dir);
2721         ext4_update_dx_flag(dir);
2722         err = ext4_mark_inode_dirty(handle, dir);
2723         if (err)
2724                 goto out_clear_inode;
2725         d_instantiate_new(dentry, inode);
2726         if (IS_DIRSYNC(dir))
2727                 ext4_handle_sync(handle);
2728
2729 out_stop:
2730         if (handle)
2731                 ext4_journal_stop(handle);
2732         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2733                 goto retry;
2734         return err;
2735 }
2736
2737 /*
2738  * routine to check that the specified directory is empty (for rmdir)
2739  */
2740 bool ext4_empty_dir(struct inode *inode)
2741 {
2742         unsigned int offset;
2743         struct buffer_head *bh;
2744         struct ext4_dir_entry_2 *de;
2745         struct super_block *sb;
2746
2747         if (ext4_has_inline_data(inode)) {
2748                 int has_inline_data = 1;
2749                 int ret;
2750
2751                 ret = empty_inline_dir(inode, &has_inline_data);
2752                 if (has_inline_data)
2753                         return ret;
2754         }
2755
2756         sb = inode->i_sb;
2757         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2758                 EXT4_ERROR_INODE(inode, "invalid size");
2759                 return true;
2760         }
2761         /* The first directory block must not be a hole,
2762          * so treat it as DIRENT_HTREE
2763          */
2764         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2765         if (IS_ERR(bh))
2766                 return true;
2767
2768         de = (struct ext4_dir_entry_2 *) bh->b_data;
2769         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2770                                  0) ||
2771             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2772                 ext4_warning_inode(inode, "directory missing '.'");
2773                 brelse(bh);
2774                 return true;
2775         }
2776         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2777         de = ext4_next_entry(de, sb->s_blocksize);
2778         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2779                                  offset) ||
2780             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2781                 ext4_warning_inode(inode, "directory missing '..'");
2782                 brelse(bh);
2783                 return true;
2784         }
2785         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2786         while (offset < inode->i_size) {
2787                 if (!(offset & (sb->s_blocksize - 1))) {
2788                         unsigned int lblock;
2789                         brelse(bh);
2790                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2791                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2792                         if (bh == NULL) {
2793                                 offset += sb->s_blocksize;
2794                                 continue;
2795                         }
2796                         if (IS_ERR(bh))
2797                                 return true;
2798                 }
2799                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2800                                         (offset & (sb->s_blocksize - 1)));
2801                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2802                                          bh->b_data, bh->b_size, offset)) {
2803                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2804                         continue;
2805                 }
2806                 if (le32_to_cpu(de->inode)) {
2807                         brelse(bh);
2808                         return false;
2809                 }
2810                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2811         }
2812         brelse(bh);
2813         return true;
2814 }
2815
2816 /*
2817  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2818  * such inodes, starting at the superblock, in case we crash before the
2819  * file is closed/deleted, or in case the inode truncate spans multiple
2820  * transactions and the last transaction is not recovered after a crash.
2821  *
2822  * At filesystem recovery time, we walk this list deleting unlinked
2823  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2824  *
2825  * Orphan list manipulation functions must be called under i_mutex unless
2826  * we are just creating the inode or deleting it.
2827  */
2828 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2829 {
2830         struct super_block *sb = inode->i_sb;
2831         struct ext4_sb_info *sbi = EXT4_SB(sb);
2832         struct ext4_iloc iloc;
2833         int err = 0, rc;
2834         bool dirty = false;
2835
2836         if (!sbi->s_journal || is_bad_inode(inode))
2837                 return 0;
2838
2839         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2840                      !inode_is_locked(inode));
2841         /*
2842          * Exit early if inode already is on orphan list. This is a big speedup
2843          * since we don't have to contend on the global s_orphan_lock.
2844          */
2845         if (!list_empty(&EXT4_I(inode)->i_orphan))
2846                 return 0;
2847
2848         /*
2849          * Orphan handling is only valid for files with data blocks
2850          * being truncated, or files being unlinked. Note that we either
2851          * hold i_mutex, or the inode can not be referenced from outside,
2852          * so i_nlink should not be bumped due to race
2853          */
2854         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2855                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2856
2857         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2858         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2859         if (err)
2860                 goto out;
2861
2862         err = ext4_reserve_inode_write(handle, inode, &iloc);
2863         if (err)
2864                 goto out;
2865
2866         mutex_lock(&sbi->s_orphan_lock);
2867         /*
2868          * Due to previous errors inode may be already a part of on-disk
2869          * orphan list. If so skip on-disk list modification.
2870          */
2871         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2872             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2873                 /* Insert this inode at the head of the on-disk orphan list */
2874                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2875                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2876                 dirty = true;
2877         }
2878         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2879         mutex_unlock(&sbi->s_orphan_lock);
2880
2881         if (dirty) {
2882                 err = ext4_handle_dirty_super(handle, sb);
2883                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2884                 if (!err)
2885                         err = rc;
2886                 if (err) {
2887                         /*
2888                          * We have to remove inode from in-memory list if
2889                          * addition to on disk orphan list failed. Stray orphan
2890                          * list entries can cause panics at unmount time.
2891                          */
2892                         mutex_lock(&sbi->s_orphan_lock);
2893                         list_del_init(&EXT4_I(inode)->i_orphan);
2894                         mutex_unlock(&sbi->s_orphan_lock);
2895                 }
2896         } else
2897                 brelse(iloc.bh);
2898
2899         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2900         jbd_debug(4, "orphan inode %lu will point to %d\n",
2901                         inode->i_ino, NEXT_ORPHAN(inode));
2902 out:
2903         ext4_std_error(sb, err);
2904         return err;
2905 }
2906
2907 /*
2908  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2909  * of such inodes stored on disk, because it is finally being cleaned up.
2910  */
2911 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2912 {
2913         struct list_head *prev;
2914         struct ext4_inode_info *ei = EXT4_I(inode);
2915         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2916         __u32 ino_next;
2917         struct ext4_iloc iloc;
2918         int err = 0;
2919
2920         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2921                 return 0;
2922
2923         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2924                      !inode_is_locked(inode));
2925         /* Do this quick check before taking global s_orphan_lock. */
2926         if (list_empty(&ei->i_orphan))
2927                 return 0;
2928
2929         if (handle) {
2930                 /* Grab inode buffer early before taking global s_orphan_lock */
2931                 err = ext4_reserve_inode_write(handle, inode, &iloc);
2932         }
2933
2934         mutex_lock(&sbi->s_orphan_lock);
2935         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2936
2937         prev = ei->i_orphan.prev;
2938         list_del_init(&ei->i_orphan);
2939
2940         /* If we're on an error path, we may not have a valid
2941          * transaction handle with which to update the orphan list on
2942          * disk, but we still need to remove the inode from the linked
2943          * list in memory. */
2944         if (!handle || err) {
2945                 mutex_unlock(&sbi->s_orphan_lock);
2946                 goto out_err;
2947         }
2948
2949         ino_next = NEXT_ORPHAN(inode);
2950         if (prev == &sbi->s_orphan) {
2951                 jbd_debug(4, "superblock will point to %u\n", ino_next);
2952                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2953                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2954                 if (err) {
2955                         mutex_unlock(&sbi->s_orphan_lock);
2956                         goto out_brelse;
2957                 }
2958                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2959                 mutex_unlock(&sbi->s_orphan_lock);
2960                 err = ext4_handle_dirty_super(handle, inode->i_sb);
2961         } else {
2962                 struct ext4_iloc iloc2;
2963                 struct inode *i_prev =
2964                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2965
2966                 jbd_debug(4, "orphan inode %lu will point to %u\n",
2967                           i_prev->i_ino, ino_next);
2968                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2969                 if (err) {
2970                         mutex_unlock(&sbi->s_orphan_lock);
2971                         goto out_brelse;
2972                 }
2973                 NEXT_ORPHAN(i_prev) = ino_next;
2974                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2975                 mutex_unlock(&sbi->s_orphan_lock);
2976         }
2977         if (err)
2978                 goto out_brelse;
2979         NEXT_ORPHAN(inode) = 0;
2980         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2981 out_err:
2982         ext4_std_error(inode->i_sb, err);
2983         return err;
2984
2985 out_brelse:
2986         brelse(iloc.bh);
2987         goto out_err;
2988 }
2989
2990 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2991 {
2992         int retval;
2993         struct inode *inode;
2994         struct buffer_head *bh;
2995         struct ext4_dir_entry_2 *de;
2996         handle_t *handle = NULL;
2997
2998         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2999                 return -EIO;
3000
3001         /* Initialize quotas before so that eventual writes go in
3002          * separate transaction */
3003         retval = dquot_initialize(dir);
3004         if (retval)
3005                 return retval;
3006         retval = dquot_initialize(d_inode(dentry));
3007         if (retval)
3008                 return retval;
3009
3010         retval = -ENOENT;
3011         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3012         if (IS_ERR(bh))
3013                 return PTR_ERR(bh);
3014         if (!bh)
3015                 goto end_rmdir;
3016
3017         inode = d_inode(dentry);
3018
3019         retval = -EFSCORRUPTED;
3020         if (le32_to_cpu(de->inode) != inode->i_ino)
3021                 goto end_rmdir;
3022
3023         retval = -ENOTEMPTY;
3024         if (!ext4_empty_dir(inode))
3025                 goto end_rmdir;
3026
3027         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3028                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3029         if (IS_ERR(handle)) {
3030                 retval = PTR_ERR(handle);
3031                 handle = NULL;
3032                 goto end_rmdir;
3033         }
3034
3035         if (IS_DIRSYNC(dir))
3036                 ext4_handle_sync(handle);
3037
3038         retval = ext4_delete_entry(handle, dir, de, bh);
3039         if (retval)
3040                 goto end_rmdir;
3041         if (!EXT4_DIR_LINK_EMPTY(inode))
3042                 ext4_warning_inode(inode,
3043                              "empty directory '%.*s' has too many links (%u)",
3044                              dentry->d_name.len, dentry->d_name.name,
3045                              inode->i_nlink);
3046         inode_inc_iversion(inode);
3047         clear_nlink(inode);
3048         /* There's no need to set i_disksize: the fact that i_nlink is
3049          * zero will ensure that the right thing happens during any
3050          * recovery. */
3051         inode->i_size = 0;
3052         ext4_orphan_add(handle, inode);
3053         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3054         ext4_mark_inode_dirty(handle, inode);
3055         ext4_dec_count(handle, dir);
3056         ext4_update_dx_flag(dir);
3057         ext4_mark_inode_dirty(handle, dir);
3058
3059 end_rmdir:
3060         brelse(bh);
3061         if (handle)
3062                 ext4_journal_stop(handle);
3063         return retval;
3064 }
3065
3066 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3067 {
3068         int retval;
3069         struct inode *inode;
3070         struct buffer_head *bh;
3071         struct ext4_dir_entry_2 *de;
3072         handle_t *handle = NULL;
3073
3074         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3075                 return -EIO;
3076
3077         trace_ext4_unlink_enter(dir, dentry);
3078         /* Initialize quotas before so that eventual writes go
3079          * in separate transaction */
3080         retval = dquot_initialize(dir);
3081         if (retval)
3082                 return retval;
3083         retval = dquot_initialize(d_inode(dentry));
3084         if (retval)
3085                 return retval;
3086
3087         retval = -ENOENT;
3088         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3089         if (IS_ERR(bh))
3090                 return PTR_ERR(bh);
3091         if (!bh)
3092                 goto end_unlink;
3093
3094         inode = d_inode(dentry);
3095
3096         retval = -EFSCORRUPTED;
3097         if (le32_to_cpu(de->inode) != inode->i_ino)
3098                 goto end_unlink;
3099
3100         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3101                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3102         if (IS_ERR(handle)) {
3103                 retval = PTR_ERR(handle);
3104                 handle = NULL;
3105                 goto end_unlink;
3106         }
3107
3108         if (IS_DIRSYNC(dir))
3109                 ext4_handle_sync(handle);
3110
3111         retval = ext4_delete_entry(handle, dir, de, bh);
3112         if (retval)
3113                 goto end_unlink;
3114         dir->i_ctime = dir->i_mtime = current_time(dir);
3115         ext4_update_dx_flag(dir);
3116         ext4_mark_inode_dirty(handle, dir);
3117         if (inode->i_nlink == 0)
3118                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3119                                    dentry->d_name.len, dentry->d_name.name);
3120         else
3121                 drop_nlink(inode);
3122         if (!inode->i_nlink)
3123                 ext4_orphan_add(handle, inode);
3124         inode->i_ctime = current_time(inode);
3125         ext4_mark_inode_dirty(handle, inode);
3126
3127 end_unlink:
3128         brelse(bh);
3129         if (handle)
3130                 ext4_journal_stop(handle);
3131         trace_ext4_unlink_exit(dentry, retval);
3132         return retval;
3133 }
3134
3135 static int ext4_symlink(struct inode *dir,
3136                         struct dentry *dentry, const char *symname)
3137 {
3138         handle_t *handle;
3139         struct inode *inode;
3140         int err, len = strlen(symname);
3141         int credits;
3142         struct fscrypt_str disk_link;
3143
3144         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3145                 return -EIO;
3146
3147         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3148                                       &disk_link);
3149         if (err)
3150                 return err;
3151
3152         err = dquot_initialize(dir);
3153         if (err)
3154                 return err;
3155
3156         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3157                 /*
3158                  * For non-fast symlinks, we just allocate inode and put it on
3159                  * orphan list in the first transaction => we need bitmap,
3160                  * group descriptor, sb, inode block, quota blocks, and
3161                  * possibly selinux xattr blocks.
3162                  */
3163                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3164                           EXT4_XATTR_TRANS_BLOCKS;
3165         } else {
3166                 /*
3167                  * Fast symlink. We have to add entry to directory
3168                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3169                  * allocate new inode (bitmap, group descriptor, inode block,
3170                  * quota blocks, sb is already counted in previous macros).
3171                  */
3172                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3173                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3174         }
3175
3176         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3177                                             &dentry->d_name, 0, NULL,
3178                                             EXT4_HT_DIR, credits);
3179         handle = ext4_journal_current_handle();
3180         if (IS_ERR(inode)) {
3181                 if (handle)
3182                         ext4_journal_stop(handle);
3183                 return PTR_ERR(inode);
3184         }
3185
3186         if (IS_ENCRYPTED(inode)) {
3187                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3188                 if (err)
3189                         goto err_drop_inode;
3190                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3191         }
3192
3193         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3194                 if (!IS_ENCRYPTED(inode))
3195                         inode->i_op = &ext4_symlink_inode_operations;
3196                 inode_nohighmem(inode);
3197                 ext4_set_aops(inode);
3198                 /*
3199                  * We cannot call page_symlink() with transaction started
3200                  * because it calls into ext4_write_begin() which can wait
3201                  * for transaction commit if we are running out of space
3202                  * and thus we deadlock. So we have to stop transaction now
3203                  * and restart it when symlink contents is written.
3204                  *
3205                  * To keep fs consistent in case of crash, we have to put inode
3206                  * to orphan list in the mean time.
3207                  */
3208                 drop_nlink(inode);
3209                 err = ext4_orphan_add(handle, inode);
3210                 ext4_journal_stop(handle);
3211                 handle = NULL;
3212                 if (err)
3213                         goto err_drop_inode;
3214                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3215                 if (err)
3216                         goto err_drop_inode;
3217                 /*
3218                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3219                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3220                  */
3221                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3222                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3223                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3224                 if (IS_ERR(handle)) {
3225                         err = PTR_ERR(handle);
3226                         handle = NULL;
3227                         goto err_drop_inode;
3228                 }
3229                 set_nlink(inode, 1);
3230                 err = ext4_orphan_del(handle, inode);
3231                 if (err)
3232                         goto err_drop_inode;
3233         } else {
3234                 /* clear the extent format for fast symlink */
3235                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3236                 if (!IS_ENCRYPTED(inode)) {
3237                         inode->i_op = &ext4_fast_symlink_inode_operations;
3238                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3239                 }
3240                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3241                        disk_link.len);
3242                 inode->i_size = disk_link.len - 1;
3243         }
3244         EXT4_I(inode)->i_disksize = inode->i_size;
3245         err = ext4_add_nondir(handle, dentry, inode);
3246         if (!err && IS_DIRSYNC(dir))
3247                 ext4_handle_sync(handle);
3248
3249         if (handle)
3250                 ext4_journal_stop(handle);
3251         goto out_free_encrypted_link;
3252
3253 err_drop_inode:
3254         if (handle)
3255                 ext4_journal_stop(handle);
3256         clear_nlink(inode);
3257         unlock_new_inode(inode);
3258         iput(inode);
3259 out_free_encrypted_link:
3260         if (disk_link.name != (unsigned char *)symname)
3261                 kfree(disk_link.name);
3262         return err;
3263 }
3264
3265 static int ext4_link(struct dentry *old_dentry,
3266                      struct inode *dir, struct dentry *dentry)
3267 {
3268         handle_t *handle;
3269         struct inode *inode = d_inode(old_dentry);
3270         int err, retries = 0;
3271
3272         if (inode->i_nlink >= EXT4_LINK_MAX)
3273                 return -EMLINK;
3274
3275         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3276         if (err)
3277                 return err;
3278
3279         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3280             (!projid_eq(EXT4_I(dir)->i_projid,
3281                         EXT4_I(old_dentry->d_inode)->i_projid)))
3282                 return -EXDEV;
3283
3284         err = dquot_initialize(dir);
3285         if (err)
3286                 return err;
3287
3288 retry:
3289         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3290                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3291                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3292         if (IS_ERR(handle))
3293                 return PTR_ERR(handle);
3294
3295         if (IS_DIRSYNC(dir))
3296                 ext4_handle_sync(handle);
3297
3298         inode->i_ctime = current_time(inode);
3299         ext4_inc_count(handle, inode);
3300         ihold(inode);
3301
3302         err = ext4_add_entry(handle, dentry, inode);
3303         if (!err) {
3304                 ext4_mark_inode_dirty(handle, inode);
3305                 /* this can happen only for tmpfile being
3306                  * linked the first time
3307                  */
3308                 if (inode->i_nlink == 1)
3309                         ext4_orphan_del(handle, inode);
3310                 d_instantiate(dentry, inode);
3311         } else {
3312                 drop_nlink(inode);
3313                 iput(inode);
3314         }
3315         ext4_journal_stop(handle);
3316         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3317                 goto retry;
3318         return err;
3319 }
3320
3321
3322 /*
3323  * Try to find buffer head where contains the parent block.
3324  * It should be the inode block if it is inlined or the 1st block
3325  * if it is a normal dir.
3326  */
3327 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3328                                         struct inode *inode,
3329                                         int *retval,
3330                                         struct ext4_dir_entry_2 **parent_de,
3331                                         int *inlined)
3332 {
3333         struct buffer_head *bh;
3334
3335         if (!ext4_has_inline_data(inode)) {
3336                 /* The first directory block must not be a hole, so
3337                  * treat it as DIRENT_HTREE
3338                  */
3339                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3340                 if (IS_ERR(bh)) {
3341                         *retval = PTR_ERR(bh);
3342                         return NULL;
3343                 }
3344                 *parent_de = ext4_next_entry(
3345                                         (struct ext4_dir_entry_2 *)bh->b_data,
3346                                         inode->i_sb->s_blocksize);
3347                 return bh;
3348         }
3349
3350         *inlined = 1;
3351         return ext4_get_first_inline_block(inode, parent_de, retval);
3352 }
3353
3354 struct ext4_renament {
3355         struct inode *dir;
3356         struct dentry *dentry;
3357         struct inode *inode;
3358         bool is_dir;
3359         int dir_nlink_delta;
3360
3361         /* entry for "dentry" */
3362         struct buffer_head *bh;
3363         struct ext4_dir_entry_2 *de;
3364         int inlined;
3365
3366         /* entry for ".." in inode if it's a directory */
3367         struct buffer_head *dir_bh;
3368         struct ext4_dir_entry_2 *parent_de;
3369         int dir_inlined;
3370 };
3371
3372 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3373 {
3374         int retval;
3375
3376         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3377                                               &retval, &ent->parent_de,
3378                                               &ent->dir_inlined);
3379         if (!ent->dir_bh)
3380                 return retval;
3381         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3382                 return -EFSCORRUPTED;
3383         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3384         return ext4_journal_get_write_access(handle, ent->dir_bh);
3385 }
3386
3387 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3388                                   unsigned dir_ino)
3389 {
3390         int retval;
3391
3392         ent->parent_de->inode = cpu_to_le32(dir_ino);
3393         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3394         if (!ent->dir_inlined) {
3395                 if (is_dx(ent->inode)) {
3396                         retval = ext4_handle_dirty_dx_node(handle,
3397                                                            ent->inode,
3398                                                            ent->dir_bh);
3399                 } else {
3400                         retval = ext4_handle_dirty_dirent_node(handle,
3401                                                                ent->inode,
3402                                                                ent->dir_bh);
3403                 }
3404         } else {
3405                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3406         }
3407         if (retval) {
3408                 ext4_std_error(ent->dir->i_sb, retval);
3409                 return retval;
3410         }
3411         return 0;
3412 }
3413
3414 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3415                        unsigned ino, unsigned file_type)
3416 {
3417         int retval;
3418
3419         BUFFER_TRACE(ent->bh, "get write access");
3420         retval = ext4_journal_get_write_access(handle, ent->bh);
3421         if (retval)
3422                 return retval;
3423         ent->de->inode = cpu_to_le32(ino);
3424         if (ext4_has_feature_filetype(ent->dir->i_sb))
3425                 ent->de->file_type = file_type;
3426         inode_inc_iversion(ent->dir);
3427         ent->dir->i_ctime = ent->dir->i_mtime =
3428                 current_time(ent->dir);
3429         ext4_mark_inode_dirty(handle, ent->dir);
3430         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3431         if (!ent->inlined) {
3432                 retval = ext4_handle_dirty_dirent_node(handle,
3433                                                        ent->dir, ent->bh);
3434                 if (unlikely(retval)) {
3435                         ext4_std_error(ent->dir->i_sb, retval);
3436                         return retval;
3437                 }
3438         }
3439
3440         return 0;
3441 }
3442
3443 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3444                           unsigned ino, unsigned file_type)
3445 {
3446         struct ext4_renament old = *ent;
3447         int retval = 0;
3448
3449         /*
3450          * old->de could have moved from under us during make indexed dir,
3451          * so the old->de may no longer valid and need to find it again
3452          * before reset old inode info.
3453          */
3454         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3455         if (IS_ERR(old.bh))
3456                 retval = PTR_ERR(old.bh);
3457         if (!old.bh)
3458                 retval = -ENOENT;
3459         if (retval) {
3460                 ext4_std_error(old.dir->i_sb, retval);
3461                 return;
3462         }
3463
3464         ext4_setent(handle, &old, ino, file_type);
3465         brelse(old.bh);
3466 }
3467
3468 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3469                                   const struct qstr *d_name)
3470 {
3471         int retval = -ENOENT;
3472         struct buffer_head *bh;
3473         struct ext4_dir_entry_2 *de;
3474
3475         bh = ext4_find_entry(dir, d_name, &de, NULL);
3476         if (IS_ERR(bh))
3477                 return PTR_ERR(bh);
3478         if (bh) {
3479                 retval = ext4_delete_entry(handle, dir, de, bh);
3480                 brelse(bh);
3481         }
3482         return retval;
3483 }
3484
3485 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3486                                int force_reread)
3487 {
3488         int retval;
3489         /*
3490          * ent->de could have moved from under us during htree split, so make
3491          * sure that we are deleting the right entry.  We might also be pointing
3492          * to a stale entry in the unused part of ent->bh so just checking inum
3493          * and the name isn't enough.
3494          */
3495         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3496             ent->de->name_len != ent->dentry->d_name.len ||
3497             strncmp(ent->de->name, ent->dentry->d_name.name,
3498                     ent->de->name_len) ||
3499             force_reread) {
3500                 retval = ext4_find_delete_entry(handle, ent->dir,
3501                                                 &ent->dentry->d_name);
3502         } else {
3503                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3504                 if (retval == -ENOENT) {
3505                         retval = ext4_find_delete_entry(handle, ent->dir,
3506                                                         &ent->dentry->d_name);
3507                 }
3508         }
3509
3510         if (retval) {
3511                 ext4_warning_inode(ent->dir,
3512                                    "Deleting old file: nlink %d, error=%d",
3513                                    ent->dir->i_nlink, retval);
3514         }
3515 }
3516
3517 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3518 {
3519         if (ent->dir_nlink_delta) {
3520                 if (ent->dir_nlink_delta == -1)
3521                         ext4_dec_count(handle, ent->dir);
3522                 else
3523                         ext4_inc_count(handle, ent->dir);
3524                 ext4_mark_inode_dirty(handle, ent->dir);
3525         }
3526 }
3527
3528 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3529                                               int credits, handle_t **h)
3530 {
3531         struct inode *wh;
3532         handle_t *handle;
3533         int retries = 0;
3534
3535         /*
3536          * for inode block, sb block, group summaries,
3537          * and inode bitmap
3538          */
3539         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3540                     EXT4_XATTR_TRANS_BLOCKS + 4);
3541 retry:
3542         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3543                                          &ent->dentry->d_name, 0, NULL,
3544                                          EXT4_HT_DIR, credits);
3545
3546         handle = ext4_journal_current_handle();
3547         if (IS_ERR(wh)) {
3548                 if (handle)
3549                         ext4_journal_stop(handle);
3550                 if (PTR_ERR(wh) == -ENOSPC &&
3551                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3552                         goto retry;
3553         } else {
3554                 *h = handle;
3555                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3556                 wh->i_op = &ext4_special_inode_operations;
3557         }
3558         return wh;
3559 }
3560
3561 /*
3562  * Anybody can rename anything with this: the permission checks are left to the
3563  * higher-level routines.
3564  *
3565  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3566  * while new_{dentry,inode) refers to the destination dentry/inode
3567  * This comes from rename(const char *oldpath, const char *newpath)
3568  */
3569 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3570                        struct inode *new_dir, struct dentry *new_dentry,
3571                        unsigned int flags)
3572 {
3573         handle_t *handle = NULL;
3574         struct ext4_renament old = {
3575                 .dir = old_dir,
3576                 .dentry = old_dentry,
3577                 .inode = d_inode(old_dentry),
3578         };
3579         struct ext4_renament new = {
3580                 .dir = new_dir,
3581                 .dentry = new_dentry,
3582                 .inode = d_inode(new_dentry),
3583         };
3584         int force_reread;
3585         int retval;
3586         struct inode *whiteout = NULL;
3587         int credits;
3588         u8 old_file_type;
3589
3590         if (new.inode && new.inode->i_nlink == 0) {
3591                 EXT4_ERROR_INODE(new.inode,
3592                                  "target of rename is already freed");
3593                 return -EFSCORRUPTED;
3594         }
3595
3596         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3597             (!projid_eq(EXT4_I(new_dir)->i_projid,
3598                         EXT4_I(old_dentry->d_inode)->i_projid)))
3599                 return -EXDEV;
3600
3601         retval = dquot_initialize(old.dir);
3602         if (retval)
3603                 return retval;
3604         retval = dquot_initialize(new.dir);
3605         if (retval)
3606                 return retval;
3607
3608         /* Initialize quotas before so that eventual writes go
3609          * in separate transaction */
3610         if (new.inode) {
3611                 retval = dquot_initialize(new.inode);
3612                 if (retval)
3613                         return retval;
3614         }
3615
3616         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3617         if (IS_ERR(old.bh))
3618                 return PTR_ERR(old.bh);
3619         /*
3620          *  Check for inode number is _not_ due to possible IO errors.
3621          *  We might rmdir the source, keep it as pwd of some process
3622          *  and merrily kill the link to whatever was created under the
3623          *  same name. Goodbye sticky bit ;-<
3624          */
3625         retval = -ENOENT;
3626         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3627                 goto release_bh;
3628
3629         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3630                                  &new.de, &new.inlined);
3631         if (IS_ERR(new.bh)) {
3632                 retval = PTR_ERR(new.bh);
3633                 new.bh = NULL;
3634                 goto release_bh;
3635         }
3636         if (new.bh) {
3637                 if (!new.inode) {
3638                         brelse(new.bh);
3639                         new.bh = NULL;
3640                 }
3641         }
3642         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3643                 ext4_alloc_da_blocks(old.inode);
3644
3645         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3646                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3647         if (!(flags & RENAME_WHITEOUT)) {
3648                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3649                 if (IS_ERR(handle)) {
3650                         retval = PTR_ERR(handle);
3651                         goto release_bh;
3652                 }
3653         } else {
3654                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3655                 if (IS_ERR(whiteout)) {
3656                         retval = PTR_ERR(whiteout);
3657                         goto release_bh;
3658                 }
3659         }
3660
3661         old_file_type = old.de->file_type;
3662         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3663                 ext4_handle_sync(handle);
3664
3665         if (S_ISDIR(old.inode->i_mode)) {
3666                 if (new.inode) {
3667                         retval = -ENOTEMPTY;
3668                         if (!ext4_empty_dir(new.inode))
3669                                 goto end_rename;
3670                 } else {
3671                         retval = -EMLINK;
3672                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3673                                 goto end_rename;
3674                 }
3675                 retval = ext4_rename_dir_prepare(handle, &old);
3676                 if (retval)
3677                         goto end_rename;
3678         }
3679         /*
3680          * If we're renaming a file within an inline_data dir and adding or
3681          * setting the new dirent causes a conversion from inline_data to
3682          * extents/blockmap, we need to force the dirent delete code to
3683          * re-read the directory, or else we end up trying to delete a dirent
3684          * from what is now the extent tree root (or a block map).
3685          */
3686         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3687                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3688
3689         if (whiteout) {
3690                 /*
3691                  * Do this before adding a new entry, so the old entry is sure
3692                  * to be still pointing to the valid old entry.
3693                  */
3694                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3695                                      EXT4_FT_CHRDEV);
3696                 if (retval)
3697                         goto end_rename;
3698                 ext4_mark_inode_dirty(handle, whiteout);
3699         }
3700         if (!new.bh) {
3701                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3702                 if (retval)
3703                         goto end_rename;
3704         } else {
3705                 retval = ext4_setent(handle, &new,
3706                                      old.inode->i_ino, old_file_type);
3707                 if (retval)
3708                         goto end_rename;
3709         }
3710         if (force_reread)
3711                 force_reread = !ext4_test_inode_flag(new.dir,
3712                                                      EXT4_INODE_INLINE_DATA);
3713
3714         /*
3715          * Like most other Unix systems, set the ctime for inodes on a
3716          * rename.
3717          */
3718         old.inode->i_ctime = current_time(old.inode);
3719         ext4_mark_inode_dirty(handle, old.inode);
3720
3721         if (!whiteout) {
3722                 /*
3723                  * ok, that's it
3724                  */
3725                 ext4_rename_delete(handle, &old, force_reread);
3726         }
3727
3728         if (new.inode) {
3729                 ext4_dec_count(handle, new.inode);
3730                 new.inode->i_ctime = current_time(new.inode);
3731         }
3732         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3733         ext4_update_dx_flag(old.dir);
3734         if (old.dir_bh) {
3735                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3736                 if (retval)
3737                         goto end_rename;
3738
3739                 ext4_dec_count(handle, old.dir);
3740                 if (new.inode) {
3741                         /* checked ext4_empty_dir above, can't have another
3742                          * parent, ext4_dec_count() won't work for many-linked
3743                          * dirs */
3744                         clear_nlink(new.inode);
3745                 } else {
3746                         ext4_inc_count(handle, new.dir);
3747                         ext4_update_dx_flag(new.dir);
3748                         ext4_mark_inode_dirty(handle, new.dir);
3749                 }
3750         }
3751         ext4_mark_inode_dirty(handle, old.dir);
3752         if (new.inode) {
3753                 ext4_mark_inode_dirty(handle, new.inode);
3754                 if (!new.inode->i_nlink)
3755                         ext4_orphan_add(handle, new.inode);
3756         }
3757         retval = 0;
3758
3759 end_rename:
3760         if (whiteout) {
3761                 if (retval) {
3762                         ext4_resetent(handle, &old,
3763                                       old.inode->i_ino, old_file_type);
3764                         drop_nlink(whiteout);
3765                         ext4_orphan_add(handle, whiteout);
3766                 }
3767                 unlock_new_inode(whiteout);
3768                 ext4_journal_stop(handle);
3769                 iput(whiteout);
3770         } else {
3771                 ext4_journal_stop(handle);
3772         }
3773 release_bh:
3774         brelse(old.dir_bh);
3775         brelse(old.bh);
3776         brelse(new.bh);
3777         return retval;
3778 }
3779
3780 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3781                              struct inode *new_dir, struct dentry *new_dentry)
3782 {
3783         handle_t *handle = NULL;
3784         struct ext4_renament old = {
3785                 .dir = old_dir,
3786                 .dentry = old_dentry,
3787                 .inode = d_inode(old_dentry),
3788         };
3789         struct ext4_renament new = {
3790                 .dir = new_dir,
3791                 .dentry = new_dentry,
3792                 .inode = d_inode(new_dentry),
3793         };
3794         u8 new_file_type;
3795         int retval;
3796         struct timespec64 ctime;
3797
3798         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3799              !projid_eq(EXT4_I(new_dir)->i_projid,
3800                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3801             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3802              !projid_eq(EXT4_I(old_dir)->i_projid,
3803                         EXT4_I(new_dentry->d_inode)->i_projid)))
3804                 return -EXDEV;
3805
3806         retval = dquot_initialize(old.dir);
3807         if (retval)
3808                 return retval;
3809         retval = dquot_initialize(new.dir);
3810         if (retval)
3811                 return retval;
3812
3813         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3814                                  &old.de, &old.inlined);
3815         if (IS_ERR(old.bh))
3816                 return PTR_ERR(old.bh);
3817         /*
3818          *  Check for inode number is _not_ due to possible IO errors.
3819          *  We might rmdir the source, keep it as pwd of some process
3820          *  and merrily kill the link to whatever was created under the
3821          *  same name. Goodbye sticky bit ;-<
3822          */
3823         retval = -ENOENT;
3824         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3825                 goto end_rename;
3826
3827         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3828                                  &new.de, &new.inlined);
3829         if (IS_ERR(new.bh)) {
3830                 retval = PTR_ERR(new.bh);
3831                 new.bh = NULL;
3832                 goto end_rename;
3833         }
3834
3835         /* RENAME_EXCHANGE case: old *and* new must both exist */
3836         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3837                 goto end_rename;
3838
3839         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3840                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3841                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3842         if (IS_ERR(handle)) {
3843                 retval = PTR_ERR(handle);
3844                 handle = NULL;
3845                 goto end_rename;
3846         }
3847
3848         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3849                 ext4_handle_sync(handle);
3850
3851         if (S_ISDIR(old.inode->i_mode)) {
3852                 old.is_dir = true;
3853                 retval = ext4_rename_dir_prepare(handle, &old);
3854                 if (retval)
3855                         goto end_rename;
3856         }
3857         if (S_ISDIR(new.inode->i_mode)) {
3858                 new.is_dir = true;
3859                 retval = ext4_rename_dir_prepare(handle, &new);
3860                 if (retval)
3861                         goto end_rename;
3862         }
3863
3864         /*
3865          * Other than the special case of overwriting a directory, parents'
3866          * nlink only needs to be modified if this is a cross directory rename.
3867          */
3868         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3869                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3870                 new.dir_nlink_delta = -old.dir_nlink_delta;
3871                 retval = -EMLINK;
3872                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3873                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3874                         goto end_rename;
3875         }
3876
3877         new_file_type = new.de->file_type;
3878         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3879         if (retval)
3880                 goto end_rename;
3881
3882         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3883         if (retval)
3884                 goto end_rename;
3885
3886         /*
3887          * Like most other Unix systems, set the ctime for inodes on a
3888          * rename.
3889          */
3890         ctime = current_time(old.inode);
3891         old.inode->i_ctime = ctime;
3892         new.inode->i_ctime = ctime;
3893         ext4_mark_inode_dirty(handle, old.inode);
3894         ext4_mark_inode_dirty(handle, new.inode);
3895
3896         if (old.dir_bh) {
3897                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3898                 if (retval)
3899                         goto end_rename;
3900         }
3901         if (new.dir_bh) {
3902                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3903                 if (retval)
3904                         goto end_rename;
3905         }
3906         ext4_update_dir_count(handle, &old);
3907         ext4_update_dir_count(handle, &new);
3908         retval = 0;
3909
3910 end_rename:
3911         brelse(old.dir_bh);
3912         brelse(new.dir_bh);
3913         brelse(old.bh);
3914         brelse(new.bh);
3915         if (handle)
3916                 ext4_journal_stop(handle);
3917         return retval;
3918 }
3919
3920 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3921                         struct inode *new_dir, struct dentry *new_dentry,
3922                         unsigned int flags)
3923 {
3924         int err;
3925
3926         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3927                 return -EIO;
3928
3929         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3930                 return -EINVAL;
3931
3932         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3933                                      flags);
3934         if (err)
3935                 return err;
3936
3937         if (flags & RENAME_EXCHANGE) {
3938                 return ext4_cross_rename(old_dir, old_dentry,
3939                                          new_dir, new_dentry);
3940         }
3941
3942         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3943 }
3944
3945 /*
3946  * directories can handle most operations...
3947  */
3948 const struct inode_operations ext4_dir_inode_operations = {
3949         .create         = ext4_create,
3950         .lookup         = ext4_lookup,
3951         .link           = ext4_link,
3952         .unlink         = ext4_unlink,
3953         .symlink        = ext4_symlink,
3954         .mkdir          = ext4_mkdir,
3955         .rmdir          = ext4_rmdir,
3956         .mknod          = ext4_mknod,
3957         .tmpfile        = ext4_tmpfile,
3958         .rename         = ext4_rename2,
3959         .setattr        = ext4_setattr,
3960         .getattr        = ext4_getattr,
3961         .listxattr      = ext4_listxattr,
3962         .get_acl        = ext4_get_acl,
3963         .set_acl        = ext4_set_acl,
3964         .fiemap         = ext4_fiemap,
3965 };
3966
3967 const struct inode_operations ext4_special_inode_operations = {
3968         .setattr        = ext4_setattr,
3969         .getattr        = ext4_getattr,
3970         .listxattr      = ext4_listxattr,
3971         .get_acl        = ext4_get_acl,
3972         .set_acl        = ext4_set_acl,
3973 };