fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         bh = ext4_bread(NULL, inode, block, 0);
 113         if (IS_ERR(bh)) {
 114                 __ext4_warning(inode->i_sb, func, line,
 115                                "inode #%lu: lblock %lu: comm %s: "
 116                                "error %ld reading directory block",
 117                                inode->i_ino, (unsigned long)block,
 118                                current->comm, PTR_ERR(bh));
 119
 120                 return bh;
 121         }
 122         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 123                 ext4_error_inode(inode, func, line, block,
 124                                  "Directory hole found for htree %s block",
 125                                  (type == INDEX) ? "index" : "leaf");
 126                 return ERR_PTR(-EFSCORRUPTED);
 127         }
 128         if (!bh)
 129                 return NULL;
 130         dirent = (struct ext4_dir_entry *) bh->b_data;
 131         /* Determine whether or not we have an index block */
 132         if (is_dx(inode)) {
 133                 if (block == 0)
 134                         is_dx_block = 1;
 135                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 136                                                 inode->i_sb->s_blocksize) ==
 137                          inode->i_sb->s_blocksize)
 138                         is_dx_block = 1;
 139         }
 140         if (!is_dx_block && type == INDEX) {
 141                 ext4_error_inode(inode, func, line, block,
 142                        "directory leaf block found instead of index block");
 143                 brelse(bh);
 144                 return ERR_PTR(-EFSCORRUPTED);
 145         }
 146         if (!ext4_has_metadata_csum(inode->i_sb) ||
 147             buffer_verified(bh))
 148                 return bh;
 149
 150         /*
 151          * An empty leaf block can get mistaken for a index block; for
 152          * this reason, we can only check the index checksum when the
 153          * caller is sure it should be an index block.
 154          */
 155         if (is_dx_block && type == INDEX) {
 156                 if (ext4_dx_csum_verify(inode, dirent))
 157                         set_buffer_verified(bh);
 158                 else {
 159                         ext4_set_errno(inode->i_sb, EFSBADCRC);
 160                         ext4_error_inode(inode, func, line, block,
 161                                          "Directory index failed checksum");
 162                         brelse(bh);
 163                         return ERR_PTR(-EFSBADCRC);
 164                 }
 165         }
 166         if (!is_dx_block) {
 167                 if (ext4_dirblock_csum_verify(inode, bh))
 168                         set_buffer_verified(bh);
 169                 else {
 170                         ext4_set_errno(inode->i_sb, EFSBADCRC);
 171                         ext4_error_inode(inode, func, line, block,
 172                                          "Directory block failed checksum");
 173                         brelse(bh);
 174                         return ERR_PTR(-EFSBADCRC);
 175                 }
 176         }
 177         return bh;
 178 }
 179
 180 #ifndef assert
 181 #define assert(test) J_ASSERT(test)
 182 #endif
 183
 184 #ifdef DX_DEBUG
 185 #define dxtrace(command) command
 186 #else
 187 #define dxtrace(command)
 188 #endif
 189
 190 struct fake_dirent
 191 {
 192         __le32 inode;
 193         __le16 rec_len;
 194         u8 name_len;
 195         u8 file_type;
 196 };
 197
 198 struct dx_countlimit
 199 {
 200         __le16 limit;
 201         __le16 count;
 202 };
 203
 204 struct dx_entry
 205 {
 206         __le32 hash;
 207         __le32 block;
 208 };
 209
 210 /*
 211  * dx_root_info is laid out so that if it should somehow get overlaid by a
 212  * dirent the two low bits of the hash version will be zero.  Therefore, the
 213  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 214  */
 215
 216 struct dx_root
 217 {
 218         struct fake_dirent dot;
 219         char dot_name[4];
 220         struct fake_dirent dotdot;
 221         char dotdot_name[4];
 222         struct dx_root_info
 223         {
 224                 __le32 reserved_zero;
 225                 u8 hash_version;
 226                 u8 info_length; /* 8 */
 227                 u8 indirect_levels;
 228                 u8 unused_flags;
 229         }
 230         info;
 231         struct dx_entry entries[0];
 232 };
 233
 234 struct dx_node
 235 {
 236         struct fake_dirent fake;
 237         struct dx_entry entries[0];
 238 };
 239
 240
 241 struct dx_frame
 242 {
 243         struct buffer_head *bh;
 244         struct dx_entry *entries;
 245         struct dx_entry *at;
 246 };
 247
 248 struct dx_map_entry
 249 {
 250         u32 hash;
 251         u16 offs;
 252         u16 size;
 253 };
 254
 255 /*
 256  * This goes at the end of each htree block.
 257  */
 258 struct dx_tail {
 259         u32 dt_reserved;
 260         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 261 };
 262
 263 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 264 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 265 static inline unsigned dx_get_hash(struct dx_entry *entry);
 266 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 267 static unsigned dx_get_count(struct dx_entry *entries);
 268 static unsigned dx_get_limit(struct dx_entry *entries);
 269 static void dx_set_count(struct dx_entry *entries, unsigned value);
 270 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 271 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 272 static unsigned dx_node_limit(struct inode *dir);
 273 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 274                                  struct inode *dir,
 275                                  struct dx_hash_info *hinfo,
 276                                  struct dx_frame *frame);
 277 static void dx_release(struct dx_frame *frames);
 278 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 279                        unsigned blocksize, struct dx_hash_info *hinfo,
 280                        struct dx_map_entry map[]);
 281 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 282 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 283                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 284 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 285 static void dx_insert_block(struct dx_frame *frame,
 286                                         u32 hash, ext4_lblk_t block);
 287 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 288                                  struct dx_frame *frame,
 289                                  struct dx_frame *frames,
 290                                  __u32 *start_hash);
 291 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 292                 struct ext4_filename *fname,
 293                 struct ext4_dir_entry_2 **res_dir);
 294 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 295                              struct inode *dir, struct inode *inode);
 296
 297 /* checksumming functions */
 298 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 299                                  unsigned int blocksize)
 300 {
 301         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 302
 303         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 304         t->det_rec_len = ext4_rec_len_to_disk(
 305                         sizeof(struct ext4_dir_entry_tail), blocksize);
 306         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 307 }
 308
 309 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 310 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 311                                                    struct buffer_head *bh)
 312 {
 313         struct ext4_dir_entry_tail *t;
 314
 315 #ifdef PARANOID
 316         struct ext4_dir_entry *d, *top;
 317
 318         d = (struct ext4_dir_entry *)bh->b_data;
 319         top = (struct ext4_dir_entry *)(bh->b_data +
 320                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 321                  sizeof(struct ext4_dir_entry_tail)));
 322         while (d < top && d->rec_len)
 323                 d = (struct ext4_dir_entry *)(((void *)d) +
 324                     le16_to_cpu(d->rec_len));
 325
 326         if (d != top)
 327                 return NULL;
 328
 329         t = (struct ext4_dir_entry_tail *)d;
 330 #else
 331         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 332 #endif
 333
 334         if (t->det_reserved_zero1 ||
 335             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 336             t->det_reserved_zero2 ||
 337             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 338                 return NULL;
 339
 340         return t;
 341 }
 342
 343 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 344 {
 345         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 346         struct ext4_inode_info *ei = EXT4_I(inode);
 347         __u32 csum;
 348
 349         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 350         return cpu_to_le32(csum);
 351 }
 352
 353 #define warn_no_space_for_csum(inode)                                   \
 354         __warn_no_space_for_csum((inode), __func__, __LINE__)
 355
 356 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 357                                      unsigned int line)
 358 {
 359         __ext4_warning_inode(inode, func, line,
 360                 "No space for directory leaf checksum. Please run e2fsck -D.");
 361 }
 362
 363 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 364 {
 365         struct ext4_dir_entry_tail *t;
 366
 367         if (!ext4_has_metadata_csum(inode->i_sb))
 368                 return 1;
 369
 370         t = get_dirent_tail(inode, bh);
 371         if (!t) {
 372                 warn_no_space_for_csum(inode);
 373                 return 0;
 374         }
 375
 376         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 377                                                   (char *)t - bh->b_data))
 378                 return 0;
 379
 380         return 1;
 381 }
 382
 383 static void ext4_dirblock_csum_set(struct inode *inode,
 384                                  struct buffer_head *bh)
 385 {
 386         struct ext4_dir_entry_tail *t;
 387
 388         if (!ext4_has_metadata_csum(inode->i_sb))
 389                 return;
 390
 391         t = get_dirent_tail(inode, bh);
 392         if (!t) {
 393                 warn_no_space_for_csum(inode);
 394                 return;
 395         }
 396
 397         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 398                                              (char *)t - bh->b_data);
 399 }
 400
 401 int ext4_handle_dirty_dirblock(handle_t *handle,
 402                                struct inode *inode,
 403                                struct buffer_head *bh)
 404 {
 405         ext4_dirblock_csum_set(inode, bh);
 406         return ext4_handle_dirty_metadata(handle, inode, bh);
 407 }
 408
 409 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 410                                                struct ext4_dir_entry *dirent,
 411                                                int *offset)
 412 {
 413         struct ext4_dir_entry *dp;
 414         struct dx_root_info *root;
 415         int count_offset;
 416
 417         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 418                 count_offset = 8;
 419         else if (le16_to_cpu(dirent->rec_len) == 12) {
 420                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 421                 if (le16_to_cpu(dp->rec_len) !=
 422                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 423                         return NULL;
 424                 root = (struct dx_root_info *)(((void *)dp + 12));
 425                 if (root->reserved_zero ||
 426                     root->info_length != sizeof(struct dx_root_info))
 427                         return NULL;
 428                 count_offset = 32;
 429         } else
 430                 return NULL;
 431
 432         if (offset)
 433                 *offset = count_offset;
 434         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 435 }
 436
 437 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 438                            int count_offset, int count, struct dx_tail *t)
 439 {
 440         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 441         struct ext4_inode_info *ei = EXT4_I(inode);
 442         __u32 csum;
 443         int size;
 444         __u32 dummy_csum = 0;
 445         int offset = offsetof(struct dx_tail, dt_checksum);
 446
 447         size = count_offset + (count * sizeof(struct dx_entry));
 448         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 449         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 450         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 451
 452         return cpu_to_le32(csum);
 453 }
 454
 455 static int ext4_dx_csum_verify(struct inode *inode,
 456                                struct ext4_dir_entry *dirent)
 457 {
 458         struct dx_countlimit *c;
 459         struct dx_tail *t;
 460         int count_offset, limit, count;
 461
 462         if (!ext4_has_metadata_csum(inode->i_sb))
 463                 return 1;
 464
 465         c = get_dx_countlimit(inode, dirent, &count_offset);
 466         if (!c) {
 467                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 468                 return 0;
 469         }
 470         limit = le16_to_cpu(c->limit);
 471         count = le16_to_cpu(c->count);
 472         if (count_offset + (limit * sizeof(struct dx_entry)) >
 473             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 474                 warn_no_space_for_csum(inode);
 475                 return 0;
 476         }
 477         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 478
 479         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 480                                             count, t))
 481                 return 0;
 482         return 1;
 483 }
 484
 485 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 486 {
 487         struct dx_countlimit *c;
 488         struct dx_tail *t;
 489         int count_offset, limit, count;
 490
 491         if (!ext4_has_metadata_csum(inode->i_sb))
 492                 return;
 493
 494         c = get_dx_countlimit(inode, dirent, &count_offset);
 495         if (!c) {
 496                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 497                 return;
 498         }
 499         limit = le16_to_cpu(c->limit);
 500         count = le16_to_cpu(c->count);
 501         if (count_offset + (limit * sizeof(struct dx_entry)) >
 502             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 503                 warn_no_space_for_csum(inode);
 504                 return;
 505         }
 506         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 507
 508         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 509 }
 510
 511 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 512                                             struct inode *inode,
 513                                             struct buffer_head *bh)
 514 {
 515         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 516         return ext4_handle_dirty_metadata(handle, inode, bh);
 517 }
 518
 519 /*
 520  * p is at least 6 bytes before the end of page
 521  */
 522 static inline struct ext4_dir_entry_2 *
 523 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 524 {
 525         return (struct ext4_dir_entry_2 *)((char *)p +
 526                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 527 }
 528
 529 /*
 530  * Future: use high four bits of block for coalesce-on-delete flags
 531  * Mask them off for now.
 532  */
 533
 534 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 535 {
 536         return le32_to_cpu(entry->block) & 0x0fffffff;
 537 }
 538
 539 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 540 {
 541         entry->block = cpu_to_le32(value);
 542 }
 543
 544 static inline unsigned dx_get_hash(struct dx_entry *entry)
 545 {
 546         return le32_to_cpu(entry->hash);
 547 }
 548
 549 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 550 {
 551         entry->hash = cpu_to_le32(value);
 552 }
 553
 554 static inline unsigned dx_get_count(struct dx_entry *entries)
 555 {
 556         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 557 }
 558
 559 static inline unsigned dx_get_limit(struct dx_entry *entries)
 560 {
 561         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 562 }
 563
 564 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 565 {
 566         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 567 }
 568
 569 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 570 {
 571         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 572 }
 573
 574 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 575 {
 576         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 577                 EXT4_DIR_REC_LEN(2) - infosize;
 578
 579         if (ext4_has_metadata_csum(dir->i_sb))
 580                 entry_space -= sizeof(struct dx_tail);
 581         return entry_space / sizeof(struct dx_entry);
 582 }
 583
 584 static inline unsigned dx_node_limit(struct inode *dir)
 585 {
 586         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 587
 588         if (ext4_has_metadata_csum(dir->i_sb))
 589                 entry_space -= sizeof(struct dx_tail);
 590         return entry_space / sizeof(struct dx_entry);
 591 }
 592
 593 /*
 594  * Debug
 595  */
 596 #ifdef DX_DEBUG
 597 static void dx_show_index(char * label, struct dx_entry *entries)
 598 {
 599         int i, n = dx_get_count (entries);
 600         printk(KERN_DEBUG "%s index", label);
 601         for (i = 0; i < n; i++) {
 602                 printk(KERN_CONT " %x->%lu",
 603                        i ? dx_get_hash(entries + i) : 0,
 604                        (unsigned long)dx_get_block(entries + i));
 605         }
 606         printk(KERN_CONT "\n");
 607 }
 608
 609 struct stats
 610 {
 611         unsigned names;
 612         unsigned space;
 613         unsigned bcount;
 614 };
 615
 616 static struct stats dx_show_leaf(struct inode *dir,
 617                                 struct dx_hash_info *hinfo,
 618                                 struct ext4_dir_entry_2 *de,
 619                                 int size, int show_names)
 620 {
 621         unsigned names = 0, space = 0;
 622         char *base = (char *) de;
 623         struct dx_hash_info h = *hinfo;
 624
 625         printk("names: ");
 626         while ((char *) de < base + size)
 627         {
 628                 if (de->inode)
 629                 {
 630                         if (show_names)
 631                         {
 632 #ifdef CONFIG_FS_ENCRYPTION
 633                                 int len;
 634                                 char *name;
 635                                 struct fscrypt_str fname_crypto_str =
 636                                         FSTR_INIT(NULL, 0);
 637                                 int res = 0;
 638
 639                                 name  = de->name;
 640                                 len = de->name_len;
 641                                 if (IS_ENCRYPTED(dir))
 642                                         res = fscrypt_get_encryption_info(dir);
 643                                 if (res) {
 644                                         printk(KERN_WARNING "Error setting up"
 645                                                " fname crypto: %d\n", res);
 646                                 }
 647                                 if (!fscrypt_has_encryption_key(dir)) {
 648                                         /* Directory is not encrypted */
 649                                         ext4fs_dirhash(dir, de->name,
 650                                                 de->name_len, &h);
 651                                         printk("%*.s:(U)%x.%u ", len,
 652                                                name, h.hash,
 653                                                (unsigned) ((char *) de
 654                                                            - base));
 655                                 } else {
 656                                         struct fscrypt_str de_name =
 657                                                 FSTR_INIT(name, len);
 658
 659                                         /* Directory is encrypted */
 660                                         res = fscrypt_fname_alloc_buffer(
 661                                                 dir, len,
 662                                                 &fname_crypto_str);
 663                                         if (res)
 664                                                 printk(KERN_WARNING "Error "
 665                                                         "allocating crypto "
 666                                                         "buffer--skipping "
 667                                                         "crypto\n");
 668                                         res = fscrypt_fname_disk_to_usr(dir,
 669                                                 0, 0, &de_name,
 670                                                 &fname_crypto_str);
 671                                         if (res) {
 672                                                 printk(KERN_WARNING "Error "
 673                                                         "converting filename "
 674                                                         "from disk to usr"
 675                                                         "\n");
 676                                                 name = "??";
 677                                                 len = 2;
 678                                         } else {
 679                                                 name = fname_crypto_str.name;
 680                                                 len = fname_crypto_str.len;
 681                                         }
 682                                         ext4fs_dirhash(dir, de->name,
 683                                                        de->name_len, &h);
 684                                         printk("%*.s:(E)%x.%u ", len, name,
 685                                                h.hash, (unsigned) ((char *) de
 686                                                                    - base));
 687                                         fscrypt_fname_free_buffer(
 688                                                         &fname_crypto_str);
 689                                 }
 690 #else
 691                                 int len = de->name_len;
 692                                 char *name = de->name;
 693                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 694                                 printk("%*.s:%x.%u ", len, name, h.hash,
 695                                        (unsigned) ((char *) de - base));
 696 #endif
 697                         }
 698                         space += EXT4_DIR_REC_LEN(de->name_len);
 699                         names++;
 700                 }
 701                 de = ext4_next_entry(de, size);
 702         }
 703         printk(KERN_CONT "(%i)\n", names);
 704         return (struct stats) { names, space, 1 };
 705 }
 706
 707 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 708                              struct dx_entry *entries, int levels)
 709 {
 710         unsigned blocksize = dir->i_sb->s_blocksize;
 711         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 712         unsigned bcount = 0;
 713         struct buffer_head *bh;
 714         printk("%i indexed blocks...\n", count);
 715         for (i = 0; i < count; i++, entries++)
 716         {
 717                 ext4_lblk_t block = dx_get_block(entries);
 718                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 719                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 720                 struct stats stats;
 721                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 722                 bh = ext4_bread(NULL,dir, block, 0);
 723                 if (!bh || IS_ERR(bh))
 724                         continue;
 725                 stats = levels?
 726                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 727                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 728                         bh->b_data, blocksize, 0);
 729                 names += stats.names;
 730                 space += stats.space;
 731                 bcount += stats.bcount;
 732                 brelse(bh);
 733         }
 734         if (bcount)
 735                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 736                        levels ? "" : "   ", names, space/bcount,
 737                        (space/bcount)*100/blocksize);
 738         return (struct stats) { names, space, bcount};
 739 }
 740 #endif /* DX_DEBUG */
 741
 742 /*
 743  * Probe for a directory leaf block to search.
 744  *
 745  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 746  * error in the directory index, and the caller should fall back to
 747  * searching the directory normally.  The callers of dx_probe **MUST**
 748  * check for this error code, and make sure it never gets reflected
 749  * back to userspace.
 750  */
 751 static struct dx_frame *
 752 dx_probe(struct ext4_filename *fname, struct inode *dir,
 753          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 754 {
 755         unsigned count, indirect;
 756         struct dx_entry *at, *entries, *p, *q, *m;
 757         struct dx_root *root;
 758         struct dx_frame *frame = frame_in;
 759         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 760         u32 hash;
 761
 762         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 763         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 764         if (IS_ERR(frame->bh))
 765                 return (struct dx_frame *) frame->bh;
 766
 767         root = (struct dx_root *) frame->bh->b_data;
 768         if (root->info.hash_version != DX_HASH_TEA &&
 769             root->info.hash_version != DX_HASH_HALF_MD4 &&
 770             root->info.hash_version != DX_HASH_LEGACY) {
 771                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 772                                    root->info.hash_version);
 773                 goto fail;
 774         }
 775         if (fname)
 776                 hinfo = &fname->hinfo;
 777         hinfo->hash_version = root->info.hash_version;
 778         if (hinfo->hash_version <= DX_HASH_TEA)
 779                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 780         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 781         if (fname && fname_name(fname))
 782                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 783         hash = hinfo->hash;
 784
 785         if (root->info.unused_flags & 1) {
 786                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 787                                    root->info.unused_flags);
 788                 goto fail;
 789         }
 790
 791         indirect = root->info.indirect_levels;
 792         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 793                 ext4_warning(dir->i_sb,
 794                              "Directory (ino: %lu) htree depth %#06x exceed"
 795                              "supported value", dir->i_ino,
 796                              ext4_dir_htree_level(dir->i_sb));
 797                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 798                         ext4_warning(dir->i_sb, "Enable large directory "
 799                                                 "feature to access it");
 800                 }
 801                 goto fail;
 802         }
 803
 804         entries = (struct dx_entry *)(((char *)&root->info) +
 805                                       root->info.info_length);
 806
 807         if (dx_get_limit(entries) != dx_root_limit(dir,
 808                                                    root->info.info_length)) {
 809                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 810                                    dx_get_limit(entries),
 811                                    dx_root_limit(dir, root->info.info_length));
 812                 goto fail;
 813         }
 814
 815         dxtrace(printk("Look up %x", hash));
 816         while (1) {
 817                 count = dx_get_count(entries);
 818                 if (!count || count > dx_get_limit(entries)) {
 819                         ext4_warning_inode(dir,
 820                                            "dx entry: count %u beyond limit %u",
 821                                            count, dx_get_limit(entries));
 822                         goto fail;
 823                 }
 824
 825                 p = entries + 1;
 826                 q = entries + count - 1;
 827                 while (p <= q) {
 828                         m = p + (q - p) / 2;
 829                         dxtrace(printk(KERN_CONT "."));
 830                         if (dx_get_hash(m) > hash)
 831                                 q = m - 1;
 832                         else
 833                                 p = m + 1;
 834                 }
 835
 836                 if (0) { // linear search cross check
 837                         unsigned n = count - 1;
 838                         at = entries;
 839                         while (n--)
 840                         {
 841                                 dxtrace(printk(KERN_CONT ","));
 842                                 if (dx_get_hash(++at) > hash)
 843                                 {
 844                                         at--;
 845                                         break;
 846                                 }
 847                         }
 848                         assert (at == p - 1);
 849                 }
 850
 851                 at = p - 1;
 852                 dxtrace(printk(KERN_CONT " %x->%u\n",
 853                                at == entries ? 0 : dx_get_hash(at),
 854                                dx_get_block(at)));
 855                 frame->entries = entries;
 856                 frame->at = at;
 857                 if (!indirect--)
 858                         return frame;
 859                 frame++;
 860                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 861                 if (IS_ERR(frame->bh)) {
 862                         ret_err = (struct dx_frame *) frame->bh;
 863                         frame->bh = NULL;
 864                         goto fail;
 865                 }
 866                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 867
 868                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 869                         ext4_warning_inode(dir,
 870                                 "dx entry: limit %u != node limit %u",
 871                                 dx_get_limit(entries), dx_node_limit(dir));
 872                         goto fail;
 873                 }
 874         }
 875 fail:
 876         while (frame >= frame_in) {
 877                 brelse(frame->bh);
 878                 frame--;
 879         }
 880
 881         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 882                 ext4_warning_inode(dir,
 883                         "Corrupt directory, running e2fsck is recommended");
 884         return ret_err;
 885 }
 886
 887 static void dx_release(struct dx_frame *frames)
 888 {
 889         struct dx_root_info *info;
 890         int i;
 891         unsigned int indirect_levels;
 892
 893         if (frames[0].bh == NULL)
 894                 return;
 895
 896         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 897         /* save local copy, "info" may be freed after brelse() */
 898         indirect_levels = info->indirect_levels;
 899         for (i = 0; i <= indirect_levels; i++) {
 900                 if (frames[i].bh == NULL)
 901                         break;
 902                 brelse(frames[i].bh);
 903                 frames[i].bh = NULL;
 904         }
 905 }
 906
 907 /*
 908  * This function increments the frame pointer to search the next leaf
 909  * block, and reads in the necessary intervening nodes if the search
 910  * should be necessary.  Whether or not the search is necessary is
 911  * controlled by the hash parameter.  If the hash value is even, then
 912  * the search is only continued if the next block starts with that
 913  * hash value.  This is used if we are searching for a specific file.
 914  *
 915  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 916  *
 917  * This function returns 1 if the caller should continue to search,
 918  * or 0 if it should not.  If there is an error reading one of the
 919  * index blocks, it will a negative error code.
 920  *
 921  * If start_hash is non-null, it will be filled in with the starting
 922  * hash of the next page.
 923  */
 924 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 925                                  struct dx_frame *frame,
 926                                  struct dx_frame *frames,
 927                                  __u32 *start_hash)
 928 {
 929         struct dx_frame *p;
 930         struct buffer_head *bh;
 931         int num_frames = 0;
 932         __u32 bhash;
 933
 934         p = frame;
 935         /*
 936          * Find the next leaf page by incrementing the frame pointer.
 937          * If we run out of entries in the interior node, loop around and
 938          * increment pointer in the parent node.  When we break out of
 939          * this loop, num_frames indicates the number of interior
 940          * nodes need to be read.
 941          */
 942         while (1) {
 943                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 944                         break;
 945                 if (p == frames)
 946                         return 0;
 947                 num_frames++;
 948                 p--;
 949         }
 950
 951         /*
 952          * If the hash is 1, then continue only if the next page has a
 953          * continuation hash of any value.  This is used for readdir
 954          * handling.  Otherwise, check to see if the hash matches the
 955          * desired contiuation hash.  If it doesn't, return since
 956          * there's no point to read in the successive index pages.
 957          */
 958         bhash = dx_get_hash(p->at);
 959         if (start_hash)
 960                 *start_hash = bhash;
 961         if ((hash & 1) == 0) {
 962                 if ((bhash & ~1) != hash)
 963                         return 0;
 964         }
 965         /*
 966          * If the hash is HASH_NB_ALWAYS, we always go to the next
 967          * block so no check is necessary
 968          */
 969         while (num_frames--) {
 970                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 971                 if (IS_ERR(bh))
 972                         return PTR_ERR(bh);
 973                 p++;
 974                 brelse(p->bh);
 975                 p->bh = bh;
 976                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 977         }
 978         return 1;
 979 }
 980
 981
 982 /*
 983  * This function fills a red-black tree with information from a
 984  * directory block.  It returns the number directory entries loaded
 985  * into the tree.  If there is an error it is returned in err.
 986  */
 987 static int htree_dirblock_to_tree(struct file *dir_file,
 988                                   struct inode *dir, ext4_lblk_t block,
 989                                   struct dx_hash_info *hinfo,
 990                                   __u32 start_hash, __u32 start_minor_hash)
 991 {
 992         struct buffer_head *bh;
 993         struct ext4_dir_entry_2 *de, *top;
 994         int err = 0, count = 0;
 995         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 996
 997         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 998                                                         (unsigned long)block));
 999         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1000         if (IS_ERR(bh))
1001                 return PTR_ERR(bh);
1002
1003         de = (struct ext4_dir_entry_2 *) bh->b_data;
1004         top = (struct ext4_dir_entry_2 *) ((char *) de +
1005                                            dir->i_sb->s_blocksize -
1006                                            EXT4_DIR_REC_LEN(0));
1007 #ifdef CONFIG_FS_ENCRYPTION
1008         /* Check if the directory is encrypted */
1009         if (IS_ENCRYPTED(dir)) {
1010                 err = fscrypt_get_encryption_info(dir);
1011                 if (err < 0) {
1012                         brelse(bh);
1013                         return err;
1014                 }
1015                 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1016                                                      &fname_crypto_str);
1017                 if (err < 0) {
1018                         brelse(bh);
1019                         return err;
1020                 }
1021         }
1022 #endif
1023         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1024                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1025                                 bh->b_data, bh->b_size,
1026                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1027                                          + ((char *)de - bh->b_data))) {
1028                         /* silently ignore the rest of the block */
1029                         break;
1030                 }
1031                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1032                 if ((hinfo->hash < start_hash) ||
1033                     ((hinfo->hash == start_hash) &&
1034                      (hinfo->minor_hash < start_minor_hash)))
1035                         continue;
1036                 if (de->inode == 0)
1037                         continue;
1038                 if (!IS_ENCRYPTED(dir)) {
1039                         tmp_str.name = de->name;
1040                         tmp_str.len = de->name_len;
1041                         err = ext4_htree_store_dirent(dir_file,
1042                                    hinfo->hash, hinfo->minor_hash, de,
1043                                    &tmp_str);
1044                 } else {
1045                         int save_len = fname_crypto_str.len;
1046                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1047                                                                 de->name_len);
1048
1049                         /* Directory is encrypted */
1050                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1051                                         hinfo->minor_hash, &de_name,
1052                                         &fname_crypto_str);
1053                         if (err) {
1054                                 count = err;
1055                                 goto errout;
1056                         }
1057                         err = ext4_htree_store_dirent(dir_file,
1058                                    hinfo->hash, hinfo->minor_hash, de,
1059                                         &fname_crypto_str);
1060                         fname_crypto_str.len = save_len;
1061                 }
1062                 if (err != 0) {
1063                         count = err;
1064                         goto errout;
1065                 }
1066                 count++;
1067         }
1068 errout:
1069         brelse(bh);
1070 #ifdef CONFIG_FS_ENCRYPTION
1071         fscrypt_fname_free_buffer(&fname_crypto_str);
1072 #endif
1073         return count;
1074 }
1075
1076
1077 /*
1078  * This function fills a red-black tree with information from a
1079  * directory.  We start scanning the directory in hash order, starting
1080  * at start_hash and start_minor_hash.
1081  *
1082  * This function returns the number of entries inserted into the tree,
1083  * or a negative error code.
1084  */
1085 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1086                          __u32 start_minor_hash, __u32 *next_hash)
1087 {
1088         struct dx_hash_info hinfo;
1089         struct ext4_dir_entry_2 *de;
1090         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1091         struct inode *dir;
1092         ext4_lblk_t block;
1093         int count = 0;
1094         int ret, err;
1095         __u32 hashval;
1096         struct fscrypt_str tmp_str;
1097
1098         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1099                        start_hash, start_minor_hash));
1100         dir = file_inode(dir_file);
1101         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1102                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1103                 if (hinfo.hash_version <= DX_HASH_TEA)
1104                         hinfo.hash_version +=
1105                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1106                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1107                 if (ext4_has_inline_data(dir)) {
1108                         int has_inline_data = 1;
1109                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1110                                                        &hinfo, start_hash,
1111                                                        start_minor_hash,
1112                                                        &has_inline_data);
1113                         if (has_inline_data) {
1114                                 *next_hash = ~0;
1115                                 return count;
1116                         }
1117                 }
1118                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1119                                                start_hash, start_minor_hash);
1120                 *next_hash = ~0;
1121                 return count;
1122         }
1123         hinfo.hash = start_hash;
1124         hinfo.minor_hash = 0;
1125         frame = dx_probe(NULL, dir, &hinfo, frames);
1126         if (IS_ERR(frame))
1127                 return PTR_ERR(frame);
1128
1129         /* Add '.' and '..' from the htree header */
1130         if (!start_hash && !start_minor_hash) {
1131                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1132                 tmp_str.name = de->name;
1133                 tmp_str.len = de->name_len;
1134                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1135                                               de, &tmp_str);
1136                 if (err != 0)
1137                         goto errout;
1138                 count++;
1139         }
1140         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1141                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1142                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1143                 tmp_str.name = de->name;
1144                 tmp_str.len = de->name_len;
1145                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1146                                               de, &tmp_str);
1147                 if (err != 0)
1148                         goto errout;
1149                 count++;
1150         }
1151
1152         while (1) {
1153                 if (fatal_signal_pending(current)) {
1154                         err = -ERESTARTSYS;
1155                         goto errout;
1156                 }
1157                 cond_resched();
1158                 block = dx_get_block(frame->at);
1159                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1160                                              start_hash, start_minor_hash);
1161                 if (ret < 0) {
1162                         err = ret;
1163                         goto errout;
1164                 }
1165                 count += ret;
1166                 hashval = ~0;
1167                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1168                                             frame, frames, &hashval);
1169                 *next_hash = hashval;
1170                 if (ret < 0) {
1171                         err = ret;
1172                         goto errout;
1173                 }
1174                 /*
1175                  * Stop if:  (a) there are no more entries, or
1176                  * (b) we have inserted at least one entry and the
1177                  * next hash value is not a continuation
1178                  */
1179                 if ((ret == 0) ||
1180                     (count && ((hashval & 1) == 0)))
1181                         break;
1182         }
1183         dx_release(frames);
1184         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1185                        "next hash: %x\n", count, *next_hash));
1186         return count;
1187 errout:
1188         dx_release(frames);
1189         return (err);
1190 }
1191
1192 static inline int search_dirblock(struct buffer_head *bh,
1193                                   struct inode *dir,
1194                                   struct ext4_filename *fname,
1195                                   unsigned int offset,
1196                                   struct ext4_dir_entry_2 **res_dir)
1197 {
1198         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1199                                fname, offset, res_dir);
1200 }
1201
1202 /*
1203  * Directory block splitting, compacting
1204  */
1205
1206 /*
1207  * Create map of hash values, offsets, and sizes, stored at end of block.
1208  * Returns number of entries mapped.
1209  */
1210 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1211                        unsigned blocksize, struct dx_hash_info *hinfo,
1212                        struct dx_map_entry *map_tail)
1213 {
1214         int count = 0;
1215         char *base = (char *) de;
1216         struct dx_hash_info h = *hinfo;
1217
1218         while ((char *) de < base + blocksize) {
1219                 if (de->name_len && de->inode) {
1220                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1221                         map_tail--;
1222                         map_tail->hash = h.hash;
1223                         map_tail->offs = ((char *) de - base)>>2;
1224                         map_tail->size = le16_to_cpu(de->rec_len);
1225                         count++;
1226                         cond_resched();
1227                 }
1228                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1229                 de = ext4_next_entry(de, blocksize);
1230         }
1231         return count;
1232 }
1233
1234 /* Sort map by hash value */
1235 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1236 {
1237         struct dx_map_entry *p, *q, *top = map + count - 1;
1238         int more;
1239         /* Combsort until bubble sort doesn't suck */
1240         while (count > 2) {
1241                 count = count*10/13;
1242                 if (count - 9 < 2) /* 9, 10 -> 11 */
1243                         count = 11;
1244                 for (p = top, q = p - count; q >= map; p--, q--)
1245                         if (p->hash < q->hash)
1246                                 swap(*p, *q);
1247         }
1248         /* Garden variety bubble sort */
1249         do {
1250                 more = 0;
1251                 q = top;
1252                 while (q-- > map) {
1253                         if (q[1].hash >= q[0].hash)
1254                                 continue;
1255                         swap(*(q+1), *q);
1256                         more = 1;
1257                 }
1258         } while(more);
1259 }
1260
1261 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1262 {
1263         struct dx_entry *entries = frame->entries;
1264         struct dx_entry *old = frame->at, *new = old + 1;
1265         int count = dx_get_count(entries);
1266
1267         assert(count < dx_get_limit(entries));
1268         assert(old < entries + count);
1269         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1270         dx_set_hash(new, hash);
1271         dx_set_block(new, block);
1272         dx_set_count(entries, count + 1);
1273 }
1274
1275 #ifdef CONFIG_UNICODE
1276 /*
1277  * Test whether a case-insensitive directory entry matches the filename
1278  * being searched for.  If quick is set, assume the name being looked up
1279  * is already in the casefolded form.
1280  *
1281  * Returns: 0 if the directory entry matches, more than 0 if it
1282  * doesn't match or less than zero on error.
1283  */
1284 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1285                     const struct qstr *entry, bool quick)
1286 {
1287         const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1288         const struct unicode_map *um = sbi->s_encoding;
1289         int ret;
1290
1291         if (quick)
1292                 ret = utf8_strncasecmp_folded(um, name, entry);
1293         else
1294                 ret = utf8_strncasecmp(um, name, entry);
1295
1296         if (ret < 0) {
1297                 /* Handle invalid character sequence as either an error
1298                  * or as an opaque byte sequence.
1299                  */
1300                 if (ext4_has_strict_mode(sbi))
1301                         return -EINVAL;
1302
1303                 if (name->len != entry->len)
1304                         return 1;
1305
1306                 return !!memcmp(name->name, entry->name, name->len);
1307         }
1308
1309         return ret;
1310 }
1311
1312 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1313                                   struct fscrypt_str *cf_name)
1314 {
1315         int len;
1316
1317         if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1318                 cf_name->name = NULL;
1319                 return;
1320         }
1321
1322         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1323         if (!cf_name->name)
1324                 return;
1325
1326         len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1327                             iname, cf_name->name,
1328                             EXT4_NAME_LEN);
1329         if (len <= 0) {
1330                 kfree(cf_name->name);
1331                 cf_name->name = NULL;
1332                 return;
1333         }
1334         cf_name->len = (unsigned) len;
1335
1336 }
1337 #endif
1338
1339 /*
1340  * Test whether a directory entry matches the filename being searched for.
1341  *
1342  * Return: %true if the directory entry matches, otherwise %false.
1343  */
1344 static inline bool ext4_match(const struct inode *parent,
1345                               const struct ext4_filename *fname,
1346                               const struct ext4_dir_entry_2 *de)
1347 {
1348         struct fscrypt_name f;
1349 #ifdef CONFIG_UNICODE
1350         const struct qstr entry = {.name = de->name, .len = de->name_len};
1351 #endif
1352
1353         if (!de->inode)
1354                 return false;
1355
1356         f.usr_fname = fname->usr_fname;
1357         f.disk_name = fname->disk_name;
1358 #ifdef CONFIG_FS_ENCRYPTION
1359         f.crypto_buf = fname->crypto_buf;
1360 #endif
1361
1362 #ifdef CONFIG_UNICODE
1363         if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1364                 if (fname->cf_name.name) {
1365                         struct qstr cf = {.name = fname->cf_name.name,
1366                                           .len = fname->cf_name.len};
1367                         return !ext4_ci_compare(parent, &cf, &entry, true);
1368                 }
1369                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1370                                         false);
1371         }
1372 #endif
1373
1374         return fscrypt_match_name(&f, de->name, de->name_len);
1375 }
1376
1377 /*
1378  * Returns 0 if not found, -1 on failure, and 1 on success
1379  */
1380 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1381                     struct inode *dir, struct ext4_filename *fname,
1382                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1383 {
1384         struct ext4_dir_entry_2 * de;
1385         char * dlimit;
1386         int de_len;
1387
1388         de = (struct ext4_dir_entry_2 *)search_buf;
1389         dlimit = search_buf + buf_size;
1390         while ((char *) de < dlimit) {
1391                 /* this code is executed quadratically often */
1392                 /* do minimal checking `by hand' */
1393                 if ((char *) de + de->name_len <= dlimit &&
1394                     ext4_match(dir, fname, de)) {
1395                         /* found a match - just to be sure, do
1396                          * a full check */
1397                         if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1398                                                  bh->b_size, offset))
1399                                 return -1;
1400                         *res_dir = de;
1401                         return 1;
1402                 }
1403                 /* prevent looping on a bad block */
1404                 de_len = ext4_rec_len_from_disk(de->rec_len,
1405                                                 dir->i_sb->s_blocksize);
1406                 if (de_len <= 0)
1407                         return -1;
1408                 offset += de_len;
1409                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1410         }
1411         return 0;
1412 }
1413
1414 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1415                                struct ext4_dir_entry *de)
1416 {
1417         struct super_block *sb = dir->i_sb;
1418
1419         if (!is_dx(dir))
1420                 return 0;
1421         if (block == 0)
1422                 return 1;
1423         if (de->inode == 0 &&
1424             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1425                         sb->s_blocksize)
1426                 return 1;
1427         return 0;
1428 }
1429
1430 /*
1431  *      __ext4_find_entry()
1432  *
1433  * finds an entry in the specified directory with the wanted name. It
1434  * returns the cache buffer in which the entry was found, and the entry
1435  * itself (as a parameter - res_dir). It does NOT read the inode of the
1436  * entry - you'll have to do that yourself if you want to.
1437  *
1438  * The returned buffer_head has ->b_count elevated.  The caller is expected
1439  * to brelse() it when appropriate.
1440  */
1441 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1442                                              struct ext4_filename *fname,
1443                                              struct ext4_dir_entry_2 **res_dir,
1444                                              int *inlined)
1445 {
1446         struct super_block *sb;
1447         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1448         struct buffer_head *bh, *ret = NULL;
1449         ext4_lblk_t start, block;
1450         const u8 *name = fname->usr_fname->name;
1451         size_t ra_max = 0;      /* Number of bh's in the readahead
1452                                    buffer, bh_use[] */
1453         size_t ra_ptr = 0;      /* Current index into readahead
1454                                    buffer */
1455         ext4_lblk_t  nblocks;
1456         int i, namelen, retval;
1457
1458         *res_dir = NULL;
1459         sb = dir->i_sb;
1460         namelen = fname->usr_fname->len;
1461         if (namelen > EXT4_NAME_LEN)
1462                 return NULL;
1463
1464         if (ext4_has_inline_data(dir)) {
1465                 int has_inline_data = 1;
1466                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1467                                              &has_inline_data);
1468                 if (has_inline_data) {
1469                         if (inlined)
1470                                 *inlined = 1;
1471                         goto cleanup_and_exit;
1472                 }
1473         }
1474
1475         if ((namelen <= 2) && (name[0] == '.') &&
1476             (name[1] == '.' || name[1] == '\0')) {
1477                 /*
1478                  * "." or ".." will only be in the first block
1479                  * NFS may look up ".."; "." should be handled by the VFS
1480                  */
1481                 block = start = 0;
1482                 nblocks = 1;
1483                 goto restart;
1484         }
1485         if (is_dx(dir)) {
1486                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1487                 /*
1488                  * On success, or if the error was file not found,
1489                  * return.  Otherwise, fall back to doing a search the
1490                  * old fashioned way.
1491                  */
1492                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1493                         goto cleanup_and_exit;
1494                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1495                                "falling back\n"));
1496                 ret = NULL;
1497         }
1498         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1499         if (!nblocks) {
1500                 ret = NULL;
1501                 goto cleanup_and_exit;
1502         }
1503         start = EXT4_I(dir)->i_dir_start_lookup;
1504         if (start >= nblocks)
1505                 start = 0;
1506         block = start;
1507 restart:
1508         do {
1509                 /*
1510                  * We deal with the read-ahead logic here.
1511                  */
1512                 if (ra_ptr >= ra_max) {
1513                         /* Refill the readahead buffer */
1514                         ra_ptr = 0;
1515                         if (block < start)
1516                                 ra_max = start - block;
1517                         else
1518                                 ra_max = nblocks - block;
1519                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1520                         retval = ext4_bread_batch(dir, block, ra_max,
1521                                                   false /* wait */, bh_use);
1522                         if (retval) {
1523                                 ret = ERR_PTR(retval);
1524                                 ra_max = 0;
1525                                 goto cleanup_and_exit;
1526                         }
1527                 }
1528                 if ((bh = bh_use[ra_ptr++]) == NULL)
1529                         goto next;
1530                 wait_on_buffer(bh);
1531                 if (!buffer_uptodate(bh)) {
1532                         ext4_set_errno(sb, EIO);
1533                         EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1534                                          (unsigned long) block);
1535                         brelse(bh);
1536                         ret = ERR_PTR(-EIO);
1537                         goto cleanup_and_exit;
1538                 }
1539                 if (!buffer_verified(bh) &&
1540                     !is_dx_internal_node(dir, block,
1541                                          (struct ext4_dir_entry *)bh->b_data) &&
1542                     !ext4_dirblock_csum_verify(dir, bh)) {
1543                         ext4_set_errno(sb, EFSBADCRC);
1544                         EXT4_ERROR_INODE(dir, "checksumming directory "
1545                                          "block %lu", (unsigned long)block);
1546                         brelse(bh);
1547                         ret = ERR_PTR(-EFSBADCRC);
1548                         goto cleanup_and_exit;
1549                 }
1550                 set_buffer_verified(bh);
1551                 i = search_dirblock(bh, dir, fname,
1552                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1553                 if (i == 1) {
1554                         EXT4_I(dir)->i_dir_start_lookup = block;
1555                         ret = bh;
1556                         goto cleanup_and_exit;
1557                 } else {
1558                         brelse(bh);
1559                         if (i < 0)
1560                                 goto cleanup_and_exit;
1561                 }
1562         next:
1563                 if (++block >= nblocks)
1564                         block = 0;
1565         } while (block != start);
1566
1567         /*
1568          * If the directory has grown while we were searching, then
1569          * search the last part of the directory before giving up.
1570          */
1571         block = nblocks;
1572         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1573         if (block < nblocks) {
1574                 start = 0;
1575                 goto restart;
1576         }
1577
1578 cleanup_and_exit:
1579         /* Clean up the read-ahead blocks */
1580         for (; ra_ptr < ra_max; ra_ptr++)
1581                 brelse(bh_use[ra_ptr]);
1582         return ret;
1583 }
1584
1585 static struct buffer_head *ext4_find_entry(struct inode *dir,
1586                                            const struct qstr *d_name,
1587                                            struct ext4_dir_entry_2 **res_dir,
1588                                            int *inlined)
1589 {
1590         int err;
1591         struct ext4_filename fname;
1592         struct buffer_head *bh;
1593
1594         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1595         if (err == -ENOENT)
1596                 return NULL;
1597         if (err)
1598                 return ERR_PTR(err);
1599
1600         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1601
1602         ext4_fname_free_filename(&fname);
1603         return bh;
1604 }
1605
1606 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1607                                              struct dentry *dentry,
1608                                              struct ext4_dir_entry_2 **res_dir)
1609 {
1610         int err;
1611         struct ext4_filename fname;
1612         struct buffer_head *bh;
1613
1614         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1615         if (err == -ENOENT)
1616                 return NULL;
1617         if (err)
1618                 return ERR_PTR(err);
1619
1620         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1621
1622         ext4_fname_free_filename(&fname);
1623         return bh;
1624 }
1625
1626 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1627                         struct ext4_filename *fname,
1628                         struct ext4_dir_entry_2 **res_dir)
1629 {
1630         struct super_block * sb = dir->i_sb;
1631         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1632         struct buffer_head *bh;
1633         ext4_lblk_t block;
1634         int retval;
1635
1636 #ifdef CONFIG_FS_ENCRYPTION
1637         *res_dir = NULL;
1638 #endif
1639         frame = dx_probe(fname, dir, NULL, frames);
1640         if (IS_ERR(frame))
1641                 return (struct buffer_head *) frame;
1642         do {
1643                 block = dx_get_block(frame->at);
1644                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1645                 if (IS_ERR(bh))
1646                         goto errout;
1647
1648                 retval = search_dirblock(bh, dir, fname,
1649                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1650                                          res_dir);
1651                 if (retval == 1)
1652                         goto success;
1653                 brelse(bh);
1654                 if (retval == -1) {
1655                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1656                         goto errout;
1657                 }
1658
1659                 /* Check to see if we should continue to search */
1660                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1661                                                frames, NULL);
1662                 if (retval < 0) {
1663                         ext4_warning_inode(dir,
1664                                 "error %d reading directory index block",
1665                                 retval);
1666                         bh = ERR_PTR(retval);
1667                         goto errout;
1668                 }
1669         } while (retval == 1);
1670
1671         bh = NULL;
1672 errout:
1673         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1674 success:
1675         dx_release(frames);
1676         return bh;
1677 }
1678
1679 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1680 {
1681         struct inode *inode;
1682         struct ext4_dir_entry_2 *de;
1683         struct buffer_head *bh;
1684
1685         if (dentry->d_name.len > EXT4_NAME_LEN)
1686                 return ERR_PTR(-ENAMETOOLONG);
1687
1688         bh = ext4_lookup_entry(dir, dentry, &de);
1689         if (IS_ERR(bh))
1690                 return ERR_CAST(bh);
1691         inode = NULL;
1692         if (bh) {
1693                 __u32 ino = le32_to_cpu(de->inode);
1694                 brelse(bh);
1695                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1696                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1697                         return ERR_PTR(-EFSCORRUPTED);
1698                 }
1699                 if (unlikely(ino == dir->i_ino)) {
1700                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1701                                          dentry);
1702                         return ERR_PTR(-EFSCORRUPTED);
1703                 }
1704                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1705                 if (inode == ERR_PTR(-ESTALE)) {
1706                         EXT4_ERROR_INODE(dir,
1707                                          "deleted inode referenced: %u",
1708                                          ino);
1709                         return ERR_PTR(-EFSCORRUPTED);
1710                 }
1711                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1712                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1713                     !fscrypt_has_permitted_context(dir, inode)) {
1714                         ext4_warning(inode->i_sb,
1715                                      "Inconsistent encryption contexts: %lu/%lu",
1716                                      dir->i_ino, inode->i_ino);
1717                         iput(inode);
1718                         return ERR_PTR(-EPERM);
1719                 }
1720         }
1721
1722 #ifdef CONFIG_UNICODE
1723         if (!inode && IS_CASEFOLDED(dir)) {
1724                 /* Eventually we want to call d_add_ci(dentry, NULL)
1725                  * for negative dentries in the encoding case as
1726                  * well.  For now, prevent the negative dentry
1727                  * from being cached.
1728                  */
1729                 return NULL;
1730         }
1731 #endif
1732         return d_splice_alias(inode, dentry);
1733 }
1734
1735
1736 struct dentry *ext4_get_parent(struct dentry *child)
1737 {
1738         __u32 ino;
1739         static const struct qstr dotdot = QSTR_INIT("..", 2);
1740         struct ext4_dir_entry_2 * de;
1741         struct buffer_head *bh;
1742
1743         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1744         if (IS_ERR(bh))
1745                 return ERR_CAST(bh);
1746         if (!bh)
1747                 return ERR_PTR(-ENOENT);
1748         ino = le32_to_cpu(de->inode);
1749         brelse(bh);
1750
1751         if (!ext4_valid_inum(child->d_sb, ino)) {
1752                 EXT4_ERROR_INODE(d_inode(child),
1753                                  "bad parent inode number: %u", ino);
1754                 return ERR_PTR(-EFSCORRUPTED);
1755         }
1756
1757         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1758 }
1759
1760 /*
1761  * Move count entries from end of map between two memory locations.
1762  * Returns pointer to last entry moved.
1763  */
1764 static struct ext4_dir_entry_2 *
1765 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1766                 unsigned blocksize)
1767 {
1768         unsigned rec_len = 0;
1769
1770         while (count--) {
1771                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1772                                                 (from + (map->offs<<2));
1773                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1774                 memcpy (to, de, rec_len);
1775                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1776                                 ext4_rec_len_to_disk(rec_len, blocksize);
1777                 de->inode = 0;
1778                 map++;
1779                 to += rec_len;
1780         }
1781         return (struct ext4_dir_entry_2 *) (to - rec_len);
1782 }
1783
1784 /*
1785  * Compact each dir entry in the range to the minimal rec_len.
1786  * Returns pointer to last entry in range.
1787  */
1788 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1789 {
1790         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1791         unsigned rec_len = 0;
1792
1793         prev = to = de;
1794         while ((char*)de < base + blocksize) {
1795                 next = ext4_next_entry(de, blocksize);
1796                 if (de->inode && de->name_len) {
1797                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1798                         if (de > to)
1799                                 memmove(to, de, rec_len);
1800                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1801                         prev = to;
1802                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1803                 }
1804                 de = next;
1805         }
1806         return prev;
1807 }
1808
1809 /*
1810  * Split a full leaf block to make room for a new dir entry.
1811  * Allocate a new block, and move entries so that they are approx. equally full.
1812  * Returns pointer to de in block into which the new entry will be inserted.
1813  */
1814 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1815                         struct buffer_head **bh,struct dx_frame *frame,
1816                         struct dx_hash_info *hinfo)
1817 {
1818         unsigned blocksize = dir->i_sb->s_blocksize;
1819         unsigned count, continued;
1820         struct buffer_head *bh2;
1821         ext4_lblk_t newblock;
1822         u32 hash2;
1823         struct dx_map_entry *map;
1824         char *data1 = (*bh)->b_data, *data2;
1825         unsigned split, move, size;
1826         struct ext4_dir_entry_2 *de = NULL, *de2;
1827         int     csum_size = 0;
1828         int     err = 0, i;
1829
1830         if (ext4_has_metadata_csum(dir->i_sb))
1831                 csum_size = sizeof(struct ext4_dir_entry_tail);
1832
1833         bh2 = ext4_append(handle, dir, &newblock);
1834         if (IS_ERR(bh2)) {
1835                 brelse(*bh);
1836                 *bh = NULL;
1837                 return (struct ext4_dir_entry_2 *) bh2;
1838         }
1839
1840         BUFFER_TRACE(*bh, "get_write_access");
1841         err = ext4_journal_get_write_access(handle, *bh);
1842         if (err)
1843                 goto journal_error;
1844
1845         BUFFER_TRACE(frame->bh, "get_write_access");
1846         err = ext4_journal_get_write_access(handle, frame->bh);
1847         if (err)
1848                 goto journal_error;
1849
1850         data2 = bh2->b_data;
1851
1852         /* create map in the end of data2 block */
1853         map = (struct dx_map_entry *) (data2 + blocksize);
1854         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1855                              blocksize, hinfo, map);
1856         map -= count;
1857         dx_sort_map(map, count);
1858         /* Split the existing block in the middle, size-wise */
1859         size = 0;
1860         move = 0;
1861         for (i = count-1; i >= 0; i--) {
1862                 /* is more than half of this entry in 2nd half of the block? */
1863                 if (size + map[i].size/2 > blocksize/2)
1864                         break;
1865                 size += map[i].size;
1866                 move++;
1867         }
1868         /* map index at which we will split */
1869         split = count - move;
1870         hash2 = map[split].hash;
1871         continued = hash2 == map[split - 1].hash;
1872         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1873                         (unsigned long)dx_get_block(frame->at),
1874                                         hash2, split, count-split));
1875
1876         /* Fancy dance to stay within two buffers */
1877         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1878                               blocksize);
1879         de = dx_pack_dirents(data1, blocksize);
1880         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1881                                            (char *) de,
1882                                            blocksize);
1883         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1884                                             (char *) de2,
1885                                             blocksize);
1886         if (csum_size) {
1887                 ext4_initialize_dirent_tail(*bh, blocksize);
1888                 ext4_initialize_dirent_tail(bh2, blocksize);
1889         }
1890
1891         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1892                         blocksize, 1));
1893         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1894                         blocksize, 1));
1895
1896         /* Which block gets the new entry? */
1897         if (hinfo->hash >= hash2) {
1898                 swap(*bh, bh2);
1899                 de = de2;
1900         }
1901         dx_insert_block(frame, hash2 + continued, newblock);
1902         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1903         if (err)
1904                 goto journal_error;
1905         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1906         if (err)
1907                 goto journal_error;
1908         brelse(bh2);
1909         dxtrace(dx_show_index("frame", frame->entries));
1910         return de;
1911
1912 journal_error:
1913         brelse(*bh);
1914         brelse(bh2);
1915         *bh = NULL;
1916         ext4_std_error(dir->i_sb, err);
1917         return ERR_PTR(err);
1918 }
1919
1920 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1921                       struct buffer_head *bh,
1922                       void *buf, int buf_size,
1923                       struct ext4_filename *fname,
1924                       struct ext4_dir_entry_2 **dest_de)
1925 {
1926         struct ext4_dir_entry_2 *de;
1927         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1928         int nlen, rlen;
1929         unsigned int offset = 0;
1930         char *top;
1931
1932         de = (struct ext4_dir_entry_2 *)buf;
1933         top = buf + buf_size - reclen;
1934         while ((char *) de <= top) {
1935                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1936                                          buf, buf_size, offset))
1937                         return -EFSCORRUPTED;
1938                 if (ext4_match(dir, fname, de))
1939                         return -EEXIST;
1940                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1941                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1942                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1943                         break;
1944                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1945                 offset += rlen;
1946         }
1947         if ((char *) de > top)
1948                 return -ENOSPC;
1949
1950         *dest_de = de;
1951         return 0;
1952 }
1953
1954 void ext4_insert_dentry(struct inode *inode,
1955                         struct ext4_dir_entry_2 *de,
1956                         int buf_size,
1957                         struct ext4_filename *fname)
1958 {
1959
1960         int nlen, rlen;
1961
1962         nlen = EXT4_DIR_REC_LEN(de->name_len);
1963         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1964         if (de->inode) {
1965                 struct ext4_dir_entry_2 *de1 =
1966                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1967                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1968                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1969                 de = de1;
1970         }
1971         de->file_type = EXT4_FT_UNKNOWN;
1972         de->inode = cpu_to_le32(inode->i_ino);
1973         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1974         de->name_len = fname_len(fname);
1975         memcpy(de->name, fname_name(fname), fname_len(fname));
1976 }
1977
1978 /*
1979  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1980  * it points to a directory entry which is guaranteed to be large
1981  * enough for new directory entry.  If de is NULL, then
1982  * add_dirent_to_buf will attempt search the directory block for
1983  * space.  It will return -ENOSPC if no space is available, and -EIO
1984  * and -EEXIST if directory entry already exists.
1985  */
1986 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1987                              struct inode *dir,
1988                              struct inode *inode, struct ext4_dir_entry_2 *de,
1989                              struct buffer_head *bh)
1990 {
1991         unsigned int    blocksize = dir->i_sb->s_blocksize;
1992         int             csum_size = 0;
1993         int             err;
1994
1995         if (ext4_has_metadata_csum(inode->i_sb))
1996                 csum_size = sizeof(struct ext4_dir_entry_tail);
1997
1998         if (!de) {
1999                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2000                                         blocksize - csum_size, fname, &de);
2001                 if (err)
2002                         return err;
2003         }
2004         BUFFER_TRACE(bh, "get_write_access");
2005         err = ext4_journal_get_write_access(handle, bh);
2006         if (err) {
2007                 ext4_std_error(dir->i_sb, err);
2008                 return err;
2009         }
2010
2011         /* By now the buffer is marked for journaling */
2012         ext4_insert_dentry(inode, de, blocksize, fname);
2013
2014         /*
2015          * XXX shouldn't update any times until successful
2016          * completion of syscall, but too many callers depend
2017          * on this.
2018          *
2019          * XXX similarly, too many callers depend on
2020          * ext4_new_inode() setting the times, but error
2021          * recovery deletes the inode, so the worst that can
2022          * happen is that the times are slightly out of date
2023          * and/or different from the directory change time.
2024          */
2025         dir->i_mtime = dir->i_ctime = current_time(dir);
2026         ext4_update_dx_flag(dir);
2027         inode_inc_iversion(dir);
2028         ext4_mark_inode_dirty(handle, dir);
2029         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2030         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2031         if (err)
2032                 ext4_std_error(dir->i_sb, err);
2033         return 0;
2034 }
2035
2036 /*
2037  * This converts a one block unindexed directory to a 3 block indexed
2038  * directory, and adds the dentry to the indexed directory.
2039  */
2040 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2041                             struct inode *dir,
2042                             struct inode *inode, struct buffer_head *bh)
2043 {
2044         struct buffer_head *bh2;
2045         struct dx_root  *root;
2046         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2047         struct dx_entry *entries;
2048         struct ext4_dir_entry_2 *de, *de2;
2049         char            *data2, *top;
2050         unsigned        len;
2051         int             retval;
2052         unsigned        blocksize;
2053         ext4_lblk_t  block;
2054         struct fake_dirent *fde;
2055         int csum_size = 0;
2056
2057         if (ext4_has_metadata_csum(inode->i_sb))
2058                 csum_size = sizeof(struct ext4_dir_entry_tail);
2059
2060         blocksize =  dir->i_sb->s_blocksize;
2061         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2062         BUFFER_TRACE(bh, "get_write_access");
2063         retval = ext4_journal_get_write_access(handle, bh);
2064         if (retval) {
2065                 ext4_std_error(dir->i_sb, retval);
2066                 brelse(bh);
2067                 return retval;
2068         }
2069         root = (struct dx_root *) bh->b_data;
2070
2071         /* The 0th block becomes the root, move the dirents out */
2072         fde = &root->dotdot;
2073         de = (struct ext4_dir_entry_2 *)((char *)fde +
2074                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2075         if ((char *) de >= (((char *) root) + blocksize)) {
2076                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2077                 brelse(bh);
2078                 return -EFSCORRUPTED;
2079         }
2080         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2081
2082         /* Allocate new block for the 0th block's dirents */
2083         bh2 = ext4_append(handle, dir, &block);
2084         if (IS_ERR(bh2)) {
2085                 brelse(bh);
2086                 return PTR_ERR(bh2);
2087         }
2088         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2089         data2 = bh2->b_data;
2090
2091         memcpy(data2, de, len);
2092         de = (struct ext4_dir_entry_2 *) data2;
2093         top = data2 + len;
2094         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2095                 de = de2;
2096         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2097                                            (char *) de, blocksize);
2098
2099         if (csum_size)
2100                 ext4_initialize_dirent_tail(bh2, blocksize);
2101
2102         /* Initialize the root; the dot dirents already exist */
2103         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2104         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2105                                            blocksize);
2106         memset (&root->info, 0, sizeof(root->info));
2107         root->info.info_length = sizeof(root->info);
2108         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2109         entries = root->entries;
2110         dx_set_block(entries, 1);
2111         dx_set_count(entries, 1);
2112         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2113
2114         /* Initialize as for dx_probe */
2115         fname->hinfo.hash_version = root->info.hash_version;
2116         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2117                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2118         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2119         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2120
2121         memset(frames, 0, sizeof(frames));
2122         frame = frames;
2123         frame->entries = entries;
2124         frame->at = entries;
2125         frame->bh = bh;
2126
2127         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2128         if (retval)
2129                 goto out_frames;
2130         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2131         if (retval)
2132                 goto out_frames;
2133
2134         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2135         if (IS_ERR(de)) {
2136                 retval = PTR_ERR(de);
2137                 goto out_frames;
2138         }
2139
2140         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2141 out_frames:
2142         /*
2143          * Even if the block split failed, we have to properly write
2144          * out all the changes we did so far. Otherwise we can end up
2145          * with corrupted filesystem.
2146          */
2147         if (retval)
2148                 ext4_mark_inode_dirty(handle, dir);
2149         dx_release(frames);
2150         brelse(bh2);
2151         return retval;
2152 }
2153
2154 /*
2155  *      ext4_add_entry()
2156  *
2157  * adds a file entry to the specified directory, using the same
2158  * semantics as ext4_find_entry(). It returns NULL if it failed.
2159  *
2160  * NOTE!! The inode part of 'de' is left at 0 - which means you
2161  * may not sleep between calling this and putting something into
2162  * the entry, as someone else might have used it while you slept.
2163  */
2164 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2165                           struct inode *inode)
2166 {
2167         struct inode *dir = d_inode(dentry->d_parent);
2168         struct buffer_head *bh = NULL;
2169         struct ext4_dir_entry_2 *de;
2170         struct super_block *sb;
2171 #ifdef CONFIG_UNICODE
2172         struct ext4_sb_info *sbi;
2173 #endif
2174         struct ext4_filename fname;
2175         int     retval;
2176         int     dx_fallback=0;
2177         unsigned blocksize;
2178         ext4_lblk_t block, blocks;
2179         int     csum_size = 0;
2180
2181         if (ext4_has_metadata_csum(inode->i_sb))
2182                 csum_size = sizeof(struct ext4_dir_entry_tail);
2183
2184         sb = dir->i_sb;
2185         blocksize = sb->s_blocksize;
2186         if (!dentry->d_name.len)
2187                 return -EINVAL;
2188
2189 #ifdef CONFIG_UNICODE
2190         sbi = EXT4_SB(sb);
2191         if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2192             sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2193                 return -EINVAL;
2194 #endif
2195
2196         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2197         if (retval)
2198                 return retval;
2199
2200         if (ext4_has_inline_data(dir)) {
2201                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2202                 if (retval < 0)
2203                         goto out;
2204                 if (retval == 1) {
2205                         retval = 0;
2206                         goto out;
2207                 }
2208         }
2209
2210         if (is_dx(dir)) {
2211                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2212                 if (!retval || (retval != ERR_BAD_DX_DIR))
2213                         goto out;
2214                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2215                 dx_fallback++;
2216                 ext4_mark_inode_dirty(handle, dir);
2217         }
2218         blocks = dir->i_size >> sb->s_blocksize_bits;
2219         for (block = 0; block < blocks; block++) {
2220                 bh = ext4_read_dirblock(dir, block, DIRENT);
2221                 if (bh == NULL) {
2222                         bh = ext4_bread(handle, dir, block,
2223                                         EXT4_GET_BLOCKS_CREATE);
2224                         goto add_to_new_block;
2225                 }
2226                 if (IS_ERR(bh)) {
2227                         retval = PTR_ERR(bh);
2228                         bh = NULL;
2229                         goto out;
2230                 }
2231                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2232                                            NULL, bh);
2233                 if (retval != -ENOSPC)
2234                         goto out;
2235
2236                 if (blocks == 1 && !dx_fallback &&
2237                     ext4_has_feature_dir_index(sb)) {
2238                         retval = make_indexed_dir(handle, &fname, dir,
2239                                                   inode, bh);
2240                         bh = NULL; /* make_indexed_dir releases bh */
2241                         goto out;
2242                 }
2243                 brelse(bh);
2244         }
2245         bh = ext4_append(handle, dir, &block);
2246 add_to_new_block:
2247         if (IS_ERR(bh)) {
2248                 retval = PTR_ERR(bh);
2249                 bh = NULL;
2250                 goto out;
2251         }
2252         de = (struct ext4_dir_entry_2 *) bh->b_data;
2253         de->inode = 0;
2254         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2255
2256         if (csum_size)
2257                 ext4_initialize_dirent_tail(bh, blocksize);
2258
2259         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2260 out:
2261         ext4_fname_free_filename(&fname);
2262         brelse(bh);
2263         if (retval == 0)
2264                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2265         return retval;
2266 }
2267
2268 /*
2269  * Returns 0 for success, or a negative error value
2270  */
2271 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2272                              struct inode *dir, struct inode *inode)
2273 {
2274         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2275         struct dx_entry *entries, *at;
2276         struct buffer_head *bh;
2277         struct super_block *sb = dir->i_sb;
2278         struct ext4_dir_entry_2 *de;
2279         int restart;
2280         int err;
2281
2282 again:
2283         restart = 0;
2284         frame = dx_probe(fname, dir, NULL, frames);
2285         if (IS_ERR(frame))
2286                 return PTR_ERR(frame);
2287         entries = frame->entries;
2288         at = frame->at;
2289         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2290         if (IS_ERR(bh)) {
2291                 err = PTR_ERR(bh);
2292                 bh = NULL;
2293                 goto cleanup;
2294         }
2295
2296         BUFFER_TRACE(bh, "get_write_access");
2297         err = ext4_journal_get_write_access(handle, bh);
2298         if (err)
2299                 goto journal_error;
2300
2301         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2302         if (err != -ENOSPC)
2303                 goto cleanup;
2304
2305         err = 0;
2306         /* Block full, should compress but for now just split */
2307         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2308                        dx_get_count(entries), dx_get_limit(entries)));
2309         /* Need to split index? */
2310         if (dx_get_count(entries) == dx_get_limit(entries)) {
2311                 ext4_lblk_t newblock;
2312                 int levels = frame - frames + 1;
2313                 unsigned int icount;
2314                 int add_level = 1;
2315                 struct dx_entry *entries2;
2316                 struct dx_node *node2;
2317                 struct buffer_head *bh2;
2318
2319                 while (frame > frames) {
2320                         if (dx_get_count((frame - 1)->entries) <
2321                             dx_get_limit((frame - 1)->entries)) {
2322                                 add_level = 0;
2323                                 break;
2324                         }
2325                         frame--; /* split higher index block */
2326                         at = frame->at;
2327                         entries = frame->entries;
2328                         restart = 1;
2329                 }
2330                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2331                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2332                                          "reach max htree level :%d",
2333                                          dir->i_ino, levels);
2334                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2335                                 ext4_warning(sb, "Large directory feature is "
2336                                                  "not enabled on this "
2337                                                  "filesystem");
2338                         }
2339                         err = -ENOSPC;
2340                         goto cleanup;
2341                 }
2342                 icount = dx_get_count(entries);
2343                 bh2 = ext4_append(handle, dir, &newblock);
2344                 if (IS_ERR(bh2)) {
2345                         err = PTR_ERR(bh2);
2346                         goto cleanup;
2347                 }
2348                 node2 = (struct dx_node *)(bh2->b_data);
2349                 entries2 = node2->entries;
2350                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2351                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2352                                                            sb->s_blocksize);
2353                 BUFFER_TRACE(frame->bh, "get_write_access");
2354                 err = ext4_journal_get_write_access(handle, frame->bh);
2355                 if (err)
2356                         goto journal_error;
2357                 if (!add_level) {
2358                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2359                         unsigned hash2 = dx_get_hash(entries + icount1);
2360                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2361                                        icount1, icount2));
2362
2363                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2364                         err = ext4_journal_get_write_access(handle,
2365                                                              (frame - 1)->bh);
2366                         if (err)
2367                                 goto journal_error;
2368
2369                         memcpy((char *) entries2, (char *) (entries + icount1),
2370                                icount2 * sizeof(struct dx_entry));
2371                         dx_set_count(entries, icount1);
2372                         dx_set_count(entries2, icount2);
2373                         dx_set_limit(entries2, dx_node_limit(dir));
2374
2375                         /* Which index block gets the new entry? */
2376                         if (at - entries >= icount1) {
2377                                 frame->at = at = at - entries - icount1 + entries2;
2378                                 frame->entries = entries = entries2;
2379                                 swap(frame->bh, bh2);
2380                         }
2381                         dx_insert_block((frame - 1), hash2, newblock);
2382                         dxtrace(dx_show_index("node", frame->entries));
2383                         dxtrace(dx_show_index("node",
2384                                ((struct dx_node *) bh2->b_data)->entries));
2385                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2386                         if (err)
2387                                 goto journal_error;
2388                         brelse (bh2);
2389                         err = ext4_handle_dirty_dx_node(handle, dir,
2390                                                    (frame - 1)->bh);
2391                         if (err)
2392                                 goto journal_error;
2393                         if (restart) {
2394                                 err = ext4_handle_dirty_dx_node(handle, dir,
2395                                                            frame->bh);
2396                                 goto journal_error;
2397                         }
2398                 } else {
2399                         struct dx_root *dxroot;
2400                         memcpy((char *) entries2, (char *) entries,
2401                                icount * sizeof(struct dx_entry));
2402                         dx_set_limit(entries2, dx_node_limit(dir));
2403
2404                         /* Set up root */
2405                         dx_set_count(entries, 1);
2406                         dx_set_block(entries + 0, newblock);
2407                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2408                         dxroot->info.indirect_levels += 1;
2409                         dxtrace(printk(KERN_DEBUG
2410                                        "Creating %d level index...\n",
2411                                        dxroot->info.indirect_levels));
2412                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2413                         if (err)
2414                                 goto journal_error;
2415                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2416                         brelse(bh2);
2417                         restart = 1;
2418                         goto journal_error;
2419                 }
2420         }
2421         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2422         if (IS_ERR(de)) {
2423                 err = PTR_ERR(de);
2424                 goto cleanup;
2425         }
2426         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2427         goto cleanup;
2428
2429 journal_error:
2430         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2431 cleanup:
2432         brelse(bh);
2433         dx_release(frames);
2434         /* @restart is true means htree-path has been changed, we need to
2435          * repeat dx_probe() to find out valid htree-path
2436          */
2437         if (restart && err == 0)
2438                 goto again;
2439         return err;
2440 }
2441
2442 /*
2443  * ext4_generic_delete_entry deletes a directory entry by merging it
2444  * with the previous entry
2445  */
2446 int ext4_generic_delete_entry(handle_t *handle,
2447                               struct inode *dir,
2448                               struct ext4_dir_entry_2 *de_del,
2449                               struct buffer_head *bh,
2450                               void *entry_buf,
2451                               int buf_size,
2452                               int csum_size)
2453 {
2454         struct ext4_dir_entry_2 *de, *pde;
2455         unsigned int blocksize = dir->i_sb->s_blocksize;
2456         int i;
2457
2458         i = 0;
2459         pde = NULL;
2460         de = (struct ext4_dir_entry_2 *)entry_buf;
2461         while (i < buf_size - csum_size) {
2462                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2463                                          bh->b_data, bh->b_size, i))
2464                         return -EFSCORRUPTED;
2465                 if (de == de_del)  {
2466                         if (pde)
2467                                 pde->rec_len = ext4_rec_len_to_disk(
2468                                         ext4_rec_len_from_disk(pde->rec_len,
2469                                                                blocksize) +
2470                                         ext4_rec_len_from_disk(de->rec_len,
2471                                                                blocksize),
2472                                         blocksize);
2473                         else
2474                                 de->inode = 0;
2475                         inode_inc_iversion(dir);
2476                         return 0;
2477                 }
2478                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2479                 pde = de;
2480                 de = ext4_next_entry(de, blocksize);
2481         }
2482         return -ENOENT;
2483 }
2484
2485 static int ext4_delete_entry(handle_t *handle,
2486                              struct inode *dir,
2487                              struct ext4_dir_entry_2 *de_del,
2488                              struct buffer_head *bh)
2489 {
2490         int err, csum_size = 0;
2491
2492         if (ext4_has_inline_data(dir)) {
2493                 int has_inline_data = 1;
2494                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2495                                                &has_inline_data);
2496                 if (has_inline_data)
2497                         return err;
2498         }
2499
2500         if (ext4_has_metadata_csum(dir->i_sb))
2501                 csum_size = sizeof(struct ext4_dir_entry_tail);
2502
2503         BUFFER_TRACE(bh, "get_write_access");
2504         err = ext4_journal_get_write_access(handle, bh);
2505         if (unlikely(err))
2506                 goto out;
2507
2508         err = ext4_generic_delete_entry(handle, dir, de_del,
2509                                         bh, bh->b_data,
2510                                         dir->i_sb->s_blocksize, csum_size);
2511         if (err)
2512                 goto out;
2513
2514         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2515         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2516         if (unlikely(err))
2517                 goto out;
2518
2519         return 0;
2520 out:
2521         if (err != -ENOENT)
2522                 ext4_std_error(dir->i_sb, err);
2523         return err;
2524 }
2525
2526 /*
2527  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2528  * since this indicates that nlinks count was previously 1 to avoid overflowing
2529  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2530  * that subdirectory link counts are not being maintained accurately.
2531  *
2532  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2533  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2534  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2535  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2536  */
2537 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2538 {
2539         inc_nlink(inode);
2540         if (is_dx(inode) &&
2541             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2542                 set_nlink(inode, 1);
2543 }
2544
2545 /*
2546  * If a directory had nlink == 1, then we should let it be 1. This indicates
2547  * directory has >EXT4_LINK_MAX subdirs.
2548  */
2549 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2550 {
2551         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2552                 drop_nlink(inode);
2553 }
2554
2555
2556 /*
2557  * Add non-directory inode to a directory. On success, the inode reference is
2558  * consumed by dentry is instantiation. This is also indicated by clearing of
2559  * *inodep pointer. On failure, the caller is responsible for dropping the
2560  * inode reference in the safe context.
2561  */
2562 static int ext4_add_nondir(handle_t *handle,
2563                 struct dentry *dentry, struct inode **inodep)
2564 {
2565         struct inode *dir = d_inode(dentry->d_parent);
2566         struct inode *inode = *inodep;
2567         int err = ext4_add_entry(handle, dentry, inode);
2568         if (!err) {
2569                 ext4_mark_inode_dirty(handle, inode);
2570                 if (IS_DIRSYNC(dir))
2571                         ext4_handle_sync(handle);
2572                 d_instantiate_new(dentry, inode);
2573                 *inodep = NULL;
2574                 return 0;
2575         }
2576         drop_nlink(inode);
2577         ext4_orphan_add(handle, inode);
2578         unlock_new_inode(inode);
2579         return err;
2580 }
2581
2582 /*
2583  * By the time this is called, we already have created
2584  * the directory cache entry for the new file, but it
2585  * is so far negative - it has no inode.
2586  *
2587  * If the create succeeds, we fill in the inode information
2588  * with d_instantiate().
2589  */
2590 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2591                        bool excl)
2592 {
2593         handle_t *handle;
2594         struct inode *inode;
2595         int err, credits, retries = 0;
2596
2597         err = dquot_initialize(dir);
2598         if (err)
2599                 return err;
2600
2601         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2602                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2603 retry:
2604         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2605                                             NULL, EXT4_HT_DIR, credits);
2606         handle = ext4_journal_current_handle();
2607         err = PTR_ERR(inode);
2608         if (!IS_ERR(inode)) {
2609                 inode->i_op = &ext4_file_inode_operations;
2610                 inode->i_fop = &ext4_file_operations;
2611                 ext4_set_aops(inode);
2612                 err = ext4_add_nondir(handle, dentry, &inode);
2613         }
2614         if (handle)
2615                 ext4_journal_stop(handle);
2616         if (!IS_ERR_OR_NULL(inode))
2617                 iput(inode);
2618         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2619                 goto retry;
2620         return err;
2621 }
2622
2623 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2624                       umode_t mode, dev_t rdev)
2625 {
2626         handle_t *handle;
2627         struct inode *inode;
2628         int err, credits, retries = 0;
2629
2630         err = dquot_initialize(dir);
2631         if (err)
2632                 return err;
2633
2634         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2635                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2636 retry:
2637         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2638                                             NULL, EXT4_HT_DIR, credits);
2639         handle = ext4_journal_current_handle();
2640         err = PTR_ERR(inode);
2641         if (!IS_ERR(inode)) {
2642                 init_special_inode(inode, inode->i_mode, rdev);
2643                 inode->i_op = &ext4_special_inode_operations;
2644                 err = ext4_add_nondir(handle, dentry, &inode);
2645         }
2646         if (handle)
2647                 ext4_journal_stop(handle);
2648         if (!IS_ERR_OR_NULL(inode))
2649                 iput(inode);
2650         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2651                 goto retry;
2652         return err;
2653 }
2654
2655 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2656 {
2657         handle_t *handle;
2658         struct inode *inode;
2659         int err, retries = 0;
2660
2661         err = dquot_initialize(dir);
2662         if (err)
2663                 return err;
2664
2665 retry:
2666         inode = ext4_new_inode_start_handle(dir, mode,
2667                                             NULL, 0, NULL,
2668                                             EXT4_HT_DIR,
2669                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2670                           4 + EXT4_XATTR_TRANS_BLOCKS);
2671         handle = ext4_journal_current_handle();
2672         err = PTR_ERR(inode);
2673         if (!IS_ERR(inode)) {
2674                 inode->i_op = &ext4_file_inode_operations;
2675                 inode->i_fop = &ext4_file_operations;
2676                 ext4_set_aops(inode);
2677                 d_tmpfile(dentry, inode);
2678                 err = ext4_orphan_add(handle, inode);
2679                 if (err)
2680                         goto err_unlock_inode;
2681                 mark_inode_dirty(inode);
2682                 unlock_new_inode(inode);
2683         }
2684         if (handle)
2685                 ext4_journal_stop(handle);
2686         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2687                 goto retry;
2688         return err;
2689 err_unlock_inode:
2690         ext4_journal_stop(handle);
2691         unlock_new_inode(inode);
2692         return err;
2693 }
2694
2695 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2696                           struct ext4_dir_entry_2 *de,
2697                           int blocksize, int csum_size,
2698                           unsigned int parent_ino, int dotdot_real_len)
2699 {
2700         de->inode = cpu_to_le32(inode->i_ino);
2701         de->name_len = 1;
2702         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2703                                            blocksize);
2704         strcpy(de->name, ".");
2705         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2706
2707         de = ext4_next_entry(de, blocksize);
2708         de->inode = cpu_to_le32(parent_ino);
2709         de->name_len = 2;
2710         if (!dotdot_real_len)
2711                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2712                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2713                                         blocksize);
2714         else
2715                 de->rec_len = ext4_rec_len_to_disk(
2716                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2717         strcpy(de->name, "..");
2718         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2719
2720         return ext4_next_entry(de, blocksize);
2721 }
2722
2723 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2724                              struct inode *inode)
2725 {
2726         struct buffer_head *dir_block = NULL;
2727         struct ext4_dir_entry_2 *de;
2728         ext4_lblk_t block = 0;
2729         unsigned int blocksize = dir->i_sb->s_blocksize;
2730         int csum_size = 0;
2731         int err;
2732
2733         if (ext4_has_metadata_csum(dir->i_sb))
2734                 csum_size = sizeof(struct ext4_dir_entry_tail);
2735
2736         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2737                 err = ext4_try_create_inline_dir(handle, dir, inode);
2738                 if (err < 0 && err != -ENOSPC)
2739                         goto out;
2740                 if (!err)
2741                         goto out;
2742         }
2743
2744         inode->i_size = 0;
2745         dir_block = ext4_append(handle, inode, &block);
2746         if (IS_ERR(dir_block))
2747                 return PTR_ERR(dir_block);
2748         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2749         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2750         set_nlink(inode, 2);
2751         if (csum_size)
2752                 ext4_initialize_dirent_tail(dir_block, blocksize);
2753
2754         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2755         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2756         if (err)
2757                 goto out;
2758         set_buffer_verified(dir_block);
2759 out:
2760         brelse(dir_block);
2761         return err;
2762 }
2763
2764 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2765 {
2766         handle_t *handle;
2767         struct inode *inode;
2768         int err, credits, retries = 0;
2769
2770         if (EXT4_DIR_LINK_MAX(dir))
2771                 return -EMLINK;
2772
2773         err = dquot_initialize(dir);
2774         if (err)
2775                 return err;
2776
2777         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2778                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2779 retry:
2780         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2781                                             &dentry->d_name,
2782                                             0, NULL, EXT4_HT_DIR, credits);
2783         handle = ext4_journal_current_handle();
2784         err = PTR_ERR(inode);
2785         if (IS_ERR(inode))
2786                 goto out_stop;
2787
2788         inode->i_op = &ext4_dir_inode_operations;
2789         inode->i_fop = &ext4_dir_operations;
2790         err = ext4_init_new_dir(handle, dir, inode);
2791         if (err)
2792                 goto out_clear_inode;
2793         err = ext4_mark_inode_dirty(handle, inode);
2794         if (!err)
2795                 err = ext4_add_entry(handle, dentry, inode);
2796         if (err) {
2797 out_clear_inode:
2798                 clear_nlink(inode);
2799                 ext4_orphan_add(handle, inode);
2800                 unlock_new_inode(inode);
2801                 ext4_mark_inode_dirty(handle, inode);
2802                 ext4_journal_stop(handle);
2803                 iput(inode);
2804                 goto out_retry;
2805         }
2806         ext4_inc_count(handle, dir);
2807         ext4_update_dx_flag(dir);
2808         err = ext4_mark_inode_dirty(handle, dir);
2809         if (err)
2810                 goto out_clear_inode;
2811         d_instantiate_new(dentry, inode);
2812         if (IS_DIRSYNC(dir))
2813                 ext4_handle_sync(handle);
2814
2815 out_stop:
2816         if (handle)
2817                 ext4_journal_stop(handle);
2818 out_retry:
2819         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2820                 goto retry;
2821         return err;
2822 }
2823
2824 /*
2825  * routine to check that the specified directory is empty (for rmdir)
2826  */
2827 bool ext4_empty_dir(struct inode *inode)
2828 {
2829         unsigned int offset;
2830         struct buffer_head *bh;
2831         struct ext4_dir_entry_2 *de;
2832         struct super_block *sb;
2833
2834         if (ext4_has_inline_data(inode)) {
2835                 int has_inline_data = 1;
2836                 int ret;
2837
2838                 ret = empty_inline_dir(inode, &has_inline_data);
2839                 if (has_inline_data)
2840                         return ret;
2841         }
2842
2843         sb = inode->i_sb;
2844         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2845                 EXT4_ERROR_INODE(inode, "invalid size");
2846                 return true;
2847         }
2848         /* The first directory block must not be a hole,
2849          * so treat it as DIRENT_HTREE
2850          */
2851         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2852         if (IS_ERR(bh))
2853                 return true;
2854
2855         de = (struct ext4_dir_entry_2 *) bh->b_data;
2856         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2857                                  0) ||
2858             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2859                 ext4_warning_inode(inode, "directory missing '.'");
2860                 brelse(bh);
2861                 return true;
2862         }
2863         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2864         de = ext4_next_entry(de, sb->s_blocksize);
2865         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2866                                  offset) ||
2867             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2868                 ext4_warning_inode(inode, "directory missing '..'");
2869                 brelse(bh);
2870                 return true;
2871         }
2872         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2873         while (offset < inode->i_size) {
2874                 if (!(offset & (sb->s_blocksize - 1))) {
2875                         unsigned int lblock;
2876                         brelse(bh);
2877                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2878                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2879                         if (bh == NULL) {
2880                                 offset += sb->s_blocksize;
2881                                 continue;
2882                         }
2883                         if (IS_ERR(bh))
2884                                 return true;
2885                 }
2886                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2887                                         (offset & (sb->s_blocksize - 1)));
2888                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2889                                          bh->b_data, bh->b_size, offset)) {
2890                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2891                         continue;
2892                 }
2893                 if (le32_to_cpu(de->inode)) {
2894                         brelse(bh);
2895                         return false;
2896                 }
2897                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2898         }
2899         brelse(bh);
2900         return true;
2901 }
2902
2903 /*
2904  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2905  * such inodes, starting at the superblock, in case we crash before the
2906  * file is closed/deleted, or in case the inode truncate spans multiple
2907  * transactions and the last transaction is not recovered after a crash.
2908  *
2909  * At filesystem recovery time, we walk this list deleting unlinked
2910  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2911  *
2912  * Orphan list manipulation functions must be called under i_mutex unless
2913  * we are just creating the inode or deleting it.
2914  */
2915 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2916 {
2917         struct super_block *sb = inode->i_sb;
2918         struct ext4_sb_info *sbi = EXT4_SB(sb);
2919         struct ext4_iloc iloc;
2920         int err = 0, rc;
2921         bool dirty = false;
2922
2923         if (!sbi->s_journal || is_bad_inode(inode))
2924                 return 0;
2925
2926         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2927                      !inode_is_locked(inode));
2928         /*
2929          * Exit early if inode already is on orphan list. This is a big speedup
2930          * since we don't have to contend on the global s_orphan_lock.
2931          */
2932         if (!list_empty(&EXT4_I(inode)->i_orphan))
2933                 return 0;
2934
2935         /*
2936          * Orphan handling is only valid for files with data blocks
2937          * being truncated, or files being unlinked. Note that we either
2938          * hold i_mutex, or the inode can not be referenced from outside,
2939          * so i_nlink should not be bumped due to race
2940          */
2941         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2942                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2943
2944         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2945         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2946         if (err)
2947                 goto out;
2948
2949         err = ext4_reserve_inode_write(handle, inode, &iloc);
2950         if (err)
2951                 goto out;
2952
2953         mutex_lock(&sbi->s_orphan_lock);
2954         /*
2955          * Due to previous errors inode may be already a part of on-disk
2956          * orphan list. If so skip on-disk list modification.
2957          */
2958         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2959             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2960                 /* Insert this inode at the head of the on-disk orphan list */
2961                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2962                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2963                 dirty = true;
2964         }
2965         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2966         mutex_unlock(&sbi->s_orphan_lock);
2967
2968         if (dirty) {
2969                 err = ext4_handle_dirty_super(handle, sb);
2970                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2971                 if (!err)
2972                         err = rc;
2973                 if (err) {
2974                         /*
2975                          * We have to remove inode from in-memory list if
2976                          * addition to on disk orphan list failed. Stray orphan
2977                          * list entries can cause panics at unmount time.
2978                          */
2979                         mutex_lock(&sbi->s_orphan_lock);
2980                         list_del_init(&EXT4_I(inode)->i_orphan);
2981                         mutex_unlock(&sbi->s_orphan_lock);
2982                 }
2983         } else
2984                 brelse(iloc.bh);
2985
2986         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2987         jbd_debug(4, "orphan inode %lu will point to %d\n",
2988                         inode->i_ino, NEXT_ORPHAN(inode));
2989 out:
2990         ext4_std_error(sb, err);
2991         return err;
2992 }
2993
2994 /*
2995  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2996  * of such inodes stored on disk, because it is finally being cleaned up.
2997  */
2998 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2999 {
3000         struct list_head *prev;
3001         struct ext4_inode_info *ei = EXT4_I(inode);
3002         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3003         __u32 ino_next;
3004         struct ext4_iloc iloc;
3005         int err = 0;
3006
3007         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3008                 return 0;
3009
3010         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3011                      !inode_is_locked(inode));
3012         /* Do this quick check before taking global s_orphan_lock. */
3013         if (list_empty(&ei->i_orphan))
3014                 return 0;
3015
3016         if (handle) {
3017                 /* Grab inode buffer early before taking global s_orphan_lock */
3018                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3019         }
3020
3021         mutex_lock(&sbi->s_orphan_lock);
3022         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3023
3024         prev = ei->i_orphan.prev;
3025         list_del_init(&ei->i_orphan);
3026
3027         /* If we're on an error path, we may not have a valid
3028          * transaction handle with which to update the orphan list on
3029          * disk, but we still need to remove the inode from the linked
3030          * list in memory. */
3031         if (!handle || err) {
3032                 mutex_unlock(&sbi->s_orphan_lock);
3033                 goto out_err;
3034         }
3035
3036         ino_next = NEXT_ORPHAN(inode);
3037         if (prev == &sbi->s_orphan) {
3038                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3039                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3040                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3041                 if (err) {
3042                         mutex_unlock(&sbi->s_orphan_lock);
3043                         goto out_brelse;
3044                 }
3045                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3046                 mutex_unlock(&sbi->s_orphan_lock);
3047                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3048         } else {
3049                 struct ext4_iloc iloc2;
3050                 struct inode *i_prev =
3051                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3052
3053                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3054                           i_prev->i_ino, ino_next);
3055                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3056                 if (err) {
3057                         mutex_unlock(&sbi->s_orphan_lock);
3058                         goto out_brelse;
3059                 }
3060                 NEXT_ORPHAN(i_prev) = ino_next;
3061                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3062                 mutex_unlock(&sbi->s_orphan_lock);
3063         }
3064         if (err)
3065                 goto out_brelse;
3066         NEXT_ORPHAN(inode) = 0;
3067         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3068 out_err:
3069         ext4_std_error(inode->i_sb, err);
3070         return err;
3071
3072 out_brelse:
3073         brelse(iloc.bh);
3074         goto out_err;
3075 }
3076
3077 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3078 {
3079         int retval;
3080         struct inode *inode;
3081         struct buffer_head *bh;
3082         struct ext4_dir_entry_2 *de;
3083         handle_t *handle = NULL;
3084
3085         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3086                 return -EIO;
3087
3088         /* Initialize quotas before so that eventual writes go in
3089          * separate transaction */
3090         retval = dquot_initialize(dir);
3091         if (retval)
3092                 return retval;
3093         retval = dquot_initialize(d_inode(dentry));
3094         if (retval)
3095                 return retval;
3096
3097         retval = -ENOENT;
3098         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3099         if (IS_ERR(bh))
3100                 return PTR_ERR(bh);
3101         if (!bh)
3102                 goto end_rmdir;
3103
3104         inode = d_inode(dentry);
3105
3106         retval = -EFSCORRUPTED;
3107         if (le32_to_cpu(de->inode) != inode->i_ino)
3108                 goto end_rmdir;
3109
3110         retval = -ENOTEMPTY;
3111         if (!ext4_empty_dir(inode))
3112                 goto end_rmdir;
3113
3114         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3115                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3116         if (IS_ERR(handle)) {
3117                 retval = PTR_ERR(handle);
3118                 handle = NULL;
3119                 goto end_rmdir;
3120         }
3121
3122         if (IS_DIRSYNC(dir))
3123                 ext4_handle_sync(handle);
3124
3125         retval = ext4_delete_entry(handle, dir, de, bh);
3126         if (retval)
3127                 goto end_rmdir;
3128         if (!EXT4_DIR_LINK_EMPTY(inode))
3129                 ext4_warning_inode(inode,
3130                              "empty directory '%.*s' has too many links (%u)",
3131                              dentry->d_name.len, dentry->d_name.name,
3132                              inode->i_nlink);
3133         inode_inc_iversion(inode);
3134         clear_nlink(inode);
3135         /* There's no need to set i_disksize: the fact that i_nlink is
3136          * zero will ensure that the right thing happens during any
3137          * recovery. */
3138         inode->i_size = 0;
3139         ext4_orphan_add(handle, inode);
3140         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3141         ext4_mark_inode_dirty(handle, inode);
3142         ext4_dec_count(handle, dir);
3143         ext4_update_dx_flag(dir);
3144         ext4_mark_inode_dirty(handle, dir);
3145
3146 #ifdef CONFIG_UNICODE
3147         /* VFS negative dentries are incompatible with Encoding and
3148          * Case-insensitiveness. Eventually we'll want avoid
3149          * invalidating the dentries here, alongside with returning the
3150          * negative dentries at ext4_lookup(), when it is better
3151          * supported by the VFS for the CI case.
3152          */
3153         if (IS_CASEFOLDED(dir))
3154                 d_invalidate(dentry);
3155 #endif
3156
3157 end_rmdir:
3158         brelse(bh);
3159         if (handle)
3160                 ext4_journal_stop(handle);
3161         return retval;
3162 }
3163
3164 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3165 {
3166         int retval;
3167         struct inode *inode;
3168         struct buffer_head *bh;
3169         struct ext4_dir_entry_2 *de;
3170         handle_t *handle = NULL;
3171
3172         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3173                 return -EIO;
3174
3175         trace_ext4_unlink_enter(dir, dentry);
3176         /* Initialize quotas before so that eventual writes go
3177          * in separate transaction */
3178         retval = dquot_initialize(dir);
3179         if (retval)
3180                 return retval;
3181         retval = dquot_initialize(d_inode(dentry));
3182         if (retval)
3183                 return retval;
3184
3185         retval = -ENOENT;
3186         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3187         if (IS_ERR(bh))
3188                 return PTR_ERR(bh);
3189         if (!bh)
3190                 goto end_unlink;
3191
3192         inode = d_inode(dentry);
3193
3194         retval = -EFSCORRUPTED;
3195         if (le32_to_cpu(de->inode) != inode->i_ino)
3196                 goto end_unlink;
3197
3198         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3199                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3200         if (IS_ERR(handle)) {
3201                 retval = PTR_ERR(handle);
3202                 handle = NULL;
3203                 goto end_unlink;
3204         }
3205
3206         if (IS_DIRSYNC(dir))
3207                 ext4_handle_sync(handle);
3208
3209         retval = ext4_delete_entry(handle, dir, de, bh);
3210         if (retval)
3211                 goto end_unlink;
3212         dir->i_ctime = dir->i_mtime = current_time(dir);
3213         ext4_update_dx_flag(dir);
3214         ext4_mark_inode_dirty(handle, dir);
3215         if (inode->i_nlink == 0)
3216                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3217                                    dentry->d_name.len, dentry->d_name.name);
3218         else
3219                 drop_nlink(inode);
3220         if (!inode->i_nlink)
3221                 ext4_orphan_add(handle, inode);
3222         inode->i_ctime = current_time(inode);
3223         ext4_mark_inode_dirty(handle, inode);
3224
3225 #ifdef CONFIG_UNICODE
3226         /* VFS negative dentries are incompatible with Encoding and
3227          * Case-insensitiveness. Eventually we'll want avoid
3228          * invalidating the dentries here, alongside with returning the
3229          * negative dentries at ext4_lookup(), when it is  better
3230          * supported by the VFS for the CI case.
3231          */
3232         if (IS_CASEFOLDED(dir))
3233                 d_invalidate(dentry);
3234 #endif
3235
3236 end_unlink:
3237         brelse(bh);
3238         if (handle)
3239                 ext4_journal_stop(handle);
3240         trace_ext4_unlink_exit(dentry, retval);
3241         return retval;
3242 }
3243
3244 static int ext4_symlink(struct inode *dir,
3245                         struct dentry *dentry, const char *symname)
3246 {
3247         handle_t *handle;
3248         struct inode *inode;
3249         int err, len = strlen(symname);
3250         int credits;
3251         struct fscrypt_str disk_link;
3252
3253         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3254                 return -EIO;
3255
3256         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3257                                       &disk_link);
3258         if (err)
3259                 return err;
3260
3261         err = dquot_initialize(dir);
3262         if (err)
3263                 return err;
3264
3265         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3266                 /*
3267                  * For non-fast symlinks, we just allocate inode and put it on
3268                  * orphan list in the first transaction => we need bitmap,
3269                  * group descriptor, sb, inode block, quota blocks, and
3270                  * possibly selinux xattr blocks.
3271                  */
3272                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3273                           EXT4_XATTR_TRANS_BLOCKS;
3274         } else {
3275                 /*
3276                  * Fast symlink. We have to add entry to directory
3277                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3278                  * allocate new inode (bitmap, group descriptor, inode block,
3279                  * quota blocks, sb is already counted in previous macros).
3280                  */
3281                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3282                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3283         }
3284
3285         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3286                                             &dentry->d_name, 0, NULL,
3287                                             EXT4_HT_DIR, credits);
3288         handle = ext4_journal_current_handle();
3289         if (IS_ERR(inode)) {
3290                 if (handle)
3291                         ext4_journal_stop(handle);
3292                 return PTR_ERR(inode);
3293         }
3294
3295         if (IS_ENCRYPTED(inode)) {
3296                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3297                 if (err)
3298                         goto err_drop_inode;
3299                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3300         }
3301
3302         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3303                 if (!IS_ENCRYPTED(inode))
3304                         inode->i_op = &ext4_symlink_inode_operations;
3305                 inode_nohighmem(inode);
3306                 ext4_set_aops(inode);
3307                 /*
3308                  * We cannot call page_symlink() with transaction started
3309                  * because it calls into ext4_write_begin() which can wait
3310                  * for transaction commit if we are running out of space
3311                  * and thus we deadlock. So we have to stop transaction now
3312                  * and restart it when symlink contents is written.
3313                  *
3314                  * To keep fs consistent in case of crash, we have to put inode
3315                  * to orphan list in the mean time.
3316                  */
3317                 drop_nlink(inode);
3318                 err = ext4_orphan_add(handle, inode);
3319                 ext4_journal_stop(handle);
3320                 handle = NULL;
3321                 if (err)
3322                         goto err_drop_inode;
3323                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3324                 if (err)
3325                         goto err_drop_inode;
3326                 /*
3327                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3328                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3329                  */
3330                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3331                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3332                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3333                 if (IS_ERR(handle)) {
3334                         err = PTR_ERR(handle);
3335                         handle = NULL;
3336                         goto err_drop_inode;
3337                 }
3338                 set_nlink(inode, 1);
3339                 err = ext4_orphan_del(handle, inode);
3340                 if (err)
3341                         goto err_drop_inode;
3342         } else {
3343                 /* clear the extent format for fast symlink */
3344                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3345                 if (!IS_ENCRYPTED(inode)) {
3346                         inode->i_op = &ext4_fast_symlink_inode_operations;
3347                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3348                 }
3349                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3350                        disk_link.len);
3351                 inode->i_size = disk_link.len - 1;
3352         }
3353         EXT4_I(inode)->i_disksize = inode->i_size;
3354         err = ext4_add_nondir(handle, dentry, &inode);
3355         if (handle)
3356                 ext4_journal_stop(handle);
3357         if (inode)
3358                 iput(inode);
3359         goto out_free_encrypted_link;
3360
3361 err_drop_inode:
3362         if (handle)
3363                 ext4_journal_stop(handle);
3364         clear_nlink(inode);
3365         unlock_new_inode(inode);
3366         iput(inode);
3367 out_free_encrypted_link:
3368         if (disk_link.name != (unsigned char *)symname)
3369                 kfree(disk_link.name);
3370         return err;
3371 }
3372
3373 static int ext4_link(struct dentry *old_dentry,
3374                      struct inode *dir, struct dentry *dentry)
3375 {
3376         handle_t *handle;
3377         struct inode *inode = d_inode(old_dentry);
3378         int err, retries = 0;
3379
3380         if (inode->i_nlink >= EXT4_LINK_MAX)
3381                 return -EMLINK;
3382
3383         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3384         if (err)
3385                 return err;
3386
3387         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3388             (!projid_eq(EXT4_I(dir)->i_projid,
3389                         EXT4_I(old_dentry->d_inode)->i_projid)))
3390                 return -EXDEV;
3391
3392         err = dquot_initialize(dir);
3393         if (err)
3394                 return err;
3395
3396 retry:
3397         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3398                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3399                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3400         if (IS_ERR(handle))
3401                 return PTR_ERR(handle);
3402
3403         if (IS_DIRSYNC(dir))
3404                 ext4_handle_sync(handle);
3405
3406         inode->i_ctime = current_time(inode);
3407         ext4_inc_count(handle, inode);
3408         ihold(inode);
3409
3410         err = ext4_add_entry(handle, dentry, inode);
3411         if (!err) {
3412                 ext4_mark_inode_dirty(handle, inode);
3413                 /* this can happen only for tmpfile being
3414                  * linked the first time
3415                  */
3416                 if (inode->i_nlink == 1)
3417                         ext4_orphan_del(handle, inode);
3418                 d_instantiate(dentry, inode);
3419         } else {
3420                 drop_nlink(inode);
3421                 iput(inode);
3422         }
3423         ext4_journal_stop(handle);
3424         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3425                 goto retry;
3426         return err;
3427 }
3428
3429
3430 /*
3431  * Try to find buffer head where contains the parent block.
3432  * It should be the inode block if it is inlined or the 1st block
3433  * if it is a normal dir.
3434  */
3435 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3436                                         struct inode *inode,
3437                                         int *retval,
3438                                         struct ext4_dir_entry_2 **parent_de,
3439                                         int *inlined)
3440 {
3441         struct buffer_head *bh;
3442
3443         if (!ext4_has_inline_data(inode)) {
3444                 /* The first directory block must not be a hole, so
3445                  * treat it as DIRENT_HTREE
3446                  */
3447                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3448                 if (IS_ERR(bh)) {
3449                         *retval = PTR_ERR(bh);
3450                         return NULL;
3451                 }
3452                 *parent_de = ext4_next_entry(
3453                                         (struct ext4_dir_entry_2 *)bh->b_data,
3454                                         inode->i_sb->s_blocksize);
3455                 return bh;
3456         }
3457
3458         *inlined = 1;
3459         return ext4_get_first_inline_block(inode, parent_de, retval);
3460 }
3461
3462 struct ext4_renament {
3463         struct inode *dir;
3464         struct dentry *dentry;
3465         struct inode *inode;
3466         bool is_dir;
3467         int dir_nlink_delta;
3468
3469         /* entry for "dentry" */
3470         struct buffer_head *bh;
3471         struct ext4_dir_entry_2 *de;
3472         int inlined;
3473
3474         /* entry for ".." in inode if it's a directory */
3475         struct buffer_head *dir_bh;
3476         struct ext4_dir_entry_2 *parent_de;
3477         int dir_inlined;
3478 };
3479
3480 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3481 {
3482         int retval;
3483
3484         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3485                                               &retval, &ent->parent_de,
3486                                               &ent->dir_inlined);
3487         if (!ent->dir_bh)
3488                 return retval;
3489         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3490                 return -EFSCORRUPTED;
3491         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3492         return ext4_journal_get_write_access(handle, ent->dir_bh);
3493 }
3494
3495 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3496                                   unsigned dir_ino)
3497 {
3498         int retval;
3499
3500         ent->parent_de->inode = cpu_to_le32(dir_ino);
3501         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3502         if (!ent->dir_inlined) {
3503                 if (is_dx(ent->inode)) {
3504                         retval = ext4_handle_dirty_dx_node(handle,
3505                                                            ent->inode,
3506                                                            ent->dir_bh);
3507                 } else {
3508                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3509                                                             ent->dir_bh);
3510                 }
3511         } else {
3512                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3513         }
3514         if (retval) {
3515                 ext4_std_error(ent->dir->i_sb, retval);
3516                 return retval;
3517         }
3518         return 0;
3519 }
3520
3521 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3522                        unsigned ino, unsigned file_type)
3523 {
3524         int retval;
3525
3526         BUFFER_TRACE(ent->bh, "get write access");
3527         retval = ext4_journal_get_write_access(handle, ent->bh);
3528         if (retval)
3529                 return retval;
3530         ent->de->inode = cpu_to_le32(ino);
3531         if (ext4_has_feature_filetype(ent->dir->i_sb))
3532                 ent->de->file_type = file_type;
3533         inode_inc_iversion(ent->dir);
3534         ent->dir->i_ctime = ent->dir->i_mtime =
3535                 current_time(ent->dir);
3536         ext4_mark_inode_dirty(handle, ent->dir);
3537         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3538         if (!ent->inlined) {
3539                 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3540                 if (unlikely(retval)) {
3541                         ext4_std_error(ent->dir->i_sb, retval);
3542                         return retval;
3543                 }
3544         }
3545         brelse(ent->bh);
3546         ent->bh = NULL;
3547
3548         return 0;
3549 }
3550
3551 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3552                                   const struct qstr *d_name)
3553 {
3554         int retval = -ENOENT;
3555         struct buffer_head *bh;
3556         struct ext4_dir_entry_2 *de;
3557
3558         bh = ext4_find_entry(dir, d_name, &de, NULL);
3559         if (IS_ERR(bh))
3560                 return PTR_ERR(bh);
3561         if (bh) {
3562                 retval = ext4_delete_entry(handle, dir, de, bh);
3563                 brelse(bh);
3564         }
3565         return retval;
3566 }
3567
3568 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3569                                int force_reread)
3570 {
3571         int retval;
3572         /*
3573          * ent->de could have moved from under us during htree split, so make
3574          * sure that we are deleting the right entry.  We might also be pointing
3575          * to a stale entry in the unused part of ent->bh so just checking inum
3576          * and the name isn't enough.
3577          */
3578         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3579             ent->de->name_len != ent->dentry->d_name.len ||
3580             strncmp(ent->de->name, ent->dentry->d_name.name,
3581                     ent->de->name_len) ||
3582             force_reread) {
3583                 retval = ext4_find_delete_entry(handle, ent->dir,
3584                                                 &ent->dentry->d_name);
3585         } else {
3586                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3587                 if (retval == -ENOENT) {
3588                         retval = ext4_find_delete_entry(handle, ent->dir,
3589                                                         &ent->dentry->d_name);
3590                 }
3591         }
3592
3593         if (retval) {
3594                 ext4_warning_inode(ent->dir,
3595                                    "Deleting old file: nlink %d, error=%d",
3596                                    ent->dir->i_nlink, retval);
3597         }
3598 }
3599
3600 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3601 {
3602         if (ent->dir_nlink_delta) {
3603                 if (ent->dir_nlink_delta == -1)
3604                         ext4_dec_count(handle, ent->dir);
3605                 else
3606                         ext4_inc_count(handle, ent->dir);
3607                 ext4_mark_inode_dirty(handle, ent->dir);
3608         }
3609 }
3610
3611 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3612                                               int credits, handle_t **h)
3613 {
3614         struct inode *wh;
3615         handle_t *handle;
3616         int retries = 0;
3617
3618         /*
3619          * for inode block, sb block, group summaries,
3620          * and inode bitmap
3621          */
3622         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3623                     EXT4_XATTR_TRANS_BLOCKS + 4);
3624 retry:
3625         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3626                                          &ent->dentry->d_name, 0, NULL,
3627                                          EXT4_HT_DIR, credits);
3628
3629         handle = ext4_journal_current_handle();
3630         if (IS_ERR(wh)) {
3631                 if (handle)
3632                         ext4_journal_stop(handle);
3633                 if (PTR_ERR(wh) == -ENOSPC &&
3634                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3635                         goto retry;
3636         } else {
3637                 *h = handle;
3638                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3639                 wh->i_op = &ext4_special_inode_operations;
3640         }
3641         return wh;
3642 }
3643
3644 /*
3645  * Anybody can rename anything with this: the permission checks are left to the
3646  * higher-level routines.
3647  *
3648  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3649  * while new_{dentry,inode) refers to the destination dentry/inode
3650  * This comes from rename(const char *oldpath, const char *newpath)
3651  */
3652 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3653                        struct inode *new_dir, struct dentry *new_dentry,
3654                        unsigned int flags)
3655 {
3656         handle_t *handle = NULL;
3657         struct ext4_renament old = {
3658                 .dir = old_dir,
3659                 .dentry = old_dentry,
3660                 .inode = d_inode(old_dentry),
3661         };
3662         struct ext4_renament new = {
3663                 .dir = new_dir,
3664                 .dentry = new_dentry,
3665                 .inode = d_inode(new_dentry),
3666         };
3667         int force_reread;
3668         int retval;
3669         struct inode *whiteout = NULL;
3670         int credits;
3671         u8 old_file_type;
3672
3673         if (new.inode && new.inode->i_nlink == 0) {
3674                 EXT4_ERROR_INODE(new.inode,
3675                                  "target of rename is already freed");
3676                 return -EFSCORRUPTED;
3677         }
3678
3679         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3680             (!projid_eq(EXT4_I(new_dir)->i_projid,
3681                         EXT4_I(old_dentry->d_inode)->i_projid)))
3682                 return -EXDEV;
3683
3684         retval = dquot_initialize(old.dir);
3685         if (retval)
3686                 return retval;
3687         retval = dquot_initialize(new.dir);
3688         if (retval)
3689                 return retval;
3690
3691         /* Initialize quotas before so that eventual writes go
3692          * in separate transaction */
3693         if (new.inode) {
3694                 retval = dquot_initialize(new.inode);
3695                 if (retval)
3696                         return retval;
3697         }
3698
3699         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3700         if (IS_ERR(old.bh))
3701                 return PTR_ERR(old.bh);
3702         /*
3703          *  Check for inode number is _not_ due to possible IO errors.
3704          *  We might rmdir the source, keep it as pwd of some process
3705          *  and merrily kill the link to whatever was created under the
3706          *  same name. Goodbye sticky bit ;-<
3707          */
3708         retval = -ENOENT;
3709         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3710                 goto end_rename;
3711
3712         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3713                                  &new.de, &new.inlined);
3714         if (IS_ERR(new.bh)) {
3715                 retval = PTR_ERR(new.bh);
3716                 new.bh = NULL;
3717                 goto end_rename;
3718         }
3719         if (new.bh) {
3720                 if (!new.inode) {
3721                         brelse(new.bh);
3722                         new.bh = NULL;
3723                 }
3724         }
3725         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3726                 ext4_alloc_da_blocks(old.inode);
3727
3728         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3729                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3730         if (!(flags & RENAME_WHITEOUT)) {
3731                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3732                 if (IS_ERR(handle)) {
3733                         retval = PTR_ERR(handle);
3734                         handle = NULL;
3735                         goto end_rename;
3736                 }
3737         } else {
3738                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3739                 if (IS_ERR(whiteout)) {
3740                         retval = PTR_ERR(whiteout);
3741                         whiteout = NULL;
3742                         goto end_rename;
3743                 }
3744         }
3745
3746         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3747                 ext4_handle_sync(handle);
3748
3749         if (S_ISDIR(old.inode->i_mode)) {
3750                 if (new.inode) {
3751                         retval = -ENOTEMPTY;
3752                         if (!ext4_empty_dir(new.inode))
3753                                 goto end_rename;
3754                 } else {
3755                         retval = -EMLINK;
3756                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3757                                 goto end_rename;
3758                 }
3759                 retval = ext4_rename_dir_prepare(handle, &old);
3760                 if (retval)
3761                         goto end_rename;
3762         }
3763         /*
3764          * If we're renaming a file within an inline_data dir and adding or
3765          * setting the new dirent causes a conversion from inline_data to
3766          * extents/blockmap, we need to force the dirent delete code to
3767          * re-read the directory, or else we end up trying to delete a dirent
3768          * from what is now the extent tree root (or a block map).
3769          */
3770         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3771                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3772
3773         old_file_type = old.de->file_type;
3774         if (whiteout) {
3775                 /*
3776                  * Do this before adding a new entry, so the old entry is sure
3777                  * to be still pointing to the valid old entry.
3778                  */
3779                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3780                                      EXT4_FT_CHRDEV);
3781                 if (retval)
3782                         goto end_rename;
3783                 ext4_mark_inode_dirty(handle, whiteout);
3784         }
3785         if (!new.bh) {
3786                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3787                 if (retval)
3788                         goto end_rename;
3789         } else {
3790                 retval = ext4_setent(handle, &new,
3791                                      old.inode->i_ino, old_file_type);
3792                 if (retval)
3793                         goto end_rename;
3794         }
3795         if (force_reread)
3796                 force_reread = !ext4_test_inode_flag(new.dir,
3797                                                      EXT4_INODE_INLINE_DATA);
3798
3799         /*
3800          * Like most other Unix systems, set the ctime for inodes on a
3801          * rename.
3802          */
3803         old.inode->i_ctime = current_time(old.inode);
3804         ext4_mark_inode_dirty(handle, old.inode);
3805
3806         if (!whiteout) {
3807                 /*
3808                  * ok, that's it
3809                  */
3810                 ext4_rename_delete(handle, &old, force_reread);
3811         }
3812
3813         if (new.inode) {
3814                 ext4_dec_count(handle, new.inode);
3815                 new.inode->i_ctime = current_time(new.inode);
3816         }
3817         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3818         ext4_update_dx_flag(old.dir);
3819         if (old.dir_bh) {
3820                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3821                 if (retval)
3822                         goto end_rename;
3823
3824                 ext4_dec_count(handle, old.dir);
3825                 if (new.inode) {
3826                         /* checked ext4_empty_dir above, can't have another
3827                          * parent, ext4_dec_count() won't work for many-linked
3828                          * dirs */
3829                         clear_nlink(new.inode);
3830                 } else {
3831                         ext4_inc_count(handle, new.dir);
3832                         ext4_update_dx_flag(new.dir);
3833                         ext4_mark_inode_dirty(handle, new.dir);
3834                 }
3835         }
3836         ext4_mark_inode_dirty(handle, old.dir);
3837         if (new.inode) {
3838                 ext4_mark_inode_dirty(handle, new.inode);
3839                 if (!new.inode->i_nlink)
3840                         ext4_orphan_add(handle, new.inode);
3841         }
3842         retval = 0;
3843
3844 end_rename:
3845         brelse(old.dir_bh);
3846         brelse(old.bh);
3847         brelse(new.bh);
3848         if (whiteout) {
3849                 if (retval)
3850                         drop_nlink(whiteout);
3851                 unlock_new_inode(whiteout);
3852                 iput(whiteout);
3853         }
3854         if (handle)
3855                 ext4_journal_stop(handle);
3856         return retval;
3857 }
3858
3859 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3860                              struct inode *new_dir, struct dentry *new_dentry)
3861 {
3862         handle_t *handle = NULL;
3863         struct ext4_renament old = {
3864                 .dir = old_dir,
3865                 .dentry = old_dentry,
3866                 .inode = d_inode(old_dentry),
3867         };
3868         struct ext4_renament new = {
3869                 .dir = new_dir,
3870                 .dentry = new_dentry,
3871                 .inode = d_inode(new_dentry),
3872         };
3873         u8 new_file_type;
3874         int retval;
3875         struct timespec64 ctime;
3876
3877         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3878              !projid_eq(EXT4_I(new_dir)->i_projid,
3879                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3880             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3881              !projid_eq(EXT4_I(old_dir)->i_projid,
3882                         EXT4_I(new_dentry->d_inode)->i_projid)))
3883                 return -EXDEV;
3884
3885         retval = dquot_initialize(old.dir);
3886         if (retval)
3887                 return retval;
3888         retval = dquot_initialize(new.dir);
3889         if (retval)
3890                 return retval;
3891
3892         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3893                                  &old.de, &old.inlined);
3894         if (IS_ERR(old.bh))
3895                 return PTR_ERR(old.bh);
3896         /*
3897          *  Check for inode number is _not_ due to possible IO errors.
3898          *  We might rmdir the source, keep it as pwd of some process
3899          *  and merrily kill the link to whatever was created under the
3900          *  same name. Goodbye sticky bit ;-<
3901          */
3902         retval = -ENOENT;
3903         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3904                 goto end_rename;
3905
3906         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3907                                  &new.de, &new.inlined);
3908         if (IS_ERR(new.bh)) {
3909                 retval = PTR_ERR(new.bh);
3910                 new.bh = NULL;
3911                 goto end_rename;
3912         }
3913
3914         /* RENAME_EXCHANGE case: old *and* new must both exist */
3915         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3916                 goto end_rename;
3917
3918         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3919                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3920                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3921         if (IS_ERR(handle)) {
3922                 retval = PTR_ERR(handle);
3923                 handle = NULL;
3924                 goto end_rename;
3925         }
3926
3927         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3928                 ext4_handle_sync(handle);
3929
3930         if (S_ISDIR(old.inode->i_mode)) {
3931                 old.is_dir = true;
3932                 retval = ext4_rename_dir_prepare(handle, &old);
3933                 if (retval)
3934                         goto end_rename;
3935         }
3936         if (S_ISDIR(new.inode->i_mode)) {
3937                 new.is_dir = true;
3938                 retval = ext4_rename_dir_prepare(handle, &new);
3939                 if (retval)
3940                         goto end_rename;
3941         }
3942
3943         /*
3944          * Other than the special case of overwriting a directory, parents'
3945          * nlink only needs to be modified if this is a cross directory rename.
3946          */
3947         if (old.dir != new.dir && old.is_dir != new.is_dir) {
3948                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3949                 new.dir_nlink_delta = -old.dir_nlink_delta;
3950                 retval = -EMLINK;
3951                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3952                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3953                         goto end_rename;
3954         }
3955
3956         new_file_type = new.de->file_type;
3957         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3958         if (retval)
3959                 goto end_rename;
3960
3961         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3962         if (retval)
3963                 goto end_rename;
3964
3965         /*
3966          * Like most other Unix systems, set the ctime for inodes on a
3967          * rename.
3968          */
3969         ctime = current_time(old.inode);
3970         old.inode->i_ctime = ctime;
3971         new.inode->i_ctime = ctime;
3972         ext4_mark_inode_dirty(handle, old.inode);
3973         ext4_mark_inode_dirty(handle, new.inode);
3974
3975         if (old.dir_bh) {
3976                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3977                 if (retval)
3978                         goto end_rename;
3979         }
3980         if (new.dir_bh) {
3981                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3982                 if (retval)
3983                         goto end_rename;
3984         }
3985         ext4_update_dir_count(handle, &old);
3986         ext4_update_dir_count(handle, &new);
3987         retval = 0;
3988
3989 end_rename:
3990         brelse(old.dir_bh);
3991         brelse(new.dir_bh);
3992         brelse(old.bh);
3993         brelse(new.bh);
3994         if (handle)
3995                 ext4_journal_stop(handle);
3996         return retval;
3997 }
3998
3999 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4000                         struct inode *new_dir, struct dentry *new_dentry,
4001                         unsigned int flags)
4002 {
4003         int err;
4004
4005         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4006                 return -EIO;
4007
4008         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4009                 return -EINVAL;
4010
4011         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4012                                      flags);
4013         if (err)
4014                 return err;
4015
4016         if (flags & RENAME_EXCHANGE) {
4017                 return ext4_cross_rename(old_dir, old_dentry,
4018                                          new_dir, new_dentry);
4019         }
4020
4021         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4022 }
4023
4024 /*
4025  * directories can handle most operations...
4026  */
4027 const struct inode_operations ext4_dir_inode_operations = {
4028         .create         = ext4_create,
4029         .lookup         = ext4_lookup,
4030         .link           = ext4_link,
4031         .unlink         = ext4_unlink,
4032         .symlink        = ext4_symlink,
4033         .mkdir          = ext4_mkdir,
4034         .rmdir          = ext4_rmdir,
4035         .mknod          = ext4_mknod,
4036         .tmpfile        = ext4_tmpfile,
4037         .rename         = ext4_rename2,
4038         .setattr        = ext4_setattr,
4039         .getattr        = ext4_getattr,
4040         .listxattr      = ext4_listxattr,
4041         .get_acl        = ext4_get_acl,
4042         .set_acl        = ext4_set_acl,
4043         .fiemap         = ext4_fiemap,
4044 };
4045
4046 const struct inode_operations ext4_special_inode_operations = {
4047         .setattr        = ext4_setattr,
4048         .getattr        = ext4_getattr,
4049         .listxattr      = ext4_listxattr,
4050         .get_acl        = ext4_get_acl,
4051         .set_acl        = ext4_set_acl,
4052 };