fs/ext4/namei.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/namei.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  from
  11  *
  12  *  linux/fs/minix/namei.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  Big-endian to little-endian byte-swapping/bitmaps by
  17  *        David S. Miller (davem@caip.rutgers.edu), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  20  *  Hash Tree Directory indexing (c)
  21  *      Daniel Phillips, 2001
  22  *  Hash Tree Directory indexing porting
  23  *      Christopher Li, 2002
  24  *  Hash Tree Directory indexing cleanup
  25  *      Theodore Ts'o, 2002
  26  */
  27
  28 #include <linux/fs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/time.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/quotaops.h>
  35 #include <linux/buffer_head.h>
  36 #include <linux/bio.h>
  37 #include <linux/iversion.h>
  38 #include <linux/unicode.h>
  39 #include "ext4.h"
  40 #include "ext4_jbd2.h"
  41
  42 #include "xattr.h"
  43 #include "acl.h"
  44
  45 #include <trace/events/ext4.h>
  46 /*
  47  * define how far ahead to read directories while searching them.
  48  */
  49 #define NAMEI_RA_CHUNKS  2
  50 #define NAMEI_RA_BLOCKS  4
  51 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  52
  53 static struct buffer_head *ext4_append(handle_t *handle,
  54                                         struct inode *inode,
  55                                         ext4_lblk_t *block)
  56 {
  57         struct buffer_head *bh;
  58         int err;
  59
  60         if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
  61                      ((inode->i_size >> 10) >=
  62                       EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
  63                 return ERR_PTR(-ENOSPC);
  64
  65         *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  66
  67         bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
  68         if (IS_ERR(bh))
  69                 return bh;
  70         inode->i_size += inode->i_sb->s_blocksize;
  71         EXT4_I(inode)->i_disksize = inode->i_size;
  72         BUFFER_TRACE(bh, "get_write_access");
  73         err = ext4_journal_get_write_access(handle, bh);
  74         if (err) {
  75                 brelse(bh);
  76                 ext4_std_error(inode->i_sb, err);
  77                 return ERR_PTR(err);
  78         }
  79         return bh;
  80 }
  81
  82 static int ext4_dx_csum_verify(struct inode *inode,
  83                                struct ext4_dir_entry *dirent);
  84
  85 /*
  86  * Hints to ext4_read_dirblock regarding whether we expect a directory
  87  * block being read to be an index block, or a block containing
  88  * directory entries (and if the latter, whether it was found via a
  89  * logical block in an htree index block).  This is used to control
  90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  91  * directory block read from the storage device.  EITHER will means
  92  * the caller doesn't know what kind of directory block will be read,
  93  * so no specific verification will be done.
  94  */
  95 typedef enum {
  96         EITHER, INDEX, DIRENT, DIRENT_HTREE
  97 } dirblock_type_t;
  98
  99 #define ext4_read_dirblock(inode, block, type) \
 100         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 101
 102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 103                                                 ext4_lblk_t block,
 104                                                 dirblock_type_t type,
 105                                                 const char *func,
 106                                                 unsigned int line)
 107 {
 108         struct buffer_head *bh;
 109         struct ext4_dir_entry *dirent;
 110         int is_dx_block = 0;
 111
 112         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 113                 bh = ERR_PTR(-EIO);
 114         else
 115                 bh = ext4_bread(NULL, inode, block, 0);
 116         if (IS_ERR(bh)) {
 117                 __ext4_warning(inode->i_sb, func, line,
 118                                "inode #%lu: lblock %lu: comm %s: "
 119                                "error %ld reading directory block",
 120                                inode->i_ino, (unsigned long)block,
 121                                current->comm, PTR_ERR(bh));
 122
 123                 return bh;
 124         }
 125         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 126                 ext4_error_inode(inode, func, line, block,
 127                                  "Directory hole found for htree %s block",
 128                                  (type == INDEX) ? "index" : "leaf");
 129                 return ERR_PTR(-EFSCORRUPTED);
 130         }
 131         if (!bh)
 132                 return NULL;
 133         dirent = (struct ext4_dir_entry *) bh->b_data;
 134         /* Determine whether or not we have an index block */
 135         if (is_dx(inode)) {
 136                 if (block == 0)
 137                         is_dx_block = 1;
 138                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 139                                                 inode->i_sb->s_blocksize) ==
 140                          inode->i_sb->s_blocksize)
 141                         is_dx_block = 1;
 142         }
 143         if (!is_dx_block && type == INDEX) {
 144                 ext4_error_inode(inode, func, line, block,
 145                        "directory leaf block found instead of index block");
 146                 brelse(bh);
 147                 return ERR_PTR(-EFSCORRUPTED);
 148         }
 149         if (!ext4_has_metadata_csum(inode->i_sb) ||
 150             buffer_verified(bh))
 151                 return bh;
 152
 153         /*
 154          * An empty leaf block can get mistaken for a index block; for
 155          * this reason, we can only check the index checksum when the
 156          * caller is sure it should be an index block.
 157          */
 158         if (is_dx_block && type == INDEX) {
 159                 if (ext4_dx_csum_verify(inode, dirent) &&
 160                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 161                         set_buffer_verified(bh);
 162                 else {
 163                         ext4_error_inode_err(inode, func, line, block,
 164                                              EFSBADCRC,
 165                                              "Directory index failed checksum");
 166                         brelse(bh);
 167                         return ERR_PTR(-EFSBADCRC);
 168                 }
 169         }
 170         if (!is_dx_block) {
 171                 if (ext4_dirblock_csum_verify(inode, bh) &&
 172                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 173                         set_buffer_verified(bh);
 174                 else {
 175                         ext4_error_inode_err(inode, func, line, block,
 176                                              EFSBADCRC,
 177                                              "Directory block failed checksum");
 178                         brelse(bh);
 179                         return ERR_PTR(-EFSBADCRC);
 180                 }
 181         }
 182         return bh;
 183 }
 184
 185 #ifndef assert
 186 #define assert(test) J_ASSERT(test)
 187 #endif
 188
 189 #ifdef DX_DEBUG
 190 #define dxtrace(command) command
 191 #else
 192 #define dxtrace(command)
 193 #endif
 194
 195 struct fake_dirent
 196 {
 197         __le32 inode;
 198         __le16 rec_len;
 199         u8 name_len;
 200         u8 file_type;
 201 };
 202
 203 struct dx_countlimit
 204 {
 205         __le16 limit;
 206         __le16 count;
 207 };
 208
 209 struct dx_entry
 210 {
 211         __le32 hash;
 212         __le32 block;
 213 };
 214
 215 /*
 216  * dx_root_info is laid out so that if it should somehow get overlaid by a
 217  * dirent the two low bits of the hash version will be zero.  Therefore, the
 218  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 219  */
 220
 221 struct dx_root
 222 {
 223         struct fake_dirent dot;
 224         char dot_name[4];
 225         struct fake_dirent dotdot;
 226         char dotdot_name[4];
 227         struct dx_root_info
 228         {
 229                 __le32 reserved_zero;
 230                 u8 hash_version;
 231                 u8 info_length; /* 8 */
 232                 u8 indirect_levels;
 233                 u8 unused_flags;
 234         }
 235         info;
 236         struct dx_entry entries[];
 237 };
 238
 239 struct dx_node
 240 {
 241         struct fake_dirent fake;
 242         struct dx_entry entries[];
 243 };
 244
 245
 246 struct dx_frame
 247 {
 248         struct buffer_head *bh;
 249         struct dx_entry *entries;
 250         struct dx_entry *at;
 251 };
 252
 253 struct dx_map_entry
 254 {
 255         u32 hash;
 256         u16 offs;
 257         u16 size;
 258 };
 259
 260 /*
 261  * This goes at the end of each htree block.
 262  */
 263 struct dx_tail {
 264         u32 dt_reserved;
 265         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 266 };
 267
 268 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 269 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 270 static inline unsigned dx_get_hash(struct dx_entry *entry);
 271 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 272 static unsigned dx_get_count(struct dx_entry *entries);
 273 static unsigned dx_get_limit(struct dx_entry *entries);
 274 static void dx_set_count(struct dx_entry *entries, unsigned value);
 275 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 276 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 277 static unsigned dx_node_limit(struct inode *dir);
 278 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 279                                  struct inode *dir,
 280                                  struct dx_hash_info *hinfo,
 281                                  struct dx_frame *frame);
 282 static void dx_release(struct dx_frame *frames);
 283 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 284                        unsigned blocksize, struct dx_hash_info *hinfo,
 285                        struct dx_map_entry map[]);
 286 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 287 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 288                 struct dx_map_entry *offsets, int count, unsigned blocksize);
 289 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 290 static void dx_insert_block(struct dx_frame *frame,
 291                                         u32 hash, ext4_lblk_t block);
 292 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 293                                  struct dx_frame *frame,
 294                                  struct dx_frame *frames,
 295                                  __u32 *start_hash);
 296 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 297                 struct ext4_filename *fname,
 298                 struct ext4_dir_entry_2 **res_dir);
 299 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 300                              struct inode *dir, struct inode *inode);
 301
 302 /* checksumming functions */
 303 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 304                                  unsigned int blocksize)
 305 {
 306         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 307
 308         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 309         t->det_rec_len = ext4_rec_len_to_disk(
 310                         sizeof(struct ext4_dir_entry_tail), blocksize);
 311         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 312 }
 313
 314 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 315 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 316                                                    struct buffer_head *bh)
 317 {
 318         struct ext4_dir_entry_tail *t;
 319
 320 #ifdef PARANOID
 321         struct ext4_dir_entry *d, *top;
 322
 323         d = (struct ext4_dir_entry *)bh->b_data;
 324         top = (struct ext4_dir_entry *)(bh->b_data +
 325                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 326                  sizeof(struct ext4_dir_entry_tail)));
 327         while (d < top && d->rec_len)
 328                 d = (struct ext4_dir_entry *)(((void *)d) +
 329                     le16_to_cpu(d->rec_len));
 330
 331         if (d != top)
 332                 return NULL;
 333
 334         t = (struct ext4_dir_entry_tail *)d;
 335 #else
 336         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 337 #endif
 338
 339         if (t->det_reserved_zero1 ||
 340             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 341             t->det_reserved_zero2 ||
 342             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 343                 return NULL;
 344
 345         return t;
 346 }
 347
 348 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 349 {
 350         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 351         struct ext4_inode_info *ei = EXT4_I(inode);
 352         __u32 csum;
 353
 354         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 355         return cpu_to_le32(csum);
 356 }
 357
 358 #define warn_no_space_for_csum(inode)                                   \
 359         __warn_no_space_for_csum((inode), __func__, __LINE__)
 360
 361 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 362                                      unsigned int line)
 363 {
 364         __ext4_warning_inode(inode, func, line,
 365                 "No space for directory leaf checksum. Please run e2fsck -D.");
 366 }
 367
 368 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 369 {
 370         struct ext4_dir_entry_tail *t;
 371
 372         if (!ext4_has_metadata_csum(inode->i_sb))
 373                 return 1;
 374
 375         t = get_dirent_tail(inode, bh);
 376         if (!t) {
 377                 warn_no_space_for_csum(inode);
 378                 return 0;
 379         }
 380
 381         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 382                                                   (char *)t - bh->b_data))
 383                 return 0;
 384
 385         return 1;
 386 }
 387
 388 static void ext4_dirblock_csum_set(struct inode *inode,
 389                                  struct buffer_head *bh)
 390 {
 391         struct ext4_dir_entry_tail *t;
 392
 393         if (!ext4_has_metadata_csum(inode->i_sb))
 394                 return;
 395
 396         t = get_dirent_tail(inode, bh);
 397         if (!t) {
 398                 warn_no_space_for_csum(inode);
 399                 return;
 400         }
 401
 402         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 403                                              (char *)t - bh->b_data);
 404 }
 405
 406 int ext4_handle_dirty_dirblock(handle_t *handle,
 407                                struct inode *inode,
 408                                struct buffer_head *bh)
 409 {
 410         ext4_dirblock_csum_set(inode, bh);
 411         return ext4_handle_dirty_metadata(handle, inode, bh);
 412 }
 413
 414 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 415                                                struct ext4_dir_entry *dirent,
 416                                                int *offset)
 417 {
 418         struct ext4_dir_entry *dp;
 419         struct dx_root_info *root;
 420         int count_offset;
 421
 422         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 423                 count_offset = 8;
 424         else if (le16_to_cpu(dirent->rec_len) == 12) {
 425                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 426                 if (le16_to_cpu(dp->rec_len) !=
 427                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 428                         return NULL;
 429                 root = (struct dx_root_info *)(((void *)dp + 12));
 430                 if (root->reserved_zero ||
 431                     root->info_length != sizeof(struct dx_root_info))
 432                         return NULL;
 433                 count_offset = 32;
 434         } else
 435                 return NULL;
 436
 437         if (offset)
 438                 *offset = count_offset;
 439         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 440 }
 441
 442 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 443                            int count_offset, int count, struct dx_tail *t)
 444 {
 445         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 446         struct ext4_inode_info *ei = EXT4_I(inode);
 447         __u32 csum;
 448         int size;
 449         __u32 dummy_csum = 0;
 450         int offset = offsetof(struct dx_tail, dt_checksum);
 451
 452         size = count_offset + (count * sizeof(struct dx_entry));
 453         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 454         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 455         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 456
 457         return cpu_to_le32(csum);
 458 }
 459
 460 static int ext4_dx_csum_verify(struct inode *inode,
 461                                struct ext4_dir_entry *dirent)
 462 {
 463         struct dx_countlimit *c;
 464         struct dx_tail *t;
 465         int count_offset, limit, count;
 466
 467         if (!ext4_has_metadata_csum(inode->i_sb))
 468                 return 1;
 469
 470         c = get_dx_countlimit(inode, dirent, &count_offset);
 471         if (!c) {
 472                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 473                 return 0;
 474         }
 475         limit = le16_to_cpu(c->limit);
 476         count = le16_to_cpu(c->count);
 477         if (count_offset + (limit * sizeof(struct dx_entry)) >
 478             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 479                 warn_no_space_for_csum(inode);
 480                 return 0;
 481         }
 482         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 483
 484         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 485                                             count, t))
 486                 return 0;
 487         return 1;
 488 }
 489
 490 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 491 {
 492         struct dx_countlimit *c;
 493         struct dx_tail *t;
 494         int count_offset, limit, count;
 495
 496         if (!ext4_has_metadata_csum(inode->i_sb))
 497                 return;
 498
 499         c = get_dx_countlimit(inode, dirent, &count_offset);
 500         if (!c) {
 501                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 502                 return;
 503         }
 504         limit = le16_to_cpu(c->limit);
 505         count = le16_to_cpu(c->count);
 506         if (count_offset + (limit * sizeof(struct dx_entry)) >
 507             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 508                 warn_no_space_for_csum(inode);
 509                 return;
 510         }
 511         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 512
 513         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 514 }
 515
 516 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 517                                             struct inode *inode,
 518                                             struct buffer_head *bh)
 519 {
 520         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 521         return ext4_handle_dirty_metadata(handle, inode, bh);
 522 }
 523
 524 /*
 525  * p is at least 6 bytes before the end of page
 526  */
 527 static inline struct ext4_dir_entry_2 *
 528 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 529 {
 530         return (struct ext4_dir_entry_2 *)((char *)p +
 531                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 532 }
 533
 534 /*
 535  * Future: use high four bits of block for coalesce-on-delete flags
 536  * Mask them off for now.
 537  */
 538
 539 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 540 {
 541         return le32_to_cpu(entry->block) & 0x0fffffff;
 542 }
 543
 544 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 545 {
 546         entry->block = cpu_to_le32(value);
 547 }
 548
 549 static inline unsigned dx_get_hash(struct dx_entry *entry)
 550 {
 551         return le32_to_cpu(entry->hash);
 552 }
 553
 554 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 555 {
 556         entry->hash = cpu_to_le32(value);
 557 }
 558
 559 static inline unsigned dx_get_count(struct dx_entry *entries)
 560 {
 561         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 562 }
 563
 564 static inline unsigned dx_get_limit(struct dx_entry *entries)
 565 {
 566         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 567 }
 568
 569 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 570 {
 571         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 572 }
 573
 574 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 575 {
 576         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 577 }
 578
 579 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 580 {
 581         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 582                 EXT4_DIR_REC_LEN(2) - infosize;
 583
 584         if (ext4_has_metadata_csum(dir->i_sb))
 585                 entry_space -= sizeof(struct dx_tail);
 586         return entry_space / sizeof(struct dx_entry);
 587 }
 588
 589 static inline unsigned dx_node_limit(struct inode *dir)
 590 {
 591         unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 592
 593         if (ext4_has_metadata_csum(dir->i_sb))
 594                 entry_space -= sizeof(struct dx_tail);
 595         return entry_space / sizeof(struct dx_entry);
 596 }
 597
 598 /*
 599  * Debug
 600  */
 601 #ifdef DX_DEBUG
 602 static void dx_show_index(char * label, struct dx_entry *entries)
 603 {
 604         int i, n = dx_get_count (entries);
 605         printk(KERN_DEBUG "%s index", label);
 606         for (i = 0; i < n; i++) {
 607                 printk(KERN_CONT " %x->%lu",
 608                        i ? dx_get_hash(entries + i) : 0,
 609                        (unsigned long)dx_get_block(entries + i));
 610         }
 611         printk(KERN_CONT "\n");
 612 }
 613
 614 struct stats
 615 {
 616         unsigned names;
 617         unsigned space;
 618         unsigned bcount;
 619 };
 620
 621 static struct stats dx_show_leaf(struct inode *dir,
 622                                 struct dx_hash_info *hinfo,
 623                                 struct ext4_dir_entry_2 *de,
 624                                 int size, int show_names)
 625 {
 626         unsigned names = 0, space = 0;
 627         char *base = (char *) de;
 628         struct dx_hash_info h = *hinfo;
 629
 630         printk("names: ");
 631         while ((char *) de < base + size)
 632         {
 633                 if (de->inode)
 634                 {
 635                         if (show_names)
 636                         {
 637 #ifdef CONFIG_FS_ENCRYPTION
 638                                 int len;
 639                                 char *name;
 640                                 struct fscrypt_str fname_crypto_str =
 641                                         FSTR_INIT(NULL, 0);
 642                                 int res = 0;
 643
 644                                 name  = de->name;
 645                                 len = de->name_len;
 646                                 if (!IS_ENCRYPTED(dir)) {
 647                                         /* Directory is not encrypted */
 648                                         ext4fs_dirhash(dir, de->name,
 649                                                 de->name_len, &h);
 650                                         printk("%*.s:(U)%x.%u ", len,
 651                                                name, h.hash,
 652                                                (unsigned) ((char *) de
 653                                                            - base));
 654                                 } else {
 655                                         struct fscrypt_str de_name =
 656                                                 FSTR_INIT(name, len);
 657
 658                                         /* Directory is encrypted */
 659                                         res = fscrypt_fname_alloc_buffer(
 660                                                 len, &fname_crypto_str);
 661                                         if (res)
 662                                                 printk(KERN_WARNING "Error "
 663                                                         "allocating crypto "
 664                                                         "buffer--skipping "
 665                                                         "crypto\n");
 666                                         res = fscrypt_fname_disk_to_usr(dir,
 667                                                 0, 0, &de_name,
 668                                                 &fname_crypto_str);
 669                                         if (res) {
 670                                                 printk(KERN_WARNING "Error "
 671                                                         "converting filename "
 672                                                         "from disk to usr"
 673                                                         "\n");
 674                                                 name = "??";
 675                                                 len = 2;
 676                                         } else {
 677                                                 name = fname_crypto_str.name;
 678                                                 len = fname_crypto_str.len;
 679                                         }
 680                                         ext4fs_dirhash(dir, de->name,
 681                                                        de->name_len, &h);
 682                                         printk("%*.s:(E)%x.%u ", len, name,
 683                                                h.hash, (unsigned) ((char *) de
 684                                                                    - base));
 685                                         fscrypt_fname_free_buffer(
 686                                                         &fname_crypto_str);
 687                                 }
 688 #else
 689                                 int len = de->name_len;
 690                                 char *name = de->name;
 691                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 692                                 printk("%*.s:%x.%u ", len, name, h.hash,
 693                                        (unsigned) ((char *) de - base));
 694 #endif
 695                         }
 696                         space += EXT4_DIR_REC_LEN(de->name_len);
 697                         names++;
 698                 }
 699                 de = ext4_next_entry(de, size);
 700         }
 701         printk(KERN_CONT "(%i)\n", names);
 702         return (struct stats) { names, space, 1 };
 703 }
 704
 705 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 706                              struct dx_entry *entries, int levels)
 707 {
 708         unsigned blocksize = dir->i_sb->s_blocksize;
 709         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 710         unsigned bcount = 0;
 711         struct buffer_head *bh;
 712         printk("%i indexed blocks...\n", count);
 713         for (i = 0; i < count; i++, entries++)
 714         {
 715                 ext4_lblk_t block = dx_get_block(entries);
 716                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 717                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 718                 struct stats stats;
 719                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 720                 bh = ext4_bread(NULL,dir, block, 0);
 721                 if (!bh || IS_ERR(bh))
 722                         continue;
 723                 stats = levels?
 724                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 725                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 726                         bh->b_data, blocksize, 0);
 727                 names += stats.names;
 728                 space += stats.space;
 729                 bcount += stats.bcount;
 730                 brelse(bh);
 731         }
 732         if (bcount)
 733                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 734                        levels ? "" : "   ", names, space/bcount,
 735                        (space/bcount)*100/blocksize);
 736         return (struct stats) { names, space, bcount};
 737 }
 738 #endif /* DX_DEBUG */
 739
 740 /*
 741  * Probe for a directory leaf block to search.
 742  *
 743  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 744  * error in the directory index, and the caller should fall back to
 745  * searching the directory normally.  The callers of dx_probe **MUST**
 746  * check for this error code, and make sure it never gets reflected
 747  * back to userspace.
 748  */
 749 static struct dx_frame *
 750 dx_probe(struct ext4_filename *fname, struct inode *dir,
 751          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 752 {
 753         unsigned count, indirect;
 754         struct dx_entry *at, *entries, *p, *q, *m;
 755         struct dx_root *root;
 756         struct dx_frame *frame = frame_in;
 757         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 758         u32 hash;
 759
 760         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 761         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 762         if (IS_ERR(frame->bh))
 763                 return (struct dx_frame *) frame->bh;
 764
 765         root = (struct dx_root *) frame->bh->b_data;
 766         if (root->info.hash_version != DX_HASH_TEA &&
 767             root->info.hash_version != DX_HASH_HALF_MD4 &&
 768             root->info.hash_version != DX_HASH_LEGACY) {
 769                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 770                                    root->info.hash_version);
 771                 goto fail;
 772         }
 773         if (fname)
 774                 hinfo = &fname->hinfo;
 775         hinfo->hash_version = root->info.hash_version;
 776         if (hinfo->hash_version <= DX_HASH_TEA)
 777                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 778         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 779         if (fname && fname_name(fname))
 780                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 781         hash = hinfo->hash;
 782
 783         if (root->info.unused_flags & 1) {
 784                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 785                                    root->info.unused_flags);
 786                 goto fail;
 787         }
 788
 789         indirect = root->info.indirect_levels;
 790         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 791                 ext4_warning(dir->i_sb,
 792                              "Directory (ino: %lu) htree depth %#06x exceed"
 793                              "supported value", dir->i_ino,
 794                              ext4_dir_htree_level(dir->i_sb));
 795                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 796                         ext4_warning(dir->i_sb, "Enable large directory "
 797                                                 "feature to access it");
 798                 }
 799                 goto fail;
 800         }
 801
 802         entries = (struct dx_entry *)(((char *)&root->info) +
 803                                       root->info.info_length);
 804
 805         if (dx_get_limit(entries) != dx_root_limit(dir,
 806                                                    root->info.info_length)) {
 807                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 808                                    dx_get_limit(entries),
 809                                    dx_root_limit(dir, root->info.info_length));
 810                 goto fail;
 811         }
 812
 813         dxtrace(printk("Look up %x", hash));
 814         while (1) {
 815                 count = dx_get_count(entries);
 816                 if (!count || count > dx_get_limit(entries)) {
 817                         ext4_warning_inode(dir,
 818                                            "dx entry: count %u beyond limit %u",
 819                                            count, dx_get_limit(entries));
 820                         goto fail;
 821                 }
 822
 823                 p = entries + 1;
 824                 q = entries + count - 1;
 825                 while (p <= q) {
 826                         m = p + (q - p) / 2;
 827                         dxtrace(printk(KERN_CONT "."));
 828                         if (dx_get_hash(m) > hash)
 829                                 q = m - 1;
 830                         else
 831                                 p = m + 1;
 832                 }
 833
 834                 if (0) { // linear search cross check
 835                         unsigned n = count - 1;
 836                         at = entries;
 837                         while (n--)
 838                         {
 839                                 dxtrace(printk(KERN_CONT ","));
 840                                 if (dx_get_hash(++at) > hash)
 841                                 {
 842                                         at--;
 843                                         break;
 844                                 }
 845                         }
 846                         assert (at == p - 1);
 847                 }
 848
 849                 at = p - 1;
 850                 dxtrace(printk(KERN_CONT " %x->%u\n",
 851                                at == entries ? 0 : dx_get_hash(at),
 852                                dx_get_block(at)));
 853                 frame->entries = entries;
 854                 frame->at = at;
 855                 if (!indirect--)
 856                         return frame;
 857                 frame++;
 858                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 859                 if (IS_ERR(frame->bh)) {
 860                         ret_err = (struct dx_frame *) frame->bh;
 861                         frame->bh = NULL;
 862                         goto fail;
 863                 }
 864                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 865
 866                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 867                         ext4_warning_inode(dir,
 868                                 "dx entry: limit %u != node limit %u",
 869                                 dx_get_limit(entries), dx_node_limit(dir));
 870                         goto fail;
 871                 }
 872         }
 873 fail:
 874         while (frame >= frame_in) {
 875                 brelse(frame->bh);
 876                 frame--;
 877         }
 878
 879         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 880                 ext4_warning_inode(dir,
 881                         "Corrupt directory, running e2fsck is recommended");
 882         return ret_err;
 883 }
 884
 885 static void dx_release(struct dx_frame *frames)
 886 {
 887         struct dx_root_info *info;
 888         int i;
 889         unsigned int indirect_levels;
 890
 891         if (frames[0].bh == NULL)
 892                 return;
 893
 894         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 895         /* save local copy, "info" may be freed after brelse() */
 896         indirect_levels = info->indirect_levels;
 897         for (i = 0; i <= indirect_levels; i++) {
 898                 if (frames[i].bh == NULL)
 899                         break;
 900                 brelse(frames[i].bh);
 901                 frames[i].bh = NULL;
 902         }
 903 }
 904
 905 /*
 906  * This function increments the frame pointer to search the next leaf
 907  * block, and reads in the necessary intervening nodes if the search
 908  * should be necessary.  Whether or not the search is necessary is
 909  * controlled by the hash parameter.  If the hash value is even, then
 910  * the search is only continued if the next block starts with that
 911  * hash value.  This is used if we are searching for a specific file.
 912  *
 913  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 914  *
 915  * This function returns 1 if the caller should continue to search,
 916  * or 0 if it should not.  If there is an error reading one of the
 917  * index blocks, it will a negative error code.
 918  *
 919  * If start_hash is non-null, it will be filled in with the starting
 920  * hash of the next page.
 921  */
 922 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 923                                  struct dx_frame *frame,
 924                                  struct dx_frame *frames,
 925                                  __u32 *start_hash)
 926 {
 927         struct dx_frame *p;
 928         struct buffer_head *bh;
 929         int num_frames = 0;
 930         __u32 bhash;
 931
 932         p = frame;
 933         /*
 934          * Find the next leaf page by incrementing the frame pointer.
 935          * If we run out of entries in the interior node, loop around and
 936          * increment pointer in the parent node.  When we break out of
 937          * this loop, num_frames indicates the number of interior
 938          * nodes need to be read.
 939          */
 940         while (1) {
 941                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 942                         break;
 943                 if (p == frames)
 944                         return 0;
 945                 num_frames++;
 946                 p--;
 947         }
 948
 949         /*
 950          * If the hash is 1, then continue only if the next page has a
 951          * continuation hash of any value.  This is used for readdir
 952          * handling.  Otherwise, check to see if the hash matches the
 953          * desired contiuation hash.  If it doesn't, return since
 954          * there's no point to read in the successive index pages.
 955          */
 956         bhash = dx_get_hash(p->at);
 957         if (start_hash)
 958                 *start_hash = bhash;
 959         if ((hash & 1) == 0) {
 960                 if ((bhash & ~1) != hash)
 961                         return 0;
 962         }
 963         /*
 964          * If the hash is HASH_NB_ALWAYS, we always go to the next
 965          * block so no check is necessary
 966          */
 967         while (num_frames--) {
 968                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 969                 if (IS_ERR(bh))
 970                         return PTR_ERR(bh);
 971                 p++;
 972                 brelse(p->bh);
 973                 p->bh = bh;
 974                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 975         }
 976         return 1;
 977 }
 978
 979
 980 /*
 981  * This function fills a red-black tree with information from a
 982  * directory block.  It returns the number directory entries loaded
 983  * into the tree.  If there is an error it is returned in err.
 984  */
 985 static int htree_dirblock_to_tree(struct file *dir_file,
 986                                   struct inode *dir, ext4_lblk_t block,
 987                                   struct dx_hash_info *hinfo,
 988                                   __u32 start_hash, __u32 start_minor_hash)
 989 {
 990         struct buffer_head *bh;
 991         struct ext4_dir_entry_2 *de, *top;
 992         int err = 0, count = 0;
 993         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
 994
 995         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 996                                                         (unsigned long)block));
 997         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
 998         if (IS_ERR(bh))
 999                 return PTR_ERR(bh);
1000
1001         de = (struct ext4_dir_entry_2 *) bh->b_data;
1002         top = (struct ext4_dir_entry_2 *) ((char *) de +
1003                                            dir->i_sb->s_blocksize -
1004                                            EXT4_DIR_REC_LEN(0));
1005         /* Check if the directory is encrypted */
1006         if (IS_ENCRYPTED(dir)) {
1007                 err = fscrypt_prepare_readdir(dir);
1008                 if (err < 0) {
1009                         brelse(bh);
1010                         return err;
1011                 }
1012                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1013                                                  &fname_crypto_str);
1014                 if (err < 0) {
1015                         brelse(bh);
1016                         return err;
1017                 }
1018         }
1019
1020         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1021                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1022                                 bh->b_data, bh->b_size,
1023                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1024                                          + ((char *)de - bh->b_data))) {
1025                         /* silently ignore the rest of the block */
1026                         break;
1027                 }
1028                 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1029                 if ((hinfo->hash < start_hash) ||
1030                     ((hinfo->hash == start_hash) &&
1031                      (hinfo->minor_hash < start_minor_hash)))
1032                         continue;
1033                 if (de->inode == 0)
1034                         continue;
1035                 if (!IS_ENCRYPTED(dir)) {
1036                         tmp_str.name = de->name;
1037                         tmp_str.len = de->name_len;
1038                         err = ext4_htree_store_dirent(dir_file,
1039                                    hinfo->hash, hinfo->minor_hash, de,
1040                                    &tmp_str);
1041                 } else {
1042                         int save_len = fname_crypto_str.len;
1043                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1044                                                                 de->name_len);
1045
1046                         /* Directory is encrypted */
1047                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1048                                         hinfo->minor_hash, &de_name,
1049                                         &fname_crypto_str);
1050                         if (err) {
1051                                 count = err;
1052                                 goto errout;
1053                         }
1054                         err = ext4_htree_store_dirent(dir_file,
1055                                    hinfo->hash, hinfo->minor_hash, de,
1056                                         &fname_crypto_str);
1057                         fname_crypto_str.len = save_len;
1058                 }
1059                 if (err != 0) {
1060                         count = err;
1061                         goto errout;
1062                 }
1063                 count++;
1064         }
1065 errout:
1066         brelse(bh);
1067         fscrypt_fname_free_buffer(&fname_crypto_str);
1068         return count;
1069 }
1070
1071
1072 /*
1073  * This function fills a red-black tree with information from a
1074  * directory.  We start scanning the directory in hash order, starting
1075  * at start_hash and start_minor_hash.
1076  *
1077  * This function returns the number of entries inserted into the tree,
1078  * or a negative error code.
1079  */
1080 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1081                          __u32 start_minor_hash, __u32 *next_hash)
1082 {
1083         struct dx_hash_info hinfo;
1084         struct ext4_dir_entry_2 *de;
1085         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1086         struct inode *dir;
1087         ext4_lblk_t block;
1088         int count = 0;
1089         int ret, err;
1090         __u32 hashval;
1091         struct fscrypt_str tmp_str;
1092
1093         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1094                        start_hash, start_minor_hash));
1095         dir = file_inode(dir_file);
1096         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1097                 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1098                 if (hinfo.hash_version <= DX_HASH_TEA)
1099                         hinfo.hash_version +=
1100                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1101                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1102                 if (ext4_has_inline_data(dir)) {
1103                         int has_inline_data = 1;
1104                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1105                                                        &hinfo, start_hash,
1106                                                        start_minor_hash,
1107                                                        &has_inline_data);
1108                         if (has_inline_data) {
1109                                 *next_hash = ~0;
1110                                 return count;
1111                         }
1112                 }
1113                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1114                                                start_hash, start_minor_hash);
1115                 *next_hash = ~0;
1116                 return count;
1117         }
1118         hinfo.hash = start_hash;
1119         hinfo.minor_hash = 0;
1120         frame = dx_probe(NULL, dir, &hinfo, frames);
1121         if (IS_ERR(frame))
1122                 return PTR_ERR(frame);
1123
1124         /* Add '.' and '..' from the htree header */
1125         if (!start_hash && !start_minor_hash) {
1126                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1127                 tmp_str.name = de->name;
1128                 tmp_str.len = de->name_len;
1129                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1130                                               de, &tmp_str);
1131                 if (err != 0)
1132                         goto errout;
1133                 count++;
1134         }
1135         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1136                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1137                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1138                 tmp_str.name = de->name;
1139                 tmp_str.len = de->name_len;
1140                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1141                                               de, &tmp_str);
1142                 if (err != 0)
1143                         goto errout;
1144                 count++;
1145         }
1146
1147         while (1) {
1148                 if (fatal_signal_pending(current)) {
1149                         err = -ERESTARTSYS;
1150                         goto errout;
1151                 }
1152                 cond_resched();
1153                 block = dx_get_block(frame->at);
1154                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1155                                              start_hash, start_minor_hash);
1156                 if (ret < 0) {
1157                         err = ret;
1158                         goto errout;
1159                 }
1160                 count += ret;
1161                 hashval = ~0;
1162                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1163                                             frame, frames, &hashval);
1164                 *next_hash = hashval;
1165                 if (ret < 0) {
1166                         err = ret;
1167                         goto errout;
1168                 }
1169                 /*
1170                  * Stop if:  (a) there are no more entries, or
1171                  * (b) we have inserted at least one entry and the
1172                  * next hash value is not a continuation
1173                  */
1174                 if ((ret == 0) ||
1175                     (count && ((hashval & 1) == 0)))
1176                         break;
1177         }
1178         dx_release(frames);
1179         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1180                        "next hash: %x\n", count, *next_hash));
1181         return count;
1182 errout:
1183         dx_release(frames);
1184         return (err);
1185 }
1186
1187 static inline int search_dirblock(struct buffer_head *bh,
1188                                   struct inode *dir,
1189                                   struct ext4_filename *fname,
1190                                   unsigned int offset,
1191                                   struct ext4_dir_entry_2 **res_dir)
1192 {
1193         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1194                                fname, offset, res_dir);
1195 }
1196
1197 /*
1198  * Directory block splitting, compacting
1199  */
1200
1201 /*
1202  * Create map of hash values, offsets, and sizes, stored at end of block.
1203  * Returns number of entries mapped.
1204  */
1205 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1206                        unsigned blocksize, struct dx_hash_info *hinfo,
1207                        struct dx_map_entry *map_tail)
1208 {
1209         int count = 0;
1210         char *base = (char *) de;
1211         struct dx_hash_info h = *hinfo;
1212
1213         while ((char *) de < base + blocksize) {
1214                 if (de->name_len && de->inode) {
1215                         ext4fs_dirhash(dir, de->name, de->name_len, &h);
1216                         map_tail--;
1217                         map_tail->hash = h.hash;
1218                         map_tail->offs = ((char *) de - base)>>2;
1219                         map_tail->size = le16_to_cpu(de->rec_len);
1220                         count++;
1221                         cond_resched();
1222                 }
1223                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1224                 de = ext4_next_entry(de, blocksize);
1225         }
1226         return count;
1227 }
1228
1229 /* Sort map by hash value */
1230 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1231 {
1232         struct dx_map_entry *p, *q, *top = map + count - 1;
1233         int more;
1234         /* Combsort until bubble sort doesn't suck */
1235         while (count > 2) {
1236                 count = count*10/13;
1237                 if (count - 9 < 2) /* 9, 10 -> 11 */
1238                         count = 11;
1239                 for (p = top, q = p - count; q >= map; p--, q--)
1240                         if (p->hash < q->hash)
1241                                 swap(*p, *q);
1242         }
1243         /* Garden variety bubble sort */
1244         do {
1245                 more = 0;
1246                 q = top;
1247                 while (q-- > map) {
1248                         if (q[1].hash >= q[0].hash)
1249                                 continue;
1250                         swap(*(q+1), *q);
1251                         more = 1;
1252                 }
1253         } while(more);
1254 }
1255
1256 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1257 {
1258         struct dx_entry *entries = frame->entries;
1259         struct dx_entry *old = frame->at, *new = old + 1;
1260         int count = dx_get_count(entries);
1261
1262         assert(count < dx_get_limit(entries));
1263         assert(old < entries + count);
1264         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1265         dx_set_hash(new, hash);
1266         dx_set_block(new, block);
1267         dx_set_count(entries, count + 1);
1268 }
1269
1270 #ifdef CONFIG_UNICODE
1271 /*
1272  * Test whether a case-insensitive directory entry matches the filename
1273  * being searched for.  If quick is set, assume the name being looked up
1274  * is already in the casefolded form.
1275  *
1276  * Returns: 0 if the directory entry matches, more than 0 if it
1277  * doesn't match or less than zero on error.
1278  */
1279 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1280                     const struct qstr *entry, bool quick)
1281 {
1282         const struct super_block *sb = parent->i_sb;
1283         const struct unicode_map *um = sb->s_encoding;
1284         int ret;
1285
1286         if (quick)
1287                 ret = utf8_strncasecmp_folded(um, name, entry);
1288         else
1289                 ret = utf8_strncasecmp(um, name, entry);
1290
1291         if (ret < 0) {
1292                 /* Handle invalid character sequence as either an error
1293                  * or as an opaque byte sequence.
1294                  */
1295                 if (sb_has_strict_encoding(sb))
1296                         return -EINVAL;
1297
1298                 if (name->len != entry->len)
1299                         return 1;
1300
1301                 return !!memcmp(name->name, entry->name, name->len);
1302         }
1303
1304         return ret;
1305 }
1306
1307 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1308                                   struct fscrypt_str *cf_name)
1309 {
1310         int len;
1311
1312         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1313                 cf_name->name = NULL;
1314                 return;
1315         }
1316
1317         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1318         if (!cf_name->name)
1319                 return;
1320
1321         len = utf8_casefold(dir->i_sb->s_encoding,
1322                             iname, cf_name->name,
1323                             EXT4_NAME_LEN);
1324         if (len <= 0) {
1325                 kfree(cf_name->name);
1326                 cf_name->name = NULL;
1327                 return;
1328         }
1329         cf_name->len = (unsigned) len;
1330
1331 }
1332 #endif
1333
1334 /*
1335  * Test whether a directory entry matches the filename being searched for.
1336  *
1337  * Return: %true if the directory entry matches, otherwise %false.
1338  */
1339 static inline bool ext4_match(const struct inode *parent,
1340                               const struct ext4_filename *fname,
1341                               const struct ext4_dir_entry_2 *de)
1342 {
1343         struct fscrypt_name f;
1344 #ifdef CONFIG_UNICODE
1345         const struct qstr entry = {.name = de->name, .len = de->name_len};
1346 #endif
1347
1348         if (!de->inode)
1349                 return false;
1350
1351         f.usr_fname = fname->usr_fname;
1352         f.disk_name = fname->disk_name;
1353 #ifdef CONFIG_FS_ENCRYPTION
1354         f.crypto_buf = fname->crypto_buf;
1355 #endif
1356
1357 #ifdef CONFIG_UNICODE
1358         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1359                 if (fname->cf_name.name) {
1360                         struct qstr cf = {.name = fname->cf_name.name,
1361                                           .len = fname->cf_name.len};
1362                         return !ext4_ci_compare(parent, &cf, &entry, true);
1363                 }
1364                 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1365                                         false);
1366         }
1367 #endif
1368
1369         return fscrypt_match_name(&f, de->name, de->name_len);
1370 }
1371
1372 /*
1373  * Returns 0 if not found, -1 on failure, and 1 on success
1374  */
1375 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1376                     struct inode *dir, struct ext4_filename *fname,
1377                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1378 {
1379         struct ext4_dir_entry_2 * de;
1380         char * dlimit;
1381         int de_len;
1382
1383         de = (struct ext4_dir_entry_2 *)search_buf;
1384         dlimit = search_buf + buf_size;
1385         while ((char *) de < dlimit) {
1386                 /* this code is executed quadratically often */
1387                 /* do minimal checking `by hand' */
1388                 if ((char *) de + de->name_len <= dlimit &&
1389                     ext4_match(dir, fname, de)) {
1390                         /* found a match - just to be sure, do
1391                          * a full check */
1392                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1393                                                  buf_size, offset))
1394                                 return -1;
1395                         *res_dir = de;
1396                         return 1;
1397                 }
1398                 /* prevent looping on a bad block */
1399                 de_len = ext4_rec_len_from_disk(de->rec_len,
1400                                                 dir->i_sb->s_blocksize);
1401                 if (de_len <= 0)
1402                         return -1;
1403                 offset += de_len;
1404                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1405         }
1406         return 0;
1407 }
1408
1409 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1410                                struct ext4_dir_entry *de)
1411 {
1412         struct super_block *sb = dir->i_sb;
1413
1414         if (!is_dx(dir))
1415                 return 0;
1416         if (block == 0)
1417                 return 1;
1418         if (de->inode == 0 &&
1419             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1420                         sb->s_blocksize)
1421                 return 1;
1422         return 0;
1423 }
1424
1425 /*
1426  *      __ext4_find_entry()
1427  *
1428  * finds an entry in the specified directory with the wanted name. It
1429  * returns the cache buffer in which the entry was found, and the entry
1430  * itself (as a parameter - res_dir). It does NOT read the inode of the
1431  * entry - you'll have to do that yourself if you want to.
1432  *
1433  * The returned buffer_head has ->b_count elevated.  The caller is expected
1434  * to brelse() it when appropriate.
1435  */
1436 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1437                                              struct ext4_filename *fname,
1438                                              struct ext4_dir_entry_2 **res_dir,
1439                                              int *inlined)
1440 {
1441         struct super_block *sb;
1442         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1443         struct buffer_head *bh, *ret = NULL;
1444         ext4_lblk_t start, block;
1445         const u8 *name = fname->usr_fname->name;
1446         size_t ra_max = 0;      /* Number of bh's in the readahead
1447                                    buffer, bh_use[] */
1448         size_t ra_ptr = 0;      /* Current index into readahead
1449                                    buffer */
1450         ext4_lblk_t  nblocks;
1451         int i, namelen, retval;
1452
1453         *res_dir = NULL;
1454         sb = dir->i_sb;
1455         namelen = fname->usr_fname->len;
1456         if (namelen > EXT4_NAME_LEN)
1457                 return NULL;
1458
1459         if (ext4_has_inline_data(dir)) {
1460                 int has_inline_data = 1;
1461                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1462                                              &has_inline_data);
1463                 if (has_inline_data) {
1464                         if (inlined)
1465                                 *inlined = 1;
1466                         goto cleanup_and_exit;
1467                 }
1468         }
1469
1470         if ((namelen <= 2) && (name[0] == '.') &&
1471             (name[1] == '.' || name[1] == '\0')) {
1472                 /*
1473                  * "." or ".." will only be in the first block
1474                  * NFS may look up ".."; "." should be handled by the VFS
1475                  */
1476                 block = start = 0;
1477                 nblocks = 1;
1478                 goto restart;
1479         }
1480         if (is_dx(dir)) {
1481                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1482                 /*
1483                  * On success, or if the error was file not found,
1484                  * return.  Otherwise, fall back to doing a search the
1485                  * old fashioned way.
1486                  */
1487                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1488                         goto cleanup_and_exit;
1489                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1490                                "falling back\n"));
1491                 ret = NULL;
1492         }
1493         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1494         if (!nblocks) {
1495                 ret = NULL;
1496                 goto cleanup_and_exit;
1497         }
1498         start = EXT4_I(dir)->i_dir_start_lookup;
1499         if (start >= nblocks)
1500                 start = 0;
1501         block = start;
1502 restart:
1503         do {
1504                 /*
1505                  * We deal with the read-ahead logic here.
1506                  */
1507                 cond_resched();
1508                 if (ra_ptr >= ra_max) {
1509                         /* Refill the readahead buffer */
1510                         ra_ptr = 0;
1511                         if (block < start)
1512                                 ra_max = start - block;
1513                         else
1514                                 ra_max = nblocks - block;
1515                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1516                         retval = ext4_bread_batch(dir, block, ra_max,
1517                                                   false /* wait */, bh_use);
1518                         if (retval) {
1519                                 ret = ERR_PTR(retval);
1520                                 ra_max = 0;
1521                                 goto cleanup_and_exit;
1522                         }
1523                 }
1524                 if ((bh = bh_use[ra_ptr++]) == NULL)
1525                         goto next;
1526                 wait_on_buffer(bh);
1527                 if (!buffer_uptodate(bh)) {
1528                         EXT4_ERROR_INODE_ERR(dir, EIO,
1529                                              "reading directory lblock %lu",
1530                                              (unsigned long) block);
1531                         brelse(bh);
1532                         ret = ERR_PTR(-EIO);
1533                         goto cleanup_and_exit;
1534                 }
1535                 if (!buffer_verified(bh) &&
1536                     !is_dx_internal_node(dir, block,
1537                                          (struct ext4_dir_entry *)bh->b_data) &&
1538                     !ext4_dirblock_csum_verify(dir, bh)) {
1539                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1540                                              "checksumming directory "
1541                                              "block %lu", (unsigned long)block);
1542                         brelse(bh);
1543                         ret = ERR_PTR(-EFSBADCRC);
1544                         goto cleanup_and_exit;
1545                 }
1546                 set_buffer_verified(bh);
1547                 i = search_dirblock(bh, dir, fname,
1548                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1549                 if (i == 1) {
1550                         EXT4_I(dir)->i_dir_start_lookup = block;
1551                         ret = bh;
1552                         goto cleanup_and_exit;
1553                 } else {
1554                         brelse(bh);
1555                         if (i < 0)
1556                                 goto cleanup_and_exit;
1557                 }
1558         next:
1559                 if (++block >= nblocks)
1560                         block = 0;
1561         } while (block != start);
1562
1563         /*
1564          * If the directory has grown while we were searching, then
1565          * search the last part of the directory before giving up.
1566          */
1567         block = nblocks;
1568         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1569         if (block < nblocks) {
1570                 start = 0;
1571                 goto restart;
1572         }
1573
1574 cleanup_and_exit:
1575         /* Clean up the read-ahead blocks */
1576         for (; ra_ptr < ra_max; ra_ptr++)
1577                 brelse(bh_use[ra_ptr]);
1578         return ret;
1579 }
1580
1581 static struct buffer_head *ext4_find_entry(struct inode *dir,
1582                                            const struct qstr *d_name,
1583                                            struct ext4_dir_entry_2 **res_dir,
1584                                            int *inlined)
1585 {
1586         int err;
1587         struct ext4_filename fname;
1588         struct buffer_head *bh;
1589
1590         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1591         if (err == -ENOENT)
1592                 return NULL;
1593         if (err)
1594                 return ERR_PTR(err);
1595
1596         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1597
1598         ext4_fname_free_filename(&fname);
1599         return bh;
1600 }
1601
1602 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1603                                              struct dentry *dentry,
1604                                              struct ext4_dir_entry_2 **res_dir)
1605 {
1606         int err;
1607         struct ext4_filename fname;
1608         struct buffer_head *bh;
1609
1610         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1611         generic_set_encrypted_ci_d_ops(dentry);
1612         if (err == -ENOENT)
1613                 return NULL;
1614         if (err)
1615                 return ERR_PTR(err);
1616
1617         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1618
1619         ext4_fname_free_filename(&fname);
1620         return bh;
1621 }
1622
1623 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1624                         struct ext4_filename *fname,
1625                         struct ext4_dir_entry_2 **res_dir)
1626 {
1627         struct super_block * sb = dir->i_sb;
1628         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1629         struct buffer_head *bh;
1630         ext4_lblk_t block;
1631         int retval;
1632
1633 #ifdef CONFIG_FS_ENCRYPTION
1634         *res_dir = NULL;
1635 #endif
1636         frame = dx_probe(fname, dir, NULL, frames);
1637         if (IS_ERR(frame))
1638                 return (struct buffer_head *) frame;
1639         do {
1640                 block = dx_get_block(frame->at);
1641                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1642                 if (IS_ERR(bh))
1643                         goto errout;
1644
1645                 retval = search_dirblock(bh, dir, fname,
1646                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1647                                          res_dir);
1648                 if (retval == 1)
1649                         goto success;
1650                 brelse(bh);
1651                 if (retval == -1) {
1652                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1653                         goto errout;
1654                 }
1655
1656                 /* Check to see if we should continue to search */
1657                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1658                                                frames, NULL);
1659                 if (retval < 0) {
1660                         ext4_warning_inode(dir,
1661                                 "error %d reading directory index block",
1662                                 retval);
1663                         bh = ERR_PTR(retval);
1664                         goto errout;
1665                 }
1666         } while (retval == 1);
1667
1668         bh = NULL;
1669 errout:
1670         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1671 success:
1672         dx_release(frames);
1673         return bh;
1674 }
1675
1676 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1677 {
1678         struct inode *inode;
1679         struct ext4_dir_entry_2 *de;
1680         struct buffer_head *bh;
1681
1682         if (dentry->d_name.len > EXT4_NAME_LEN)
1683                 return ERR_PTR(-ENAMETOOLONG);
1684
1685         bh = ext4_lookup_entry(dir, dentry, &de);
1686         if (IS_ERR(bh))
1687                 return ERR_CAST(bh);
1688         inode = NULL;
1689         if (bh) {
1690                 __u32 ino = le32_to_cpu(de->inode);
1691                 brelse(bh);
1692                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1693                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1694                         return ERR_PTR(-EFSCORRUPTED);
1695                 }
1696                 if (unlikely(ino == dir->i_ino)) {
1697                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1698                                          dentry);
1699                         return ERR_PTR(-EFSCORRUPTED);
1700                 }
1701                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1702                 if (inode == ERR_PTR(-ESTALE)) {
1703                         EXT4_ERROR_INODE(dir,
1704                                          "deleted inode referenced: %u",
1705                                          ino);
1706                         return ERR_PTR(-EFSCORRUPTED);
1707                 }
1708                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1709                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1710                     !fscrypt_has_permitted_context(dir, inode)) {
1711                         ext4_warning(inode->i_sb,
1712                                      "Inconsistent encryption contexts: %lu/%lu",
1713                                      dir->i_ino, inode->i_ino);
1714                         iput(inode);
1715                         return ERR_PTR(-EPERM);
1716                 }
1717         }
1718
1719 #ifdef CONFIG_UNICODE
1720         if (!inode && IS_CASEFOLDED(dir)) {
1721                 /* Eventually we want to call d_add_ci(dentry, NULL)
1722                  * for negative dentries in the encoding case as
1723                  * well.  For now, prevent the negative dentry
1724                  * from being cached.
1725                  */
1726                 return NULL;
1727         }
1728 #endif
1729         return d_splice_alias(inode, dentry);
1730 }
1731
1732
1733 struct dentry *ext4_get_parent(struct dentry *child)
1734 {
1735         __u32 ino;
1736         static const struct qstr dotdot = QSTR_INIT("..", 2);
1737         struct ext4_dir_entry_2 * de;
1738         struct buffer_head *bh;
1739
1740         bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1741         if (IS_ERR(bh))
1742                 return ERR_CAST(bh);
1743         if (!bh)
1744                 return ERR_PTR(-ENOENT);
1745         ino = le32_to_cpu(de->inode);
1746         brelse(bh);
1747
1748         if (!ext4_valid_inum(child->d_sb, ino)) {
1749                 EXT4_ERROR_INODE(d_inode(child),
1750                                  "bad parent inode number: %u", ino);
1751                 return ERR_PTR(-EFSCORRUPTED);
1752         }
1753
1754         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1755 }
1756
1757 /*
1758  * Move count entries from end of map between two memory locations.
1759  * Returns pointer to last entry moved.
1760  */
1761 static struct ext4_dir_entry_2 *
1762 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1763                 unsigned blocksize)
1764 {
1765         unsigned rec_len = 0;
1766
1767         while (count--) {
1768                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1769                                                 (from + (map->offs<<2));
1770                 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1771                 memcpy (to, de, rec_len);
1772                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1773                                 ext4_rec_len_to_disk(rec_len, blocksize);
1774                 de->inode = 0;
1775                 map++;
1776                 to += rec_len;
1777         }
1778         return (struct ext4_dir_entry_2 *) (to - rec_len);
1779 }
1780
1781 /*
1782  * Compact each dir entry in the range to the minimal rec_len.
1783  * Returns pointer to last entry in range.
1784  */
1785 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1786 {
1787         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1788         unsigned rec_len = 0;
1789
1790         prev = to = de;
1791         while ((char*)de < base + blocksize) {
1792                 next = ext4_next_entry(de, blocksize);
1793                 if (de->inode && de->name_len) {
1794                         rec_len = EXT4_DIR_REC_LEN(de->name_len);
1795                         if (de > to)
1796                                 memmove(to, de, rec_len);
1797                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1798                         prev = to;
1799                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1800                 }
1801                 de = next;
1802         }
1803         return prev;
1804 }
1805
1806 /*
1807  * Split a full leaf block to make room for a new dir entry.
1808  * Allocate a new block, and move entries so that they are approx. equally full.
1809  * Returns pointer to de in block into which the new entry will be inserted.
1810  */
1811 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1812                         struct buffer_head **bh,struct dx_frame *frame,
1813                         struct dx_hash_info *hinfo)
1814 {
1815         unsigned blocksize = dir->i_sb->s_blocksize;
1816         unsigned count, continued;
1817         struct buffer_head *bh2;
1818         ext4_lblk_t newblock;
1819         u32 hash2;
1820         struct dx_map_entry *map;
1821         char *data1 = (*bh)->b_data, *data2;
1822         unsigned split, move, size;
1823         struct ext4_dir_entry_2 *de = NULL, *de2;
1824         int     csum_size = 0;
1825         int     err = 0, i;
1826
1827         if (ext4_has_metadata_csum(dir->i_sb))
1828                 csum_size = sizeof(struct ext4_dir_entry_tail);
1829
1830         bh2 = ext4_append(handle, dir, &newblock);
1831         if (IS_ERR(bh2)) {
1832                 brelse(*bh);
1833                 *bh = NULL;
1834                 return (struct ext4_dir_entry_2 *) bh2;
1835         }
1836
1837         BUFFER_TRACE(*bh, "get_write_access");
1838         err = ext4_journal_get_write_access(handle, *bh);
1839         if (err)
1840                 goto journal_error;
1841
1842         BUFFER_TRACE(frame->bh, "get_write_access");
1843         err = ext4_journal_get_write_access(handle, frame->bh);
1844         if (err)
1845                 goto journal_error;
1846
1847         data2 = bh2->b_data;
1848
1849         /* create map in the end of data2 block */
1850         map = (struct dx_map_entry *) (data2 + blocksize);
1851         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1852                              blocksize, hinfo, map);
1853         map -= count;
1854         dx_sort_map(map, count);
1855         /* Ensure that neither split block is over half full */
1856         size = 0;
1857         move = 0;
1858         for (i = count-1; i >= 0; i--) {
1859                 /* is more than half of this entry in 2nd half of the block? */
1860                 if (size + map[i].size/2 > blocksize/2)
1861                         break;
1862                 size += map[i].size;
1863                 move++;
1864         }
1865         /*
1866          * map index at which we will split
1867          *
1868          * If the sum of active entries didn't exceed half the block size, just
1869          * split it in half by count; each resulting block will have at least
1870          * half the space free.
1871          */
1872         if (i > 0)
1873                 split = count - move;
1874         else
1875                 split = count/2;
1876
1877         hash2 = map[split].hash;
1878         continued = hash2 == map[split - 1].hash;
1879         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1880                         (unsigned long)dx_get_block(frame->at),
1881                                         hash2, split, count-split));
1882
1883         /* Fancy dance to stay within two buffers */
1884         de2 = dx_move_dirents(data1, data2, map + split, count - split,
1885                               blocksize);
1886         de = dx_pack_dirents(data1, blocksize);
1887         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1888                                            (char *) de,
1889                                            blocksize);
1890         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1891                                             (char *) de2,
1892                                             blocksize);
1893         if (csum_size) {
1894                 ext4_initialize_dirent_tail(*bh, blocksize);
1895                 ext4_initialize_dirent_tail(bh2, blocksize);
1896         }
1897
1898         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1899                         blocksize, 1));
1900         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1901                         blocksize, 1));
1902
1903         /* Which block gets the new entry? */
1904         if (hinfo->hash >= hash2) {
1905                 swap(*bh, bh2);
1906                 de = de2;
1907         }
1908         dx_insert_block(frame, hash2 + continued, newblock);
1909         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1910         if (err)
1911                 goto journal_error;
1912         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1913         if (err)
1914                 goto journal_error;
1915         brelse(bh2);
1916         dxtrace(dx_show_index("frame", frame->entries));
1917         return de;
1918
1919 journal_error:
1920         brelse(*bh);
1921         brelse(bh2);
1922         *bh = NULL;
1923         ext4_std_error(dir->i_sb, err);
1924         return ERR_PTR(err);
1925 }
1926
1927 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1928                       struct buffer_head *bh,
1929                       void *buf, int buf_size,
1930                       struct ext4_filename *fname,
1931                       struct ext4_dir_entry_2 **dest_de)
1932 {
1933         struct ext4_dir_entry_2 *de;
1934         unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1935         int nlen, rlen;
1936         unsigned int offset = 0;
1937         char *top;
1938
1939         de = (struct ext4_dir_entry_2 *)buf;
1940         top = buf + buf_size - reclen;
1941         while ((char *) de <= top) {
1942                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1943                                          buf, buf_size, offset))
1944                         return -EFSCORRUPTED;
1945                 if (ext4_match(dir, fname, de))
1946                         return -EEXIST;
1947                 nlen = EXT4_DIR_REC_LEN(de->name_len);
1948                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1949                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1950                         break;
1951                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1952                 offset += rlen;
1953         }
1954         if ((char *) de > top)
1955                 return -ENOSPC;
1956
1957         *dest_de = de;
1958         return 0;
1959 }
1960
1961 void ext4_insert_dentry(struct inode *inode,
1962                         struct ext4_dir_entry_2 *de,
1963                         int buf_size,
1964                         struct ext4_filename *fname)
1965 {
1966
1967         int nlen, rlen;
1968
1969         nlen = EXT4_DIR_REC_LEN(de->name_len);
1970         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1971         if (de->inode) {
1972                 struct ext4_dir_entry_2 *de1 =
1973                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
1974                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1975                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1976                 de = de1;
1977         }
1978         de->file_type = EXT4_FT_UNKNOWN;
1979         de->inode = cpu_to_le32(inode->i_ino);
1980         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1981         de->name_len = fname_len(fname);
1982         memcpy(de->name, fname_name(fname), fname_len(fname));
1983 }
1984
1985 /*
1986  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1987  * it points to a directory entry which is guaranteed to be large
1988  * enough for new directory entry.  If de is NULL, then
1989  * add_dirent_to_buf will attempt search the directory block for
1990  * space.  It will return -ENOSPC if no space is available, and -EIO
1991  * and -EEXIST if directory entry already exists.
1992  */
1993 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1994                              struct inode *dir,
1995                              struct inode *inode, struct ext4_dir_entry_2 *de,
1996                              struct buffer_head *bh)
1997 {
1998         unsigned int    blocksize = dir->i_sb->s_blocksize;
1999         int             csum_size = 0;
2000         int             err, err2;
2001
2002         if (ext4_has_metadata_csum(inode->i_sb))
2003                 csum_size = sizeof(struct ext4_dir_entry_tail);
2004
2005         if (!de) {
2006                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2007                                         blocksize - csum_size, fname, &de);
2008                 if (err)
2009                         return err;
2010         }
2011         BUFFER_TRACE(bh, "get_write_access");
2012         err = ext4_journal_get_write_access(handle, bh);
2013         if (err) {
2014                 ext4_std_error(dir->i_sb, err);
2015                 return err;
2016         }
2017
2018         /* By now the buffer is marked for journaling */
2019         ext4_insert_dentry(inode, de, blocksize, fname);
2020
2021         /*
2022          * XXX shouldn't update any times until successful
2023          * completion of syscall, but too many callers depend
2024          * on this.
2025          *
2026          * XXX similarly, too many callers depend on
2027          * ext4_new_inode() setting the times, but error
2028          * recovery deletes the inode, so the worst that can
2029          * happen is that the times are slightly out of date
2030          * and/or different from the directory change time.
2031          */
2032         dir->i_mtime = dir->i_ctime = current_time(dir);
2033         ext4_update_dx_flag(dir);
2034         inode_inc_iversion(dir);
2035         err2 = ext4_mark_inode_dirty(handle, dir);
2036         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2037         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2038         if (err)
2039                 ext4_std_error(dir->i_sb, err);
2040         return err ? err : err2;
2041 }
2042
2043 /*
2044  * This converts a one block unindexed directory to a 3 block indexed
2045  * directory, and adds the dentry to the indexed directory.
2046  */
2047 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2048                             struct inode *dir,
2049                             struct inode *inode, struct buffer_head *bh)
2050 {
2051         struct buffer_head *bh2;
2052         struct dx_root  *root;
2053         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2054         struct dx_entry *entries;
2055         struct ext4_dir_entry_2 *de, *de2;
2056         char            *data2, *top;
2057         unsigned        len;
2058         int             retval;
2059         unsigned        blocksize;
2060         ext4_lblk_t  block;
2061         struct fake_dirent *fde;
2062         int csum_size = 0;
2063
2064         if (ext4_has_metadata_csum(inode->i_sb))
2065                 csum_size = sizeof(struct ext4_dir_entry_tail);
2066
2067         blocksize =  dir->i_sb->s_blocksize;
2068         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2069         BUFFER_TRACE(bh, "get_write_access");
2070         retval = ext4_journal_get_write_access(handle, bh);
2071         if (retval) {
2072                 ext4_std_error(dir->i_sb, retval);
2073                 brelse(bh);
2074                 return retval;
2075         }
2076         root = (struct dx_root *) bh->b_data;
2077
2078         /* The 0th block becomes the root, move the dirents out */
2079         fde = &root->dotdot;
2080         de = (struct ext4_dir_entry_2 *)((char *)fde +
2081                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2082         if ((char *) de >= (((char *) root) + blocksize)) {
2083                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2084                 brelse(bh);
2085                 return -EFSCORRUPTED;
2086         }
2087         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2088
2089         /* Allocate new block for the 0th block's dirents */
2090         bh2 = ext4_append(handle, dir, &block);
2091         if (IS_ERR(bh2)) {
2092                 brelse(bh);
2093                 return PTR_ERR(bh2);
2094         }
2095         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2096         data2 = bh2->b_data;
2097
2098         memcpy(data2, de, len);
2099         de = (struct ext4_dir_entry_2 *) data2;
2100         top = data2 + len;
2101         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2102                 de = de2;
2103         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2104                                            (char *) de, blocksize);
2105
2106         if (csum_size)
2107                 ext4_initialize_dirent_tail(bh2, blocksize);
2108
2109         /* Initialize the root; the dot dirents already exist */
2110         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2111         de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2112                                            blocksize);
2113         memset (&root->info, 0, sizeof(root->info));
2114         root->info.info_length = sizeof(root->info);
2115         root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2116         entries = root->entries;
2117         dx_set_block(entries, 1);
2118         dx_set_count(entries, 1);
2119         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2120
2121         /* Initialize as for dx_probe */
2122         fname->hinfo.hash_version = root->info.hash_version;
2123         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2124                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2125         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2126         ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2127
2128         memset(frames, 0, sizeof(frames));
2129         frame = frames;
2130         frame->entries = entries;
2131         frame->at = entries;
2132         frame->bh = bh;
2133
2134         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2135         if (retval)
2136                 goto out_frames;
2137         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2138         if (retval)
2139                 goto out_frames;
2140
2141         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2142         if (IS_ERR(de)) {
2143                 retval = PTR_ERR(de);
2144                 goto out_frames;
2145         }
2146
2147         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2148 out_frames:
2149         /*
2150          * Even if the block split failed, we have to properly write
2151          * out all the changes we did so far. Otherwise we can end up
2152          * with corrupted filesystem.
2153          */
2154         if (retval)
2155                 ext4_mark_inode_dirty(handle, dir);
2156         dx_release(frames);
2157         brelse(bh2);
2158         return retval;
2159 }
2160
2161 /*
2162  *      ext4_add_entry()
2163  *
2164  * adds a file entry to the specified directory, using the same
2165  * semantics as ext4_find_entry(). It returns NULL if it failed.
2166  *
2167  * NOTE!! The inode part of 'de' is left at 0 - which means you
2168  * may not sleep between calling this and putting something into
2169  * the entry, as someone else might have used it while you slept.
2170  */
2171 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2172                           struct inode *inode)
2173 {
2174         struct inode *dir = d_inode(dentry->d_parent);
2175         struct buffer_head *bh = NULL;
2176         struct ext4_dir_entry_2 *de;
2177         struct super_block *sb;
2178         struct ext4_filename fname;
2179         int     retval;
2180         int     dx_fallback=0;
2181         unsigned blocksize;
2182         ext4_lblk_t block, blocks;
2183         int     csum_size = 0;
2184
2185         if (ext4_has_metadata_csum(inode->i_sb))
2186                 csum_size = sizeof(struct ext4_dir_entry_tail);
2187
2188         sb = dir->i_sb;
2189         blocksize = sb->s_blocksize;
2190         if (!dentry->d_name.len)
2191                 return -EINVAL;
2192
2193         if (fscrypt_is_nokey_name(dentry))
2194                 return -ENOKEY;
2195
2196 #ifdef CONFIG_UNICODE
2197         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2198             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2199                 return -EINVAL;
2200 #endif
2201
2202         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2203         if (retval)
2204                 return retval;
2205
2206         if (ext4_has_inline_data(dir)) {
2207                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2208                 if (retval < 0)
2209                         goto out;
2210                 if (retval == 1) {
2211                         retval = 0;
2212                         goto out;
2213                 }
2214         }
2215
2216         if (is_dx(dir)) {
2217                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2218                 if (!retval || (retval != ERR_BAD_DX_DIR))
2219                         goto out;
2220                 /* Can we just ignore htree data? */
2221                 if (ext4_has_metadata_csum(sb)) {
2222                         EXT4_ERROR_INODE(dir,
2223                                 "Directory has corrupted htree index.");
2224                         retval = -EFSCORRUPTED;
2225                         goto out;
2226                 }
2227                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2228                 dx_fallback++;
2229                 retval = ext4_mark_inode_dirty(handle, dir);
2230                 if (unlikely(retval))
2231                         goto out;
2232         }
2233         blocks = dir->i_size >> sb->s_blocksize_bits;
2234         for (block = 0; block < blocks; block++) {
2235                 bh = ext4_read_dirblock(dir, block, DIRENT);
2236                 if (bh == NULL) {
2237                         bh = ext4_bread(handle, dir, block,
2238                                         EXT4_GET_BLOCKS_CREATE);
2239                         goto add_to_new_block;
2240                 }
2241                 if (IS_ERR(bh)) {
2242                         retval = PTR_ERR(bh);
2243                         bh = NULL;
2244                         goto out;
2245                 }
2246                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2247                                            NULL, bh);
2248                 if (retval != -ENOSPC)
2249                         goto out;
2250
2251                 if (blocks == 1 && !dx_fallback &&
2252                     ext4_has_feature_dir_index(sb)) {
2253                         retval = make_indexed_dir(handle, &fname, dir,
2254                                                   inode, bh);
2255                         bh = NULL; /* make_indexed_dir releases bh */
2256                         goto out;
2257                 }
2258                 brelse(bh);
2259         }
2260         bh = ext4_append(handle, dir, &block);
2261 add_to_new_block:
2262         if (IS_ERR(bh)) {
2263                 retval = PTR_ERR(bh);
2264                 bh = NULL;
2265                 goto out;
2266         }
2267         de = (struct ext4_dir_entry_2 *) bh->b_data;
2268         de->inode = 0;
2269         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2270
2271         if (csum_size)
2272                 ext4_initialize_dirent_tail(bh, blocksize);
2273
2274         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2275 out:
2276         ext4_fname_free_filename(&fname);
2277         brelse(bh);
2278         if (retval == 0)
2279                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2280         return retval;
2281 }
2282
2283 /*
2284  * Returns 0 for success, or a negative error value
2285  */
2286 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2287                              struct inode *dir, struct inode *inode)
2288 {
2289         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2290         struct dx_entry *entries, *at;
2291         struct buffer_head *bh;
2292         struct super_block *sb = dir->i_sb;
2293         struct ext4_dir_entry_2 *de;
2294         int restart;
2295         int err;
2296
2297 again:
2298         restart = 0;
2299         frame = dx_probe(fname, dir, NULL, frames);
2300         if (IS_ERR(frame))
2301                 return PTR_ERR(frame);
2302         entries = frame->entries;
2303         at = frame->at;
2304         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2305         if (IS_ERR(bh)) {
2306                 err = PTR_ERR(bh);
2307                 bh = NULL;
2308                 goto cleanup;
2309         }
2310
2311         BUFFER_TRACE(bh, "get_write_access");
2312         err = ext4_journal_get_write_access(handle, bh);
2313         if (err)
2314                 goto journal_error;
2315
2316         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2317         if (err != -ENOSPC)
2318                 goto cleanup;
2319
2320         err = 0;
2321         /* Block full, should compress but for now just split */
2322         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2323                        dx_get_count(entries), dx_get_limit(entries)));
2324         /* Need to split index? */
2325         if (dx_get_count(entries) == dx_get_limit(entries)) {
2326                 ext4_lblk_t newblock;
2327                 int levels = frame - frames + 1;
2328                 unsigned int icount;
2329                 int add_level = 1;
2330                 struct dx_entry *entries2;
2331                 struct dx_node *node2;
2332                 struct buffer_head *bh2;
2333
2334                 while (frame > frames) {
2335                         if (dx_get_count((frame - 1)->entries) <
2336                             dx_get_limit((frame - 1)->entries)) {
2337                                 add_level = 0;
2338                                 break;
2339                         }
2340                         frame--; /* split higher index block */
2341                         at = frame->at;
2342                         entries = frame->entries;
2343                         restart = 1;
2344                 }
2345                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2346                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2347                                          "reach max htree level :%d",
2348                                          dir->i_ino, levels);
2349                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2350                                 ext4_warning(sb, "Large directory feature is "
2351                                                  "not enabled on this "
2352                                                  "filesystem");
2353                         }
2354                         err = -ENOSPC;
2355                         goto cleanup;
2356                 }
2357                 icount = dx_get_count(entries);
2358                 bh2 = ext4_append(handle, dir, &newblock);
2359                 if (IS_ERR(bh2)) {
2360                         err = PTR_ERR(bh2);
2361                         goto cleanup;
2362                 }
2363                 node2 = (struct dx_node *)(bh2->b_data);
2364                 entries2 = node2->entries;
2365                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2366                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2367                                                            sb->s_blocksize);
2368                 BUFFER_TRACE(frame->bh, "get_write_access");
2369                 err = ext4_journal_get_write_access(handle, frame->bh);
2370                 if (err)
2371                         goto journal_error;
2372                 if (!add_level) {
2373                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2374                         unsigned hash2 = dx_get_hash(entries + icount1);
2375                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2376                                        icount1, icount2));
2377
2378                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2379                         err = ext4_journal_get_write_access(handle,
2380                                                              (frame - 1)->bh);
2381                         if (err)
2382                                 goto journal_error;
2383
2384                         memcpy((char *) entries2, (char *) (entries + icount1),
2385                                icount2 * sizeof(struct dx_entry));
2386                         dx_set_count(entries, icount1);
2387                         dx_set_count(entries2, icount2);
2388                         dx_set_limit(entries2, dx_node_limit(dir));
2389
2390                         /* Which index block gets the new entry? */
2391                         if (at - entries >= icount1) {
2392                                 frame->at = at = at - entries - icount1 + entries2;
2393                                 frame->entries = entries = entries2;
2394                                 swap(frame->bh, bh2);
2395                         }
2396                         dx_insert_block((frame - 1), hash2, newblock);
2397                         dxtrace(dx_show_index("node", frame->entries));
2398                         dxtrace(dx_show_index("node",
2399                                ((struct dx_node *) bh2->b_data)->entries));
2400                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2401                         if (err)
2402                                 goto journal_error;
2403                         brelse (bh2);
2404                         err = ext4_handle_dirty_dx_node(handle, dir,
2405                                                    (frame - 1)->bh);
2406                         if (err)
2407                                 goto journal_error;
2408                         if (restart) {
2409                                 err = ext4_handle_dirty_dx_node(handle, dir,
2410                                                            frame->bh);
2411                                 goto journal_error;
2412                         }
2413                 } else {
2414                         struct dx_root *dxroot;
2415                         memcpy((char *) entries2, (char *) entries,
2416                                icount * sizeof(struct dx_entry));
2417                         dx_set_limit(entries2, dx_node_limit(dir));
2418
2419                         /* Set up root */
2420                         dx_set_count(entries, 1);
2421                         dx_set_block(entries + 0, newblock);
2422                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2423                         dxroot->info.indirect_levels += 1;
2424                         dxtrace(printk(KERN_DEBUG
2425                                        "Creating %d level index...\n",
2426                                        dxroot->info.indirect_levels));
2427                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2428                         if (err)
2429                                 goto journal_error;
2430                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2431                         brelse(bh2);
2432                         restart = 1;
2433                         goto journal_error;
2434                 }
2435         }
2436         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2437         if (IS_ERR(de)) {
2438                 err = PTR_ERR(de);
2439                 goto cleanup;
2440         }
2441         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2442         goto cleanup;
2443
2444 journal_error:
2445         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2446 cleanup:
2447         brelse(bh);
2448         dx_release(frames);
2449         /* @restart is true means htree-path has been changed, we need to
2450          * repeat dx_probe() to find out valid htree-path
2451          */
2452         if (restart && err == 0)
2453                 goto again;
2454         return err;
2455 }
2456
2457 /*
2458  * ext4_generic_delete_entry deletes a directory entry by merging it
2459  * with the previous entry
2460  */
2461 int ext4_generic_delete_entry(struct inode *dir,
2462                               struct ext4_dir_entry_2 *de_del,
2463                               struct buffer_head *bh,
2464                               void *entry_buf,
2465                               int buf_size,
2466                               int csum_size)
2467 {
2468         struct ext4_dir_entry_2 *de, *pde;
2469         unsigned int blocksize = dir->i_sb->s_blocksize;
2470         int i;
2471
2472         i = 0;
2473         pde = NULL;
2474         de = (struct ext4_dir_entry_2 *)entry_buf;
2475         while (i < buf_size - csum_size) {
2476                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2477                                          entry_buf, buf_size, i))
2478                         return -EFSCORRUPTED;
2479                 if (de == de_del)  {
2480                         if (pde)
2481                                 pde->rec_len = ext4_rec_len_to_disk(
2482                                         ext4_rec_len_from_disk(pde->rec_len,
2483                                                                blocksize) +
2484                                         ext4_rec_len_from_disk(de->rec_len,
2485                                                                blocksize),
2486                                         blocksize);
2487                         else
2488                                 de->inode = 0;
2489                         inode_inc_iversion(dir);
2490                         return 0;
2491                 }
2492                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2493                 pde = de;
2494                 de = ext4_next_entry(de, blocksize);
2495         }
2496         return -ENOENT;
2497 }
2498
2499 static int ext4_delete_entry(handle_t *handle,
2500                              struct inode *dir,
2501                              struct ext4_dir_entry_2 *de_del,
2502                              struct buffer_head *bh)
2503 {
2504         int err, csum_size = 0;
2505
2506         if (ext4_has_inline_data(dir)) {
2507                 int has_inline_data = 1;
2508                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2509                                                &has_inline_data);
2510                 if (has_inline_data)
2511                         return err;
2512         }
2513
2514         if (ext4_has_metadata_csum(dir->i_sb))
2515                 csum_size = sizeof(struct ext4_dir_entry_tail);
2516
2517         BUFFER_TRACE(bh, "get_write_access");
2518         err = ext4_journal_get_write_access(handle, bh);
2519         if (unlikely(err))
2520                 goto out;
2521
2522         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2523                                         dir->i_sb->s_blocksize, csum_size);
2524         if (err)
2525                 goto out;
2526
2527         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2528         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2529         if (unlikely(err))
2530                 goto out;
2531
2532         return 0;
2533 out:
2534         if (err != -ENOENT)
2535                 ext4_std_error(dir->i_sb, err);
2536         return err;
2537 }
2538
2539 /*
2540  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2541  * since this indicates that nlinks count was previously 1 to avoid overflowing
2542  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2543  * that subdirectory link counts are not being maintained accurately.
2544  *
2545  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2546  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2547  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2548  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2549  */
2550 static void ext4_inc_count(struct inode *inode)
2551 {
2552         inc_nlink(inode);
2553         if (is_dx(inode) &&
2554             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2555                 set_nlink(inode, 1);
2556 }
2557
2558 /*
2559  * If a directory had nlink == 1, then we should let it be 1. This indicates
2560  * directory has >EXT4_LINK_MAX subdirs.
2561  */
2562 static void ext4_dec_count(struct inode *inode)
2563 {
2564         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2565                 drop_nlink(inode);
2566 }
2567
2568
2569 /*
2570  * Add non-directory inode to a directory. On success, the inode reference is
2571  * consumed by dentry is instantiation. This is also indicated by clearing of
2572  * *inodep pointer. On failure, the caller is responsible for dropping the
2573  * inode reference in the safe context.
2574  */
2575 static int ext4_add_nondir(handle_t *handle,
2576                 struct dentry *dentry, struct inode **inodep)
2577 {
2578         struct inode *dir = d_inode(dentry->d_parent);
2579         struct inode *inode = *inodep;
2580         int err = ext4_add_entry(handle, dentry, inode);
2581         if (!err) {
2582                 err = ext4_mark_inode_dirty(handle, inode);
2583                 if (IS_DIRSYNC(dir))
2584                         ext4_handle_sync(handle);
2585                 d_instantiate_new(dentry, inode);
2586                 *inodep = NULL;
2587                 return err;
2588         }
2589         drop_nlink(inode);
2590         ext4_orphan_add(handle, inode);
2591         unlock_new_inode(inode);
2592         return err;
2593 }
2594
2595 /*
2596  * By the time this is called, we already have created
2597  * the directory cache entry for the new file, but it
2598  * is so far negative - it has no inode.
2599  *
2600  * If the create succeeds, we fill in the inode information
2601  * with d_instantiate().
2602  */
2603 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2604                        bool excl)
2605 {
2606         handle_t *handle;
2607         struct inode *inode;
2608         int err, credits, retries = 0;
2609
2610         err = dquot_initialize(dir);
2611         if (err)
2612                 return err;
2613
2614         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2615                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2616 retry:
2617         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2618                                             NULL, EXT4_HT_DIR, credits);
2619         handle = ext4_journal_current_handle();
2620         err = PTR_ERR(inode);
2621         if (!IS_ERR(inode)) {
2622                 inode->i_op = &ext4_file_inode_operations;
2623                 inode->i_fop = &ext4_file_operations;
2624                 ext4_set_aops(inode);
2625                 err = ext4_add_nondir(handle, dentry, &inode);
2626                 if (!err)
2627                         ext4_fc_track_create(handle, dentry);
2628         }
2629         if (handle)
2630                 ext4_journal_stop(handle);
2631         if (!IS_ERR_OR_NULL(inode))
2632                 iput(inode);
2633         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2634                 goto retry;
2635         return err;
2636 }
2637
2638 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2639                       umode_t mode, dev_t rdev)
2640 {
2641         handle_t *handle;
2642         struct inode *inode;
2643         int err, credits, retries = 0;
2644
2645         err = dquot_initialize(dir);
2646         if (err)
2647                 return err;
2648
2649         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2650                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2651 retry:
2652         inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2653                                             NULL, EXT4_HT_DIR, credits);
2654         handle = ext4_journal_current_handle();
2655         err = PTR_ERR(inode);
2656         if (!IS_ERR(inode)) {
2657                 init_special_inode(inode, inode->i_mode, rdev);
2658                 inode->i_op = &ext4_special_inode_operations;
2659                 err = ext4_add_nondir(handle, dentry, &inode);
2660                 if (!err)
2661                         ext4_fc_track_create(handle, dentry);
2662         }
2663         if (handle)
2664                 ext4_journal_stop(handle);
2665         if (!IS_ERR_OR_NULL(inode))
2666                 iput(inode);
2667         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2668                 goto retry;
2669         return err;
2670 }
2671
2672 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2673 {
2674         handle_t *handle;
2675         struct inode *inode;
2676         int err, retries = 0;
2677
2678         err = dquot_initialize(dir);
2679         if (err)
2680                 return err;
2681
2682 retry:
2683         inode = ext4_new_inode_start_handle(dir, mode,
2684                                             NULL, 0, NULL,
2685                                             EXT4_HT_DIR,
2686                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2687                           4 + EXT4_XATTR_TRANS_BLOCKS);
2688         handle = ext4_journal_current_handle();
2689         err = PTR_ERR(inode);
2690         if (!IS_ERR(inode)) {
2691                 inode->i_op = &ext4_file_inode_operations;
2692                 inode->i_fop = &ext4_file_operations;
2693                 ext4_set_aops(inode);
2694                 d_tmpfile(dentry, inode);
2695                 err = ext4_orphan_add(handle, inode);
2696                 if (err)
2697                         goto err_unlock_inode;
2698                 mark_inode_dirty(inode);
2699                 unlock_new_inode(inode);
2700         }
2701         if (handle)
2702                 ext4_journal_stop(handle);
2703         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2704                 goto retry;
2705         return err;
2706 err_unlock_inode:
2707         ext4_journal_stop(handle);
2708         unlock_new_inode(inode);
2709         return err;
2710 }
2711
2712 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2713                           struct ext4_dir_entry_2 *de,
2714                           int blocksize, int csum_size,
2715                           unsigned int parent_ino, int dotdot_real_len)
2716 {
2717         de->inode = cpu_to_le32(inode->i_ino);
2718         de->name_len = 1;
2719         de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2720                                            blocksize);
2721         strcpy(de->name, ".");
2722         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2723
2724         de = ext4_next_entry(de, blocksize);
2725         de->inode = cpu_to_le32(parent_ino);
2726         de->name_len = 2;
2727         if (!dotdot_real_len)
2728                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2729                                         (csum_size + EXT4_DIR_REC_LEN(1)),
2730                                         blocksize);
2731         else
2732                 de->rec_len = ext4_rec_len_to_disk(
2733                                 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2734         strcpy(de->name, "..");
2735         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2736
2737         return ext4_next_entry(de, blocksize);
2738 }
2739
2740 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2741                              struct inode *inode)
2742 {
2743         struct buffer_head *dir_block = NULL;
2744         struct ext4_dir_entry_2 *de;
2745         ext4_lblk_t block = 0;
2746         unsigned int blocksize = dir->i_sb->s_blocksize;
2747         int csum_size = 0;
2748         int err;
2749
2750         if (ext4_has_metadata_csum(dir->i_sb))
2751                 csum_size = sizeof(struct ext4_dir_entry_tail);
2752
2753         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2754                 err = ext4_try_create_inline_dir(handle, dir, inode);
2755                 if (err < 0 && err != -ENOSPC)
2756                         goto out;
2757                 if (!err)
2758                         goto out;
2759         }
2760
2761         inode->i_size = 0;
2762         dir_block = ext4_append(handle, inode, &block);
2763         if (IS_ERR(dir_block))
2764                 return PTR_ERR(dir_block);
2765         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2766         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2767         set_nlink(inode, 2);
2768         if (csum_size)
2769                 ext4_initialize_dirent_tail(dir_block, blocksize);
2770
2771         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2772         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2773         if (err)
2774                 goto out;
2775         set_buffer_verified(dir_block);
2776 out:
2777         brelse(dir_block);
2778         return err;
2779 }
2780
2781 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2782 {
2783         handle_t *handle;
2784         struct inode *inode;
2785         int err, err2 = 0, credits, retries = 0;
2786
2787         if (EXT4_DIR_LINK_MAX(dir))
2788                 return -EMLINK;
2789
2790         err = dquot_initialize(dir);
2791         if (err)
2792                 return err;
2793
2794         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2795                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2796 retry:
2797         inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2798                                             &dentry->d_name,
2799                                             0, NULL, EXT4_HT_DIR, credits);
2800         handle = ext4_journal_current_handle();
2801         err = PTR_ERR(inode);
2802         if (IS_ERR(inode))
2803                 goto out_stop;
2804
2805         inode->i_op = &ext4_dir_inode_operations;
2806         inode->i_fop = &ext4_dir_operations;
2807         err = ext4_init_new_dir(handle, dir, inode);
2808         if (err)
2809                 goto out_clear_inode;
2810         err = ext4_mark_inode_dirty(handle, inode);
2811         if (!err)
2812                 err = ext4_add_entry(handle, dentry, inode);
2813         if (err) {
2814 out_clear_inode:
2815                 clear_nlink(inode);
2816                 ext4_orphan_add(handle, inode);
2817                 unlock_new_inode(inode);
2818                 err2 = ext4_mark_inode_dirty(handle, inode);
2819                 if (unlikely(err2))
2820                         err = err2;
2821                 ext4_journal_stop(handle);
2822                 iput(inode);
2823                 goto out_retry;
2824         }
2825         ext4_inc_count(dir);
2826
2827         ext4_update_dx_flag(dir);
2828         err = ext4_mark_inode_dirty(handle, dir);
2829         if (err)
2830                 goto out_clear_inode;
2831         d_instantiate_new(dentry, inode);
2832         ext4_fc_track_create(handle, dentry);
2833         if (IS_DIRSYNC(dir))
2834                 ext4_handle_sync(handle);
2835
2836 out_stop:
2837         if (handle)
2838                 ext4_journal_stop(handle);
2839 out_retry:
2840         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2841                 goto retry;
2842         return err;
2843 }
2844
2845 /*
2846  * routine to check that the specified directory is empty (for rmdir)
2847  */
2848 bool ext4_empty_dir(struct inode *inode)
2849 {
2850         unsigned int offset;
2851         struct buffer_head *bh;
2852         struct ext4_dir_entry_2 *de;
2853         struct super_block *sb;
2854
2855         if (ext4_has_inline_data(inode)) {
2856                 int has_inline_data = 1;
2857                 int ret;
2858
2859                 ret = empty_inline_dir(inode, &has_inline_data);
2860                 if (has_inline_data)
2861                         return ret;
2862         }
2863
2864         sb = inode->i_sb;
2865         if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2866                 EXT4_ERROR_INODE(inode, "invalid size");
2867                 return true;
2868         }
2869         /* The first directory block must not be a hole,
2870          * so treat it as DIRENT_HTREE
2871          */
2872         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2873         if (IS_ERR(bh))
2874                 return true;
2875
2876         de = (struct ext4_dir_entry_2 *) bh->b_data;
2877         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2878                                  0) ||
2879             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2880                 ext4_warning_inode(inode, "directory missing '.'");
2881                 brelse(bh);
2882                 return true;
2883         }
2884         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2885         de = ext4_next_entry(de, sb->s_blocksize);
2886         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2887                                  offset) ||
2888             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2889                 ext4_warning_inode(inode, "directory missing '..'");
2890                 brelse(bh);
2891                 return true;
2892         }
2893         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2894         while (offset < inode->i_size) {
2895                 if (!(offset & (sb->s_blocksize - 1))) {
2896                         unsigned int lblock;
2897                         brelse(bh);
2898                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2899                         bh = ext4_read_dirblock(inode, lblock, EITHER);
2900                         if (bh == NULL) {
2901                                 offset += sb->s_blocksize;
2902                                 continue;
2903                         }
2904                         if (IS_ERR(bh))
2905                                 return true;
2906                 }
2907                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2908                                         (offset & (sb->s_blocksize - 1)));
2909                 if (ext4_check_dir_entry(inode, NULL, de, bh,
2910                                          bh->b_data, bh->b_size, offset)) {
2911                         offset = (offset | (sb->s_blocksize - 1)) + 1;
2912                         continue;
2913                 }
2914                 if (le32_to_cpu(de->inode)) {
2915                         brelse(bh);
2916                         return false;
2917                 }
2918                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2919         }
2920         brelse(bh);
2921         return true;
2922 }
2923
2924 /*
2925  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2926  * such inodes, starting at the superblock, in case we crash before the
2927  * file is closed/deleted, or in case the inode truncate spans multiple
2928  * transactions and the last transaction is not recovered after a crash.
2929  *
2930  * At filesystem recovery time, we walk this list deleting unlinked
2931  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2932  *
2933  * Orphan list manipulation functions must be called under i_mutex unless
2934  * we are just creating the inode or deleting it.
2935  */
2936 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2937 {
2938         struct super_block *sb = inode->i_sb;
2939         struct ext4_sb_info *sbi = EXT4_SB(sb);
2940         struct ext4_iloc iloc;
2941         int err = 0, rc;
2942         bool dirty = false;
2943
2944         if (!sbi->s_journal || is_bad_inode(inode))
2945                 return 0;
2946
2947         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2948                      !inode_is_locked(inode));
2949         /*
2950          * Exit early if inode already is on orphan list. This is a big speedup
2951          * since we don't have to contend on the global s_orphan_lock.
2952          */
2953         if (!list_empty(&EXT4_I(inode)->i_orphan))
2954                 return 0;
2955
2956         /*
2957          * Orphan handling is only valid for files with data blocks
2958          * being truncated, or files being unlinked. Note that we either
2959          * hold i_mutex, or the inode can not be referenced from outside,
2960          * so i_nlink should not be bumped due to race
2961          */
2962         J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2963                   S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2964
2965         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2966         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2967         if (err)
2968                 goto out;
2969
2970         err = ext4_reserve_inode_write(handle, inode, &iloc);
2971         if (err)
2972                 goto out;
2973
2974         mutex_lock(&sbi->s_orphan_lock);
2975         /*
2976          * Due to previous errors inode may be already a part of on-disk
2977          * orphan list. If so skip on-disk list modification.
2978          */
2979         if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2980             (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2981                 /* Insert this inode at the head of the on-disk orphan list */
2982                 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2983                 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2984                 dirty = true;
2985         }
2986         list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2987         mutex_unlock(&sbi->s_orphan_lock);
2988
2989         if (dirty) {
2990                 err = ext4_handle_dirty_super(handle, sb);
2991                 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2992                 if (!err)
2993                         err = rc;
2994                 if (err) {
2995                         /*
2996                          * We have to remove inode from in-memory list if
2997                          * addition to on disk orphan list failed. Stray orphan
2998                          * list entries can cause panics at unmount time.
2999                          */
3000                         mutex_lock(&sbi->s_orphan_lock);
3001                         list_del_init(&EXT4_I(inode)->i_orphan);
3002                         mutex_unlock(&sbi->s_orphan_lock);
3003                 }
3004         } else
3005                 brelse(iloc.bh);
3006
3007         jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3008         jbd_debug(4, "orphan inode %lu will point to %d\n",
3009                         inode->i_ino, NEXT_ORPHAN(inode));
3010 out:
3011         ext4_std_error(sb, err);
3012         return err;
3013 }
3014
3015 /*
3016  * ext4_orphan_del() removes an unlinked or truncated inode from the list
3017  * of such inodes stored on disk, because it is finally being cleaned up.
3018  */
3019 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3020 {
3021         struct list_head *prev;
3022         struct ext4_inode_info *ei = EXT4_I(inode);
3023         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3024         __u32 ino_next;
3025         struct ext4_iloc iloc;
3026         int err = 0;
3027
3028         if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3029                 return 0;
3030
3031         WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3032                      !inode_is_locked(inode));
3033         /* Do this quick check before taking global s_orphan_lock. */
3034         if (list_empty(&ei->i_orphan))
3035                 return 0;
3036
3037         if (handle) {
3038                 /* Grab inode buffer early before taking global s_orphan_lock */
3039                 err = ext4_reserve_inode_write(handle, inode, &iloc);
3040         }
3041
3042         mutex_lock(&sbi->s_orphan_lock);
3043         jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3044
3045         prev = ei->i_orphan.prev;
3046         list_del_init(&ei->i_orphan);
3047
3048         /* If we're on an error path, we may not have a valid
3049          * transaction handle with which to update the orphan list on
3050          * disk, but we still need to remove the inode from the linked
3051          * list in memory. */
3052         if (!handle || err) {
3053                 mutex_unlock(&sbi->s_orphan_lock);
3054                 goto out_err;
3055         }
3056
3057         ino_next = NEXT_ORPHAN(inode);
3058         if (prev == &sbi->s_orphan) {
3059                 jbd_debug(4, "superblock will point to %u\n", ino_next);
3060                 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3061                 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3062                 if (err) {
3063                         mutex_unlock(&sbi->s_orphan_lock);
3064                         goto out_brelse;
3065                 }
3066                 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3067                 mutex_unlock(&sbi->s_orphan_lock);
3068                 err = ext4_handle_dirty_super(handle, inode->i_sb);
3069         } else {
3070                 struct ext4_iloc iloc2;
3071                 struct inode *i_prev =
3072                         &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3073
3074                 jbd_debug(4, "orphan inode %lu will point to %u\n",
3075                           i_prev->i_ino, ino_next);
3076                 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3077                 if (err) {
3078                         mutex_unlock(&sbi->s_orphan_lock);
3079                         goto out_brelse;
3080                 }
3081                 NEXT_ORPHAN(i_prev) = ino_next;
3082                 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3083                 mutex_unlock(&sbi->s_orphan_lock);
3084         }
3085         if (err)
3086                 goto out_brelse;
3087         NEXT_ORPHAN(inode) = 0;
3088         err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3089 out_err:
3090         ext4_std_error(inode->i_sb, err);
3091         return err;
3092
3093 out_brelse:
3094         brelse(iloc.bh);
3095         goto out_err;
3096 }
3097
3098 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3099 {
3100         int retval;
3101         struct inode *inode;
3102         struct buffer_head *bh;
3103         struct ext4_dir_entry_2 *de;
3104         handle_t *handle = NULL;
3105
3106         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3107                 return -EIO;
3108
3109         /* Initialize quotas before so that eventual writes go in
3110          * separate transaction */
3111         retval = dquot_initialize(dir);
3112         if (retval)
3113                 return retval;
3114         retval = dquot_initialize(d_inode(dentry));
3115         if (retval)
3116                 return retval;
3117
3118         retval = -ENOENT;
3119         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3120         if (IS_ERR(bh))
3121                 return PTR_ERR(bh);
3122         if (!bh)
3123                 goto end_rmdir;
3124
3125         inode = d_inode(dentry);
3126
3127         retval = -EFSCORRUPTED;
3128         if (le32_to_cpu(de->inode) != inode->i_ino)
3129                 goto end_rmdir;
3130
3131         retval = -ENOTEMPTY;
3132         if (!ext4_empty_dir(inode))
3133                 goto end_rmdir;
3134
3135         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3136                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3137         if (IS_ERR(handle)) {
3138                 retval = PTR_ERR(handle);
3139                 handle = NULL;
3140                 goto end_rmdir;
3141         }
3142
3143         if (IS_DIRSYNC(dir))
3144                 ext4_handle_sync(handle);
3145
3146         retval = ext4_delete_entry(handle, dir, de, bh);
3147         if (retval)
3148                 goto end_rmdir;
3149         if (!EXT4_DIR_LINK_EMPTY(inode))
3150                 ext4_warning_inode(inode,
3151                              "empty directory '%.*s' has too many links (%u)",
3152                              dentry->d_name.len, dentry->d_name.name,
3153                              inode->i_nlink);
3154         inode_inc_iversion(inode);
3155         clear_nlink(inode);
3156         /* There's no need to set i_disksize: the fact that i_nlink is
3157          * zero will ensure that the right thing happens during any
3158          * recovery. */
3159         inode->i_size = 0;
3160         ext4_orphan_add(handle, inode);
3161         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3162         retval = ext4_mark_inode_dirty(handle, inode);
3163         if (retval)
3164                 goto end_rmdir;
3165         ext4_dec_count(dir);
3166         ext4_update_dx_flag(dir);
3167         ext4_fc_track_unlink(handle, dentry);
3168         retval = ext4_mark_inode_dirty(handle, dir);
3169
3170 #ifdef CONFIG_UNICODE
3171         /* VFS negative dentries are incompatible with Encoding and
3172          * Case-insensitiveness. Eventually we'll want avoid
3173          * invalidating the dentries here, alongside with returning the
3174          * negative dentries at ext4_lookup(), when it is better
3175          * supported by the VFS for the CI case.
3176          */
3177         if (IS_CASEFOLDED(dir))
3178                 d_invalidate(dentry);
3179 #endif
3180
3181 end_rmdir:
3182         brelse(bh);
3183         if (handle)
3184                 ext4_journal_stop(handle);
3185         return retval;
3186 }
3187
3188 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3189                   struct inode *inode)
3190 {
3191         int retval = -ENOENT;
3192         struct buffer_head *bh;
3193         struct ext4_dir_entry_2 *de;
3194         int skip_remove_dentry = 0;
3195
3196         bh = ext4_find_entry(dir, d_name, &de, NULL);
3197         if (IS_ERR(bh))
3198                 return PTR_ERR(bh);
3199
3200         if (!bh)
3201                 return -ENOENT;
3202
3203         if (le32_to_cpu(de->inode) != inode->i_ino) {
3204                 /*
3205                  * It's okay if we find dont find dentry which matches
3206                  * the inode. That's because it might have gotten
3207                  * renamed to a different inode number
3208                  */
3209                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3210                         skip_remove_dentry = 1;
3211                 else
3212                         goto out;
3213         }
3214
3215         if (IS_DIRSYNC(dir))
3216                 ext4_handle_sync(handle);
3217
3218         if (!skip_remove_dentry) {
3219                 retval = ext4_delete_entry(handle, dir, de, bh);
3220                 if (retval)
3221                         goto out;
3222                 dir->i_ctime = dir->i_mtime = current_time(dir);
3223                 ext4_update_dx_flag(dir);
3224                 retval = ext4_mark_inode_dirty(handle, dir);
3225                 if (retval)
3226                         goto out;
3227         } else {
3228                 retval = 0;
3229         }
3230         if (inode->i_nlink == 0)
3231                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3232                                    d_name->len, d_name->name);
3233         else
3234                 drop_nlink(inode);
3235         if (!inode->i_nlink)
3236                 ext4_orphan_add(handle, inode);
3237         inode->i_ctime = current_time(inode);
3238         retval = ext4_mark_inode_dirty(handle, inode);
3239
3240 out:
3241         brelse(bh);
3242         return retval;
3243 }
3244
3245 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3246 {
3247         handle_t *handle;
3248         int retval;
3249
3250         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3251                 return -EIO;
3252
3253         trace_ext4_unlink_enter(dir, dentry);
3254         /*
3255          * Initialize quotas before so that eventual writes go
3256          * in separate transaction
3257          */
3258         retval = dquot_initialize(dir);
3259         if (retval)
3260                 goto out_trace;
3261         retval = dquot_initialize(d_inode(dentry));
3262         if (retval)
3263                 goto out_trace;
3264
3265         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3266                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3267         if (IS_ERR(handle)) {
3268                 retval = PTR_ERR(handle);
3269                 goto out_trace;
3270         }
3271
3272         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3273         if (!retval)
3274                 ext4_fc_track_unlink(handle, dentry);
3275 #ifdef CONFIG_UNICODE
3276         /* VFS negative dentries are incompatible with Encoding and
3277          * Case-insensitiveness. Eventually we'll want avoid
3278          * invalidating the dentries here, alongside with returning the
3279          * negative dentries at ext4_lookup(), when it is  better
3280          * supported by the VFS for the CI case.
3281          */
3282         if (IS_CASEFOLDED(dir))
3283                 d_invalidate(dentry);
3284 #endif
3285         if (handle)
3286                 ext4_journal_stop(handle);
3287
3288 out_trace:
3289         trace_ext4_unlink_exit(dentry, retval);
3290         return retval;
3291 }
3292
3293 static int ext4_symlink(struct inode *dir,
3294                         struct dentry *dentry, const char *symname)
3295 {
3296         handle_t *handle;
3297         struct inode *inode;
3298         int err, len = strlen(symname);
3299         int credits;
3300         struct fscrypt_str disk_link;
3301
3302         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3303                 return -EIO;
3304
3305         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3306                                       &disk_link);
3307         if (err)
3308                 return err;
3309
3310         err = dquot_initialize(dir);
3311         if (err)
3312                 return err;
3313
3314         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3315                 /*
3316                  * For non-fast symlinks, we just allocate inode and put it on
3317                  * orphan list in the first transaction => we need bitmap,
3318                  * group descriptor, sb, inode block, quota blocks, and
3319                  * possibly selinux xattr blocks.
3320                  */
3321                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3322                           EXT4_XATTR_TRANS_BLOCKS;
3323         } else {
3324                 /*
3325                  * Fast symlink. We have to add entry to directory
3326                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3327                  * allocate new inode (bitmap, group descriptor, inode block,
3328                  * quota blocks, sb is already counted in previous macros).
3329                  */
3330                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3331                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3332         }
3333
3334         inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3335                                             &dentry->d_name, 0, NULL,
3336                                             EXT4_HT_DIR, credits);
3337         handle = ext4_journal_current_handle();
3338         if (IS_ERR(inode)) {
3339                 if (handle)
3340                         ext4_journal_stop(handle);
3341                 return PTR_ERR(inode);
3342         }
3343
3344         if (IS_ENCRYPTED(inode)) {
3345                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3346                 if (err)
3347                         goto err_drop_inode;
3348                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3349         }
3350
3351         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3352                 if (!IS_ENCRYPTED(inode))
3353                         inode->i_op = &ext4_symlink_inode_operations;
3354                 inode_nohighmem(inode);
3355                 ext4_set_aops(inode);
3356                 /*
3357                  * We cannot call page_symlink() with transaction started
3358                  * because it calls into ext4_write_begin() which can wait
3359                  * for transaction commit if we are running out of space
3360                  * and thus we deadlock. So we have to stop transaction now
3361                  * and restart it when symlink contents is written.
3362                  *
3363                  * To keep fs consistent in case of crash, we have to put inode
3364                  * to orphan list in the mean time.
3365                  */
3366                 drop_nlink(inode);
3367                 err = ext4_orphan_add(handle, inode);
3368                 if (handle)
3369                         ext4_journal_stop(handle);
3370                 handle = NULL;
3371                 if (err)
3372                         goto err_drop_inode;
3373                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3374                 if (err)
3375                         goto err_drop_inode;
3376                 /*
3377                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3378                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3379                  */
3380                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3381                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3382                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3383                 if (IS_ERR(handle)) {
3384                         err = PTR_ERR(handle);
3385                         handle = NULL;
3386                         goto err_drop_inode;
3387                 }
3388                 set_nlink(inode, 1);
3389                 err = ext4_orphan_del(handle, inode);
3390                 if (err)
3391                         goto err_drop_inode;
3392         } else {
3393                 /* clear the extent format for fast symlink */
3394                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3395                 if (!IS_ENCRYPTED(inode)) {
3396                         inode->i_op = &ext4_fast_symlink_inode_operations;
3397                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3398                 }
3399                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3400                        disk_link.len);
3401                 inode->i_size = disk_link.len - 1;
3402         }
3403         EXT4_I(inode)->i_disksize = inode->i_size;
3404         err = ext4_add_nondir(handle, dentry, &inode);
3405         if (handle)
3406                 ext4_journal_stop(handle);
3407         if (inode)
3408                 iput(inode);
3409         goto out_free_encrypted_link;
3410
3411 err_drop_inode:
3412         if (handle)
3413                 ext4_journal_stop(handle);
3414         clear_nlink(inode);
3415         unlock_new_inode(inode);
3416         iput(inode);
3417 out_free_encrypted_link:
3418         if (disk_link.name != (unsigned char *)symname)
3419                 kfree(disk_link.name);
3420         return err;
3421 }
3422
3423 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3424 {
3425         handle_t *handle;
3426         int err, retries = 0;
3427 retry:
3428         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3429                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3430                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3431         if (IS_ERR(handle))
3432                 return PTR_ERR(handle);
3433
3434         if (IS_DIRSYNC(dir))
3435                 ext4_handle_sync(handle);
3436
3437         inode->i_ctime = current_time(inode);
3438         ext4_inc_count(inode);
3439         ihold(inode);
3440
3441         err = ext4_add_entry(handle, dentry, inode);
3442         if (!err) {
3443                 err = ext4_mark_inode_dirty(handle, inode);
3444                 /* this can happen only for tmpfile being
3445                  * linked the first time
3446                  */
3447                 if (inode->i_nlink == 1)
3448                         ext4_orphan_del(handle, inode);
3449                 d_instantiate(dentry, inode);
3450                 ext4_fc_track_link(handle, dentry);
3451         } else {
3452                 drop_nlink(inode);
3453                 iput(inode);
3454         }
3455         ext4_journal_stop(handle);
3456         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3457                 goto retry;
3458         return err;
3459 }
3460
3461 static int ext4_link(struct dentry *old_dentry,
3462                      struct inode *dir, struct dentry *dentry)
3463 {
3464         struct inode *inode = d_inode(old_dentry);
3465         int err;
3466
3467         if (inode->i_nlink >= EXT4_LINK_MAX)
3468                 return -EMLINK;
3469
3470         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3471         if (err)
3472                 return err;
3473
3474         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3475             (!projid_eq(EXT4_I(dir)->i_projid,
3476                         EXT4_I(old_dentry->d_inode)->i_projid)))
3477                 return -EXDEV;
3478
3479         err = dquot_initialize(dir);
3480         if (err)
3481                 return err;
3482         return __ext4_link(dir, inode, dentry);
3483 }
3484
3485 /*
3486  * Try to find buffer head where contains the parent block.
3487  * It should be the inode block if it is inlined or the 1st block
3488  * if it is a normal dir.
3489  */
3490 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3491                                         struct inode *inode,
3492                                         int *retval,
3493                                         struct ext4_dir_entry_2 **parent_de,
3494                                         int *inlined)
3495 {
3496         struct buffer_head *bh;
3497
3498         if (!ext4_has_inline_data(inode)) {
3499                 /* The first directory block must not be a hole, so
3500                  * treat it as DIRENT_HTREE
3501                  */
3502                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3503                 if (IS_ERR(bh)) {
3504                         *retval = PTR_ERR(bh);
3505                         return NULL;
3506                 }
3507                 *parent_de = ext4_next_entry(
3508                                         (struct ext4_dir_entry_2 *)bh->b_data,
3509                                         inode->i_sb->s_blocksize);
3510                 return bh;
3511         }
3512
3513         *inlined = 1;
3514         return ext4_get_first_inline_block(inode, parent_de, retval);
3515 }
3516
3517 struct ext4_renament {
3518         struct inode *dir;
3519         struct dentry *dentry;
3520         struct inode *inode;
3521         bool is_dir;
3522         int dir_nlink_delta;
3523
3524         /* entry for "dentry" */
3525         struct buffer_head *bh;
3526         struct ext4_dir_entry_2 *de;
3527         int inlined;
3528
3529         /* entry for ".." in inode if it's a directory */
3530         struct buffer_head *dir_bh;
3531         struct ext4_dir_entry_2 *parent_de;
3532         int dir_inlined;
3533 };
3534
3535 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3536 {
3537         int retval;
3538
3539         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3540                                               &retval, &ent->parent_de,
3541                                               &ent->dir_inlined);
3542         if (!ent->dir_bh)
3543                 return retval;
3544         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3545                 return -EFSCORRUPTED;
3546         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3547         return ext4_journal_get_write_access(handle, ent->dir_bh);
3548 }
3549
3550 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3551                                   unsigned dir_ino)
3552 {
3553         int retval;
3554
3555         ent->parent_de->inode = cpu_to_le32(dir_ino);
3556         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3557         if (!ent->dir_inlined) {
3558                 if (is_dx(ent->inode)) {
3559                         retval = ext4_handle_dirty_dx_node(handle,
3560                                                            ent->inode,
3561                                                            ent->dir_bh);
3562                 } else {
3563                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3564                                                             ent->dir_bh);
3565                 }
3566         } else {
3567                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3568         }
3569         if (retval) {
3570                 ext4_std_error(ent->dir->i_sb, retval);
3571                 return retval;
3572         }
3573         return 0;
3574 }
3575
3576 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3577                        unsigned ino, unsigned file_type)
3578 {
3579         int retval, retval2;
3580
3581         BUFFER_TRACE(ent->bh, "get write access");
3582         retval = ext4_journal_get_write_access(handle, ent->bh);
3583         if (retval)
3584                 return retval;
3585         ent->de->inode = cpu_to_le32(ino);
3586         if (ext4_has_feature_filetype(ent->dir->i_sb))
3587                 ent->de->file_type = file_type;
3588         inode_inc_iversion(ent->dir);
3589         ent->dir->i_ctime = ent->dir->i_mtime =
3590                 current_time(ent->dir);
3591         retval = ext4_mark_inode_dirty(handle, ent->dir);
3592         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3593         if (!ent->inlined) {
3594                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3595                 if (unlikely(retval2)) {
3596                         ext4_std_error(ent->dir->i_sb, retval2);
3597                         return retval2;
3598                 }
3599         }
3600         brelse(ent->bh);
3601         ent->bh = NULL;
3602
3603         return retval;
3604 }
3605
3606 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3607                                   const struct qstr *d_name)
3608 {
3609         int retval = -ENOENT;
3610         struct buffer_head *bh;
3611         struct ext4_dir_entry_2 *de;
3612
3613         bh = ext4_find_entry(dir, d_name, &de, NULL);
3614         if (IS_ERR(bh))
3615                 return PTR_ERR(bh);
3616         if (bh) {
3617                 retval = ext4_delete_entry(handle, dir, de, bh);
3618                 brelse(bh);
3619         }
3620         return retval;
3621 }
3622
3623 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3624                                int force_reread)
3625 {
3626         int retval;
3627         /*
3628          * ent->de could have moved from under us during htree split, so make
3629          * sure that we are deleting the right entry.  We might also be pointing
3630          * to a stale entry in the unused part of ent->bh so just checking inum
3631          * and the name isn't enough.
3632          */
3633         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3634             ent->de->name_len != ent->dentry->d_name.len ||
3635             strncmp(ent->de->name, ent->dentry->d_name.name,
3636                     ent->de->name_len) ||
3637             force_reread) {
3638                 retval = ext4_find_delete_entry(handle, ent->dir,
3639                                                 &ent->dentry->d_name);
3640         } else {
3641                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3642                 if (retval == -ENOENT) {
3643                         retval = ext4_find_delete_entry(handle, ent->dir,
3644                                                         &ent->dentry->d_name);
3645                 }
3646         }
3647
3648         if (retval) {
3649                 ext4_warning_inode(ent->dir,
3650                                    "Deleting old file: nlink %d, error=%d",
3651                                    ent->dir->i_nlink, retval);
3652         }
3653 }
3654
3655 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3656 {
3657         if (ent->dir_nlink_delta) {
3658                 if (ent->dir_nlink_delta == -1)
3659                         ext4_dec_count(ent->dir);
3660                 else
3661                         ext4_inc_count(ent->dir);
3662                 ext4_mark_inode_dirty(handle, ent->dir);
3663         }
3664 }
3665
3666 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3667                                               int credits, handle_t **h)
3668 {
3669         struct inode *wh;
3670         handle_t *handle;
3671         int retries = 0;
3672
3673         /*
3674          * for inode block, sb block, group summaries,
3675          * and inode bitmap
3676          */
3677         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3678                     EXT4_XATTR_TRANS_BLOCKS + 4);
3679 retry:
3680         wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3681                                          &ent->dentry->d_name, 0, NULL,
3682                                          EXT4_HT_DIR, credits);
3683
3684         handle = ext4_journal_current_handle();
3685         if (IS_ERR(wh)) {
3686                 if (handle)
3687                         ext4_journal_stop(handle);
3688                 if (PTR_ERR(wh) == -ENOSPC &&
3689                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3690                         goto retry;
3691         } else {
3692                 *h = handle;
3693                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3694                 wh->i_op = &ext4_special_inode_operations;
3695         }
3696         return wh;
3697 }
3698
3699 /*
3700  * Anybody can rename anything with this: the permission checks are left to the
3701  * higher-level routines.
3702  *
3703  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3704  * while new_{dentry,inode) refers to the destination dentry/inode
3705  * This comes from rename(const char *oldpath, const char *newpath)
3706  */
3707 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3708                        struct inode *new_dir, struct dentry *new_dentry,
3709                        unsigned int flags)
3710 {
3711         handle_t *handle = NULL;
3712         struct ext4_renament old = {
3713                 .dir = old_dir,
3714                 .dentry = old_dentry,
3715                 .inode = d_inode(old_dentry),
3716         };
3717         struct ext4_renament new = {
3718                 .dir = new_dir,
3719                 .dentry = new_dentry,
3720                 .inode = d_inode(new_dentry),
3721         };
3722         int force_reread;
3723         int retval;
3724         struct inode *whiteout = NULL;
3725         int credits;
3726         u8 old_file_type;
3727
3728         if (new.inode && new.inode->i_nlink == 0) {
3729                 EXT4_ERROR_INODE(new.inode,
3730                                  "target of rename is already freed");
3731                 return -EFSCORRUPTED;
3732         }
3733
3734         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3735             (!projid_eq(EXT4_I(new_dir)->i_projid,
3736                         EXT4_I(old_dentry->d_inode)->i_projid)))
3737                 return -EXDEV;
3738
3739         retval = dquot_initialize(old.dir);
3740         if (retval)
3741                 return retval;
3742         retval = dquot_initialize(new.dir);
3743         if (retval)
3744                 return retval;
3745
3746         /* Initialize quotas before so that eventual writes go
3747          * in separate transaction */
3748         if (new.inode) {
3749                 retval = dquot_initialize(new.inode);
3750                 if (retval)
3751                         return retval;
3752         }
3753
3754         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3755         if (IS_ERR(old.bh))
3756                 return PTR_ERR(old.bh);
3757         /*
3758          *  Check for inode number is _not_ due to possible IO errors.
3759          *  We might rmdir the source, keep it as pwd of some process
3760          *  and merrily kill the link to whatever was created under the
3761          *  same name. Goodbye sticky bit ;-<
3762          */
3763         retval = -ENOENT;
3764         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3765                 goto end_rename;
3766
3767         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3768                                  &new.de, &new.inlined);
3769         if (IS_ERR(new.bh)) {
3770                 retval = PTR_ERR(new.bh);
3771                 new.bh = NULL;
3772                 goto end_rename;
3773         }
3774         if (new.bh) {
3775                 if (!new.inode) {
3776                         brelse(new.bh);
3777                         new.bh = NULL;
3778                 }
3779         }
3780         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3781                 ext4_alloc_da_blocks(old.inode);
3782
3783         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3784                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3785         if (!(flags & RENAME_WHITEOUT)) {
3786                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3787                 if (IS_ERR(handle)) {
3788                         retval = PTR_ERR(handle);
3789                         handle = NULL;
3790                         goto end_rename;
3791                 }
3792         } else {
3793                 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3794                 if (IS_ERR(whiteout)) {
3795                         retval = PTR_ERR(whiteout);
3796                         whiteout = NULL;
3797                         goto end_rename;
3798                 }
3799         }
3800
3801         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3802                 ext4_handle_sync(handle);
3803
3804         if (S_ISDIR(old.inode->i_mode)) {
3805                 if (new.inode) {
3806                         retval = -ENOTEMPTY;
3807                         if (!ext4_empty_dir(new.inode))
3808                                 goto end_rename;
3809                 } else {
3810                         retval = -EMLINK;
3811                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3812                                 goto end_rename;
3813                 }
3814                 retval = ext4_rename_dir_prepare(handle, &old);
3815                 if (retval)
3816                         goto end_rename;
3817         }
3818         /*
3819          * If we're renaming a file within an inline_data dir and adding or
3820          * setting the new dirent causes a conversion from inline_data to
3821          * extents/blockmap, we need to force the dirent delete code to
3822          * re-read the directory, or else we end up trying to delete a dirent
3823          * from what is now the extent tree root (or a block map).
3824          */
3825         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3826                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3827
3828         old_file_type = old.de->file_type;
3829         if (whiteout) {
3830                 /*
3831                  * Do this before adding a new entry, so the old entry is sure
3832                  * to be still pointing to the valid old entry.
3833                  */
3834                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3835                                      EXT4_FT_CHRDEV);
3836                 if (retval)
3837                         goto end_rename;
3838                 retval = ext4_mark_inode_dirty(handle, whiteout);
3839                 if (unlikely(retval))
3840                         goto end_rename;
3841         }
3842         if (!new.bh) {
3843                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3844                 if (retval)
3845                         goto end_rename;
3846         } else {
3847                 retval = ext4_setent(handle, &new,
3848                                      old.inode->i_ino, old_file_type);
3849                 if (retval)
3850                         goto end_rename;
3851         }
3852         if (force_reread)
3853                 force_reread = !ext4_test_inode_flag(new.dir,
3854                                                      EXT4_INODE_INLINE_DATA);
3855
3856         /*
3857          * Like most other Unix systems, set the ctime for inodes on a
3858          * rename.
3859          */
3860         old.inode->i_ctime = current_time(old.inode);
3861         retval = ext4_mark_inode_dirty(handle, old.inode);
3862         if (unlikely(retval))
3863                 goto end_rename;
3864
3865         if (!whiteout) {
3866                 /*
3867                  * ok, that's it
3868                  */
3869                 ext4_rename_delete(handle, &old, force_reread);
3870         }
3871
3872         if (new.inode) {
3873                 ext4_dec_count(new.inode);
3874                 new.inode->i_ctime = current_time(new.inode);
3875         }
3876         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3877         ext4_update_dx_flag(old.dir);
3878         if (old.dir_bh) {
3879                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3880                 if (retval)
3881                         goto end_rename;
3882
3883                 ext4_dec_count(old.dir);
3884                 if (new.inode) {
3885                         /* checked ext4_empty_dir above, can't have another
3886                          * parent, ext4_dec_count() won't work for many-linked
3887                          * dirs */
3888                         clear_nlink(new.inode);
3889                 } else {
3890                         ext4_inc_count(new.dir);
3891                         ext4_update_dx_flag(new.dir);
3892                         retval = ext4_mark_inode_dirty(handle, new.dir);
3893                         if (unlikely(retval))
3894                                 goto end_rename;
3895                 }
3896         }
3897         retval = ext4_mark_inode_dirty(handle, old.dir);
3898         if (unlikely(retval))
3899                 goto end_rename;
3900
3901         if (S_ISDIR(old.inode->i_mode)) {
3902                 /*
3903                  * We disable fast commits here that's because the
3904                  * replay code is not yet capable of changing dot dot
3905                  * dirents in directories.
3906                  */
3907                 ext4_fc_mark_ineligible(old.inode->i_sb,
3908                         EXT4_FC_REASON_RENAME_DIR);
3909         } else {
3910                 if (new.inode)
3911                         ext4_fc_track_unlink(handle, new.dentry);
3912                 __ext4_fc_track_link(handle, old.inode, new.dentry);
3913                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3914         }
3915
3916         if (new.inode) {
3917                 retval = ext4_mark_inode_dirty(handle, new.inode);
3918                 if (unlikely(retval))
3919                         goto end_rename;
3920                 if (!new.inode->i_nlink)
3921                         ext4_orphan_add(handle, new.inode);
3922         }
3923         retval = 0;
3924
3925 end_rename:
3926         brelse(old.dir_bh);
3927         brelse(old.bh);
3928         brelse(new.bh);
3929         if (whiteout) {
3930                 if (retval)
3931                         drop_nlink(whiteout);
3932                 unlock_new_inode(whiteout);
3933                 iput(whiteout);
3934         }
3935         if (handle)
3936                 ext4_journal_stop(handle);
3937         return retval;
3938 }
3939
3940 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3941                              struct inode *new_dir, struct dentry *new_dentry)
3942 {
3943         handle_t *handle = NULL;
3944         struct ext4_renament old = {
3945                 .dir = old_dir,
3946                 .dentry = old_dentry,
3947                 .inode = d_inode(old_dentry),
3948         };
3949         struct ext4_renament new = {
3950                 .dir = new_dir,
3951                 .dentry = new_dentry,
3952                 .inode = d_inode(new_dentry),
3953         };
3954         u8 new_file_type;
3955         int retval;
3956         struct timespec64 ctime;
3957
3958         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3959              !projid_eq(EXT4_I(new_dir)->i_projid,
3960                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3961             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3962              !projid_eq(EXT4_I(old_dir)->i_projid,
3963                         EXT4_I(new_dentry->d_inode)->i_projid)))
3964                 return -EXDEV;
3965
3966         retval = dquot_initialize(old.dir);
3967         if (retval)
3968                 return retval;
3969         retval = dquot_initialize(new.dir);
3970         if (retval)
3971                 return retval;
3972
3973         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3974                                  &old.de, &old.inlined);
3975         if (IS_ERR(old.bh))
3976                 return PTR_ERR(old.bh);
3977         /*
3978          *  Check for inode number is _not_ due to possible IO errors.
3979          *  We might rmdir the source, keep it as pwd of some process
3980          *  and merrily kill the link to whatever was created under the
3981          *  same name. Goodbye sticky bit ;-<
3982          */
3983         retval = -ENOENT;
3984         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3985                 goto end_rename;
3986
3987         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3988                                  &new.de, &new.inlined);
3989         if (IS_ERR(new.bh)) {
3990                 retval = PTR_ERR(new.bh);
3991                 new.bh = NULL;
3992                 goto end_rename;
3993         }
3994
3995         /* RENAME_EXCHANGE case: old *and* new must both exist */
3996         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3997                 goto end_rename;
3998
3999         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4000                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4001                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4002         if (IS_ERR(handle)) {
4003                 retval = PTR_ERR(handle);
4004                 handle = NULL;
4005                 goto end_rename;
4006         }
4007
4008         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4009                 ext4_handle_sync(handle);
4010
4011         if (S_ISDIR(old.inode->i_mode)) {
4012                 old.is_dir = true;
4013                 retval = ext4_rename_dir_prepare(handle, &old);
4014                 if (retval)
4015                         goto end_rename;
4016         }
4017         if (S_ISDIR(new.inode->i_mode)) {
4018                 new.is_dir = true;
4019                 retval = ext4_rename_dir_prepare(handle, &new);
4020                 if (retval)
4021                         goto end_rename;
4022         }
4023
4024         /*
4025          * Other than the special case of overwriting a directory, parents'
4026          * nlink only needs to be modified if this is a cross directory rename.
4027          */
4028         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4029                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4030                 new.dir_nlink_delta = -old.dir_nlink_delta;
4031                 retval = -EMLINK;
4032                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4033                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4034                         goto end_rename;
4035         }
4036
4037         new_file_type = new.de->file_type;
4038         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4039         if (retval)
4040                 goto end_rename;
4041
4042         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4043         if (retval)
4044                 goto end_rename;
4045
4046         /*
4047          * Like most other Unix systems, set the ctime for inodes on a
4048          * rename.
4049          */
4050         ctime = current_time(old.inode);
4051         old.inode->i_ctime = ctime;
4052         new.inode->i_ctime = ctime;
4053         retval = ext4_mark_inode_dirty(handle, old.inode);
4054         if (unlikely(retval))
4055                 goto end_rename;
4056         retval = ext4_mark_inode_dirty(handle, new.inode);
4057         if (unlikely(retval))
4058                 goto end_rename;
4059         ext4_fc_mark_ineligible(new.inode->i_sb,
4060                                 EXT4_FC_REASON_CROSS_RENAME);
4061         if (old.dir_bh) {
4062                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4063                 if (retval)
4064                         goto end_rename;
4065         }
4066         if (new.dir_bh) {
4067                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4068                 if (retval)
4069                         goto end_rename;
4070         }
4071         ext4_update_dir_count(handle, &old);
4072         ext4_update_dir_count(handle, &new);
4073         retval = 0;
4074
4075 end_rename:
4076         brelse(old.dir_bh);
4077         brelse(new.dir_bh);
4078         brelse(old.bh);
4079         brelse(new.bh);
4080         if (handle)
4081                 ext4_journal_stop(handle);
4082         return retval;
4083 }
4084
4085 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4086                         struct inode *new_dir, struct dentry *new_dentry,
4087                         unsigned int flags)
4088 {
4089         int err;
4090
4091         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4092                 return -EIO;
4093
4094         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4095                 return -EINVAL;
4096
4097         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4098                                      flags);
4099         if (err)
4100                 return err;
4101
4102         if (flags & RENAME_EXCHANGE) {
4103                 return ext4_cross_rename(old_dir, old_dentry,
4104                                          new_dir, new_dentry);
4105         }
4106
4107         return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4108 }
4109
4110 /*
4111  * directories can handle most operations...
4112  */
4113 const struct inode_operations ext4_dir_inode_operations = {
4114         .create         = ext4_create,
4115         .lookup         = ext4_lookup,
4116         .link           = ext4_link,
4117         .unlink         = ext4_unlink,
4118         .symlink        = ext4_symlink,
4119         .mkdir          = ext4_mkdir,
4120         .rmdir          = ext4_rmdir,
4121         .mknod          = ext4_mknod,
4122         .tmpfile        = ext4_tmpfile,
4123         .rename         = ext4_rename2,
4124         .setattr        = ext4_setattr,
4125         .getattr        = ext4_getattr,
4126         .listxattr      = ext4_listxattr,
4127         .get_acl        = ext4_get_acl,
4128         .set_acl        = ext4_set_acl,
4129         .fiemap         = ext4_fiemap,
4130 };
4131
4132 const struct inode_operations ext4_special_inode_operations = {
4133         .setattr        = ext4_setattr,
4134         .getattr        = ext4_getattr,
4135         .listxattr      = ext4_listxattr,
4136         .get_acl        = ext4_get_acl,
4137         .set_acl        = ext4_set_acl,
4138 };