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, inode->i_sb, bh,
  74                                             EXT4_JTR_NONE);
  75         if (err) {
  76                 brelse(bh);
  77                 ext4_std_error(inode->i_sb, err);
  78                 return ERR_PTR(err);
  79         }
  80         return bh;
  81 }
  82
  83 static int ext4_dx_csum_verify(struct inode *inode,
  84                                struct ext4_dir_entry *dirent);
  85
  86 /*
  87  * Hints to ext4_read_dirblock regarding whether we expect a directory
  88  * block being read to be an index block, or a block containing
  89  * directory entries (and if the latter, whether it was found via a
  90  * logical block in an htree index block).  This is used to control
  91  * what sort of sanity checkinig ext4_read_dirblock() will do on the
  92  * directory block read from the storage device.  EITHER will means
  93  * the caller doesn't know what kind of directory block will be read,
  94  * so no specific verification will be done.
  95  */
  96 typedef enum {
  97         EITHER, INDEX, DIRENT, DIRENT_HTREE
  98 } dirblock_type_t;
  99
 100 #define ext4_read_dirblock(inode, block, type) \
 101         __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
 102
 103 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
 104                                                 ext4_lblk_t block,
 105                                                 dirblock_type_t type,
 106                                                 const char *func,
 107                                                 unsigned int line)
 108 {
 109         struct buffer_head *bh;
 110         struct ext4_dir_entry *dirent;
 111         int is_dx_block = 0;
 112
 113         if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
 114                 bh = ERR_PTR(-EIO);
 115         else
 116                 bh = ext4_bread(NULL, inode, block, 0);
 117         if (IS_ERR(bh)) {
 118                 __ext4_warning(inode->i_sb, func, line,
 119                                "inode #%lu: lblock %lu: comm %s: "
 120                                "error %ld reading directory block",
 121                                inode->i_ino, (unsigned long)block,
 122                                current->comm, PTR_ERR(bh));
 123
 124                 return bh;
 125         }
 126         if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
 127                 ext4_error_inode(inode, func, line, block,
 128                                  "Directory hole found for htree %s block",
 129                                  (type == INDEX) ? "index" : "leaf");
 130                 return ERR_PTR(-EFSCORRUPTED);
 131         }
 132         if (!bh)
 133                 return NULL;
 134         dirent = (struct ext4_dir_entry *) bh->b_data;
 135         /* Determine whether or not we have an index block */
 136         if (is_dx(inode)) {
 137                 if (block == 0)
 138                         is_dx_block = 1;
 139                 else if (ext4_rec_len_from_disk(dirent->rec_len,
 140                                                 inode->i_sb->s_blocksize) ==
 141                          inode->i_sb->s_blocksize)
 142                         is_dx_block = 1;
 143         }
 144         if (!is_dx_block && type == INDEX) {
 145                 ext4_error_inode(inode, func, line, block,
 146                        "directory leaf block found instead of index block");
 147                 brelse(bh);
 148                 return ERR_PTR(-EFSCORRUPTED);
 149         }
 150         if (!ext4_has_metadata_csum(inode->i_sb) ||
 151             buffer_verified(bh))
 152                 return bh;
 153
 154         /*
 155          * An empty leaf block can get mistaken for a index block; for
 156          * this reason, we can only check the index checksum when the
 157          * caller is sure it should be an index block.
 158          */
 159         if (is_dx_block && type == INDEX) {
 160                 if (ext4_dx_csum_verify(inode, dirent) &&
 161                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 162                         set_buffer_verified(bh);
 163                 else {
 164                         ext4_error_inode_err(inode, func, line, block,
 165                                              EFSBADCRC,
 166                                              "Directory index failed checksum");
 167                         brelse(bh);
 168                         return ERR_PTR(-EFSBADCRC);
 169                 }
 170         }
 171         if (!is_dx_block) {
 172                 if (ext4_dirblock_csum_verify(inode, bh) &&
 173                     !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
 174                         set_buffer_verified(bh);
 175                 else {
 176                         ext4_error_inode_err(inode, func, line, block,
 177                                              EFSBADCRC,
 178                                              "Directory block failed checksum");
 179                         brelse(bh);
 180                         return ERR_PTR(-EFSBADCRC);
 181                 }
 182         }
 183         return bh;
 184 }
 185
 186 #ifdef DX_DEBUG
 187 #define dxtrace(command) command
 188 #else
 189 #define dxtrace(command)
 190 #endif
 191
 192 struct fake_dirent
 193 {
 194         __le32 inode;
 195         __le16 rec_len;
 196         u8 name_len;
 197         u8 file_type;
 198 };
 199
 200 struct dx_countlimit
 201 {
 202         __le16 limit;
 203         __le16 count;
 204 };
 205
 206 struct dx_entry
 207 {
 208         __le32 hash;
 209         __le32 block;
 210 };
 211
 212 /*
 213  * dx_root_info is laid out so that if it should somehow get overlaid by a
 214  * dirent the two low bits of the hash version will be zero.  Therefore, the
 215  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 216  */
 217
 218 struct dx_root
 219 {
 220         struct fake_dirent dot;
 221         char dot_name[4];
 222         struct fake_dirent dotdot;
 223         char dotdot_name[4];
 224         struct dx_root_info
 225         {
 226                 __le32 reserved_zero;
 227                 u8 hash_version;
 228                 u8 info_length; /* 8 */
 229                 u8 indirect_levels;
 230                 u8 unused_flags;
 231         }
 232         info;
 233         struct dx_entry entries[];
 234 };
 235
 236 struct dx_node
 237 {
 238         struct fake_dirent fake;
 239         struct dx_entry entries[];
 240 };
 241
 242
 243 struct dx_frame
 244 {
 245         struct buffer_head *bh;
 246         struct dx_entry *entries;
 247         struct dx_entry *at;
 248 };
 249
 250 struct dx_map_entry
 251 {
 252         u32 hash;
 253         u16 offs;
 254         u16 size;
 255 };
 256
 257 /*
 258  * This goes at the end of each htree block.
 259  */
 260 struct dx_tail {
 261         u32 dt_reserved;
 262         __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 263 };
 264
 265 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 266 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 267 static inline unsigned dx_get_hash(struct dx_entry *entry);
 268 static void dx_set_hash(struct dx_entry *entry, unsigned value);
 269 static unsigned dx_get_count(struct dx_entry *entries);
 270 static unsigned dx_get_limit(struct dx_entry *entries);
 271 static void dx_set_count(struct dx_entry *entries, unsigned value);
 272 static void dx_set_limit(struct dx_entry *entries, unsigned value);
 273 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 274 static unsigned dx_node_limit(struct inode *dir);
 275 static struct dx_frame *dx_probe(struct ext4_filename *fname,
 276                                  struct inode *dir,
 277                                  struct dx_hash_info *hinfo,
 278                                  struct dx_frame *frame);
 279 static void dx_release(struct dx_frame *frames);
 280 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 281                        unsigned blocksize, struct dx_hash_info *hinfo,
 282                        struct dx_map_entry map[]);
 283 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 284 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
 285                                         char *to, struct dx_map_entry *offsets,
 286                                         int count, unsigned int blocksize);
 287 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 288                                                 unsigned int blocksize);
 289 static void dx_insert_block(struct dx_frame *frame,
 290                                         u32 hash, ext4_lblk_t block);
 291 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 292                                  struct dx_frame *frame,
 293                                  struct dx_frame *frames,
 294                                  __u32 *start_hash);
 295 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 296                 struct ext4_filename *fname,
 297                 struct ext4_dir_entry_2 **res_dir);
 298 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
 299                              struct inode *dir, struct inode *inode);
 300
 301 /* checksumming functions */
 302 void ext4_initialize_dirent_tail(struct buffer_head *bh,
 303                                  unsigned int blocksize)
 304 {
 305         struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
 306
 307         memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 308         t->det_rec_len = ext4_rec_len_to_disk(
 309                         sizeof(struct ext4_dir_entry_tail), blocksize);
 310         t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 311 }
 312
 313 /* Walk through a dirent block to find a checksum "dirent" at the tail */
 314 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 315                                                    struct buffer_head *bh)
 316 {
 317         struct ext4_dir_entry_tail *t;
 318
 319 #ifdef PARANOID
 320         struct ext4_dir_entry *d, *top;
 321
 322         d = (struct ext4_dir_entry *)bh->b_data;
 323         top = (struct ext4_dir_entry *)(bh->b_data +
 324                 (EXT4_BLOCK_SIZE(inode->i_sb) -
 325                  sizeof(struct ext4_dir_entry_tail)));
 326         while (d < top && d->rec_len)
 327                 d = (struct ext4_dir_entry *)(((void *)d) +
 328                     le16_to_cpu(d->rec_len));
 329
 330         if (d != top)
 331                 return NULL;
 332
 333         t = (struct ext4_dir_entry_tail *)d;
 334 #else
 335         t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
 336 #endif
 337
 338         if (t->det_reserved_zero1 ||
 339             le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 340             t->det_reserved_zero2 ||
 341             t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 342                 return NULL;
 343
 344         return t;
 345 }
 346
 347 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
 348 {
 349         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 350         struct ext4_inode_info *ei = EXT4_I(inode);
 351         __u32 csum;
 352
 353         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 354         return cpu_to_le32(csum);
 355 }
 356
 357 #define warn_no_space_for_csum(inode)                                   \
 358         __warn_no_space_for_csum((inode), __func__, __LINE__)
 359
 360 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
 361                                      unsigned int line)
 362 {
 363         __ext4_warning_inode(inode, func, line,
 364                 "No space for directory leaf checksum. Please run e2fsck -D.");
 365 }
 366
 367 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
 368 {
 369         struct ext4_dir_entry_tail *t;
 370
 371         if (!ext4_has_metadata_csum(inode->i_sb))
 372                 return 1;
 373
 374         t = get_dirent_tail(inode, bh);
 375         if (!t) {
 376                 warn_no_space_for_csum(inode);
 377                 return 0;
 378         }
 379
 380         if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
 381                                                   (char *)t - bh->b_data))
 382                 return 0;
 383
 384         return 1;
 385 }
 386
 387 static void ext4_dirblock_csum_set(struct inode *inode,
 388                                  struct buffer_head *bh)
 389 {
 390         struct ext4_dir_entry_tail *t;
 391
 392         if (!ext4_has_metadata_csum(inode->i_sb))
 393                 return;
 394
 395         t = get_dirent_tail(inode, bh);
 396         if (!t) {
 397                 warn_no_space_for_csum(inode);
 398                 return;
 399         }
 400
 401         t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
 402                                              (char *)t - bh->b_data);
 403 }
 404
 405 int ext4_handle_dirty_dirblock(handle_t *handle,
 406                                struct inode *inode,
 407                                struct buffer_head *bh)
 408 {
 409         ext4_dirblock_csum_set(inode, bh);
 410         return ext4_handle_dirty_metadata(handle, inode, bh);
 411 }
 412
 413 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 414                                                struct ext4_dir_entry *dirent,
 415                                                int *offset)
 416 {
 417         struct ext4_dir_entry *dp;
 418         struct dx_root_info *root;
 419         int count_offset;
 420
 421         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 422                 count_offset = 8;
 423         else if (le16_to_cpu(dirent->rec_len) == 12) {
 424                 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 425                 if (le16_to_cpu(dp->rec_len) !=
 426                     EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 427                         return NULL;
 428                 root = (struct dx_root_info *)(((void *)dp + 12));
 429                 if (root->reserved_zero ||
 430                     root->info_length != sizeof(struct dx_root_info))
 431                         return NULL;
 432                 count_offset = 32;
 433         } else
 434                 return NULL;
 435
 436         if (offset)
 437                 *offset = count_offset;
 438         return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 439 }
 440
 441 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 442                            int count_offset, int count, struct dx_tail *t)
 443 {
 444         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 445         struct ext4_inode_info *ei = EXT4_I(inode);
 446         __u32 csum;
 447         int size;
 448         __u32 dummy_csum = 0;
 449         int offset = offsetof(struct dx_tail, dt_checksum);
 450
 451         size = count_offset + (count * sizeof(struct dx_entry));
 452         csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 453         csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
 454         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 455
 456         return cpu_to_le32(csum);
 457 }
 458
 459 static int ext4_dx_csum_verify(struct inode *inode,
 460                                struct ext4_dir_entry *dirent)
 461 {
 462         struct dx_countlimit *c;
 463         struct dx_tail *t;
 464         int count_offset, limit, count;
 465
 466         if (!ext4_has_metadata_csum(inode->i_sb))
 467                 return 1;
 468
 469         c = get_dx_countlimit(inode, dirent, &count_offset);
 470         if (!c) {
 471                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 472                 return 0;
 473         }
 474         limit = le16_to_cpu(c->limit);
 475         count = le16_to_cpu(c->count);
 476         if (count_offset + (limit * sizeof(struct dx_entry)) >
 477             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 478                 warn_no_space_for_csum(inode);
 479                 return 0;
 480         }
 481         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 482
 483         if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 484                                             count, t))
 485                 return 0;
 486         return 1;
 487 }
 488
 489 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 490 {
 491         struct dx_countlimit *c;
 492         struct dx_tail *t;
 493         int count_offset, limit, count;
 494
 495         if (!ext4_has_metadata_csum(inode->i_sb))
 496                 return;
 497
 498         c = get_dx_countlimit(inode, dirent, &count_offset);
 499         if (!c) {
 500                 EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 501                 return;
 502         }
 503         limit = le16_to_cpu(c->limit);
 504         count = le16_to_cpu(c->count);
 505         if (count_offset + (limit * sizeof(struct dx_entry)) >
 506             EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 507                 warn_no_space_for_csum(inode);
 508                 return;
 509         }
 510         t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 511
 512         t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 513 }
 514
 515 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 516                                             struct inode *inode,
 517                                             struct buffer_head *bh)
 518 {
 519         ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 520         return ext4_handle_dirty_metadata(handle, inode, bh);
 521 }
 522
 523 /*
 524  * p is at least 6 bytes before the end of page
 525  */
 526 static inline struct ext4_dir_entry_2 *
 527 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 528 {
 529         return (struct ext4_dir_entry_2 *)((char *)p +
 530                 ext4_rec_len_from_disk(p->rec_len, blocksize));
 531 }
 532
 533 /*
 534  * Future: use high four bits of block for coalesce-on-delete flags
 535  * Mask them off for now.
 536  */
 537
 538 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 539 {
 540         return le32_to_cpu(entry->block) & 0x0fffffff;
 541 }
 542
 543 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 544 {
 545         entry->block = cpu_to_le32(value);
 546 }
 547
 548 static inline unsigned dx_get_hash(struct dx_entry *entry)
 549 {
 550         return le32_to_cpu(entry->hash);
 551 }
 552
 553 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 554 {
 555         entry->hash = cpu_to_le32(value);
 556 }
 557
 558 static inline unsigned dx_get_count(struct dx_entry *entries)
 559 {
 560         return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 561 }
 562
 563 static inline unsigned dx_get_limit(struct dx_entry *entries)
 564 {
 565         return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 566 }
 567
 568 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 569 {
 570         ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 571 }
 572
 573 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 574 {
 575         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 576 }
 577
 578 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 579 {
 580         unsigned int entry_space = dir->i_sb->s_blocksize -
 581                         ext4_dir_rec_len(1, NULL) -
 582                         ext4_dir_rec_len(2, NULL) - 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 int entry_space = dir->i_sb->s_blocksize -
 592                         ext4_dir_rec_len(0, dir);
 593
 594         if (ext4_has_metadata_csum(dir->i_sb))
 595                 entry_space -= sizeof(struct dx_tail);
 596         return entry_space / sizeof(struct dx_entry);
 597 }
 598
 599 /*
 600  * Debug
 601  */
 602 #ifdef DX_DEBUG
 603 static void dx_show_index(char * label, struct dx_entry *entries)
 604 {
 605         int i, n = dx_get_count (entries);
 606         printk(KERN_DEBUG "%s index", label);
 607         for (i = 0; i < n; i++) {
 608                 printk(KERN_CONT " %x->%lu",
 609                        i ? dx_get_hash(entries + i) : 0,
 610                        (unsigned long)dx_get_block(entries + i));
 611         }
 612         printk(KERN_CONT "\n");
 613 }
 614
 615 struct stats
 616 {
 617         unsigned names;
 618         unsigned space;
 619         unsigned bcount;
 620 };
 621
 622 static struct stats dx_show_leaf(struct inode *dir,
 623                                 struct dx_hash_info *hinfo,
 624                                 struct ext4_dir_entry_2 *de,
 625                                 int size, int show_names)
 626 {
 627         unsigned names = 0, space = 0;
 628         char *base = (char *) de;
 629         struct dx_hash_info h = *hinfo;
 630
 631         printk("names: ");
 632         while ((char *) de < base + size)
 633         {
 634                 if (de->inode)
 635                 {
 636                         if (show_names)
 637                         {
 638 #ifdef CONFIG_FS_ENCRYPTION
 639                                 int len;
 640                                 char *name;
 641                                 struct fscrypt_str fname_crypto_str =
 642                                         FSTR_INIT(NULL, 0);
 643                                 int res = 0;
 644
 645                                 name  = de->name;
 646                                 len = de->name_len;
 647                                 if (!IS_ENCRYPTED(dir)) {
 648                                         /* Directory is not encrypted */
 649                                         ext4fs_dirhash(dir, de->name,
 650                                                 de->name_len, &h);
 651                                         printk("%*.s:(U)%x.%u ", len,
 652                                                name, h.hash,
 653                                                (unsigned) ((char *) de
 654                                                            - base));
 655                                 } else {
 656                                         struct fscrypt_str de_name =
 657                                                 FSTR_INIT(name, len);
 658
 659                                         /* Directory is encrypted */
 660                                         res = fscrypt_fname_alloc_buffer(
 661                                                 len, &fname_crypto_str);
 662                                         if (res)
 663                                                 printk(KERN_WARNING "Error "
 664                                                         "allocating crypto "
 665                                                         "buffer--skipping "
 666                                                         "crypto\n");
 667                                         res = fscrypt_fname_disk_to_usr(dir,
 668                                                 0, 0, &de_name,
 669                                                 &fname_crypto_str);
 670                                         if (res) {
 671                                                 printk(KERN_WARNING "Error "
 672                                                         "converting filename "
 673                                                         "from disk to usr"
 674                                                         "\n");
 675                                                 name = "??";
 676                                                 len = 2;
 677                                         } else {
 678                                                 name = fname_crypto_str.name;
 679                                                 len = fname_crypto_str.len;
 680                                         }
 681                                         if (IS_CASEFOLDED(dir))
 682                                                 h.hash = EXT4_DIRENT_HASH(de);
 683                                         else
 684                                                 ext4fs_dirhash(dir, de->name,
 685                                                        de->name_len, &h);
 686                                         printk("%*.s:(E)%x.%u ", len, name,
 687                                                h.hash, (unsigned) ((char *) de
 688                                                                    - base));
 689                                         fscrypt_fname_free_buffer(
 690                                                         &fname_crypto_str);
 691                                 }
 692 #else
 693                                 int len = de->name_len;
 694                                 char *name = de->name;
 695                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
 696                                 printk("%*.s:%x.%u ", len, name, h.hash,
 697                                        (unsigned) ((char *) de - base));
 698 #endif
 699                         }
 700                         space += ext4_dir_rec_len(de->name_len, dir);
 701                         names++;
 702                 }
 703                 de = ext4_next_entry(de, size);
 704         }
 705         printk(KERN_CONT "(%i)\n", names);
 706         return (struct stats) { names, space, 1 };
 707 }
 708
 709 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 710                              struct dx_entry *entries, int levels)
 711 {
 712         unsigned blocksize = dir->i_sb->s_blocksize;
 713         unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 714         unsigned bcount = 0;
 715         struct buffer_head *bh;
 716         printk("%i indexed blocks...\n", count);
 717         for (i = 0; i < count; i++, entries++)
 718         {
 719                 ext4_lblk_t block = dx_get_block(entries);
 720                 ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 721                 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 722                 struct stats stats;
 723                 printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 724                 bh = ext4_bread(NULL,dir, block, 0);
 725                 if (!bh || IS_ERR(bh))
 726                         continue;
 727                 stats = levels?
 728                    dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 729                    dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
 730                         bh->b_data, blocksize, 0);
 731                 names += stats.names;
 732                 space += stats.space;
 733                 bcount += stats.bcount;
 734                 brelse(bh);
 735         }
 736         if (bcount)
 737                 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 738                        levels ? "" : "   ", names, space/bcount,
 739                        (space/bcount)*100/blocksize);
 740         return (struct stats) { names, space, bcount};
 741 }
 742
 743 /*
 744  * Linear search cross check
 745  */
 746 static inline void htree_rep_invariant_check(struct dx_entry *at,
 747                                              struct dx_entry *target,
 748                                              u32 hash, unsigned int n)
 749 {
 750         while (n--) {
 751                 dxtrace(printk(KERN_CONT ","));
 752                 if (dx_get_hash(++at) > hash) {
 753                         at--;
 754                         break;
 755                 }
 756         }
 757         ASSERT(at == target - 1);
 758 }
 759 #else /* DX_DEBUG */
 760 static inline void htree_rep_invariant_check(struct dx_entry *at,
 761                                              struct dx_entry *target,
 762                                              u32 hash, unsigned int n)
 763 {
 764 }
 765 #endif /* DX_DEBUG */
 766
 767 /*
 768  * Probe for a directory leaf block to search.
 769  *
 770  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 771  * error in the directory index, and the caller should fall back to
 772  * searching the directory normally.  The callers of dx_probe **MUST**
 773  * check for this error code, and make sure it never gets reflected
 774  * back to userspace.
 775  */
 776 static struct dx_frame *
 777 dx_probe(struct ext4_filename *fname, struct inode *dir,
 778          struct dx_hash_info *hinfo, struct dx_frame *frame_in)
 779 {
 780         unsigned count, indirect;
 781         struct dx_entry *at, *entries, *p, *q, *m;
 782         struct dx_root *root;
 783         struct dx_frame *frame = frame_in;
 784         struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
 785         u32 hash;
 786
 787         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 788         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 789         if (IS_ERR(frame->bh))
 790                 return (struct dx_frame *) frame->bh;
 791
 792         root = (struct dx_root *) frame->bh->b_data;
 793         if (root->info.hash_version != DX_HASH_TEA &&
 794             root->info.hash_version != DX_HASH_HALF_MD4 &&
 795             root->info.hash_version != DX_HASH_LEGACY &&
 796             root->info.hash_version != DX_HASH_SIPHASH) {
 797                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 798                                    root->info.hash_version);
 799                 goto fail;
 800         }
 801         if (ext4_hash_in_dirent(dir)) {
 802                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 803                         ext4_warning_inode(dir,
 804                                 "Hash in dirent, but hash is not SIPHASH");
 805                         goto fail;
 806                 }
 807         } else {
 808                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 809                         ext4_warning_inode(dir,
 810                                 "Hash code is SIPHASH, but hash not in dirent");
 811                         goto fail;
 812                 }
 813         }
 814         if (fname)
 815                 hinfo = &fname->hinfo;
 816         hinfo->hash_version = root->info.hash_version;
 817         if (hinfo->hash_version <= DX_HASH_TEA)
 818                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 819         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 820         /* hash is already computed for encrypted casefolded directory */
 821         if (fname && fname_name(fname) &&
 822                                 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
 823                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 824         hash = hinfo->hash;
 825
 826         if (root->info.unused_flags & 1) {
 827                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 828                                    root->info.unused_flags);
 829                 goto fail;
 830         }
 831
 832         indirect = root->info.indirect_levels;
 833         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 834                 ext4_warning(dir->i_sb,
 835                              "Directory (ino: %lu) htree depth %#06x exceed"
 836                              "supported value", dir->i_ino,
 837                              ext4_dir_htree_level(dir->i_sb));
 838                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 839                         ext4_warning(dir->i_sb, "Enable large directory "
 840                                                 "feature to access it");
 841                 }
 842                 goto fail;
 843         }
 844
 845         entries = (struct dx_entry *)(((char *)&root->info) +
 846                                       root->info.info_length);
 847
 848         if (dx_get_limit(entries) != dx_root_limit(dir,
 849                                                    root->info.info_length)) {
 850                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 851                                    dx_get_limit(entries),
 852                                    dx_root_limit(dir, root->info.info_length));
 853                 goto fail;
 854         }
 855
 856         dxtrace(printk("Look up %x", hash));
 857         while (1) {
 858                 count = dx_get_count(entries);
 859                 if (!count || count > dx_get_limit(entries)) {
 860                         ext4_warning_inode(dir,
 861                                            "dx entry: count %u beyond limit %u",
 862                                            count, dx_get_limit(entries));
 863                         goto fail;
 864                 }
 865
 866                 p = entries + 1;
 867                 q = entries + count - 1;
 868                 while (p <= q) {
 869                         m = p + (q - p) / 2;
 870                         dxtrace(printk(KERN_CONT "."));
 871                         if (dx_get_hash(m) > hash)
 872                                 q = m - 1;
 873                         else
 874                                 p = m + 1;
 875                 }
 876
 877                 htree_rep_invariant_check(entries, p, hash, count - 1);
 878
 879                 at = p - 1;
 880                 dxtrace(printk(KERN_CONT " %x->%u\n",
 881                                at == entries ? 0 : dx_get_hash(at),
 882                                dx_get_block(at)));
 883                 frame->entries = entries;
 884                 frame->at = at;
 885                 if (!indirect--)
 886                         return frame;
 887                 frame++;
 888                 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
 889                 if (IS_ERR(frame->bh)) {
 890                         ret_err = (struct dx_frame *) frame->bh;
 891                         frame->bh = NULL;
 892                         goto fail;
 893                 }
 894                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 895
 896                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 897                         ext4_warning_inode(dir,
 898                                 "dx entry: limit %u != node limit %u",
 899                                 dx_get_limit(entries), dx_node_limit(dir));
 900                         goto fail;
 901                 }
 902         }
 903 fail:
 904         while (frame >= frame_in) {
 905                 brelse(frame->bh);
 906                 frame--;
 907         }
 908
 909         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 910                 ext4_warning_inode(dir,
 911                         "Corrupt directory, running e2fsck is recommended");
 912         return ret_err;
 913 }
 914
 915 static void dx_release(struct dx_frame *frames)
 916 {
 917         struct dx_root_info *info;
 918         int i;
 919         unsigned int indirect_levels;
 920
 921         if (frames[0].bh == NULL)
 922                 return;
 923
 924         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 925         /* save local copy, "info" may be freed after brelse() */
 926         indirect_levels = info->indirect_levels;
 927         for (i = 0; i <= indirect_levels; i++) {
 928                 if (frames[i].bh == NULL)
 929                         break;
 930                 brelse(frames[i].bh);
 931                 frames[i].bh = NULL;
 932         }
 933 }
 934
 935 /*
 936  * This function increments the frame pointer to search the next leaf
 937  * block, and reads in the necessary intervening nodes if the search
 938  * should be necessary.  Whether or not the search is necessary is
 939  * controlled by the hash parameter.  If the hash value is even, then
 940  * the search is only continued if the next block starts with that
 941  * hash value.  This is used if we are searching for a specific file.
 942  *
 943  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 944  *
 945  * This function returns 1 if the caller should continue to search,
 946  * or 0 if it should not.  If there is an error reading one of the
 947  * index blocks, it will a negative error code.
 948  *
 949  * If start_hash is non-null, it will be filled in with the starting
 950  * hash of the next page.
 951  */
 952 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 953                                  struct dx_frame *frame,
 954                                  struct dx_frame *frames,
 955                                  __u32 *start_hash)
 956 {
 957         struct dx_frame *p;
 958         struct buffer_head *bh;
 959         int num_frames = 0;
 960         __u32 bhash;
 961
 962         p = frame;
 963         /*
 964          * Find the next leaf page by incrementing the frame pointer.
 965          * If we run out of entries in the interior node, loop around and
 966          * increment pointer in the parent node.  When we break out of
 967          * this loop, num_frames indicates the number of interior
 968          * nodes need to be read.
 969          */
 970         while (1) {
 971                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 972                         break;
 973                 if (p == frames)
 974                         return 0;
 975                 num_frames++;
 976                 p--;
 977         }
 978
 979         /*
 980          * If the hash is 1, then continue only if the next page has a
 981          * continuation hash of any value.  This is used for readdir
 982          * handling.  Otherwise, check to see if the hash matches the
 983          * desired continuation hash.  If it doesn't, return since
 984          * there's no point to read in the successive index pages.
 985          */
 986         bhash = dx_get_hash(p->at);
 987         if (start_hash)
 988                 *start_hash = bhash;
 989         if ((hash & 1) == 0) {
 990                 if ((bhash & ~1) != hash)
 991                         return 0;
 992         }
 993         /*
 994          * If the hash is HASH_NB_ALWAYS, we always go to the next
 995          * block so no check is necessary
 996          */
 997         while (num_frames--) {
 998                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
 999                 if (IS_ERR(bh))
1000                         return PTR_ERR(bh);
1001                 p++;
1002                 brelse(p->bh);
1003                 p->bh = bh;
1004                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1005         }
1006         return 1;
1007 }
1008
1009
1010 /*
1011  * This function fills a red-black tree with information from a
1012  * directory block.  It returns the number directory entries loaded
1013  * into the tree.  If there is an error it is returned in err.
1014  */
1015 static int htree_dirblock_to_tree(struct file *dir_file,
1016                                   struct inode *dir, ext4_lblk_t block,
1017                                   struct dx_hash_info *hinfo,
1018                                   __u32 start_hash, __u32 start_minor_hash)
1019 {
1020         struct buffer_head *bh;
1021         struct ext4_dir_entry_2 *de, *top;
1022         int err = 0, count = 0;
1023         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1024         int csum = ext4_has_metadata_csum(dir->i_sb);
1025
1026         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1027                                                         (unsigned long)block));
1028         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1029         if (IS_ERR(bh))
1030                 return PTR_ERR(bh);
1031
1032         de = (struct ext4_dir_entry_2 *) bh->b_data;
1033         /* csum entries are not larger in the casefolded encrypted case */
1034         top = (struct ext4_dir_entry_2 *) ((char *) de +
1035                                            dir->i_sb->s_blocksize -
1036                                            ext4_dir_rec_len(0,
1037                                                            csum ? NULL : dir));
1038         /* Check if the directory is encrypted */
1039         if (IS_ENCRYPTED(dir)) {
1040                 err = fscrypt_prepare_readdir(dir);
1041                 if (err < 0) {
1042                         brelse(bh);
1043                         return err;
1044                 }
1045                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1046                                                  &fname_crypto_str);
1047                 if (err < 0) {
1048                         brelse(bh);
1049                         return err;
1050                 }
1051         }
1052
1053         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1054                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1055                                 bh->b_data, bh->b_size,
1056                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1057                                          + ((char *)de - bh->b_data))) {
1058                         /* silently ignore the rest of the block */
1059                         break;
1060                 }
1061                 if (ext4_hash_in_dirent(dir)) {
1062                         if (de->name_len && de->inode) {
1063                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1064                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1065                         } else {
1066                                 hinfo->hash = 0;
1067                                 hinfo->minor_hash = 0;
1068                         }
1069                 } else {
1070                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1071                 }
1072                 if ((hinfo->hash < start_hash) ||
1073                     ((hinfo->hash == start_hash) &&
1074                      (hinfo->minor_hash < start_minor_hash)))
1075                         continue;
1076                 if (de->inode == 0)
1077                         continue;
1078                 if (!IS_ENCRYPTED(dir)) {
1079                         tmp_str.name = de->name;
1080                         tmp_str.len = de->name_len;
1081                         err = ext4_htree_store_dirent(dir_file,
1082                                    hinfo->hash, hinfo->minor_hash, de,
1083                                    &tmp_str);
1084                 } else {
1085                         int save_len = fname_crypto_str.len;
1086                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1087                                                                 de->name_len);
1088
1089                         /* Directory is encrypted */
1090                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1091                                         hinfo->minor_hash, &de_name,
1092                                         &fname_crypto_str);
1093                         if (err) {
1094                                 count = err;
1095                                 goto errout;
1096                         }
1097                         err = ext4_htree_store_dirent(dir_file,
1098                                    hinfo->hash, hinfo->minor_hash, de,
1099                                         &fname_crypto_str);
1100                         fname_crypto_str.len = save_len;
1101                 }
1102                 if (err != 0) {
1103                         count = err;
1104                         goto errout;
1105                 }
1106                 count++;
1107         }
1108 errout:
1109         brelse(bh);
1110         fscrypt_fname_free_buffer(&fname_crypto_str);
1111         return count;
1112 }
1113
1114
1115 /*
1116  * This function fills a red-black tree with information from a
1117  * directory.  We start scanning the directory in hash order, starting
1118  * at start_hash and start_minor_hash.
1119  *
1120  * This function returns the number of entries inserted into the tree,
1121  * or a negative error code.
1122  */
1123 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1124                          __u32 start_minor_hash, __u32 *next_hash)
1125 {
1126         struct dx_hash_info hinfo;
1127         struct ext4_dir_entry_2 *de;
1128         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1129         struct inode *dir;
1130         ext4_lblk_t block;
1131         int count = 0;
1132         int ret, err;
1133         __u32 hashval;
1134         struct fscrypt_str tmp_str;
1135
1136         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1137                        start_hash, start_minor_hash));
1138         dir = file_inode(dir_file);
1139         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1140                 if (ext4_hash_in_dirent(dir))
1141                         hinfo.hash_version = DX_HASH_SIPHASH;
1142                 else
1143                         hinfo.hash_version =
1144                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1145                 if (hinfo.hash_version <= DX_HASH_TEA)
1146                         hinfo.hash_version +=
1147                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1148                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1149                 if (ext4_has_inline_data(dir)) {
1150                         int has_inline_data = 1;
1151                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1152                                                        &hinfo, start_hash,
1153                                                        start_minor_hash,
1154                                                        &has_inline_data);
1155                         if (has_inline_data) {
1156                                 *next_hash = ~0;
1157                                 return count;
1158                         }
1159                 }
1160                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1161                                                start_hash, start_minor_hash);
1162                 *next_hash = ~0;
1163                 return count;
1164         }
1165         hinfo.hash = start_hash;
1166         hinfo.minor_hash = 0;
1167         frame = dx_probe(NULL, dir, &hinfo, frames);
1168         if (IS_ERR(frame))
1169                 return PTR_ERR(frame);
1170
1171         /* Add '.' and '..' from the htree header */
1172         if (!start_hash && !start_minor_hash) {
1173                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1174                 tmp_str.name = de->name;
1175                 tmp_str.len = de->name_len;
1176                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1177                                               de, &tmp_str);
1178                 if (err != 0)
1179                         goto errout;
1180                 count++;
1181         }
1182         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1183                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1184                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1185                 tmp_str.name = de->name;
1186                 tmp_str.len = de->name_len;
1187                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1188                                               de, &tmp_str);
1189                 if (err != 0)
1190                         goto errout;
1191                 count++;
1192         }
1193
1194         while (1) {
1195                 if (fatal_signal_pending(current)) {
1196                         err = -ERESTARTSYS;
1197                         goto errout;
1198                 }
1199                 cond_resched();
1200                 block = dx_get_block(frame->at);
1201                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1202                                              start_hash, start_minor_hash);
1203                 if (ret < 0) {
1204                         err = ret;
1205                         goto errout;
1206                 }
1207                 count += ret;
1208                 hashval = ~0;
1209                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1210                                             frame, frames, &hashval);
1211                 *next_hash = hashval;
1212                 if (ret < 0) {
1213                         err = ret;
1214                         goto errout;
1215                 }
1216                 /*
1217                  * Stop if:  (a) there are no more entries, or
1218                  * (b) we have inserted at least one entry and the
1219                  * next hash value is not a continuation
1220                  */
1221                 if ((ret == 0) ||
1222                     (count && ((hashval & 1) == 0)))
1223                         break;
1224         }
1225         dx_release(frames);
1226         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1227                        "next hash: %x\n", count, *next_hash));
1228         return count;
1229 errout:
1230         dx_release(frames);
1231         return (err);
1232 }
1233
1234 static inline int search_dirblock(struct buffer_head *bh,
1235                                   struct inode *dir,
1236                                   struct ext4_filename *fname,
1237                                   unsigned int offset,
1238                                   struct ext4_dir_entry_2 **res_dir)
1239 {
1240         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1241                                fname, offset, res_dir);
1242 }
1243
1244 /*
1245  * Directory block splitting, compacting
1246  */
1247
1248 /*
1249  * Create map of hash values, offsets, and sizes, stored at end of block.
1250  * Returns number of entries mapped.
1251  */
1252 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1253                        unsigned blocksize, struct dx_hash_info *hinfo,
1254                        struct dx_map_entry *map_tail)
1255 {
1256         int count = 0;
1257         char *base = (char *) de;
1258         struct dx_hash_info h = *hinfo;
1259
1260         while ((char *) de < base + blocksize) {
1261                 if (de->name_len && de->inode) {
1262                         if (ext4_hash_in_dirent(dir))
1263                                 h.hash = EXT4_DIRENT_HASH(de);
1264                         else
1265                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1266                         map_tail--;
1267                         map_tail->hash = h.hash;
1268                         map_tail->offs = ((char *) de - base)>>2;
1269                         map_tail->size = le16_to_cpu(de->rec_len);
1270                         count++;
1271                         cond_resched();
1272                 }
1273                 /* XXX: do we need to check rec_len == 0 case? -Chris */
1274                 de = ext4_next_entry(de, blocksize);
1275         }
1276         return count;
1277 }
1278
1279 /* Sort map by hash value */
1280 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1281 {
1282         struct dx_map_entry *p, *q, *top = map + count - 1;
1283         int more;
1284         /* Combsort until bubble sort doesn't suck */
1285         while (count > 2) {
1286                 count = count*10/13;
1287                 if (count - 9 < 2) /* 9, 10 -> 11 */
1288                         count = 11;
1289                 for (p = top, q = p - count; q >= map; p--, q--)
1290                         if (p->hash < q->hash)
1291                                 swap(*p, *q);
1292         }
1293         /* Garden variety bubble sort */
1294         do {
1295                 more = 0;
1296                 q = top;
1297                 while (q-- > map) {
1298                         if (q[1].hash >= q[0].hash)
1299                                 continue;
1300                         swap(*(q+1), *q);
1301                         more = 1;
1302                 }
1303         } while(more);
1304 }
1305
1306 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1307 {
1308         struct dx_entry *entries = frame->entries;
1309         struct dx_entry *old = frame->at, *new = old + 1;
1310         int count = dx_get_count(entries);
1311
1312         ASSERT(count < dx_get_limit(entries));
1313         ASSERT(old < entries + count);
1314         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1315         dx_set_hash(new, hash);
1316         dx_set_block(new, block);
1317         dx_set_count(entries, count + 1);
1318 }
1319
1320 #ifdef CONFIG_UNICODE
1321 /*
1322  * Test whether a case-insensitive directory entry matches the filename
1323  * being searched for.  If quick is set, assume the name being looked up
1324  * is already in the casefolded form.
1325  *
1326  * Returns: 0 if the directory entry matches, more than 0 if it
1327  * doesn't match or less than zero on error.
1328  */
1329 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1330                            u8 *de_name, size_t de_name_len, bool quick)
1331 {
1332         const struct super_block *sb = parent->i_sb;
1333         const struct unicode_map *um = sb->s_encoding;
1334         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1335         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1336         int ret;
1337
1338         if (IS_ENCRYPTED(parent)) {
1339                 const struct fscrypt_str encrypted_name =
1340                                 FSTR_INIT(de_name, de_name_len);
1341
1342                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1343                 if (!decrypted_name.name)
1344                         return -ENOMEM;
1345                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1346                                                 &decrypted_name);
1347                 if (ret < 0)
1348                         goto out;
1349                 entry.name = decrypted_name.name;
1350                 entry.len = decrypted_name.len;
1351         }
1352
1353         if (quick)
1354                 ret = utf8_strncasecmp_folded(um, name, &entry);
1355         else
1356                 ret = utf8_strncasecmp(um, name, &entry);
1357         if (ret < 0) {
1358                 /* Handle invalid character sequence as either an error
1359                  * or as an opaque byte sequence.
1360                  */
1361                 if (sb_has_strict_encoding(sb))
1362                         ret = -EINVAL;
1363                 else if (name->len != entry.len)
1364                         ret = 1;
1365                 else
1366                         ret = !!memcmp(name->name, entry.name, entry.len);
1367         }
1368 out:
1369         kfree(decrypted_name.name);
1370         return ret;
1371 }
1372
1373 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1374                                   struct ext4_filename *name)
1375 {
1376         struct fscrypt_str *cf_name = &name->cf_name;
1377         struct dx_hash_info *hinfo = &name->hinfo;
1378         int len;
1379
1380         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1381             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1382                 cf_name->name = NULL;
1383                 return 0;
1384         }
1385
1386         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1387         if (!cf_name->name)
1388                 return -ENOMEM;
1389
1390         len = utf8_casefold(dir->i_sb->s_encoding,
1391                             iname, cf_name->name,
1392                             EXT4_NAME_LEN);
1393         if (len <= 0) {
1394                 kfree(cf_name->name);
1395                 cf_name->name = NULL;
1396         }
1397         cf_name->len = (unsigned) len;
1398         if (!IS_ENCRYPTED(dir))
1399                 return 0;
1400
1401         hinfo->hash_version = DX_HASH_SIPHASH;
1402         hinfo->seed = NULL;
1403         if (cf_name->name)
1404                 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1405         else
1406                 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1407         return 0;
1408 }
1409 #endif
1410
1411 /*
1412  * Test whether a directory entry matches the filename being searched for.
1413  *
1414  * Return: %true if the directory entry matches, otherwise %false.
1415  */
1416 static bool ext4_match(struct inode *parent,
1417                               const struct ext4_filename *fname,
1418                               struct ext4_dir_entry_2 *de)
1419 {
1420         struct fscrypt_name f;
1421
1422         if (!de->inode)
1423                 return false;
1424
1425         f.usr_fname = fname->usr_fname;
1426         f.disk_name = fname->disk_name;
1427 #ifdef CONFIG_FS_ENCRYPTION
1428         f.crypto_buf = fname->crypto_buf;
1429 #endif
1430
1431 #ifdef CONFIG_UNICODE
1432         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1433             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1434                 if (fname->cf_name.name) {
1435                         struct qstr cf = {.name = fname->cf_name.name,
1436                                           .len = fname->cf_name.len};
1437                         if (IS_ENCRYPTED(parent)) {
1438                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1439                                         fname->hinfo.minor_hash !=
1440                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1441
1442                                         return 0;
1443                                 }
1444                         }
1445                         return !ext4_ci_compare(parent, &cf, de->name,
1446                                                         de->name_len, true);
1447                 }
1448                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1449                                                 de->name_len, false);
1450         }
1451 #endif
1452
1453         return fscrypt_match_name(&f, de->name, de->name_len);
1454 }
1455
1456 /*
1457  * Returns 0 if not found, -1 on failure, and 1 on success
1458  */
1459 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1460                     struct inode *dir, struct ext4_filename *fname,
1461                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1462 {
1463         struct ext4_dir_entry_2 * de;
1464         char * dlimit;
1465         int de_len;
1466
1467         de = (struct ext4_dir_entry_2 *)search_buf;
1468         dlimit = search_buf + buf_size;
1469         while ((char *) de < dlimit) {
1470                 /* this code is executed quadratically often */
1471                 /* do minimal checking `by hand' */
1472                 if ((char *) de + de->name_len <= dlimit &&
1473                     ext4_match(dir, fname, de)) {
1474                         /* found a match - just to be sure, do
1475                          * a full check */
1476                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1477                                                  buf_size, offset))
1478                                 return -1;
1479                         *res_dir = de;
1480                         return 1;
1481                 }
1482                 /* prevent looping on a bad block */
1483                 de_len = ext4_rec_len_from_disk(de->rec_len,
1484                                                 dir->i_sb->s_blocksize);
1485                 if (de_len <= 0)
1486                         return -1;
1487                 offset += de_len;
1488                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1489         }
1490         return 0;
1491 }
1492
1493 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1494                                struct ext4_dir_entry *de)
1495 {
1496         struct super_block *sb = dir->i_sb;
1497
1498         if (!is_dx(dir))
1499                 return 0;
1500         if (block == 0)
1501                 return 1;
1502         if (de->inode == 0 &&
1503             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1504                         sb->s_blocksize)
1505                 return 1;
1506         return 0;
1507 }
1508
1509 /*
1510  *      __ext4_find_entry()
1511  *
1512  * finds an entry in the specified directory with the wanted name. It
1513  * returns the cache buffer in which the entry was found, and the entry
1514  * itself (as a parameter - res_dir). It does NOT read the inode of the
1515  * entry - you'll have to do that yourself if you want to.
1516  *
1517  * The returned buffer_head has ->b_count elevated.  The caller is expected
1518  * to brelse() it when appropriate.
1519  */
1520 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1521                                              struct ext4_filename *fname,
1522                                              struct ext4_dir_entry_2 **res_dir,
1523                                              int *inlined)
1524 {
1525         struct super_block *sb;
1526         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1527         struct buffer_head *bh, *ret = NULL;
1528         ext4_lblk_t start, block;
1529         const u8 *name = fname->usr_fname->name;
1530         size_t ra_max = 0;      /* Number of bh's in the readahead
1531                                    buffer, bh_use[] */
1532         size_t ra_ptr = 0;      /* Current index into readahead
1533                                    buffer */
1534         ext4_lblk_t  nblocks;
1535         int i, namelen, retval;
1536
1537         *res_dir = NULL;
1538         sb = dir->i_sb;
1539         namelen = fname->usr_fname->len;
1540         if (namelen > EXT4_NAME_LEN)
1541                 return NULL;
1542
1543         if (ext4_has_inline_data(dir)) {
1544                 int has_inline_data = 1;
1545                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1546                                              &has_inline_data);
1547                 if (has_inline_data) {
1548                         if (inlined)
1549                                 *inlined = 1;
1550                         goto cleanup_and_exit;
1551                 }
1552         }
1553
1554         if ((namelen <= 2) && (name[0] == '.') &&
1555             (name[1] == '.' || name[1] == '\0')) {
1556                 /*
1557                  * "." or ".." will only be in the first block
1558                  * NFS may look up ".."; "." should be handled by the VFS
1559                  */
1560                 block = start = 0;
1561                 nblocks = 1;
1562                 goto restart;
1563         }
1564         if (is_dx(dir)) {
1565                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1566                 /*
1567                  * On success, or if the error was file not found,
1568                  * return.  Otherwise, fall back to doing a search the
1569                  * old fashioned way.
1570                  */
1571                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1572                         goto cleanup_and_exit;
1573                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1574                                "falling back\n"));
1575                 ret = NULL;
1576         }
1577         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1578         if (!nblocks) {
1579                 ret = NULL;
1580                 goto cleanup_and_exit;
1581         }
1582         start = EXT4_I(dir)->i_dir_start_lookup;
1583         if (start >= nblocks)
1584                 start = 0;
1585         block = start;
1586 restart:
1587         do {
1588                 /*
1589                  * We deal with the read-ahead logic here.
1590                  */
1591                 cond_resched();
1592                 if (ra_ptr >= ra_max) {
1593                         /* Refill the readahead buffer */
1594                         ra_ptr = 0;
1595                         if (block < start)
1596                                 ra_max = start - block;
1597                         else
1598                                 ra_max = nblocks - block;
1599                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1600                         retval = ext4_bread_batch(dir, block, ra_max,
1601                                                   false /* wait */, bh_use);
1602                         if (retval) {
1603                                 ret = ERR_PTR(retval);
1604                                 ra_max = 0;
1605                                 goto cleanup_and_exit;
1606                         }
1607                 }
1608                 if ((bh = bh_use[ra_ptr++]) == NULL)
1609                         goto next;
1610                 wait_on_buffer(bh);
1611                 if (!buffer_uptodate(bh)) {
1612                         EXT4_ERROR_INODE_ERR(dir, EIO,
1613                                              "reading directory lblock %lu",
1614                                              (unsigned long) block);
1615                         brelse(bh);
1616                         ret = ERR_PTR(-EIO);
1617                         goto cleanup_and_exit;
1618                 }
1619                 if (!buffer_verified(bh) &&
1620                     !is_dx_internal_node(dir, block,
1621                                          (struct ext4_dir_entry *)bh->b_data) &&
1622                     !ext4_dirblock_csum_verify(dir, bh)) {
1623                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1624                                              "checksumming directory "
1625                                              "block %lu", (unsigned long)block);
1626                         brelse(bh);
1627                         ret = ERR_PTR(-EFSBADCRC);
1628                         goto cleanup_and_exit;
1629                 }
1630                 set_buffer_verified(bh);
1631                 i = search_dirblock(bh, dir, fname,
1632                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1633                 if (i == 1) {
1634                         EXT4_I(dir)->i_dir_start_lookup = block;
1635                         ret = bh;
1636                         goto cleanup_and_exit;
1637                 } else {
1638                         brelse(bh);
1639                         if (i < 0)
1640                                 goto cleanup_and_exit;
1641                 }
1642         next:
1643                 if (++block >= nblocks)
1644                         block = 0;
1645         } while (block != start);
1646
1647         /*
1648          * If the directory has grown while we were searching, then
1649          * search the last part of the directory before giving up.
1650          */
1651         block = nblocks;
1652         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1653         if (block < nblocks) {
1654                 start = 0;
1655                 goto restart;
1656         }
1657
1658 cleanup_and_exit:
1659         /* Clean up the read-ahead blocks */
1660         for (; ra_ptr < ra_max; ra_ptr++)
1661                 brelse(bh_use[ra_ptr]);
1662         return ret;
1663 }
1664
1665 static struct buffer_head *ext4_find_entry(struct inode *dir,
1666                                            const struct qstr *d_name,
1667                                            struct ext4_dir_entry_2 **res_dir,
1668                                            int *inlined)
1669 {
1670         int err;
1671         struct ext4_filename fname;
1672         struct buffer_head *bh;
1673
1674         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1675         if (err == -ENOENT)
1676                 return NULL;
1677         if (err)
1678                 return ERR_PTR(err);
1679
1680         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1681
1682         ext4_fname_free_filename(&fname);
1683         return bh;
1684 }
1685
1686 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1687                                              struct dentry *dentry,
1688                                              struct ext4_dir_entry_2 **res_dir)
1689 {
1690         int err;
1691         struct ext4_filename fname;
1692         struct buffer_head *bh;
1693
1694         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1695         generic_set_encrypted_ci_d_ops(dentry);
1696         if (err == -ENOENT)
1697                 return NULL;
1698         if (err)
1699                 return ERR_PTR(err);
1700
1701         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1702
1703         ext4_fname_free_filename(&fname);
1704         return bh;
1705 }
1706
1707 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1708                         struct ext4_filename *fname,
1709                         struct ext4_dir_entry_2 **res_dir)
1710 {
1711         struct super_block * sb = dir->i_sb;
1712         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1713         struct buffer_head *bh;
1714         ext4_lblk_t block;
1715         int retval;
1716
1717 #ifdef CONFIG_FS_ENCRYPTION
1718         *res_dir = NULL;
1719 #endif
1720         frame = dx_probe(fname, dir, NULL, frames);
1721         if (IS_ERR(frame))
1722                 return (struct buffer_head *) frame;
1723         do {
1724                 block = dx_get_block(frame->at);
1725                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1726                 if (IS_ERR(bh))
1727                         goto errout;
1728
1729                 retval = search_dirblock(bh, dir, fname,
1730                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1731                                          res_dir);
1732                 if (retval == 1)
1733                         goto success;
1734                 brelse(bh);
1735                 if (retval == -1) {
1736                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1737                         goto errout;
1738                 }
1739
1740                 /* Check to see if we should continue to search */
1741                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1742                                                frames, NULL);
1743                 if (retval < 0) {
1744                         ext4_warning_inode(dir,
1745                                 "error %d reading directory index block",
1746                                 retval);
1747                         bh = ERR_PTR(retval);
1748                         goto errout;
1749                 }
1750         } while (retval == 1);
1751
1752         bh = NULL;
1753 errout:
1754         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1755 success:
1756         dx_release(frames);
1757         return bh;
1758 }
1759
1760 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1761 {
1762         struct inode *inode;
1763         struct ext4_dir_entry_2 *de;
1764         struct buffer_head *bh;
1765
1766         if (dentry->d_name.len > EXT4_NAME_LEN)
1767                 return ERR_PTR(-ENAMETOOLONG);
1768
1769         bh = ext4_lookup_entry(dir, dentry, &de);
1770         if (IS_ERR(bh))
1771                 return ERR_CAST(bh);
1772         inode = NULL;
1773         if (bh) {
1774                 __u32 ino = le32_to_cpu(de->inode);
1775                 brelse(bh);
1776                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1777                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1778                         return ERR_PTR(-EFSCORRUPTED);
1779                 }
1780                 if (unlikely(ino == dir->i_ino)) {
1781                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1782                                          dentry);
1783                         return ERR_PTR(-EFSCORRUPTED);
1784                 }
1785                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1786                 if (inode == ERR_PTR(-ESTALE)) {
1787                         EXT4_ERROR_INODE(dir,
1788                                          "deleted inode referenced: %u",
1789                                          ino);
1790                         return ERR_PTR(-EFSCORRUPTED);
1791                 }
1792                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1793                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1794                     !fscrypt_has_permitted_context(dir, inode)) {
1795                         ext4_warning(inode->i_sb,
1796                                      "Inconsistent encryption contexts: %lu/%lu",
1797                                      dir->i_ino, inode->i_ino);
1798                         iput(inode);
1799                         return ERR_PTR(-EPERM);
1800                 }
1801         }
1802
1803 #ifdef CONFIG_UNICODE
1804         if (!inode && IS_CASEFOLDED(dir)) {
1805                 /* Eventually we want to call d_add_ci(dentry, NULL)
1806                  * for negative dentries in the encoding case as
1807                  * well.  For now, prevent the negative dentry
1808                  * from being cached.
1809                  */
1810                 return NULL;
1811         }
1812 #endif
1813         return d_splice_alias(inode, dentry);
1814 }
1815
1816
1817 struct dentry *ext4_get_parent(struct dentry *child)
1818 {
1819         __u32 ino;
1820         struct ext4_dir_entry_2 * de;
1821         struct buffer_head *bh;
1822
1823         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1824         if (IS_ERR(bh))
1825                 return ERR_CAST(bh);
1826         if (!bh)
1827                 return ERR_PTR(-ENOENT);
1828         ino = le32_to_cpu(de->inode);
1829         brelse(bh);
1830
1831         if (!ext4_valid_inum(child->d_sb, ino)) {
1832                 EXT4_ERROR_INODE(d_inode(child),
1833                                  "bad parent inode number: %u", ino);
1834                 return ERR_PTR(-EFSCORRUPTED);
1835         }
1836
1837         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1838 }
1839
1840 /*
1841  * Move count entries from end of map between two memory locations.
1842  * Returns pointer to last entry moved.
1843  */
1844 static struct ext4_dir_entry_2 *
1845 dx_move_dirents(struct inode *dir, char *from, char *to,
1846                 struct dx_map_entry *map, int count,
1847                 unsigned blocksize)
1848 {
1849         unsigned rec_len = 0;
1850
1851         while (count--) {
1852                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1853                                                 (from + (map->offs<<2));
1854                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1855
1856                 memcpy (to, de, rec_len);
1857                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1858                                 ext4_rec_len_to_disk(rec_len, blocksize);
1859
1860                 /* wipe dir_entry excluding the rec_len field */
1861                 de->inode = 0;
1862                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1863                                                                 blocksize) -
1864                                          offsetof(struct ext4_dir_entry_2,
1865                                                                 name_len));
1866
1867                 map++;
1868                 to += rec_len;
1869         }
1870         return (struct ext4_dir_entry_2 *) (to - rec_len);
1871 }
1872
1873 /*
1874  * Compact each dir entry in the range to the minimal rec_len.
1875  * Returns pointer to last entry in range.
1876  */
1877 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1878                                                         unsigned int blocksize)
1879 {
1880         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1881         unsigned rec_len = 0;
1882
1883         prev = to = de;
1884         while ((char*)de < base + blocksize) {
1885                 next = ext4_next_entry(de, blocksize);
1886                 if (de->inode && de->name_len) {
1887                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1888                         if (de > to)
1889                                 memmove(to, de, rec_len);
1890                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1891                         prev = to;
1892                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1893                 }
1894                 de = next;
1895         }
1896         return prev;
1897 }
1898
1899 /*
1900  * Split a full leaf block to make room for a new dir entry.
1901  * Allocate a new block, and move entries so that they are approx. equally full.
1902  * Returns pointer to de in block into which the new entry will be inserted.
1903  */
1904 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1905                         struct buffer_head **bh,struct dx_frame *frame,
1906                         struct dx_hash_info *hinfo)
1907 {
1908         unsigned blocksize = dir->i_sb->s_blocksize;
1909         unsigned count, continued;
1910         struct buffer_head *bh2;
1911         ext4_lblk_t newblock;
1912         u32 hash2;
1913         struct dx_map_entry *map;
1914         char *data1 = (*bh)->b_data, *data2;
1915         unsigned split, move, size;
1916         struct ext4_dir_entry_2 *de = NULL, *de2;
1917         int     csum_size = 0;
1918         int     err = 0, i;
1919
1920         if (ext4_has_metadata_csum(dir->i_sb))
1921                 csum_size = sizeof(struct ext4_dir_entry_tail);
1922
1923         bh2 = ext4_append(handle, dir, &newblock);
1924         if (IS_ERR(bh2)) {
1925                 brelse(*bh);
1926                 *bh = NULL;
1927                 return (struct ext4_dir_entry_2 *) bh2;
1928         }
1929
1930         BUFFER_TRACE(*bh, "get_write_access");
1931         err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1932                                             EXT4_JTR_NONE);
1933         if (err)
1934                 goto journal_error;
1935
1936         BUFFER_TRACE(frame->bh, "get_write_access");
1937         err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1938                                             EXT4_JTR_NONE);
1939         if (err)
1940                 goto journal_error;
1941
1942         data2 = bh2->b_data;
1943
1944         /* create map in the end of data2 block */
1945         map = (struct dx_map_entry *) (data2 + blocksize);
1946         count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1947                              blocksize, hinfo, map);
1948         map -= count;
1949         dx_sort_map(map, count);
1950         /* Ensure that neither split block is over half full */
1951         size = 0;
1952         move = 0;
1953         for (i = count-1; i >= 0; i--) {
1954                 /* is more than half of this entry in 2nd half of the block? */
1955                 if (size + map[i].size/2 > blocksize/2)
1956                         break;
1957                 size += map[i].size;
1958                 move++;
1959         }
1960         /*
1961          * map index at which we will split
1962          *
1963          * If the sum of active entries didn't exceed half the block size, just
1964          * split it in half by count; each resulting block will have at least
1965          * half the space free.
1966          */
1967         if (i > 0)
1968                 split = count - move;
1969         else
1970                 split = count/2;
1971
1972         hash2 = map[split].hash;
1973         continued = hash2 == map[split - 1].hash;
1974         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1975                         (unsigned long)dx_get_block(frame->at),
1976                                         hash2, split, count-split));
1977
1978         /* Fancy dance to stay within two buffers */
1979         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
1980                               blocksize);
1981         de = dx_pack_dirents(dir, data1, blocksize);
1982         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1983                                            (char *) de,
1984                                            blocksize);
1985         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1986                                             (char *) de2,
1987                                             blocksize);
1988         if (csum_size) {
1989                 ext4_initialize_dirent_tail(*bh, blocksize);
1990                 ext4_initialize_dirent_tail(bh2, blocksize);
1991         }
1992
1993         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1994                         blocksize, 1));
1995         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1996                         blocksize, 1));
1997
1998         /* Which block gets the new entry? */
1999         if (hinfo->hash >= hash2) {
2000                 swap(*bh, bh2);
2001                 de = de2;
2002         }
2003         dx_insert_block(frame, hash2 + continued, newblock);
2004         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2005         if (err)
2006                 goto journal_error;
2007         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2008         if (err)
2009                 goto journal_error;
2010         brelse(bh2);
2011         dxtrace(dx_show_index("frame", frame->entries));
2012         return de;
2013
2014 journal_error:
2015         brelse(*bh);
2016         brelse(bh2);
2017         *bh = NULL;
2018         ext4_std_error(dir->i_sb, err);
2019         return ERR_PTR(err);
2020 }
2021
2022 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2023                       struct buffer_head *bh,
2024                       void *buf, int buf_size,
2025                       struct ext4_filename *fname,
2026                       struct ext4_dir_entry_2 **dest_de)
2027 {
2028         struct ext4_dir_entry_2 *de;
2029         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2030         int nlen, rlen;
2031         unsigned int offset = 0;
2032         char *top;
2033
2034         de = (struct ext4_dir_entry_2 *)buf;
2035         top = buf + buf_size - reclen;
2036         while ((char *) de <= top) {
2037                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2038                                          buf, buf_size, offset))
2039                         return -EFSCORRUPTED;
2040                 if (ext4_match(dir, fname, de))
2041                         return -EEXIST;
2042                 nlen = ext4_dir_rec_len(de->name_len, dir);
2043                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2044                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2045                         break;
2046                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2047                 offset += rlen;
2048         }
2049         if ((char *) de > top)
2050                 return -ENOSPC;
2051
2052         *dest_de = de;
2053         return 0;
2054 }
2055
2056 void ext4_insert_dentry(struct inode *dir,
2057                         struct inode *inode,
2058                         struct ext4_dir_entry_2 *de,
2059                         int buf_size,
2060                         struct ext4_filename *fname)
2061 {
2062
2063         int nlen, rlen;
2064
2065         nlen = ext4_dir_rec_len(de->name_len, dir);
2066         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2067         if (de->inode) {
2068                 struct ext4_dir_entry_2 *de1 =
2069                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2070                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2071                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2072                 de = de1;
2073         }
2074         de->file_type = EXT4_FT_UNKNOWN;
2075         de->inode = cpu_to_le32(inode->i_ino);
2076         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2077         de->name_len = fname_len(fname);
2078         memcpy(de->name, fname_name(fname), fname_len(fname));
2079         if (ext4_hash_in_dirent(dir)) {
2080                 struct dx_hash_info *hinfo = &fname->hinfo;
2081
2082                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2083                 EXT4_DIRENT_HASHES(de)->minor_hash =
2084                                                 cpu_to_le32(hinfo->minor_hash);
2085         }
2086 }
2087
2088 /*
2089  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2090  * it points to a directory entry which is guaranteed to be large
2091  * enough for new directory entry.  If de is NULL, then
2092  * add_dirent_to_buf will attempt search the directory block for
2093  * space.  It will return -ENOSPC if no space is available, and -EIO
2094  * and -EEXIST if directory entry already exists.
2095  */
2096 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2097                              struct inode *dir,
2098                              struct inode *inode, struct ext4_dir_entry_2 *de,
2099                              struct buffer_head *bh)
2100 {
2101         unsigned int    blocksize = dir->i_sb->s_blocksize;
2102         int             csum_size = 0;
2103         int             err, err2;
2104
2105         if (ext4_has_metadata_csum(inode->i_sb))
2106                 csum_size = sizeof(struct ext4_dir_entry_tail);
2107
2108         if (!de) {
2109                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2110                                         blocksize - csum_size, fname, &de);
2111                 if (err)
2112                         return err;
2113         }
2114         BUFFER_TRACE(bh, "get_write_access");
2115         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2116                                             EXT4_JTR_NONE);
2117         if (err) {
2118                 ext4_std_error(dir->i_sb, err);
2119                 return err;
2120         }
2121
2122         /* By now the buffer is marked for journaling */
2123         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2124
2125         /*
2126          * XXX shouldn't update any times until successful
2127          * completion of syscall, but too many callers depend
2128          * on this.
2129          *
2130          * XXX similarly, too many callers depend on
2131          * ext4_new_inode() setting the times, but error
2132          * recovery deletes the inode, so the worst that can
2133          * happen is that the times are slightly out of date
2134          * and/or different from the directory change time.
2135          */
2136         dir->i_mtime = dir->i_ctime = current_time(dir);
2137         ext4_update_dx_flag(dir);
2138         inode_inc_iversion(dir);
2139         err2 = ext4_mark_inode_dirty(handle, dir);
2140         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2141         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2142         if (err)
2143                 ext4_std_error(dir->i_sb, err);
2144         return err ? err : err2;
2145 }
2146
2147 /*
2148  * This converts a one block unindexed directory to a 3 block indexed
2149  * directory, and adds the dentry to the indexed directory.
2150  */
2151 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2152                             struct inode *dir,
2153                             struct inode *inode, struct buffer_head *bh)
2154 {
2155         struct buffer_head *bh2;
2156         struct dx_root  *root;
2157         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2158         struct dx_entry *entries;
2159         struct ext4_dir_entry_2 *de, *de2;
2160         char            *data2, *top;
2161         unsigned        len;
2162         int             retval;
2163         unsigned        blocksize;
2164         ext4_lblk_t  block;
2165         struct fake_dirent *fde;
2166         int csum_size = 0;
2167
2168         if (ext4_has_metadata_csum(inode->i_sb))
2169                 csum_size = sizeof(struct ext4_dir_entry_tail);
2170
2171         blocksize =  dir->i_sb->s_blocksize;
2172         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2173         BUFFER_TRACE(bh, "get_write_access");
2174         retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2175                                                EXT4_JTR_NONE);
2176         if (retval) {
2177                 ext4_std_error(dir->i_sb, retval);
2178                 brelse(bh);
2179                 return retval;
2180         }
2181         root = (struct dx_root *) bh->b_data;
2182
2183         /* The 0th block becomes the root, move the dirents out */
2184         fde = &root->dotdot;
2185         de = (struct ext4_dir_entry_2 *)((char *)fde +
2186                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2187         if ((char *) de >= (((char *) root) + blocksize)) {
2188                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2189                 brelse(bh);
2190                 return -EFSCORRUPTED;
2191         }
2192         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2193
2194         /* Allocate new block for the 0th block's dirents */
2195         bh2 = ext4_append(handle, dir, &block);
2196         if (IS_ERR(bh2)) {
2197                 brelse(bh);
2198                 return PTR_ERR(bh2);
2199         }
2200         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2201         data2 = bh2->b_data;
2202
2203         memcpy(data2, de, len);
2204         memset(de, 0, len); /* wipe old data */
2205         de = (struct ext4_dir_entry_2 *) data2;
2206         top = data2 + len;
2207         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2208                 de = de2;
2209         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2210                                            (char *) de, blocksize);
2211
2212         if (csum_size)
2213                 ext4_initialize_dirent_tail(bh2, blocksize);
2214
2215         /* Initialize the root; the dot dirents already exist */
2216         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2217         de->rec_len = ext4_rec_len_to_disk(
2218                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2219         memset (&root->info, 0, sizeof(root->info));
2220         root->info.info_length = sizeof(root->info);
2221         if (ext4_hash_in_dirent(dir))
2222                 root->info.hash_version = DX_HASH_SIPHASH;
2223         else
2224                 root->info.hash_version =
2225                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2226
2227         entries = root->entries;
2228         dx_set_block(entries, 1);
2229         dx_set_count(entries, 1);
2230         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2231
2232         /* Initialize as for dx_probe */
2233         fname->hinfo.hash_version = root->info.hash_version;
2234         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2235                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2236         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2237
2238         /* casefolded encrypted hashes are computed on fname setup */
2239         if (!ext4_hash_in_dirent(dir))
2240                 ext4fs_dirhash(dir, fname_name(fname),
2241                                 fname_len(fname), &fname->hinfo);
2242
2243         memset(frames, 0, sizeof(frames));
2244         frame = frames;
2245         frame->entries = entries;
2246         frame->at = entries;
2247         frame->bh = bh;
2248
2249         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2250         if (retval)
2251                 goto out_frames;
2252         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2253         if (retval)
2254                 goto out_frames;
2255
2256         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2257         if (IS_ERR(de)) {
2258                 retval = PTR_ERR(de);
2259                 goto out_frames;
2260         }
2261
2262         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2263 out_frames:
2264         /*
2265          * Even if the block split failed, we have to properly write
2266          * out all the changes we did so far. Otherwise we can end up
2267          * with corrupted filesystem.
2268          */
2269         if (retval)
2270                 ext4_mark_inode_dirty(handle, dir);
2271         dx_release(frames);
2272         brelse(bh2);
2273         return retval;
2274 }
2275
2276 /*
2277  *      ext4_add_entry()
2278  *
2279  * adds a file entry to the specified directory, using the same
2280  * semantics as ext4_find_entry(). It returns NULL if it failed.
2281  *
2282  * NOTE!! The inode part of 'de' is left at 0 - which means you
2283  * may not sleep between calling this and putting something into
2284  * the entry, as someone else might have used it while you slept.
2285  */
2286 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2287                           struct inode *inode)
2288 {
2289         struct inode *dir = d_inode(dentry->d_parent);
2290         struct buffer_head *bh = NULL;
2291         struct ext4_dir_entry_2 *de;
2292         struct super_block *sb;
2293         struct ext4_filename fname;
2294         int     retval;
2295         int     dx_fallback=0;
2296         unsigned blocksize;
2297         ext4_lblk_t block, blocks;
2298         int     csum_size = 0;
2299
2300         if (ext4_has_metadata_csum(inode->i_sb))
2301                 csum_size = sizeof(struct ext4_dir_entry_tail);
2302
2303         sb = dir->i_sb;
2304         blocksize = sb->s_blocksize;
2305         if (!dentry->d_name.len)
2306                 return -EINVAL;
2307
2308         if (fscrypt_is_nokey_name(dentry))
2309                 return -ENOKEY;
2310
2311 #ifdef CONFIG_UNICODE
2312         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2313             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2314                 return -EINVAL;
2315 #endif
2316
2317         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2318         if (retval)
2319                 return retval;
2320
2321         if (ext4_has_inline_data(dir)) {
2322                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2323                 if (retval < 0)
2324                         goto out;
2325                 if (retval == 1) {
2326                         retval = 0;
2327                         goto out;
2328                 }
2329         }
2330
2331         if (is_dx(dir)) {
2332                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2333                 if (!retval || (retval != ERR_BAD_DX_DIR))
2334                         goto out;
2335                 /* Can we just ignore htree data? */
2336                 if (ext4_has_metadata_csum(sb)) {
2337                         EXT4_ERROR_INODE(dir,
2338                                 "Directory has corrupted htree index.");
2339                         retval = -EFSCORRUPTED;
2340                         goto out;
2341                 }
2342                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2343                 dx_fallback++;
2344                 retval = ext4_mark_inode_dirty(handle, dir);
2345                 if (unlikely(retval))
2346                         goto out;
2347         }
2348         blocks = dir->i_size >> sb->s_blocksize_bits;
2349         for (block = 0; block < blocks; block++) {
2350                 bh = ext4_read_dirblock(dir, block, DIRENT);
2351                 if (bh == NULL) {
2352                         bh = ext4_bread(handle, dir, block,
2353                                         EXT4_GET_BLOCKS_CREATE);
2354                         goto add_to_new_block;
2355                 }
2356                 if (IS_ERR(bh)) {
2357                         retval = PTR_ERR(bh);
2358                         bh = NULL;
2359                         goto out;
2360                 }
2361                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2362                                            NULL, bh);
2363                 if (retval != -ENOSPC)
2364                         goto out;
2365
2366                 if (blocks == 1 && !dx_fallback &&
2367                     ext4_has_feature_dir_index(sb)) {
2368                         retval = make_indexed_dir(handle, &fname, dir,
2369                                                   inode, bh);
2370                         bh = NULL; /* make_indexed_dir releases bh */
2371                         goto out;
2372                 }
2373                 brelse(bh);
2374         }
2375         bh = ext4_append(handle, dir, &block);
2376 add_to_new_block:
2377         if (IS_ERR(bh)) {
2378                 retval = PTR_ERR(bh);
2379                 bh = NULL;
2380                 goto out;
2381         }
2382         de = (struct ext4_dir_entry_2 *) bh->b_data;
2383         de->inode = 0;
2384         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2385
2386         if (csum_size)
2387                 ext4_initialize_dirent_tail(bh, blocksize);
2388
2389         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2390 out:
2391         ext4_fname_free_filename(&fname);
2392         brelse(bh);
2393         if (retval == 0)
2394                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2395         return retval;
2396 }
2397
2398 /*
2399  * Returns 0 for success, or a negative error value
2400  */
2401 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2402                              struct inode *dir, struct inode *inode)
2403 {
2404         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2405         struct dx_entry *entries, *at;
2406         struct buffer_head *bh;
2407         struct super_block *sb = dir->i_sb;
2408         struct ext4_dir_entry_2 *de;
2409         int restart;
2410         int err;
2411
2412 again:
2413         restart = 0;
2414         frame = dx_probe(fname, dir, NULL, frames);
2415         if (IS_ERR(frame))
2416                 return PTR_ERR(frame);
2417         entries = frame->entries;
2418         at = frame->at;
2419         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2420         if (IS_ERR(bh)) {
2421                 err = PTR_ERR(bh);
2422                 bh = NULL;
2423                 goto cleanup;
2424         }
2425
2426         BUFFER_TRACE(bh, "get_write_access");
2427         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2428         if (err)
2429                 goto journal_error;
2430
2431         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2432         if (err != -ENOSPC)
2433                 goto cleanup;
2434
2435         err = 0;
2436         /* Block full, should compress but for now just split */
2437         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2438                        dx_get_count(entries), dx_get_limit(entries)));
2439         /* Need to split index? */
2440         if (dx_get_count(entries) == dx_get_limit(entries)) {
2441                 ext4_lblk_t newblock;
2442                 int levels = frame - frames + 1;
2443                 unsigned int icount;
2444                 int add_level = 1;
2445                 struct dx_entry *entries2;
2446                 struct dx_node *node2;
2447                 struct buffer_head *bh2;
2448
2449                 while (frame > frames) {
2450                         if (dx_get_count((frame - 1)->entries) <
2451                             dx_get_limit((frame - 1)->entries)) {
2452                                 add_level = 0;
2453                                 break;
2454                         }
2455                         frame--; /* split higher index block */
2456                         at = frame->at;
2457                         entries = frame->entries;
2458                         restart = 1;
2459                 }
2460                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2461                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2462                                          "reach max htree level :%d",
2463                                          dir->i_ino, levels);
2464                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2465                                 ext4_warning(sb, "Large directory feature is "
2466                                                  "not enabled on this "
2467                                                  "filesystem");
2468                         }
2469                         err = -ENOSPC;
2470                         goto cleanup;
2471                 }
2472                 icount = dx_get_count(entries);
2473                 bh2 = ext4_append(handle, dir, &newblock);
2474                 if (IS_ERR(bh2)) {
2475                         err = PTR_ERR(bh2);
2476                         goto cleanup;
2477                 }
2478                 node2 = (struct dx_node *)(bh2->b_data);
2479                 entries2 = node2->entries;
2480                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2481                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2482                                                            sb->s_blocksize);
2483                 BUFFER_TRACE(frame->bh, "get_write_access");
2484                 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2485                                                     EXT4_JTR_NONE);
2486                 if (err)
2487                         goto journal_error;
2488                 if (!add_level) {
2489                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2490                         unsigned hash2 = dx_get_hash(entries + icount1);
2491                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2492                                        icount1, icount2));
2493
2494                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2495                         err = ext4_journal_get_write_access(handle, sb,
2496                                                             (frame - 1)->bh,
2497                                                             EXT4_JTR_NONE);
2498                         if (err)
2499                                 goto journal_error;
2500
2501                         memcpy((char *) entries2, (char *) (entries + icount1),
2502                                icount2 * sizeof(struct dx_entry));
2503                         dx_set_count(entries, icount1);
2504                         dx_set_count(entries2, icount2);
2505                         dx_set_limit(entries2, dx_node_limit(dir));
2506
2507                         /* Which index block gets the new entry? */
2508                         if (at - entries >= icount1) {
2509                                 frame->at = at - entries - icount1 + entries2;
2510                                 frame->entries = entries = entries2;
2511                                 swap(frame->bh, bh2);
2512                         }
2513                         dx_insert_block((frame - 1), hash2, newblock);
2514                         dxtrace(dx_show_index("node", frame->entries));
2515                         dxtrace(dx_show_index("node",
2516                                ((struct dx_node *) bh2->b_data)->entries));
2517                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2518                         if (err)
2519                                 goto journal_error;
2520                         brelse (bh2);
2521                         err = ext4_handle_dirty_dx_node(handle, dir,
2522                                                    (frame - 1)->bh);
2523                         if (err)
2524                                 goto journal_error;
2525                         err = ext4_handle_dirty_dx_node(handle, dir,
2526                                                         frame->bh);
2527                         if (restart || err)
2528                                 goto journal_error;
2529                 } else {
2530                         struct dx_root *dxroot;
2531                         memcpy((char *) entries2, (char *) entries,
2532                                icount * sizeof(struct dx_entry));
2533                         dx_set_limit(entries2, dx_node_limit(dir));
2534
2535                         /* Set up root */
2536                         dx_set_count(entries, 1);
2537                         dx_set_block(entries + 0, newblock);
2538                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2539                         dxroot->info.indirect_levels += 1;
2540                         dxtrace(printk(KERN_DEBUG
2541                                        "Creating %d level index...\n",
2542                                        dxroot->info.indirect_levels));
2543                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2544                         if (err)
2545                                 goto journal_error;
2546                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2547                         brelse(bh2);
2548                         restart = 1;
2549                         goto journal_error;
2550                 }
2551         }
2552         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2553         if (IS_ERR(de)) {
2554                 err = PTR_ERR(de);
2555                 goto cleanup;
2556         }
2557         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2558         goto cleanup;
2559
2560 journal_error:
2561         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2562 cleanup:
2563         brelse(bh);
2564         dx_release(frames);
2565         /* @restart is true means htree-path has been changed, we need to
2566          * repeat dx_probe() to find out valid htree-path
2567          */
2568         if (restart && err == 0)
2569                 goto again;
2570         return err;
2571 }
2572
2573 /*
2574  * ext4_generic_delete_entry deletes a directory entry by merging it
2575  * with the previous entry
2576  */
2577 int ext4_generic_delete_entry(struct inode *dir,
2578                               struct ext4_dir_entry_2 *de_del,
2579                               struct buffer_head *bh,
2580                               void *entry_buf,
2581                               int buf_size,
2582                               int csum_size)
2583 {
2584         struct ext4_dir_entry_2 *de, *pde;
2585         unsigned int blocksize = dir->i_sb->s_blocksize;
2586         int i;
2587
2588         i = 0;
2589         pde = NULL;
2590         de = (struct ext4_dir_entry_2 *)entry_buf;
2591         while (i < buf_size - csum_size) {
2592                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2593                                          entry_buf, buf_size, i))
2594                         return -EFSCORRUPTED;
2595                 if (de == de_del)  {
2596                         if (pde) {
2597                                 pde->rec_len = ext4_rec_len_to_disk(
2598                                         ext4_rec_len_from_disk(pde->rec_len,
2599                                                                blocksize) +
2600                                         ext4_rec_len_from_disk(de->rec_len,
2601                                                                blocksize),
2602                                         blocksize);
2603
2604                                 /* wipe entire dir_entry */
2605                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2606                                                                 blocksize));
2607                         } else {
2608                                 /* wipe dir_entry excluding the rec_len field */
2609                                 de->inode = 0;
2610                                 memset(&de->name_len, 0,
2611                                         ext4_rec_len_from_disk(de->rec_len,
2612                                                                 blocksize) -
2613                                         offsetof(struct ext4_dir_entry_2,
2614                                                                 name_len));
2615                         }
2616
2617                         inode_inc_iversion(dir);
2618                         return 0;
2619                 }
2620                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2621                 pde = de;
2622                 de = ext4_next_entry(de, blocksize);
2623         }
2624         return -ENOENT;
2625 }
2626
2627 static int ext4_delete_entry(handle_t *handle,
2628                              struct inode *dir,
2629                              struct ext4_dir_entry_2 *de_del,
2630                              struct buffer_head *bh)
2631 {
2632         int err, csum_size = 0;
2633
2634         if (ext4_has_inline_data(dir)) {
2635                 int has_inline_data = 1;
2636                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2637                                                &has_inline_data);
2638                 if (has_inline_data)
2639                         return err;
2640         }
2641
2642         if (ext4_has_metadata_csum(dir->i_sb))
2643                 csum_size = sizeof(struct ext4_dir_entry_tail);
2644
2645         BUFFER_TRACE(bh, "get_write_access");
2646         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2647                                             EXT4_JTR_NONE);
2648         if (unlikely(err))
2649                 goto out;
2650
2651         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2652                                         dir->i_sb->s_blocksize, csum_size);
2653         if (err)
2654                 goto out;
2655
2656         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2657         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2658         if (unlikely(err))
2659                 goto out;
2660
2661         return 0;
2662 out:
2663         if (err != -ENOENT)
2664                 ext4_std_error(dir->i_sb, err);
2665         return err;
2666 }
2667
2668 /*
2669  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2670  * since this indicates that nlinks count was previously 1 to avoid overflowing
2671  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2672  * that subdirectory link counts are not being maintained accurately.
2673  *
2674  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2675  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2676  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2677  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2678  */
2679 static void ext4_inc_count(struct inode *inode)
2680 {
2681         inc_nlink(inode);
2682         if (is_dx(inode) &&
2683             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2684                 set_nlink(inode, 1);
2685 }
2686
2687 /*
2688  * If a directory had nlink == 1, then we should let it be 1. This indicates
2689  * directory has >EXT4_LINK_MAX subdirs.
2690  */
2691 static void ext4_dec_count(struct inode *inode)
2692 {
2693         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2694                 drop_nlink(inode);
2695 }
2696
2697
2698 /*
2699  * Add non-directory inode to a directory. On success, the inode reference is
2700  * consumed by dentry is instantiation. This is also indicated by clearing of
2701  * *inodep pointer. On failure, the caller is responsible for dropping the
2702  * inode reference in the safe context.
2703  */
2704 static int ext4_add_nondir(handle_t *handle,
2705                 struct dentry *dentry, struct inode **inodep)
2706 {
2707         struct inode *dir = d_inode(dentry->d_parent);
2708         struct inode *inode = *inodep;
2709         int err = ext4_add_entry(handle, dentry, inode);
2710         if (!err) {
2711                 err = ext4_mark_inode_dirty(handle, inode);
2712                 if (IS_DIRSYNC(dir))
2713                         ext4_handle_sync(handle);
2714                 d_instantiate_new(dentry, inode);
2715                 *inodep = NULL;
2716                 return err;
2717         }
2718         drop_nlink(inode);
2719         ext4_orphan_add(handle, inode);
2720         unlock_new_inode(inode);
2721         return err;
2722 }
2723
2724 /*
2725  * By the time this is called, we already have created
2726  * the directory cache entry for the new file, but it
2727  * is so far negative - it has no inode.
2728  *
2729  * If the create succeeds, we fill in the inode information
2730  * with d_instantiate().
2731  */
2732 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2733                        struct dentry *dentry, umode_t mode, bool excl)
2734 {
2735         handle_t *handle;
2736         struct inode *inode;
2737         int err, credits, retries = 0;
2738
2739         err = dquot_initialize(dir);
2740         if (err)
2741                 return err;
2742
2743         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2744                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2745 retry:
2746         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2747                                             0, NULL, EXT4_HT_DIR, credits);
2748         handle = ext4_journal_current_handle();
2749         err = PTR_ERR(inode);
2750         if (!IS_ERR(inode)) {
2751                 inode->i_op = &ext4_file_inode_operations;
2752                 inode->i_fop = &ext4_file_operations;
2753                 ext4_set_aops(inode);
2754                 err = ext4_add_nondir(handle, dentry, &inode);
2755                 if (!err)
2756                         ext4_fc_track_create(handle, dentry);
2757         }
2758         if (handle)
2759                 ext4_journal_stop(handle);
2760         if (!IS_ERR_OR_NULL(inode))
2761                 iput(inode);
2762         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2763                 goto retry;
2764         return err;
2765 }
2766
2767 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2768                       struct dentry *dentry, umode_t mode, dev_t rdev)
2769 {
2770         handle_t *handle;
2771         struct inode *inode;
2772         int err, credits, retries = 0;
2773
2774         err = dquot_initialize(dir);
2775         if (err)
2776                 return err;
2777
2778         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2779                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2780 retry:
2781         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2782                                             0, NULL, EXT4_HT_DIR, credits);
2783         handle = ext4_journal_current_handle();
2784         err = PTR_ERR(inode);
2785         if (!IS_ERR(inode)) {
2786                 init_special_inode(inode, inode->i_mode, rdev);
2787                 inode->i_op = &ext4_special_inode_operations;
2788                 err = ext4_add_nondir(handle, dentry, &inode);
2789                 if (!err)
2790                         ext4_fc_track_create(handle, dentry);
2791         }
2792         if (handle)
2793                 ext4_journal_stop(handle);
2794         if (!IS_ERR_OR_NULL(inode))
2795                 iput(inode);
2796         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2797                 goto retry;
2798         return err;
2799 }
2800
2801 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2802                         struct dentry *dentry, umode_t mode)
2803 {
2804         handle_t *handle;
2805         struct inode *inode;
2806         int err, retries = 0;
2807
2808         err = dquot_initialize(dir);
2809         if (err)
2810                 return err;
2811
2812 retry:
2813         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2814                                             NULL, 0, NULL,
2815                                             EXT4_HT_DIR,
2816                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2817                           4 + EXT4_XATTR_TRANS_BLOCKS);
2818         handle = ext4_journal_current_handle();
2819         err = PTR_ERR(inode);
2820         if (!IS_ERR(inode)) {
2821                 inode->i_op = &ext4_file_inode_operations;
2822                 inode->i_fop = &ext4_file_operations;
2823                 ext4_set_aops(inode);
2824                 d_tmpfile(dentry, inode);
2825                 err = ext4_orphan_add(handle, inode);
2826                 if (err)
2827                         goto err_unlock_inode;
2828                 mark_inode_dirty(inode);
2829                 unlock_new_inode(inode);
2830         }
2831         if (handle)
2832                 ext4_journal_stop(handle);
2833         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2834                 goto retry;
2835         return err;
2836 err_unlock_inode:
2837         ext4_journal_stop(handle);
2838         unlock_new_inode(inode);
2839         return err;
2840 }
2841
2842 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2843                           struct ext4_dir_entry_2 *de,
2844                           int blocksize, int csum_size,
2845                           unsigned int parent_ino, int dotdot_real_len)
2846 {
2847         de->inode = cpu_to_le32(inode->i_ino);
2848         de->name_len = 1;
2849         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2850                                            blocksize);
2851         strcpy(de->name, ".");
2852         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2853
2854         de = ext4_next_entry(de, blocksize);
2855         de->inode = cpu_to_le32(parent_ino);
2856         de->name_len = 2;
2857         if (!dotdot_real_len)
2858                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2859                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2860                                         blocksize);
2861         else
2862                 de->rec_len = ext4_rec_len_to_disk(
2863                                         ext4_dir_rec_len(de->name_len, NULL),
2864                                         blocksize);
2865         strcpy(de->name, "..");
2866         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2867
2868         return ext4_next_entry(de, blocksize);
2869 }
2870
2871 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2872                              struct inode *inode)
2873 {
2874         struct buffer_head *dir_block = NULL;
2875         struct ext4_dir_entry_2 *de;
2876         ext4_lblk_t block = 0;
2877         unsigned int blocksize = dir->i_sb->s_blocksize;
2878         int csum_size = 0;
2879         int err;
2880
2881         if (ext4_has_metadata_csum(dir->i_sb))
2882                 csum_size = sizeof(struct ext4_dir_entry_tail);
2883
2884         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2885                 err = ext4_try_create_inline_dir(handle, dir, inode);
2886                 if (err < 0 && err != -ENOSPC)
2887                         goto out;
2888                 if (!err)
2889                         goto out;
2890         }
2891
2892         inode->i_size = 0;
2893         dir_block = ext4_append(handle, inode, &block);
2894         if (IS_ERR(dir_block))
2895                 return PTR_ERR(dir_block);
2896         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2897         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2898         set_nlink(inode, 2);
2899         if (csum_size)
2900                 ext4_initialize_dirent_tail(dir_block, blocksize);
2901
2902         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2903         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2904         if (err)
2905                 goto out;
2906         set_buffer_verified(dir_block);
2907 out:
2908         brelse(dir_block);
2909         return err;
2910 }
2911
2912 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2913                       struct dentry *dentry, umode_t mode)
2914 {
2915         handle_t *handle;
2916         struct inode *inode;
2917         int err, err2 = 0, credits, retries = 0;
2918
2919         if (EXT4_DIR_LINK_MAX(dir))
2920                 return -EMLINK;
2921
2922         err = dquot_initialize(dir);
2923         if (err)
2924                 return err;
2925
2926         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2927                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2928 retry:
2929         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2930                                             &dentry->d_name,
2931                                             0, NULL, EXT4_HT_DIR, credits);
2932         handle = ext4_journal_current_handle();
2933         err = PTR_ERR(inode);
2934         if (IS_ERR(inode))
2935                 goto out_stop;
2936
2937         inode->i_op = &ext4_dir_inode_operations;
2938         inode->i_fop = &ext4_dir_operations;
2939         err = ext4_init_new_dir(handle, dir, inode);
2940         if (err)
2941                 goto out_clear_inode;
2942         err = ext4_mark_inode_dirty(handle, inode);
2943         if (!err)
2944                 err = ext4_add_entry(handle, dentry, inode);
2945         if (err) {
2946 out_clear_inode:
2947                 clear_nlink(inode);
2948                 ext4_orphan_add(handle, inode);
2949                 unlock_new_inode(inode);
2950                 err2 = ext4_mark_inode_dirty(handle, inode);
2951                 if (unlikely(err2))
2952                         err = err2;
2953                 ext4_journal_stop(handle);
2954                 iput(inode);
2955                 goto out_retry;
2956         }
2957         ext4_inc_count(dir);
2958
2959         ext4_update_dx_flag(dir);
2960         err = ext4_mark_inode_dirty(handle, dir);
2961         if (err)
2962                 goto out_clear_inode;
2963         d_instantiate_new(dentry, inode);
2964         ext4_fc_track_create(handle, dentry);
2965         if (IS_DIRSYNC(dir))
2966                 ext4_handle_sync(handle);
2967
2968 out_stop:
2969         if (handle)
2970                 ext4_journal_stop(handle);
2971 out_retry:
2972         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2973                 goto retry;
2974         return err;
2975 }
2976
2977 /*
2978  * routine to check that the specified directory is empty (for rmdir)
2979  */
2980 bool ext4_empty_dir(struct inode *inode)
2981 {
2982         unsigned int offset;
2983         struct buffer_head *bh;
2984         struct ext4_dir_entry_2 *de;
2985         struct super_block *sb;
2986
2987         if (ext4_has_inline_data(inode)) {
2988                 int has_inline_data = 1;
2989                 int ret;
2990
2991                 ret = empty_inline_dir(inode, &has_inline_data);
2992                 if (has_inline_data)
2993                         return ret;
2994         }
2995
2996         sb = inode->i_sb;
2997         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
2998                                         ext4_dir_rec_len(2, NULL)) {
2999                 EXT4_ERROR_INODE(inode, "invalid size");
3000                 return true;
3001         }
3002         /* The first directory block must not be a hole,
3003          * so treat it as DIRENT_HTREE
3004          */
3005         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3006         if (IS_ERR(bh))
3007                 return true;
3008
3009         de = (struct ext4_dir_entry_2 *) bh->b_data;
3010         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3011                                  0) ||
3012             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3013                 ext4_warning_inode(inode, "directory missing '.'");
3014                 brelse(bh);
3015                 return true;
3016         }
3017         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3018         de = ext4_next_entry(de, sb->s_blocksize);
3019         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3020                                  offset) ||
3021             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3022                 ext4_warning_inode(inode, "directory missing '..'");
3023                 brelse(bh);
3024                 return true;
3025         }
3026         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3027         while (offset < inode->i_size) {
3028                 if (!(offset & (sb->s_blocksize - 1))) {
3029                         unsigned int lblock;
3030                         brelse(bh);
3031                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3032                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3033                         if (bh == NULL) {
3034                                 offset += sb->s_blocksize;
3035                                 continue;
3036                         }
3037                         if (IS_ERR(bh))
3038                                 return true;
3039                 }
3040                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3041                                         (offset & (sb->s_blocksize - 1)));
3042                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3043                                          bh->b_data, bh->b_size, offset)) {
3044                         offset = (offset | (sb->s_blocksize - 1)) + 1;
3045                         continue;
3046                 }
3047                 if (le32_to_cpu(de->inode)) {
3048                         brelse(bh);
3049                         return false;
3050                 }
3051                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3052         }
3053         brelse(bh);
3054         return true;
3055 }
3056
3057 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3058 {
3059         int retval;
3060         struct inode *inode;
3061         struct buffer_head *bh;
3062         struct ext4_dir_entry_2 *de;
3063         handle_t *handle = NULL;
3064
3065         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3066                 return -EIO;
3067
3068         /* Initialize quotas before so that eventual writes go in
3069          * separate transaction */
3070         retval = dquot_initialize(dir);
3071         if (retval)
3072                 return retval;
3073         retval = dquot_initialize(d_inode(dentry));
3074         if (retval)
3075                 return retval;
3076
3077         retval = -ENOENT;
3078         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3079         if (IS_ERR(bh))
3080                 return PTR_ERR(bh);
3081         if (!bh)
3082                 goto end_rmdir;
3083
3084         inode = d_inode(dentry);
3085
3086         retval = -EFSCORRUPTED;
3087         if (le32_to_cpu(de->inode) != inode->i_ino)
3088                 goto end_rmdir;
3089
3090         retval = -ENOTEMPTY;
3091         if (!ext4_empty_dir(inode))
3092                 goto end_rmdir;
3093
3094         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3095                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3096         if (IS_ERR(handle)) {
3097                 retval = PTR_ERR(handle);
3098                 handle = NULL;
3099                 goto end_rmdir;
3100         }
3101
3102         if (IS_DIRSYNC(dir))
3103                 ext4_handle_sync(handle);
3104
3105         retval = ext4_delete_entry(handle, dir, de, bh);
3106         if (retval)
3107                 goto end_rmdir;
3108         if (!EXT4_DIR_LINK_EMPTY(inode))
3109                 ext4_warning_inode(inode,
3110                              "empty directory '%.*s' has too many links (%u)",
3111                              dentry->d_name.len, dentry->d_name.name,
3112                              inode->i_nlink);
3113         inode_inc_iversion(inode);
3114         clear_nlink(inode);
3115         /* There's no need to set i_disksize: the fact that i_nlink is
3116          * zero will ensure that the right thing happens during any
3117          * recovery. */
3118         inode->i_size = 0;
3119         ext4_orphan_add(handle, inode);
3120         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3121         retval = ext4_mark_inode_dirty(handle, inode);
3122         if (retval)
3123                 goto end_rmdir;
3124         ext4_dec_count(dir);
3125         ext4_update_dx_flag(dir);
3126         ext4_fc_track_unlink(handle, dentry);
3127         retval = ext4_mark_inode_dirty(handle, dir);
3128
3129 #ifdef CONFIG_UNICODE
3130         /* VFS negative dentries are incompatible with Encoding and
3131          * Case-insensitiveness. Eventually we'll want avoid
3132          * invalidating the dentries here, alongside with returning the
3133          * negative dentries at ext4_lookup(), when it is better
3134          * supported by the VFS for the CI case.
3135          */
3136         if (IS_CASEFOLDED(dir))
3137                 d_invalidate(dentry);
3138 #endif
3139
3140 end_rmdir:
3141         brelse(bh);
3142         if (handle)
3143                 ext4_journal_stop(handle);
3144         return retval;
3145 }
3146
3147 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3148                   struct inode *inode)
3149 {
3150         int retval = -ENOENT;
3151         struct buffer_head *bh;
3152         struct ext4_dir_entry_2 *de;
3153         int skip_remove_dentry = 0;
3154
3155         bh = ext4_find_entry(dir, d_name, &de, NULL);
3156         if (IS_ERR(bh))
3157                 return PTR_ERR(bh);
3158
3159         if (!bh)
3160                 return -ENOENT;
3161
3162         if (le32_to_cpu(de->inode) != inode->i_ino) {
3163                 /*
3164                  * It's okay if we find dont find dentry which matches
3165                  * the inode. That's because it might have gotten
3166                  * renamed to a different inode number
3167                  */
3168                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3169                         skip_remove_dentry = 1;
3170                 else
3171                         goto out;
3172         }
3173
3174         if (IS_DIRSYNC(dir))
3175                 ext4_handle_sync(handle);
3176
3177         if (!skip_remove_dentry) {
3178                 retval = ext4_delete_entry(handle, dir, de, bh);
3179                 if (retval)
3180                         goto out;
3181                 dir->i_ctime = dir->i_mtime = current_time(dir);
3182                 ext4_update_dx_flag(dir);
3183                 retval = ext4_mark_inode_dirty(handle, dir);
3184                 if (retval)
3185                         goto out;
3186         } else {
3187                 retval = 0;
3188         }
3189         if (inode->i_nlink == 0)
3190                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3191                                    d_name->len, d_name->name);
3192         else
3193                 drop_nlink(inode);
3194         if (!inode->i_nlink)
3195                 ext4_orphan_add(handle, inode);
3196         inode->i_ctime = current_time(inode);
3197         retval = ext4_mark_inode_dirty(handle, inode);
3198
3199 out:
3200         brelse(bh);
3201         return retval;
3202 }
3203
3204 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3205 {
3206         handle_t *handle;
3207         int retval;
3208
3209         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3210                 return -EIO;
3211
3212         trace_ext4_unlink_enter(dir, dentry);
3213         /*
3214          * Initialize quotas before so that eventual writes go
3215          * in separate transaction
3216          */
3217         retval = dquot_initialize(dir);
3218         if (retval)
3219                 goto out_trace;
3220         retval = dquot_initialize(d_inode(dentry));
3221         if (retval)
3222                 goto out_trace;
3223
3224         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3225                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3226         if (IS_ERR(handle)) {
3227                 retval = PTR_ERR(handle);
3228                 goto out_trace;
3229         }
3230
3231         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3232         if (!retval)
3233                 ext4_fc_track_unlink(handle, dentry);
3234 #ifdef CONFIG_UNICODE
3235         /* VFS negative dentries are incompatible with Encoding and
3236          * Case-insensitiveness. Eventually we'll want avoid
3237          * invalidating the dentries here, alongside with returning the
3238          * negative dentries at ext4_lookup(), when it is  better
3239          * supported by the VFS for the CI case.
3240          */
3241         if (IS_CASEFOLDED(dir))
3242                 d_invalidate(dentry);
3243 #endif
3244         if (handle)
3245                 ext4_journal_stop(handle);
3246
3247 out_trace:
3248         trace_ext4_unlink_exit(dentry, retval);
3249         return retval;
3250 }
3251
3252 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3253                         struct dentry *dentry, const char *symname)
3254 {
3255         handle_t *handle;
3256         struct inode *inode;
3257         int err, len = strlen(symname);
3258         int credits;
3259         struct fscrypt_str disk_link;
3260
3261         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3262                 return -EIO;
3263
3264         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3265                                       &disk_link);
3266         if (err)
3267                 return err;
3268
3269         err = dquot_initialize(dir);
3270         if (err)
3271                 return err;
3272
3273         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3274                 /*
3275                  * For non-fast symlinks, we just allocate inode and put it on
3276                  * orphan list in the first transaction => we need bitmap,
3277                  * group descriptor, sb, inode block, quota blocks, and
3278                  * possibly selinux xattr blocks.
3279                  */
3280                 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3281                           EXT4_XATTR_TRANS_BLOCKS;
3282         } else {
3283                 /*
3284                  * Fast symlink. We have to add entry to directory
3285                  * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3286                  * allocate new inode (bitmap, group descriptor, inode block,
3287                  * quota blocks, sb is already counted in previous macros).
3288                  */
3289                 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3290                           EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3291         }
3292
3293         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3294                                             &dentry->d_name, 0, NULL,
3295                                             EXT4_HT_DIR, credits);
3296         handle = ext4_journal_current_handle();
3297         if (IS_ERR(inode)) {
3298                 if (handle)
3299                         ext4_journal_stop(handle);
3300                 return PTR_ERR(inode);
3301         }
3302
3303         if (IS_ENCRYPTED(inode)) {
3304                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3305                 if (err)
3306                         goto err_drop_inode;
3307                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3308         }
3309
3310         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3311                 if (!IS_ENCRYPTED(inode))
3312                         inode->i_op = &ext4_symlink_inode_operations;
3313                 inode_nohighmem(inode);
3314                 ext4_set_aops(inode);
3315                 /*
3316                  * We cannot call page_symlink() with transaction started
3317                  * because it calls into ext4_write_begin() which can wait
3318                  * for transaction commit if we are running out of space
3319                  * and thus we deadlock. So we have to stop transaction now
3320                  * and restart it when symlink contents is written.
3321                  *
3322                  * To keep fs consistent in case of crash, we have to put inode
3323                  * to orphan list in the mean time.
3324                  */
3325                 drop_nlink(inode);
3326                 err = ext4_orphan_add(handle, inode);
3327                 if (handle)
3328                         ext4_journal_stop(handle);
3329                 handle = NULL;
3330                 if (err)
3331                         goto err_drop_inode;
3332                 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3333                 if (err)
3334                         goto err_drop_inode;
3335                 /*
3336                  * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3337                  * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3338                  */
3339                 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3340                                 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3341                                 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3342                 if (IS_ERR(handle)) {
3343                         err = PTR_ERR(handle);
3344                         handle = NULL;
3345                         goto err_drop_inode;
3346                 }
3347                 set_nlink(inode, 1);
3348                 err = ext4_orphan_del(handle, inode);
3349                 if (err)
3350                         goto err_drop_inode;
3351         } else {
3352                 /* clear the extent format for fast symlink */
3353                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3354                 if (!IS_ENCRYPTED(inode)) {
3355                         inode->i_op = &ext4_fast_symlink_inode_operations;
3356                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3357                 }
3358                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3359                        disk_link.len);
3360                 inode->i_size = disk_link.len - 1;
3361         }
3362         EXT4_I(inode)->i_disksize = inode->i_size;
3363         err = ext4_add_nondir(handle, dentry, &inode);
3364         if (handle)
3365                 ext4_journal_stop(handle);
3366         if (inode)
3367                 iput(inode);
3368         goto out_free_encrypted_link;
3369
3370 err_drop_inode:
3371         if (handle)
3372                 ext4_journal_stop(handle);
3373         clear_nlink(inode);
3374         unlock_new_inode(inode);
3375         iput(inode);
3376 out_free_encrypted_link:
3377         if (disk_link.name != (unsigned char *)symname)
3378                 kfree(disk_link.name);
3379         return err;
3380 }
3381
3382 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3383 {
3384         handle_t *handle;
3385         int err, retries = 0;
3386 retry:
3387         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3388                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3389                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3390         if (IS_ERR(handle))
3391                 return PTR_ERR(handle);
3392
3393         if (IS_DIRSYNC(dir))
3394                 ext4_handle_sync(handle);
3395
3396         inode->i_ctime = current_time(inode);
3397         ext4_inc_count(inode);
3398         ihold(inode);
3399
3400         err = ext4_add_entry(handle, dentry, inode);
3401         if (!err) {
3402                 err = ext4_mark_inode_dirty(handle, inode);
3403                 /* this can happen only for tmpfile being
3404                  * linked the first time
3405                  */
3406                 if (inode->i_nlink == 1)
3407                         ext4_orphan_del(handle, inode);
3408                 d_instantiate(dentry, inode);
3409                 ext4_fc_track_link(handle, dentry);
3410         } else {
3411                 drop_nlink(inode);
3412                 iput(inode);
3413         }
3414         ext4_journal_stop(handle);
3415         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3416                 goto retry;
3417         return err;
3418 }
3419
3420 static int ext4_link(struct dentry *old_dentry,
3421                      struct inode *dir, struct dentry *dentry)
3422 {
3423         struct inode *inode = d_inode(old_dentry);
3424         int err;
3425
3426         if (inode->i_nlink >= EXT4_LINK_MAX)
3427                 return -EMLINK;
3428
3429         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3430         if (err)
3431                 return err;
3432
3433         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3434             (!projid_eq(EXT4_I(dir)->i_projid,
3435                         EXT4_I(old_dentry->d_inode)->i_projid)))
3436                 return -EXDEV;
3437
3438         err = dquot_initialize(dir);
3439         if (err)
3440                 return err;
3441         return __ext4_link(dir, inode, dentry);
3442 }
3443
3444 /*
3445  * Try to find buffer head where contains the parent block.
3446  * It should be the inode block if it is inlined or the 1st block
3447  * if it is a normal dir.
3448  */
3449 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3450                                         struct inode *inode,
3451                                         int *retval,
3452                                         struct ext4_dir_entry_2 **parent_de,
3453                                         int *inlined)
3454 {
3455         struct buffer_head *bh;
3456
3457         if (!ext4_has_inline_data(inode)) {
3458                 /* The first directory block must not be a hole, so
3459                  * treat it as DIRENT_HTREE
3460                  */
3461                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3462                 if (IS_ERR(bh)) {
3463                         *retval = PTR_ERR(bh);
3464                         return NULL;
3465                 }
3466                 *parent_de = ext4_next_entry(
3467                                         (struct ext4_dir_entry_2 *)bh->b_data,
3468                                         inode->i_sb->s_blocksize);
3469                 return bh;
3470         }
3471
3472         *inlined = 1;
3473         return ext4_get_first_inline_block(inode, parent_de, retval);
3474 }
3475
3476 struct ext4_renament {
3477         struct inode *dir;
3478         struct dentry *dentry;
3479         struct inode *inode;
3480         bool is_dir;
3481         int dir_nlink_delta;
3482
3483         /* entry for "dentry" */
3484         struct buffer_head *bh;
3485         struct ext4_dir_entry_2 *de;
3486         int inlined;
3487
3488         /* entry for ".." in inode if it's a directory */
3489         struct buffer_head *dir_bh;
3490         struct ext4_dir_entry_2 *parent_de;
3491         int dir_inlined;
3492 };
3493
3494 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3495 {
3496         int retval;
3497
3498         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3499                                               &retval, &ent->parent_de,
3500                                               &ent->dir_inlined);
3501         if (!ent->dir_bh)
3502                 return retval;
3503         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3504                 return -EFSCORRUPTED;
3505         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3506         return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3507                                              ent->dir_bh, EXT4_JTR_NONE);
3508 }
3509
3510 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3511                                   unsigned dir_ino)
3512 {
3513         int retval;
3514
3515         ent->parent_de->inode = cpu_to_le32(dir_ino);
3516         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3517         if (!ent->dir_inlined) {
3518                 if (is_dx(ent->inode)) {
3519                         retval = ext4_handle_dirty_dx_node(handle,
3520                                                            ent->inode,
3521                                                            ent->dir_bh);
3522                 } else {
3523                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3524                                                             ent->dir_bh);
3525                 }
3526         } else {
3527                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3528         }
3529         if (retval) {
3530                 ext4_std_error(ent->dir->i_sb, retval);
3531                 return retval;
3532         }
3533         return 0;
3534 }
3535
3536 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3537                        unsigned ino, unsigned file_type)
3538 {
3539         int retval, retval2;
3540
3541         BUFFER_TRACE(ent->bh, "get write access");
3542         retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3543                                                EXT4_JTR_NONE);
3544         if (retval)
3545                 return retval;
3546         ent->de->inode = cpu_to_le32(ino);
3547         if (ext4_has_feature_filetype(ent->dir->i_sb))
3548                 ent->de->file_type = file_type;
3549         inode_inc_iversion(ent->dir);
3550         ent->dir->i_ctime = ent->dir->i_mtime =
3551                 current_time(ent->dir);
3552         retval = ext4_mark_inode_dirty(handle, ent->dir);
3553         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3554         if (!ent->inlined) {
3555                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3556                 if (unlikely(retval2)) {
3557                         ext4_std_error(ent->dir->i_sb, retval2);
3558                         return retval2;
3559                 }
3560         }
3561         return retval;
3562 }
3563
3564 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3565                           unsigned ino, unsigned file_type)
3566 {
3567         struct ext4_renament old = *ent;
3568         int retval = 0;
3569
3570         /*
3571          * old->de could have moved from under us during make indexed dir,
3572          * so the old->de may no longer valid and need to find it again
3573          * before reset old inode info.
3574          */
3575         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3576         if (IS_ERR(old.bh))
3577                 retval = PTR_ERR(old.bh);
3578         if (!old.bh)
3579                 retval = -ENOENT;
3580         if (retval) {
3581                 ext4_std_error(old.dir->i_sb, retval);
3582                 return;
3583         }
3584
3585         ext4_setent(handle, &old, ino, file_type);
3586         brelse(old.bh);
3587 }
3588
3589 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3590                                   const struct qstr *d_name)
3591 {
3592         int retval = -ENOENT;
3593         struct buffer_head *bh;
3594         struct ext4_dir_entry_2 *de;
3595
3596         bh = ext4_find_entry(dir, d_name, &de, NULL);
3597         if (IS_ERR(bh))
3598                 return PTR_ERR(bh);
3599         if (bh) {
3600                 retval = ext4_delete_entry(handle, dir, de, bh);
3601                 brelse(bh);
3602         }
3603         return retval;
3604 }
3605
3606 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3607                                int force_reread)
3608 {
3609         int retval;
3610         /*
3611          * ent->de could have moved from under us during htree split, so make
3612          * sure that we are deleting the right entry.  We might also be pointing
3613          * to a stale entry in the unused part of ent->bh so just checking inum
3614          * and the name isn't enough.
3615          */
3616         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3617             ent->de->name_len != ent->dentry->d_name.len ||
3618             strncmp(ent->de->name, ent->dentry->d_name.name,
3619                     ent->de->name_len) ||
3620             force_reread) {
3621                 retval = ext4_find_delete_entry(handle, ent->dir,
3622                                                 &ent->dentry->d_name);
3623         } else {
3624                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3625                 if (retval == -ENOENT) {
3626                         retval = ext4_find_delete_entry(handle, ent->dir,
3627                                                         &ent->dentry->d_name);
3628                 }
3629         }
3630
3631         if (retval) {
3632                 ext4_warning_inode(ent->dir,
3633                                    "Deleting old file: nlink %d, error=%d",
3634                                    ent->dir->i_nlink, retval);
3635         }
3636 }
3637
3638 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3639 {
3640         if (ent->dir_nlink_delta) {
3641                 if (ent->dir_nlink_delta == -1)
3642                         ext4_dec_count(ent->dir);
3643                 else
3644                         ext4_inc_count(ent->dir);
3645                 ext4_mark_inode_dirty(handle, ent->dir);
3646         }
3647 }
3648
3649 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3650                                               struct ext4_renament *ent,
3651                                               int credits, handle_t **h)
3652 {
3653         struct inode *wh;
3654         handle_t *handle;
3655         int retries = 0;
3656
3657         /*
3658          * for inode block, sb block, group summaries,
3659          * and inode bitmap
3660          */
3661         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3662                     EXT4_XATTR_TRANS_BLOCKS + 4);
3663 retry:
3664         wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3665                                          S_IFCHR | WHITEOUT_MODE,
3666                                          &ent->dentry->d_name, 0, NULL,
3667                                          EXT4_HT_DIR, credits);
3668
3669         handle = ext4_journal_current_handle();
3670         if (IS_ERR(wh)) {
3671                 if (handle)
3672                         ext4_journal_stop(handle);
3673                 if (PTR_ERR(wh) == -ENOSPC &&
3674                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3675                         goto retry;
3676         } else {
3677                 *h = handle;
3678                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3679                 wh->i_op = &ext4_special_inode_operations;
3680         }
3681         return wh;
3682 }
3683
3684 /*
3685  * Anybody can rename anything with this: the permission checks are left to the
3686  * higher-level routines.
3687  *
3688  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3689  * while new_{dentry,inode) refers to the destination dentry/inode
3690  * This comes from rename(const char *oldpath, const char *newpath)
3691  */
3692 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3693                        struct dentry *old_dentry, struct inode *new_dir,
3694                        struct dentry *new_dentry, unsigned int flags)
3695 {
3696         handle_t *handle = NULL;
3697         struct ext4_renament old = {
3698                 .dir = old_dir,
3699                 .dentry = old_dentry,
3700                 .inode = d_inode(old_dentry),
3701         };
3702         struct ext4_renament new = {
3703                 .dir = new_dir,
3704                 .dentry = new_dentry,
3705                 .inode = d_inode(new_dentry),
3706         };
3707         int force_reread;
3708         int retval;
3709         struct inode *whiteout = NULL;
3710         int credits;
3711         u8 old_file_type;
3712
3713         if (new.inode && new.inode->i_nlink == 0) {
3714                 EXT4_ERROR_INODE(new.inode,
3715                                  "target of rename is already freed");
3716                 return -EFSCORRUPTED;
3717         }
3718
3719         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3720             (!projid_eq(EXT4_I(new_dir)->i_projid,
3721                         EXT4_I(old_dentry->d_inode)->i_projid)))
3722                 return -EXDEV;
3723
3724         retval = dquot_initialize(old.dir);
3725         if (retval)
3726                 return retval;
3727         retval = dquot_initialize(new.dir);
3728         if (retval)
3729                 return retval;
3730
3731         /* Initialize quotas before so that eventual writes go
3732          * in separate transaction */
3733         if (new.inode) {
3734                 retval = dquot_initialize(new.inode);
3735                 if (retval)
3736                         return retval;
3737         }
3738
3739         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3740         if (IS_ERR(old.bh))
3741                 return PTR_ERR(old.bh);
3742         /*
3743          *  Check for inode number is _not_ due to possible IO errors.
3744          *  We might rmdir the source, keep it as pwd of some process
3745          *  and merrily kill the link to whatever was created under the
3746          *  same name. Goodbye sticky bit ;-<
3747          */
3748         retval = -ENOENT;
3749         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3750                 goto release_bh;
3751
3752         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3753                                  &new.de, &new.inlined);
3754         if (IS_ERR(new.bh)) {
3755                 retval = PTR_ERR(new.bh);
3756                 new.bh = NULL;
3757                 goto release_bh;
3758         }
3759         if (new.bh) {
3760                 if (!new.inode) {
3761                         brelse(new.bh);
3762                         new.bh = NULL;
3763                 }
3764         }
3765         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3766                 ext4_alloc_da_blocks(old.inode);
3767
3768         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3769                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3770         if (!(flags & RENAME_WHITEOUT)) {
3771                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3772                 if (IS_ERR(handle)) {
3773                         retval = PTR_ERR(handle);
3774                         goto release_bh;
3775                 }
3776         } else {
3777                 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3778                 if (IS_ERR(whiteout)) {
3779                         retval = PTR_ERR(whiteout);
3780                         goto release_bh;
3781                 }
3782         }
3783
3784         old_file_type = old.de->file_type;
3785         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3786                 ext4_handle_sync(handle);
3787
3788         if (S_ISDIR(old.inode->i_mode)) {
3789                 if (new.inode) {
3790                         retval = -ENOTEMPTY;
3791                         if (!ext4_empty_dir(new.inode))
3792                                 goto end_rename;
3793                 } else {
3794                         retval = -EMLINK;
3795                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3796                                 goto end_rename;
3797                 }
3798                 retval = ext4_rename_dir_prepare(handle, &old);
3799                 if (retval)
3800                         goto end_rename;
3801         }
3802         /*
3803          * If we're renaming a file within an inline_data dir and adding or
3804          * setting the new dirent causes a conversion from inline_data to
3805          * extents/blockmap, we need to force the dirent delete code to
3806          * re-read the directory, or else we end up trying to delete a dirent
3807          * from what is now the extent tree root (or a block map).
3808          */
3809         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3810                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3811
3812         if (whiteout) {
3813                 /*
3814                  * Do this before adding a new entry, so the old entry is sure
3815                  * to be still pointing to the valid old entry.
3816                  */
3817                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3818                                      EXT4_FT_CHRDEV);
3819                 if (retval)
3820                         goto end_rename;
3821                 retval = ext4_mark_inode_dirty(handle, whiteout);
3822                 if (unlikely(retval))
3823                         goto end_rename;
3824
3825         }
3826         if (!new.bh) {
3827                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3828                 if (retval)
3829                         goto end_rename;
3830         } else {
3831                 retval = ext4_setent(handle, &new,
3832                                      old.inode->i_ino, old_file_type);
3833                 if (retval)
3834                         goto end_rename;
3835         }
3836         if (force_reread)
3837                 force_reread = !ext4_test_inode_flag(new.dir,
3838                                                      EXT4_INODE_INLINE_DATA);
3839
3840         /*
3841          * Like most other Unix systems, set the ctime for inodes on a
3842          * rename.
3843          */
3844         old.inode->i_ctime = current_time(old.inode);
3845         retval = ext4_mark_inode_dirty(handle, old.inode);
3846         if (unlikely(retval))
3847                 goto end_rename;
3848
3849         if (!whiteout) {
3850                 /*
3851                  * ok, that's it
3852                  */
3853                 ext4_rename_delete(handle, &old, force_reread);
3854         }
3855
3856         if (new.inode) {
3857                 ext4_dec_count(new.inode);
3858                 new.inode->i_ctime = current_time(new.inode);
3859         }
3860         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3861         ext4_update_dx_flag(old.dir);
3862         if (old.dir_bh) {
3863                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3864                 if (retval)
3865                         goto end_rename;
3866
3867                 ext4_dec_count(old.dir);
3868                 if (new.inode) {
3869                         /* checked ext4_empty_dir above, can't have another
3870                          * parent, ext4_dec_count() won't work for many-linked
3871                          * dirs */
3872                         clear_nlink(new.inode);
3873                 } else {
3874                         ext4_inc_count(new.dir);
3875                         ext4_update_dx_flag(new.dir);
3876                         retval = ext4_mark_inode_dirty(handle, new.dir);
3877                         if (unlikely(retval))
3878                                 goto end_rename;
3879                 }
3880         }
3881         retval = ext4_mark_inode_dirty(handle, old.dir);
3882         if (unlikely(retval))
3883                 goto end_rename;
3884
3885         if (S_ISDIR(old.inode->i_mode)) {
3886                 /*
3887                  * We disable fast commits here that's because the
3888                  * replay code is not yet capable of changing dot dot
3889                  * dirents in directories.
3890                  */
3891                 ext4_fc_mark_ineligible(old.inode->i_sb,
3892                         EXT4_FC_REASON_RENAME_DIR);
3893         } else {
3894                 if (new.inode)
3895                         ext4_fc_track_unlink(handle, new.dentry);
3896                 __ext4_fc_track_link(handle, old.inode, new.dentry);
3897                 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3898                 if (whiteout)
3899                         __ext4_fc_track_create(handle, whiteout, old.dentry);
3900         }
3901
3902         if (new.inode) {
3903                 retval = ext4_mark_inode_dirty(handle, new.inode);
3904                 if (unlikely(retval))
3905                         goto end_rename;
3906                 if (!new.inode->i_nlink)
3907                         ext4_orphan_add(handle, new.inode);
3908         }
3909         retval = 0;
3910
3911 end_rename:
3912         if (whiteout) {
3913                 if (retval) {
3914                         ext4_resetent(handle, &old,
3915                                       old.inode->i_ino, old_file_type);
3916                         drop_nlink(whiteout);
3917                         ext4_orphan_add(handle, whiteout);
3918                 }
3919                 unlock_new_inode(whiteout);
3920                 ext4_journal_stop(handle);
3921                 iput(whiteout);
3922         } else {
3923                 ext4_journal_stop(handle);
3924         }
3925 release_bh:
3926         brelse(old.dir_bh);
3927         brelse(old.bh);
3928         brelse(new.bh);
3929         return retval;
3930 }
3931
3932 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3933                              struct inode *new_dir, struct dentry *new_dentry)
3934 {
3935         handle_t *handle = NULL;
3936         struct ext4_renament old = {
3937                 .dir = old_dir,
3938                 .dentry = old_dentry,
3939                 .inode = d_inode(old_dentry),
3940         };
3941         struct ext4_renament new = {
3942                 .dir = new_dir,
3943                 .dentry = new_dentry,
3944                 .inode = d_inode(new_dentry),
3945         };
3946         u8 new_file_type;
3947         int retval;
3948         struct timespec64 ctime;
3949
3950         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3951              !projid_eq(EXT4_I(new_dir)->i_projid,
3952                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3953             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3954              !projid_eq(EXT4_I(old_dir)->i_projid,
3955                         EXT4_I(new_dentry->d_inode)->i_projid)))
3956                 return -EXDEV;
3957
3958         retval = dquot_initialize(old.dir);
3959         if (retval)
3960                 return retval;
3961         retval = dquot_initialize(new.dir);
3962         if (retval)
3963                 return retval;
3964
3965         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3966                                  &old.de, &old.inlined);
3967         if (IS_ERR(old.bh))
3968                 return PTR_ERR(old.bh);
3969         /*
3970          *  Check for inode number is _not_ due to possible IO errors.
3971          *  We might rmdir the source, keep it as pwd of some process
3972          *  and merrily kill the link to whatever was created under the
3973          *  same name. Goodbye sticky bit ;-<
3974          */
3975         retval = -ENOENT;
3976         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3977                 goto end_rename;
3978
3979         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3980                                  &new.de, &new.inlined);
3981         if (IS_ERR(new.bh)) {
3982                 retval = PTR_ERR(new.bh);
3983                 new.bh = NULL;
3984                 goto end_rename;
3985         }
3986
3987         /* RENAME_EXCHANGE case: old *and* new must both exist */
3988         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3989                 goto end_rename;
3990
3991         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3992                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3993                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3994         if (IS_ERR(handle)) {
3995                 retval = PTR_ERR(handle);
3996                 handle = NULL;
3997                 goto end_rename;
3998         }
3999
4000         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4001                 ext4_handle_sync(handle);
4002
4003         if (S_ISDIR(old.inode->i_mode)) {
4004                 old.is_dir = true;
4005                 retval = ext4_rename_dir_prepare(handle, &old);
4006                 if (retval)
4007                         goto end_rename;
4008         }
4009         if (S_ISDIR(new.inode->i_mode)) {
4010                 new.is_dir = true;
4011                 retval = ext4_rename_dir_prepare(handle, &new);
4012                 if (retval)
4013                         goto end_rename;
4014         }
4015
4016         /*
4017          * Other than the special case of overwriting a directory, parents'
4018          * nlink only needs to be modified if this is a cross directory rename.
4019          */
4020         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4021                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4022                 new.dir_nlink_delta = -old.dir_nlink_delta;
4023                 retval = -EMLINK;
4024                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4025                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4026                         goto end_rename;
4027         }
4028
4029         new_file_type = new.de->file_type;
4030         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4031         if (retval)
4032                 goto end_rename;
4033
4034         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4035         if (retval)
4036                 goto end_rename;
4037
4038         /*
4039          * Like most other Unix systems, set the ctime for inodes on a
4040          * rename.
4041          */
4042         ctime = current_time(old.inode);
4043         old.inode->i_ctime = ctime;
4044         new.inode->i_ctime = ctime;
4045         retval = ext4_mark_inode_dirty(handle, old.inode);
4046         if (unlikely(retval))
4047                 goto end_rename;
4048         retval = ext4_mark_inode_dirty(handle, new.inode);
4049         if (unlikely(retval))
4050                 goto end_rename;
4051         ext4_fc_mark_ineligible(new.inode->i_sb,
4052                                 EXT4_FC_REASON_CROSS_RENAME);
4053         if (old.dir_bh) {
4054                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4055                 if (retval)
4056                         goto end_rename;
4057         }
4058         if (new.dir_bh) {
4059                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4060                 if (retval)
4061                         goto end_rename;
4062         }
4063         ext4_update_dir_count(handle, &old);
4064         ext4_update_dir_count(handle, &new);
4065         retval = 0;
4066
4067 end_rename:
4068         brelse(old.dir_bh);
4069         brelse(new.dir_bh);
4070         brelse(old.bh);
4071         brelse(new.bh);
4072         if (handle)
4073                 ext4_journal_stop(handle);
4074         return retval;
4075 }
4076
4077 static int ext4_rename2(struct user_namespace *mnt_userns,
4078                         struct inode *old_dir, struct dentry *old_dentry,
4079                         struct inode *new_dir, struct dentry *new_dentry,
4080                         unsigned int flags)
4081 {
4082         int err;
4083
4084         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4085                 return -EIO;
4086
4087         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4088                 return -EINVAL;
4089
4090         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4091                                      flags);
4092         if (err)
4093                 return err;
4094
4095         if (flags & RENAME_EXCHANGE) {
4096                 return ext4_cross_rename(old_dir, old_dentry,
4097                                          new_dir, new_dentry);
4098         }
4099
4100         return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4101 }
4102
4103 /*
4104  * directories can handle most operations...
4105  */
4106 const struct inode_operations ext4_dir_inode_operations = {
4107         .create         = ext4_create,
4108         .lookup         = ext4_lookup,
4109         .link           = ext4_link,
4110         .unlink         = ext4_unlink,
4111         .symlink        = ext4_symlink,
4112         .mkdir          = ext4_mkdir,
4113         .rmdir          = ext4_rmdir,
4114         .mknod          = ext4_mknod,
4115         .tmpfile        = ext4_tmpfile,
4116         .rename         = ext4_rename2,
4117         .setattr        = ext4_setattr,
4118         .getattr        = ext4_getattr,
4119         .listxattr      = ext4_listxattr,
4120         .get_acl        = ext4_get_acl,
4121         .set_acl        = ext4_set_acl,
4122         .fiemap         = ext4_fiemap,
4123         .fileattr_get   = ext4_fileattr_get,
4124         .fileattr_set   = ext4_fileattr_set,
4125 };
4126
4127 const struct inode_operations ext4_special_inode_operations = {
4128         .setattr        = ext4_setattr,
4129         .getattr        = ext4_getattr,
4130         .listxattr      = ext4_listxattr,
4131         .get_acl        = ext4_get_acl,
4132         .set_acl        = ext4_set_acl,
4133 };