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 ([email protected])
   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 ([email protected]), 1995
  18  *  Directory entry file type support and forward compatibility hooks
  19  *      for B-tree directories by Theodore Ts'o ([email protected]), 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 buffer_head *bh,
 281                        struct dx_hash_info *hinfo,
 282                        struct dx_map_entry *map_tail);
 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, level, i;
 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         ext4_lblk_t block;
 787         ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
 788
 789         memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
 790         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
 791         if (IS_ERR(frame->bh))
 792                 return (struct dx_frame *) frame->bh;
 793
 794         root = (struct dx_root *) frame->bh->b_data;
 795         if (root->info.hash_version != DX_HASH_TEA &&
 796             root->info.hash_version != DX_HASH_HALF_MD4 &&
 797             root->info.hash_version != DX_HASH_LEGACY &&
 798             root->info.hash_version != DX_HASH_SIPHASH) {
 799                 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
 800                                    root->info.hash_version);
 801                 goto fail;
 802         }
 803         if (ext4_hash_in_dirent(dir)) {
 804                 if (root->info.hash_version != DX_HASH_SIPHASH) {
 805                         ext4_warning_inode(dir,
 806                                 "Hash in dirent, but hash is not SIPHASH");
 807                         goto fail;
 808                 }
 809         } else {
 810                 if (root->info.hash_version == DX_HASH_SIPHASH) {
 811                         ext4_warning_inode(dir,
 812                                 "Hash code is SIPHASH, but hash not in dirent");
 813                         goto fail;
 814                 }
 815         }
 816         if (fname)
 817                 hinfo = &fname->hinfo;
 818         hinfo->hash_version = root->info.hash_version;
 819         if (hinfo->hash_version <= DX_HASH_TEA)
 820                 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 821         hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 822         /* hash is already computed for encrypted casefolded directory */
 823         if (fname && fname_name(fname) &&
 824                                 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
 825                 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
 826         hash = hinfo->hash;
 827
 828         if (root->info.unused_flags & 1) {
 829                 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
 830                                    root->info.unused_flags);
 831                 goto fail;
 832         }
 833
 834         indirect = root->info.indirect_levels;
 835         if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
 836                 ext4_warning(dir->i_sb,
 837                              "Directory (ino: %lu) htree depth %#06x exceed"
 838                              "supported value", dir->i_ino,
 839                              ext4_dir_htree_level(dir->i_sb));
 840                 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
 841                         ext4_warning(dir->i_sb, "Enable large directory "
 842                                                 "feature to access it");
 843                 }
 844                 goto fail;
 845         }
 846
 847         entries = (struct dx_entry *)(((char *)&root->info) +
 848                                       root->info.info_length);
 849
 850         if (dx_get_limit(entries) != dx_root_limit(dir,
 851                                                    root->info.info_length)) {
 852                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 853                                    dx_get_limit(entries),
 854                                    dx_root_limit(dir, root->info.info_length));
 855                 goto fail;
 856         }
 857
 858         dxtrace(printk("Look up %x", hash));
 859         level = 0;
 860         blocks[0] = 0;
 861         while (1) {
 862                 count = dx_get_count(entries);
 863                 if (!count || count > dx_get_limit(entries)) {
 864                         ext4_warning_inode(dir,
 865                                            "dx entry: count %u beyond limit %u",
 866                                            count, dx_get_limit(entries));
 867                         goto fail;
 868                 }
 869
 870                 p = entries + 1;
 871                 q = entries + count - 1;
 872                 while (p <= q) {
 873                         m = p + (q - p) / 2;
 874                         dxtrace(printk(KERN_CONT "."));
 875                         if (dx_get_hash(m) > hash)
 876                                 q = m - 1;
 877                         else
 878                                 p = m + 1;
 879                 }
 880
 881                 htree_rep_invariant_check(entries, p, hash, count - 1);
 882
 883                 at = p - 1;
 884                 dxtrace(printk(KERN_CONT " %x->%u\n",
 885                                at == entries ? 0 : dx_get_hash(at),
 886                                dx_get_block(at)));
 887                 frame->entries = entries;
 888                 frame->at = at;
 889
 890                 block = dx_get_block(at);
 891                 for (i = 0; i <= level; i++) {
 892                         if (blocks[i] == block) {
 893                                 ext4_warning_inode(dir,
 894                                         "dx entry: tree cycle block %u points back to block %u",
 895                                         blocks[level], block);
 896                                 goto fail;
 897                         }
 898                 }
 899                 if (++level > indirect)
 900                         return frame;
 901                 blocks[level] = block;
 902                 frame++;
 903                 frame->bh = ext4_read_dirblock(dir, block, INDEX);
 904                 if (IS_ERR(frame->bh)) {
 905                         ret_err = (struct dx_frame *) frame->bh;
 906                         frame->bh = NULL;
 907                         goto fail;
 908                 }
 909
 910                 entries = ((struct dx_node *) frame->bh->b_data)->entries;
 911
 912                 if (dx_get_limit(entries) != dx_node_limit(dir)) {
 913                         ext4_warning_inode(dir,
 914                                 "dx entry: limit %u != node limit %u",
 915                                 dx_get_limit(entries), dx_node_limit(dir));
 916                         goto fail;
 917                 }
 918         }
 919 fail:
 920         while (frame >= frame_in) {
 921                 brelse(frame->bh);
 922                 frame--;
 923         }
 924
 925         if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
 926                 ext4_warning_inode(dir,
 927                         "Corrupt directory, running e2fsck is recommended");
 928         return ret_err;
 929 }
 930
 931 static void dx_release(struct dx_frame *frames)
 932 {
 933         struct dx_root_info *info;
 934         int i;
 935         unsigned int indirect_levels;
 936
 937         if (frames[0].bh == NULL)
 938                 return;
 939
 940         info = &((struct dx_root *)frames[0].bh->b_data)->info;
 941         /* save local copy, "info" may be freed after brelse() */
 942         indirect_levels = info->indirect_levels;
 943         for (i = 0; i <= indirect_levels; i++) {
 944                 if (frames[i].bh == NULL)
 945                         break;
 946                 brelse(frames[i].bh);
 947                 frames[i].bh = NULL;
 948         }
 949 }
 950
 951 /*
 952  * This function increments the frame pointer to search the next leaf
 953  * block, and reads in the necessary intervening nodes if the search
 954  * should be necessary.  Whether or not the search is necessary is
 955  * controlled by the hash parameter.  If the hash value is even, then
 956  * the search is only continued if the next block starts with that
 957  * hash value.  This is used if we are searching for a specific file.
 958  *
 959  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 960  *
 961  * This function returns 1 if the caller should continue to search,
 962  * or 0 if it should not.  If there is an error reading one of the
 963  * index blocks, it will a negative error code.
 964  *
 965  * If start_hash is non-null, it will be filled in with the starting
 966  * hash of the next page.
 967  */
 968 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 969                                  struct dx_frame *frame,
 970                                  struct dx_frame *frames,
 971                                  __u32 *start_hash)
 972 {
 973         struct dx_frame *p;
 974         struct buffer_head *bh;
 975         int num_frames = 0;
 976         __u32 bhash;
 977
 978         p = frame;
 979         /*
 980          * Find the next leaf page by incrementing the frame pointer.
 981          * If we run out of entries in the interior node, loop around and
 982          * increment pointer in the parent node.  When we break out of
 983          * this loop, num_frames indicates the number of interior
 984          * nodes need to be read.
 985          */
 986         while (1) {
 987                 if (++(p->at) < p->entries + dx_get_count(p->entries))
 988                         break;
 989                 if (p == frames)
 990                         return 0;
 991                 num_frames++;
 992                 p--;
 993         }
 994
 995         /*
 996          * If the hash is 1, then continue only if the next page has a
 997          * continuation hash of any value.  This is used for readdir
 998          * handling.  Otherwise, check to see if the hash matches the
 999          * desired continuation hash.  If it doesn't, return since
1000          * there's no point to read in the successive index pages.
1001          */
1002         bhash = dx_get_hash(p->at);
1003         if (start_hash)
1004                 *start_hash = bhash;
1005         if ((hash & 1) == 0) {
1006                 if ((bhash & ~1) != hash)
1007                         return 0;
1008         }
1009         /*
1010          * If the hash is HASH_NB_ALWAYS, we always go to the next
1011          * block so no check is necessary
1012          */
1013         while (num_frames--) {
1014                 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1015                 if (IS_ERR(bh))
1016                         return PTR_ERR(bh);
1017                 p++;
1018                 brelse(p->bh);
1019                 p->bh = bh;
1020                 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1021         }
1022         return 1;
1023 }
1024
1025
1026 /*
1027  * This function fills a red-black tree with information from a
1028  * directory block.  It returns the number directory entries loaded
1029  * into the tree.  If there is an error it is returned in err.
1030  */
1031 static int htree_dirblock_to_tree(struct file *dir_file,
1032                                   struct inode *dir, ext4_lblk_t block,
1033                                   struct dx_hash_info *hinfo,
1034                                   __u32 start_hash, __u32 start_minor_hash)
1035 {
1036         struct buffer_head *bh;
1037         struct ext4_dir_entry_2 *de, *top;
1038         int err = 0, count = 0;
1039         struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1040         int csum = ext4_has_metadata_csum(dir->i_sb);
1041
1042         dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1043                                                         (unsigned long)block));
1044         bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1045         if (IS_ERR(bh))
1046                 return PTR_ERR(bh);
1047
1048         de = (struct ext4_dir_entry_2 *) bh->b_data;
1049         /* csum entries are not larger in the casefolded encrypted case */
1050         top = (struct ext4_dir_entry_2 *) ((char *) de +
1051                                            dir->i_sb->s_blocksize -
1052                                            ext4_dir_rec_len(0,
1053                                                            csum ? NULL : dir));
1054         /* Check if the directory is encrypted */
1055         if (IS_ENCRYPTED(dir)) {
1056                 err = fscrypt_prepare_readdir(dir);
1057                 if (err < 0) {
1058                         brelse(bh);
1059                         return err;
1060                 }
1061                 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1062                                                  &fname_crypto_str);
1063                 if (err < 0) {
1064                         brelse(bh);
1065                         return err;
1066                 }
1067         }
1068
1069         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1070                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1071                                 bh->b_data, bh->b_size,
1072                                 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1073                                          + ((char *)de - bh->b_data))) {
1074                         /* silently ignore the rest of the block */
1075                         break;
1076                 }
1077                 if (ext4_hash_in_dirent(dir)) {
1078                         if (de->name_len && de->inode) {
1079                                 hinfo->hash = EXT4_DIRENT_HASH(de);
1080                                 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1081                         } else {
1082                                 hinfo->hash = 0;
1083                                 hinfo->minor_hash = 0;
1084                         }
1085                 } else {
1086                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1087                 }
1088                 if ((hinfo->hash < start_hash) ||
1089                     ((hinfo->hash == start_hash) &&
1090                      (hinfo->minor_hash < start_minor_hash)))
1091                         continue;
1092                 if (de->inode == 0)
1093                         continue;
1094                 if (!IS_ENCRYPTED(dir)) {
1095                         tmp_str.name = de->name;
1096                         tmp_str.len = de->name_len;
1097                         err = ext4_htree_store_dirent(dir_file,
1098                                    hinfo->hash, hinfo->minor_hash, de,
1099                                    &tmp_str);
1100                 } else {
1101                         int save_len = fname_crypto_str.len;
1102                         struct fscrypt_str de_name = FSTR_INIT(de->name,
1103                                                                 de->name_len);
1104
1105                         /* Directory is encrypted */
1106                         err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1107                                         hinfo->minor_hash, &de_name,
1108                                         &fname_crypto_str);
1109                         if (err) {
1110                                 count = err;
1111                                 goto errout;
1112                         }
1113                         err = ext4_htree_store_dirent(dir_file,
1114                                    hinfo->hash, hinfo->minor_hash, de,
1115                                         &fname_crypto_str);
1116                         fname_crypto_str.len = save_len;
1117                 }
1118                 if (err != 0) {
1119                         count = err;
1120                         goto errout;
1121                 }
1122                 count++;
1123         }
1124 errout:
1125         brelse(bh);
1126         fscrypt_fname_free_buffer(&fname_crypto_str);
1127         return count;
1128 }
1129
1130
1131 /*
1132  * This function fills a red-black tree with information from a
1133  * directory.  We start scanning the directory in hash order, starting
1134  * at start_hash and start_minor_hash.
1135  *
1136  * This function returns the number of entries inserted into the tree,
1137  * or a negative error code.
1138  */
1139 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1140                          __u32 start_minor_hash, __u32 *next_hash)
1141 {
1142         struct dx_hash_info hinfo;
1143         struct ext4_dir_entry_2 *de;
1144         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1145         struct inode *dir;
1146         ext4_lblk_t block;
1147         int count = 0;
1148         int ret, err;
1149         __u32 hashval;
1150         struct fscrypt_str tmp_str;
1151
1152         dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1153                        start_hash, start_minor_hash));
1154         dir = file_inode(dir_file);
1155         if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1156                 if (ext4_hash_in_dirent(dir))
1157                         hinfo.hash_version = DX_HASH_SIPHASH;
1158                 else
1159                         hinfo.hash_version =
1160                                         EXT4_SB(dir->i_sb)->s_def_hash_version;
1161                 if (hinfo.hash_version <= DX_HASH_TEA)
1162                         hinfo.hash_version +=
1163                                 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1164                 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1165                 if (ext4_has_inline_data(dir)) {
1166                         int has_inline_data = 1;
1167                         count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1168                                                        &hinfo, start_hash,
1169                                                        start_minor_hash,
1170                                                        &has_inline_data);
1171                         if (has_inline_data) {
1172                                 *next_hash = ~0;
1173                                 return count;
1174                         }
1175                 }
1176                 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1177                                                start_hash, start_minor_hash);
1178                 *next_hash = ~0;
1179                 return count;
1180         }
1181         hinfo.hash = start_hash;
1182         hinfo.minor_hash = 0;
1183         frame = dx_probe(NULL, dir, &hinfo, frames);
1184         if (IS_ERR(frame))
1185                 return PTR_ERR(frame);
1186
1187         /* Add '.' and '..' from the htree header */
1188         if (!start_hash && !start_minor_hash) {
1189                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1190                 tmp_str.name = de->name;
1191                 tmp_str.len = de->name_len;
1192                 err = ext4_htree_store_dirent(dir_file, 0, 0,
1193                                               de, &tmp_str);
1194                 if (err != 0)
1195                         goto errout;
1196                 count++;
1197         }
1198         if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1199                 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1200                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1201                 tmp_str.name = de->name;
1202                 tmp_str.len = de->name_len;
1203                 err = ext4_htree_store_dirent(dir_file, 2, 0,
1204                                               de, &tmp_str);
1205                 if (err != 0)
1206                         goto errout;
1207                 count++;
1208         }
1209
1210         while (1) {
1211                 if (fatal_signal_pending(current)) {
1212                         err = -ERESTARTSYS;
1213                         goto errout;
1214                 }
1215                 cond_resched();
1216                 block = dx_get_block(frame->at);
1217                 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1218                                              start_hash, start_minor_hash);
1219                 if (ret < 0) {
1220                         err = ret;
1221                         goto errout;
1222                 }
1223                 count += ret;
1224                 hashval = ~0;
1225                 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1226                                             frame, frames, &hashval);
1227                 *next_hash = hashval;
1228                 if (ret < 0) {
1229                         err = ret;
1230                         goto errout;
1231                 }
1232                 /*
1233                  * Stop if:  (a) there are no more entries, or
1234                  * (b) we have inserted at least one entry and the
1235                  * next hash value is not a continuation
1236                  */
1237                 if ((ret == 0) ||
1238                     (count && ((hashval & 1) == 0)))
1239                         break;
1240         }
1241         dx_release(frames);
1242         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1243                        "next hash: %x\n", count, *next_hash));
1244         return count;
1245 errout:
1246         dx_release(frames);
1247         return (err);
1248 }
1249
1250 static inline int search_dirblock(struct buffer_head *bh,
1251                                   struct inode *dir,
1252                                   struct ext4_filename *fname,
1253                                   unsigned int offset,
1254                                   struct ext4_dir_entry_2 **res_dir)
1255 {
1256         return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1257                                fname, offset, res_dir);
1258 }
1259
1260 /*
1261  * Directory block splitting, compacting
1262  */
1263
1264 /*
1265  * Create map of hash values, offsets, and sizes, stored at end of block.
1266  * Returns number of entries mapped.
1267  */
1268 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1269                        struct dx_hash_info *hinfo,
1270                        struct dx_map_entry *map_tail)
1271 {
1272         int count = 0;
1273         struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1274         unsigned int buflen = bh->b_size;
1275         char *base = bh->b_data;
1276         struct dx_hash_info h = *hinfo;
1277
1278         if (ext4_has_metadata_csum(dir->i_sb))
1279                 buflen -= sizeof(struct ext4_dir_entry_tail);
1280
1281         while ((char *) de < base + buflen) {
1282                 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1283                                          ((char *)de) - base))
1284                         return -EFSCORRUPTED;
1285                 if (de->name_len && de->inode) {
1286                         if (ext4_hash_in_dirent(dir))
1287                                 h.hash = EXT4_DIRENT_HASH(de);
1288                         else
1289                                 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1290                         map_tail--;
1291                         map_tail->hash = h.hash;
1292                         map_tail->offs = ((char *) de - base)>>2;
1293                         map_tail->size = le16_to_cpu(de->rec_len);
1294                         count++;
1295                         cond_resched();
1296                 }
1297                 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1298         }
1299         return count;
1300 }
1301
1302 /* Sort map by hash value */
1303 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1304 {
1305         struct dx_map_entry *p, *q, *top = map + count - 1;
1306         int more;
1307         /* Combsort until bubble sort doesn't suck */
1308         while (count > 2) {
1309                 count = count*10/13;
1310                 if (count - 9 < 2) /* 9, 10 -> 11 */
1311                         count = 11;
1312                 for (p = top, q = p - count; q >= map; p--, q--)
1313                         if (p->hash < q->hash)
1314                                 swap(*p, *q);
1315         }
1316         /* Garden variety bubble sort */
1317         do {
1318                 more = 0;
1319                 q = top;
1320                 while (q-- > map) {
1321                         if (q[1].hash >= q[0].hash)
1322                                 continue;
1323                         swap(*(q+1), *q);
1324                         more = 1;
1325                 }
1326         } while(more);
1327 }
1328
1329 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1330 {
1331         struct dx_entry *entries = frame->entries;
1332         struct dx_entry *old = frame->at, *new = old + 1;
1333         int count = dx_get_count(entries);
1334
1335         ASSERT(count < dx_get_limit(entries));
1336         ASSERT(old < entries + count);
1337         memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1338         dx_set_hash(new, hash);
1339         dx_set_block(new, block);
1340         dx_set_count(entries, count + 1);
1341 }
1342
1343 #if IS_ENABLED(CONFIG_UNICODE)
1344 /*
1345  * Test whether a case-insensitive directory entry matches the filename
1346  * being searched for.  If quick is set, assume the name being looked up
1347  * is already in the casefolded form.
1348  *
1349  * Returns: 0 if the directory entry matches, more than 0 if it
1350  * doesn't match or less than zero on error.
1351  */
1352 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1353                            u8 *de_name, size_t de_name_len, bool quick)
1354 {
1355         const struct super_block *sb = parent->i_sb;
1356         const struct unicode_map *um = sb->s_encoding;
1357         struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1358         struct qstr entry = QSTR_INIT(de_name, de_name_len);
1359         int ret;
1360
1361         if (IS_ENCRYPTED(parent)) {
1362                 const struct fscrypt_str encrypted_name =
1363                                 FSTR_INIT(de_name, de_name_len);
1364
1365                 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1366                 if (!decrypted_name.name)
1367                         return -ENOMEM;
1368                 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1369                                                 &decrypted_name);
1370                 if (ret < 0)
1371                         goto out;
1372                 entry.name = decrypted_name.name;
1373                 entry.len = decrypted_name.len;
1374         }
1375
1376         if (quick)
1377                 ret = utf8_strncasecmp_folded(um, name, &entry);
1378         else
1379                 ret = utf8_strncasecmp(um, name, &entry);
1380         if (ret < 0) {
1381                 /* Handle invalid character sequence as either an error
1382                  * or as an opaque byte sequence.
1383                  */
1384                 if (sb_has_strict_encoding(sb))
1385                         ret = -EINVAL;
1386                 else if (name->len != entry.len)
1387                         ret = 1;
1388                 else
1389                         ret = !!memcmp(name->name, entry.name, entry.len);
1390         }
1391 out:
1392         kfree(decrypted_name.name);
1393         return ret;
1394 }
1395
1396 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1397                                   struct ext4_filename *name)
1398 {
1399         struct fscrypt_str *cf_name = &name->cf_name;
1400         struct dx_hash_info *hinfo = &name->hinfo;
1401         int len;
1402
1403         if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1404             (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1405                 cf_name->name = NULL;
1406                 return 0;
1407         }
1408
1409         cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1410         if (!cf_name->name)
1411                 return -ENOMEM;
1412
1413         len = utf8_casefold(dir->i_sb->s_encoding,
1414                             iname, cf_name->name,
1415                             EXT4_NAME_LEN);
1416         if (len <= 0) {
1417                 kfree(cf_name->name);
1418                 cf_name->name = NULL;
1419         }
1420         cf_name->len = (unsigned) len;
1421         if (!IS_ENCRYPTED(dir))
1422                 return 0;
1423
1424         hinfo->hash_version = DX_HASH_SIPHASH;
1425         hinfo->seed = NULL;
1426         if (cf_name->name)
1427                 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1428         else
1429                 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1430         return 0;
1431 }
1432 #endif
1433
1434 /*
1435  * Test whether a directory entry matches the filename being searched for.
1436  *
1437  * Return: %true if the directory entry matches, otherwise %false.
1438  */
1439 static bool ext4_match(struct inode *parent,
1440                               const struct ext4_filename *fname,
1441                               struct ext4_dir_entry_2 *de)
1442 {
1443         struct fscrypt_name f;
1444
1445         if (!de->inode)
1446                 return false;
1447
1448         f.usr_fname = fname->usr_fname;
1449         f.disk_name = fname->disk_name;
1450 #ifdef CONFIG_FS_ENCRYPTION
1451         f.crypto_buf = fname->crypto_buf;
1452 #endif
1453
1454 #if IS_ENABLED(CONFIG_UNICODE)
1455         if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1456             (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1457                 if (fname->cf_name.name) {
1458                         struct qstr cf = {.name = fname->cf_name.name,
1459                                           .len = fname->cf_name.len};
1460                         if (IS_ENCRYPTED(parent)) {
1461                                 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1462                                         fname->hinfo.minor_hash !=
1463                                                 EXT4_DIRENT_MINOR_HASH(de)) {
1464
1465                                         return false;
1466                                 }
1467                         }
1468                         return !ext4_ci_compare(parent, &cf, de->name,
1469                                                         de->name_len, true);
1470                 }
1471                 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1472                                                 de->name_len, false);
1473         }
1474 #endif
1475
1476         return fscrypt_match_name(&f, de->name, de->name_len);
1477 }
1478
1479 /*
1480  * Returns 0 if not found, -1 on failure, and 1 on success
1481  */
1482 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1483                     struct inode *dir, struct ext4_filename *fname,
1484                     unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1485 {
1486         struct ext4_dir_entry_2 * de;
1487         char * dlimit;
1488         int de_len;
1489
1490         de = (struct ext4_dir_entry_2 *)search_buf;
1491         dlimit = search_buf + buf_size;
1492         while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1493                 /* this code is executed quadratically often */
1494                 /* do minimal checking `by hand' */
1495                 if (de->name + de->name_len <= dlimit &&
1496                     ext4_match(dir, fname, de)) {
1497                         /* found a match - just to be sure, do
1498                          * a full check */
1499                         if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1500                                                  buf_size, offset))
1501                                 return -1;
1502                         *res_dir = de;
1503                         return 1;
1504                 }
1505                 /* prevent looping on a bad block */
1506                 de_len = ext4_rec_len_from_disk(de->rec_len,
1507                                                 dir->i_sb->s_blocksize);
1508                 if (de_len <= 0)
1509                         return -1;
1510                 offset += de_len;
1511                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1512         }
1513         return 0;
1514 }
1515
1516 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1517                                struct ext4_dir_entry *de)
1518 {
1519         struct super_block *sb = dir->i_sb;
1520
1521         if (!is_dx(dir))
1522                 return 0;
1523         if (block == 0)
1524                 return 1;
1525         if (de->inode == 0 &&
1526             ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1527                         sb->s_blocksize)
1528                 return 1;
1529         return 0;
1530 }
1531
1532 /*
1533  *      __ext4_find_entry()
1534  *
1535  * finds an entry in the specified directory with the wanted name. It
1536  * returns the cache buffer in which the entry was found, and the entry
1537  * itself (as a parameter - res_dir). It does NOT read the inode of the
1538  * entry - you'll have to do that yourself if you want to.
1539  *
1540  * The returned buffer_head has ->b_count elevated.  The caller is expected
1541  * to brelse() it when appropriate.
1542  */
1543 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1544                                              struct ext4_filename *fname,
1545                                              struct ext4_dir_entry_2 **res_dir,
1546                                              int *inlined)
1547 {
1548         struct super_block *sb;
1549         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1550         struct buffer_head *bh, *ret = NULL;
1551         ext4_lblk_t start, block;
1552         const u8 *name = fname->usr_fname->name;
1553         size_t ra_max = 0;      /* Number of bh's in the readahead
1554                                    buffer, bh_use[] */
1555         size_t ra_ptr = 0;      /* Current index into readahead
1556                                    buffer */
1557         ext4_lblk_t  nblocks;
1558         int i, namelen, retval;
1559
1560         *res_dir = NULL;
1561         sb = dir->i_sb;
1562         namelen = fname->usr_fname->len;
1563         if (namelen > EXT4_NAME_LEN)
1564                 return NULL;
1565
1566         if (ext4_has_inline_data(dir)) {
1567                 int has_inline_data = 1;
1568                 ret = ext4_find_inline_entry(dir, fname, res_dir,
1569                                              &has_inline_data);
1570                 if (has_inline_data) {
1571                         if (inlined)
1572                                 *inlined = 1;
1573                         goto cleanup_and_exit;
1574                 }
1575         }
1576
1577         if ((namelen <= 2) && (name[0] == '.') &&
1578             (name[1] == '.' || name[1] == '\0')) {
1579                 /*
1580                  * "." or ".." will only be in the first block
1581                  * NFS may look up ".."; "." should be handled by the VFS
1582                  */
1583                 block = start = 0;
1584                 nblocks = 1;
1585                 goto restart;
1586         }
1587         if (is_dx(dir)) {
1588                 ret = ext4_dx_find_entry(dir, fname, res_dir);
1589                 /*
1590                  * On success, or if the error was file not found,
1591                  * return.  Otherwise, fall back to doing a search the
1592                  * old fashioned way.
1593                  */
1594                 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1595                         goto cleanup_and_exit;
1596                 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1597                                "falling back\n"));
1598                 ret = NULL;
1599         }
1600         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1601         if (!nblocks) {
1602                 ret = NULL;
1603                 goto cleanup_and_exit;
1604         }
1605         start = EXT4_I(dir)->i_dir_start_lookup;
1606         if (start >= nblocks)
1607                 start = 0;
1608         block = start;
1609 restart:
1610         do {
1611                 /*
1612                  * We deal with the read-ahead logic here.
1613                  */
1614                 cond_resched();
1615                 if (ra_ptr >= ra_max) {
1616                         /* Refill the readahead buffer */
1617                         ra_ptr = 0;
1618                         if (block < start)
1619                                 ra_max = start - block;
1620                         else
1621                                 ra_max = nblocks - block;
1622                         ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1623                         retval = ext4_bread_batch(dir, block, ra_max,
1624                                                   false /* wait */, bh_use);
1625                         if (retval) {
1626                                 ret = ERR_PTR(retval);
1627                                 ra_max = 0;
1628                                 goto cleanup_and_exit;
1629                         }
1630                 }
1631                 if ((bh = bh_use[ra_ptr++]) == NULL)
1632                         goto next;
1633                 wait_on_buffer(bh);
1634                 if (!buffer_uptodate(bh)) {
1635                         EXT4_ERROR_INODE_ERR(dir, EIO,
1636                                              "reading directory lblock %lu",
1637                                              (unsigned long) block);
1638                         brelse(bh);
1639                         ret = ERR_PTR(-EIO);
1640                         goto cleanup_and_exit;
1641                 }
1642                 if (!buffer_verified(bh) &&
1643                     !is_dx_internal_node(dir, block,
1644                                          (struct ext4_dir_entry *)bh->b_data) &&
1645                     !ext4_dirblock_csum_verify(dir, bh)) {
1646                         EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1647                                              "checksumming directory "
1648                                              "block %lu", (unsigned long)block);
1649                         brelse(bh);
1650                         ret = ERR_PTR(-EFSBADCRC);
1651                         goto cleanup_and_exit;
1652                 }
1653                 set_buffer_verified(bh);
1654                 i = search_dirblock(bh, dir, fname,
1655                             block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1656                 if (i == 1) {
1657                         EXT4_I(dir)->i_dir_start_lookup = block;
1658                         ret = bh;
1659                         goto cleanup_and_exit;
1660                 } else {
1661                         brelse(bh);
1662                         if (i < 0)
1663                                 goto cleanup_and_exit;
1664                 }
1665         next:
1666                 if (++block >= nblocks)
1667                         block = 0;
1668         } while (block != start);
1669
1670         /*
1671          * If the directory has grown while we were searching, then
1672          * search the last part of the directory before giving up.
1673          */
1674         block = nblocks;
1675         nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1676         if (block < nblocks) {
1677                 start = 0;
1678                 goto restart;
1679         }
1680
1681 cleanup_and_exit:
1682         /* Clean up the read-ahead blocks */
1683         for (; ra_ptr < ra_max; ra_ptr++)
1684                 brelse(bh_use[ra_ptr]);
1685         return ret;
1686 }
1687
1688 static struct buffer_head *ext4_find_entry(struct inode *dir,
1689                                            const struct qstr *d_name,
1690                                            struct ext4_dir_entry_2 **res_dir,
1691                                            int *inlined)
1692 {
1693         int err;
1694         struct ext4_filename fname;
1695         struct buffer_head *bh;
1696
1697         err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1698         if (err == -ENOENT)
1699                 return NULL;
1700         if (err)
1701                 return ERR_PTR(err);
1702
1703         bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1704
1705         ext4_fname_free_filename(&fname);
1706         return bh;
1707 }
1708
1709 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1710                                              struct dentry *dentry,
1711                                              struct ext4_dir_entry_2 **res_dir)
1712 {
1713         int err;
1714         struct ext4_filename fname;
1715         struct buffer_head *bh;
1716
1717         err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1718         generic_set_encrypted_ci_d_ops(dentry);
1719         if (err == -ENOENT)
1720                 return NULL;
1721         if (err)
1722                 return ERR_PTR(err);
1723
1724         bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1725
1726         ext4_fname_free_filename(&fname);
1727         return bh;
1728 }
1729
1730 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1731                         struct ext4_filename *fname,
1732                         struct ext4_dir_entry_2 **res_dir)
1733 {
1734         struct super_block * sb = dir->i_sb;
1735         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1736         struct buffer_head *bh;
1737         ext4_lblk_t block;
1738         int retval;
1739
1740 #ifdef CONFIG_FS_ENCRYPTION
1741         *res_dir = NULL;
1742 #endif
1743         frame = dx_probe(fname, dir, NULL, frames);
1744         if (IS_ERR(frame))
1745                 return (struct buffer_head *) frame;
1746         do {
1747                 block = dx_get_block(frame->at);
1748                 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1749                 if (IS_ERR(bh))
1750                         goto errout;
1751
1752                 retval = search_dirblock(bh, dir, fname,
1753                                          block << EXT4_BLOCK_SIZE_BITS(sb),
1754                                          res_dir);
1755                 if (retval == 1)
1756                         goto success;
1757                 brelse(bh);
1758                 if (retval == -1) {
1759                         bh = ERR_PTR(ERR_BAD_DX_DIR);
1760                         goto errout;
1761                 }
1762
1763                 /* Check to see if we should continue to search */
1764                 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1765                                                frames, NULL);
1766                 if (retval < 0) {
1767                         ext4_warning_inode(dir,
1768                                 "error %d reading directory index block",
1769                                 retval);
1770                         bh = ERR_PTR(retval);
1771                         goto errout;
1772                 }
1773         } while (retval == 1);
1774
1775         bh = NULL;
1776 errout:
1777         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1778 success:
1779         dx_release(frames);
1780         return bh;
1781 }
1782
1783 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1784 {
1785         struct inode *inode;
1786         struct ext4_dir_entry_2 *de;
1787         struct buffer_head *bh;
1788
1789         if (dentry->d_name.len > EXT4_NAME_LEN)
1790                 return ERR_PTR(-ENAMETOOLONG);
1791
1792         bh = ext4_lookup_entry(dir, dentry, &de);
1793         if (IS_ERR(bh))
1794                 return ERR_CAST(bh);
1795         inode = NULL;
1796         if (bh) {
1797                 __u32 ino = le32_to_cpu(de->inode);
1798                 brelse(bh);
1799                 if (!ext4_valid_inum(dir->i_sb, ino)) {
1800                         EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1801                         return ERR_PTR(-EFSCORRUPTED);
1802                 }
1803                 if (unlikely(ino == dir->i_ino)) {
1804                         EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1805                                          dentry);
1806                         return ERR_PTR(-EFSCORRUPTED);
1807                 }
1808                 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1809                 if (inode == ERR_PTR(-ESTALE)) {
1810                         EXT4_ERROR_INODE(dir,
1811                                          "deleted inode referenced: %u",
1812                                          ino);
1813                         return ERR_PTR(-EFSCORRUPTED);
1814                 }
1815                 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1816                     (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1817                     !fscrypt_has_permitted_context(dir, inode)) {
1818                         ext4_warning(inode->i_sb,
1819                                      "Inconsistent encryption contexts: %lu/%lu",
1820                                      dir->i_ino, inode->i_ino);
1821                         iput(inode);
1822                         return ERR_PTR(-EPERM);
1823                 }
1824         }
1825
1826 #if IS_ENABLED(CONFIG_UNICODE)
1827         if (!inode && IS_CASEFOLDED(dir)) {
1828                 /* Eventually we want to call d_add_ci(dentry, NULL)
1829                  * for negative dentries in the encoding case as
1830                  * well.  For now, prevent the negative dentry
1831                  * from being cached.
1832                  */
1833                 return NULL;
1834         }
1835 #endif
1836         return d_splice_alias(inode, dentry);
1837 }
1838
1839
1840 struct dentry *ext4_get_parent(struct dentry *child)
1841 {
1842         __u32 ino;
1843         struct ext4_dir_entry_2 * de;
1844         struct buffer_head *bh;
1845
1846         bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1847         if (IS_ERR(bh))
1848                 return ERR_CAST(bh);
1849         if (!bh)
1850                 return ERR_PTR(-ENOENT);
1851         ino = le32_to_cpu(de->inode);
1852         brelse(bh);
1853
1854         if (!ext4_valid_inum(child->d_sb, ino)) {
1855                 EXT4_ERROR_INODE(d_inode(child),
1856                                  "bad parent inode number: %u", ino);
1857                 return ERR_PTR(-EFSCORRUPTED);
1858         }
1859
1860         return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1861 }
1862
1863 /*
1864  * Move count entries from end of map between two memory locations.
1865  * Returns pointer to last entry moved.
1866  */
1867 static struct ext4_dir_entry_2 *
1868 dx_move_dirents(struct inode *dir, char *from, char *to,
1869                 struct dx_map_entry *map, int count,
1870                 unsigned blocksize)
1871 {
1872         unsigned rec_len = 0;
1873
1874         while (count--) {
1875                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1876                                                 (from + (map->offs<<2));
1877                 rec_len = ext4_dir_rec_len(de->name_len, dir);
1878
1879                 memcpy (to, de, rec_len);
1880                 ((struct ext4_dir_entry_2 *) to)->rec_len =
1881                                 ext4_rec_len_to_disk(rec_len, blocksize);
1882
1883                 /* wipe dir_entry excluding the rec_len field */
1884                 de->inode = 0;
1885                 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1886                                                                 blocksize) -
1887                                          offsetof(struct ext4_dir_entry_2,
1888                                                                 name_len));
1889
1890                 map++;
1891                 to += rec_len;
1892         }
1893         return (struct ext4_dir_entry_2 *) (to - rec_len);
1894 }
1895
1896 /*
1897  * Compact each dir entry in the range to the minimal rec_len.
1898  * Returns pointer to last entry in range.
1899  */
1900 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1901                                                         unsigned int blocksize)
1902 {
1903         struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1904         unsigned rec_len = 0;
1905
1906         prev = to = de;
1907         while ((char*)de < base + blocksize) {
1908                 next = ext4_next_entry(de, blocksize);
1909                 if (de->inode && de->name_len) {
1910                         rec_len = ext4_dir_rec_len(de->name_len, dir);
1911                         if (de > to)
1912                                 memmove(to, de, rec_len);
1913                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1914                         prev = to;
1915                         to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1916                 }
1917                 de = next;
1918         }
1919         return prev;
1920 }
1921
1922 /*
1923  * Split a full leaf block to make room for a new dir entry.
1924  * Allocate a new block, and move entries so that they are approx. equally full.
1925  * Returns pointer to de in block into which the new entry will be inserted.
1926  */
1927 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1928                         struct buffer_head **bh,struct dx_frame *frame,
1929                         struct dx_hash_info *hinfo)
1930 {
1931         unsigned blocksize = dir->i_sb->s_blocksize;
1932         unsigned continued;
1933         int count;
1934         struct buffer_head *bh2;
1935         ext4_lblk_t newblock;
1936         u32 hash2;
1937         struct dx_map_entry *map;
1938         char *data1 = (*bh)->b_data, *data2;
1939         unsigned split, move, size;
1940         struct ext4_dir_entry_2 *de = NULL, *de2;
1941         int     csum_size = 0;
1942         int     err = 0, i;
1943
1944         if (ext4_has_metadata_csum(dir->i_sb))
1945                 csum_size = sizeof(struct ext4_dir_entry_tail);
1946
1947         bh2 = ext4_append(handle, dir, &newblock);
1948         if (IS_ERR(bh2)) {
1949                 brelse(*bh);
1950                 *bh = NULL;
1951                 return (struct ext4_dir_entry_2 *) bh2;
1952         }
1953
1954         BUFFER_TRACE(*bh, "get_write_access");
1955         err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1956                                             EXT4_JTR_NONE);
1957         if (err)
1958                 goto journal_error;
1959
1960         BUFFER_TRACE(frame->bh, "get_write_access");
1961         err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1962                                             EXT4_JTR_NONE);
1963         if (err)
1964                 goto journal_error;
1965
1966         data2 = bh2->b_data;
1967
1968         /* create map in the end of data2 block */
1969         map = (struct dx_map_entry *) (data2 + blocksize);
1970         count = dx_make_map(dir, *bh, hinfo, map);
1971         if (count < 0) {
1972                 err = count;
1973                 goto journal_error;
1974         }
1975         map -= count;
1976         dx_sort_map(map, count);
1977         /* Ensure that neither split block is over half full */
1978         size = 0;
1979         move = 0;
1980         for (i = count-1; i >= 0; i--) {
1981                 /* is more than half of this entry in 2nd half of the block? */
1982                 if (size + map[i].size/2 > blocksize/2)
1983                         break;
1984                 size += map[i].size;
1985                 move++;
1986         }
1987         /*
1988          * map index at which we will split
1989          *
1990          * If the sum of active entries didn't exceed half the block size, just
1991          * split it in half by count; each resulting block will have at least
1992          * half the space free.
1993          */
1994         if (i > 0)
1995                 split = count - move;
1996         else
1997                 split = count/2;
1998
1999         hash2 = map[split].hash;
2000         continued = hash2 == map[split - 1].hash;
2001         dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2002                         (unsigned long)dx_get_block(frame->at),
2003                                         hash2, split, count-split));
2004
2005         /* Fancy dance to stay within two buffers */
2006         de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2007                               blocksize);
2008         de = dx_pack_dirents(dir, data1, blocksize);
2009         de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2010                                            (char *) de,
2011                                            blocksize);
2012         de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2013                                             (char *) de2,
2014                                             blocksize);
2015         if (csum_size) {
2016                 ext4_initialize_dirent_tail(*bh, blocksize);
2017                 ext4_initialize_dirent_tail(bh2, blocksize);
2018         }
2019
2020         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2021                         blocksize, 1));
2022         dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2023                         blocksize, 1));
2024
2025         /* Which block gets the new entry? */
2026         if (hinfo->hash >= hash2) {
2027                 swap(*bh, bh2);
2028                 de = de2;
2029         }
2030         dx_insert_block(frame, hash2 + continued, newblock);
2031         err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2032         if (err)
2033                 goto journal_error;
2034         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2035         if (err)
2036                 goto journal_error;
2037         brelse(bh2);
2038         dxtrace(dx_show_index("frame", frame->entries));
2039         return de;
2040
2041 journal_error:
2042         brelse(*bh);
2043         brelse(bh2);
2044         *bh = NULL;
2045         ext4_std_error(dir->i_sb, err);
2046         return ERR_PTR(err);
2047 }
2048
2049 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2050                       struct buffer_head *bh,
2051                       void *buf, int buf_size,
2052                       struct ext4_filename *fname,
2053                       struct ext4_dir_entry_2 **dest_de)
2054 {
2055         struct ext4_dir_entry_2 *de;
2056         unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2057         int nlen, rlen;
2058         unsigned int offset = 0;
2059         char *top;
2060
2061         de = buf;
2062         top = buf + buf_size - reclen;
2063         while ((char *) de <= top) {
2064                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2065                                          buf, buf_size, offset))
2066                         return -EFSCORRUPTED;
2067                 if (ext4_match(dir, fname, de))
2068                         return -EEXIST;
2069                 nlen = ext4_dir_rec_len(de->name_len, dir);
2070                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2071                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2072                         break;
2073                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2074                 offset += rlen;
2075         }
2076         if ((char *) de > top)
2077                 return -ENOSPC;
2078
2079         *dest_de = de;
2080         return 0;
2081 }
2082
2083 void ext4_insert_dentry(struct inode *dir,
2084                         struct inode *inode,
2085                         struct ext4_dir_entry_2 *de,
2086                         int buf_size,
2087                         struct ext4_filename *fname)
2088 {
2089
2090         int nlen, rlen;
2091
2092         nlen = ext4_dir_rec_len(de->name_len, dir);
2093         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2094         if (de->inode) {
2095                 struct ext4_dir_entry_2 *de1 =
2096                         (struct ext4_dir_entry_2 *)((char *)de + nlen);
2097                 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2098                 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2099                 de = de1;
2100         }
2101         de->file_type = EXT4_FT_UNKNOWN;
2102         de->inode = cpu_to_le32(inode->i_ino);
2103         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2104         de->name_len = fname_len(fname);
2105         memcpy(de->name, fname_name(fname), fname_len(fname));
2106         if (ext4_hash_in_dirent(dir)) {
2107                 struct dx_hash_info *hinfo = &fname->hinfo;
2108
2109                 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2110                 EXT4_DIRENT_HASHES(de)->minor_hash =
2111                                                 cpu_to_le32(hinfo->minor_hash);
2112         }
2113 }
2114
2115 /*
2116  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
2117  * it points to a directory entry which is guaranteed to be large
2118  * enough for new directory entry.  If de is NULL, then
2119  * add_dirent_to_buf will attempt search the directory block for
2120  * space.  It will return -ENOSPC if no space is available, and -EIO
2121  * and -EEXIST if directory entry already exists.
2122  */
2123 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2124                              struct inode *dir,
2125                              struct inode *inode, struct ext4_dir_entry_2 *de,
2126                              struct buffer_head *bh)
2127 {
2128         unsigned int    blocksize = dir->i_sb->s_blocksize;
2129         int             csum_size = 0;
2130         int             err, err2;
2131
2132         if (ext4_has_metadata_csum(inode->i_sb))
2133                 csum_size = sizeof(struct ext4_dir_entry_tail);
2134
2135         if (!de) {
2136                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2137                                         blocksize - csum_size, fname, &de);
2138                 if (err)
2139                         return err;
2140         }
2141         BUFFER_TRACE(bh, "get_write_access");
2142         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2143                                             EXT4_JTR_NONE);
2144         if (err) {
2145                 ext4_std_error(dir->i_sb, err);
2146                 return err;
2147         }
2148
2149         /* By now the buffer is marked for journaling */
2150         ext4_insert_dentry(dir, inode, de, blocksize, fname);
2151
2152         /*
2153          * XXX shouldn't update any times until successful
2154          * completion of syscall, but too many callers depend
2155          * on this.
2156          *
2157          * XXX similarly, too many callers depend on
2158          * ext4_new_inode() setting the times, but error
2159          * recovery deletes the inode, so the worst that can
2160          * happen is that the times are slightly out of date
2161          * and/or different from the directory change time.
2162          */
2163         dir->i_mtime = dir->i_ctime = current_time(dir);
2164         ext4_update_dx_flag(dir);
2165         inode_inc_iversion(dir);
2166         err2 = ext4_mark_inode_dirty(handle, dir);
2167         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2168         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2169         if (err)
2170                 ext4_std_error(dir->i_sb, err);
2171         return err ? err : err2;
2172 }
2173
2174 /*
2175  * This converts a one block unindexed directory to a 3 block indexed
2176  * directory, and adds the dentry to the indexed directory.
2177  */
2178 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2179                             struct inode *dir,
2180                             struct inode *inode, struct buffer_head *bh)
2181 {
2182         struct buffer_head *bh2;
2183         struct dx_root  *root;
2184         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2185         struct dx_entry *entries;
2186         struct ext4_dir_entry_2 *de, *de2;
2187         char            *data2, *top;
2188         unsigned        len;
2189         int             retval;
2190         unsigned        blocksize;
2191         ext4_lblk_t  block;
2192         struct fake_dirent *fde;
2193         int csum_size = 0;
2194
2195         if (ext4_has_metadata_csum(inode->i_sb))
2196                 csum_size = sizeof(struct ext4_dir_entry_tail);
2197
2198         blocksize =  dir->i_sb->s_blocksize;
2199         dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2200         BUFFER_TRACE(bh, "get_write_access");
2201         retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2202                                                EXT4_JTR_NONE);
2203         if (retval) {
2204                 ext4_std_error(dir->i_sb, retval);
2205                 brelse(bh);
2206                 return retval;
2207         }
2208         root = (struct dx_root *) bh->b_data;
2209
2210         /* The 0th block becomes the root, move the dirents out */
2211         fde = &root->dotdot;
2212         de = (struct ext4_dir_entry_2 *)((char *)fde +
2213                 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2214         if ((char *) de >= (((char *) root) + blocksize)) {
2215                 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2216                 brelse(bh);
2217                 return -EFSCORRUPTED;
2218         }
2219         len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2220
2221         /* Allocate new block for the 0th block's dirents */
2222         bh2 = ext4_append(handle, dir, &block);
2223         if (IS_ERR(bh2)) {
2224                 brelse(bh);
2225                 return PTR_ERR(bh2);
2226         }
2227         ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2228         data2 = bh2->b_data;
2229
2230         memcpy(data2, de, len);
2231         memset(de, 0, len); /* wipe old data */
2232         de = (struct ext4_dir_entry_2 *) data2;
2233         top = data2 + len;
2234         while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2235                 de = de2;
2236         de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2237                                            (char *) de, blocksize);
2238
2239         if (csum_size)
2240                 ext4_initialize_dirent_tail(bh2, blocksize);
2241
2242         /* Initialize the root; the dot dirents already exist */
2243         de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2244         de->rec_len = ext4_rec_len_to_disk(
2245                         blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2246         memset (&root->info, 0, sizeof(root->info));
2247         root->info.info_length = sizeof(root->info);
2248         if (ext4_hash_in_dirent(dir))
2249                 root->info.hash_version = DX_HASH_SIPHASH;
2250         else
2251                 root->info.hash_version =
2252                                 EXT4_SB(dir->i_sb)->s_def_hash_version;
2253
2254         entries = root->entries;
2255         dx_set_block(entries, 1);
2256         dx_set_count(entries, 1);
2257         dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2258
2259         /* Initialize as for dx_probe */
2260         fname->hinfo.hash_version = root->info.hash_version;
2261         if (fname->hinfo.hash_version <= DX_HASH_TEA)
2262                 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2263         fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2264
2265         /* casefolded encrypted hashes are computed on fname setup */
2266         if (!ext4_hash_in_dirent(dir))
2267                 ext4fs_dirhash(dir, fname_name(fname),
2268                                 fname_len(fname), &fname->hinfo);
2269
2270         memset(frames, 0, sizeof(frames));
2271         frame = frames;
2272         frame->entries = entries;
2273         frame->at = entries;
2274         frame->bh = bh;
2275
2276         retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2277         if (retval)
2278                 goto out_frames;
2279         retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2280         if (retval)
2281                 goto out_frames;
2282
2283         de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2284         if (IS_ERR(de)) {
2285                 retval = PTR_ERR(de);
2286                 goto out_frames;
2287         }
2288
2289         retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2290 out_frames:
2291         /*
2292          * Even if the block split failed, we have to properly write
2293          * out all the changes we did so far. Otherwise we can end up
2294          * with corrupted filesystem.
2295          */
2296         if (retval)
2297                 ext4_mark_inode_dirty(handle, dir);
2298         dx_release(frames);
2299         brelse(bh2);
2300         return retval;
2301 }
2302
2303 /*
2304  *      ext4_add_entry()
2305  *
2306  * adds a file entry to the specified directory, using the same
2307  * semantics as ext4_find_entry(). It returns NULL if it failed.
2308  *
2309  * NOTE!! The inode part of 'de' is left at 0 - which means you
2310  * may not sleep between calling this and putting something into
2311  * the entry, as someone else might have used it while you slept.
2312  */
2313 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2314                           struct inode *inode)
2315 {
2316         struct inode *dir = d_inode(dentry->d_parent);
2317         struct buffer_head *bh = NULL;
2318         struct ext4_dir_entry_2 *de;
2319         struct super_block *sb;
2320         struct ext4_filename fname;
2321         int     retval;
2322         int     dx_fallback=0;
2323         unsigned blocksize;
2324         ext4_lblk_t block, blocks;
2325         int     csum_size = 0;
2326
2327         if (ext4_has_metadata_csum(inode->i_sb))
2328                 csum_size = sizeof(struct ext4_dir_entry_tail);
2329
2330         sb = dir->i_sb;
2331         blocksize = sb->s_blocksize;
2332         if (!dentry->d_name.len)
2333                 return -EINVAL;
2334
2335         if (fscrypt_is_nokey_name(dentry))
2336                 return -ENOKEY;
2337
2338 #if IS_ENABLED(CONFIG_UNICODE)
2339         if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2340             sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2341                 return -EINVAL;
2342 #endif
2343
2344         retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2345         if (retval)
2346                 return retval;
2347
2348         if (ext4_has_inline_data(dir)) {
2349                 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2350                 if (retval < 0)
2351                         goto out;
2352                 if (retval == 1) {
2353                         retval = 0;
2354                         goto out;
2355                 }
2356         }
2357
2358         if (is_dx(dir)) {
2359                 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2360                 if (!retval || (retval != ERR_BAD_DX_DIR))
2361                         goto out;
2362                 /* Can we just ignore htree data? */
2363                 if (ext4_has_metadata_csum(sb)) {
2364                         EXT4_ERROR_INODE(dir,
2365                                 "Directory has corrupted htree index.");
2366                         retval = -EFSCORRUPTED;
2367                         goto out;
2368                 }
2369                 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2370                 dx_fallback++;
2371                 retval = ext4_mark_inode_dirty(handle, dir);
2372                 if (unlikely(retval))
2373                         goto out;
2374         }
2375         blocks = dir->i_size >> sb->s_blocksize_bits;
2376         for (block = 0; block < blocks; block++) {
2377                 bh = ext4_read_dirblock(dir, block, DIRENT);
2378                 if (bh == NULL) {
2379                         bh = ext4_bread(handle, dir, block,
2380                                         EXT4_GET_BLOCKS_CREATE);
2381                         goto add_to_new_block;
2382                 }
2383                 if (IS_ERR(bh)) {
2384                         retval = PTR_ERR(bh);
2385                         bh = NULL;
2386                         goto out;
2387                 }
2388                 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2389                                            NULL, bh);
2390                 if (retval != -ENOSPC)
2391                         goto out;
2392
2393                 if (blocks == 1 && !dx_fallback &&
2394                     ext4_has_feature_dir_index(sb)) {
2395                         retval = make_indexed_dir(handle, &fname, dir,
2396                                                   inode, bh);
2397                         bh = NULL; /* make_indexed_dir releases bh */
2398                         goto out;
2399                 }
2400                 brelse(bh);
2401         }
2402         bh = ext4_append(handle, dir, &block);
2403 add_to_new_block:
2404         if (IS_ERR(bh)) {
2405                 retval = PTR_ERR(bh);
2406                 bh = NULL;
2407                 goto out;
2408         }
2409         de = (struct ext4_dir_entry_2 *) bh->b_data;
2410         de->inode = 0;
2411         de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2412
2413         if (csum_size)
2414                 ext4_initialize_dirent_tail(bh, blocksize);
2415
2416         retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2417 out:
2418         ext4_fname_free_filename(&fname);
2419         brelse(bh);
2420         if (retval == 0)
2421                 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2422         return retval;
2423 }
2424
2425 /*
2426  * Returns 0 for success, or a negative error value
2427  */
2428 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2429                              struct inode *dir, struct inode *inode)
2430 {
2431         struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2432         struct dx_entry *entries, *at;
2433         struct buffer_head *bh;
2434         struct super_block *sb = dir->i_sb;
2435         struct ext4_dir_entry_2 *de;
2436         int restart;
2437         int err;
2438
2439 again:
2440         restart = 0;
2441         frame = dx_probe(fname, dir, NULL, frames);
2442         if (IS_ERR(frame))
2443                 return PTR_ERR(frame);
2444         entries = frame->entries;
2445         at = frame->at;
2446         bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2447         if (IS_ERR(bh)) {
2448                 err = PTR_ERR(bh);
2449                 bh = NULL;
2450                 goto cleanup;
2451         }
2452
2453         BUFFER_TRACE(bh, "get_write_access");
2454         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2455         if (err)
2456                 goto journal_error;
2457
2458         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2459         if (err != -ENOSPC)
2460                 goto cleanup;
2461
2462         err = 0;
2463         /* Block full, should compress but for now just split */
2464         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2465                        dx_get_count(entries), dx_get_limit(entries)));
2466         /* Need to split index? */
2467         if (dx_get_count(entries) == dx_get_limit(entries)) {
2468                 ext4_lblk_t newblock;
2469                 int levels = frame - frames + 1;
2470                 unsigned int icount;
2471                 int add_level = 1;
2472                 struct dx_entry *entries2;
2473                 struct dx_node *node2;
2474                 struct buffer_head *bh2;
2475
2476                 while (frame > frames) {
2477                         if (dx_get_count((frame - 1)->entries) <
2478                             dx_get_limit((frame - 1)->entries)) {
2479                                 add_level = 0;
2480                                 break;
2481                         }
2482                         frame--; /* split higher index block */
2483                         at = frame->at;
2484                         entries = frame->entries;
2485                         restart = 1;
2486                 }
2487                 if (add_level && levels == ext4_dir_htree_level(sb)) {
2488                         ext4_warning(sb, "Directory (ino: %lu) index full, "
2489                                          "reach max htree level :%d",
2490                                          dir->i_ino, levels);
2491                         if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2492                                 ext4_warning(sb, "Large directory feature is "
2493                                                  "not enabled on this "
2494                                                  "filesystem");
2495                         }
2496                         err = -ENOSPC;
2497                         goto cleanup;
2498                 }
2499                 icount = dx_get_count(entries);
2500                 bh2 = ext4_append(handle, dir, &newblock);
2501                 if (IS_ERR(bh2)) {
2502                         err = PTR_ERR(bh2);
2503                         goto cleanup;
2504                 }
2505                 node2 = (struct dx_node *)(bh2->b_data);
2506                 entries2 = node2->entries;
2507                 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2508                 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2509                                                            sb->s_blocksize);
2510                 BUFFER_TRACE(frame->bh, "get_write_access");
2511                 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2512                                                     EXT4_JTR_NONE);
2513                 if (err)
2514                         goto journal_error;
2515                 if (!add_level) {
2516                         unsigned icount1 = icount/2, icount2 = icount - icount1;
2517                         unsigned hash2 = dx_get_hash(entries + icount1);
2518                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2519                                        icount1, icount2));
2520
2521                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2522                         err = ext4_journal_get_write_access(handle, sb,
2523                                                             (frame - 1)->bh,
2524                                                             EXT4_JTR_NONE);
2525                         if (err)
2526                                 goto journal_error;
2527
2528                         memcpy((char *) entries2, (char *) (entries + icount1),
2529                                icount2 * sizeof(struct dx_entry));
2530                         dx_set_count(entries, icount1);
2531                         dx_set_count(entries2, icount2);
2532                         dx_set_limit(entries2, dx_node_limit(dir));
2533
2534                         /* Which index block gets the new entry? */
2535                         if (at - entries >= icount1) {
2536                                 frame->at = at - entries - icount1 + entries2;
2537                                 frame->entries = entries = entries2;
2538                                 swap(frame->bh, bh2);
2539                         }
2540                         dx_insert_block((frame - 1), hash2, newblock);
2541                         dxtrace(dx_show_index("node", frame->entries));
2542                         dxtrace(dx_show_index("node",
2543                                ((struct dx_node *) bh2->b_data)->entries));
2544                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2545                         if (err)
2546                                 goto journal_error;
2547                         brelse (bh2);
2548                         err = ext4_handle_dirty_dx_node(handle, dir,
2549                                                    (frame - 1)->bh);
2550                         if (err)
2551                                 goto journal_error;
2552                         err = ext4_handle_dirty_dx_node(handle, dir,
2553                                                         frame->bh);
2554                         if (restart || err)
2555                                 goto journal_error;
2556                 } else {
2557                         struct dx_root *dxroot;
2558                         memcpy((char *) entries2, (char *) entries,
2559                                icount * sizeof(struct dx_entry));
2560                         dx_set_limit(entries2, dx_node_limit(dir));
2561
2562                         /* Set up root */
2563                         dx_set_count(entries, 1);
2564                         dx_set_block(entries + 0, newblock);
2565                         dxroot = (struct dx_root *)frames[0].bh->b_data;
2566                         dxroot->info.indirect_levels += 1;
2567                         dxtrace(printk(KERN_DEBUG
2568                                        "Creating %d level index...\n",
2569                                        dxroot->info.indirect_levels));
2570                         err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2571                         if (err)
2572                                 goto journal_error;
2573                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2574                         brelse(bh2);
2575                         restart = 1;
2576                         goto journal_error;
2577                 }
2578         }
2579         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2580         if (IS_ERR(de)) {
2581                 err = PTR_ERR(de);
2582                 goto cleanup;
2583         }
2584         err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2585         goto cleanup;
2586
2587 journal_error:
2588         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2589 cleanup:
2590         brelse(bh);
2591         dx_release(frames);
2592         /* @restart is true means htree-path has been changed, we need to
2593          * repeat dx_probe() to find out valid htree-path
2594          */
2595         if (restart && err == 0)
2596                 goto again;
2597         return err;
2598 }
2599
2600 /*
2601  * ext4_generic_delete_entry deletes a directory entry by merging it
2602  * with the previous entry
2603  */
2604 int ext4_generic_delete_entry(struct inode *dir,
2605                               struct ext4_dir_entry_2 *de_del,
2606                               struct buffer_head *bh,
2607                               void *entry_buf,
2608                               int buf_size,
2609                               int csum_size)
2610 {
2611         struct ext4_dir_entry_2 *de, *pde;
2612         unsigned int blocksize = dir->i_sb->s_blocksize;
2613         int i;
2614
2615         i = 0;
2616         pde = NULL;
2617         de = entry_buf;
2618         while (i < buf_size - csum_size) {
2619                 if (ext4_check_dir_entry(dir, NULL, de, bh,
2620                                          entry_buf, buf_size, i))
2621                         return -EFSCORRUPTED;
2622                 if (de == de_del)  {
2623                         if (pde) {
2624                                 pde->rec_len = ext4_rec_len_to_disk(
2625                                         ext4_rec_len_from_disk(pde->rec_len,
2626                                                                blocksize) +
2627                                         ext4_rec_len_from_disk(de->rec_len,
2628                                                                blocksize),
2629                                         blocksize);
2630
2631                                 /* wipe entire dir_entry */
2632                                 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2633                                                                 blocksize));
2634                         } else {
2635                                 /* wipe dir_entry excluding the rec_len field */
2636                                 de->inode = 0;
2637                                 memset(&de->name_len, 0,
2638                                         ext4_rec_len_from_disk(de->rec_len,
2639                                                                 blocksize) -
2640                                         offsetof(struct ext4_dir_entry_2,
2641                                                                 name_len));
2642                         }
2643
2644                         inode_inc_iversion(dir);
2645                         return 0;
2646                 }
2647                 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2648                 pde = de;
2649                 de = ext4_next_entry(de, blocksize);
2650         }
2651         return -ENOENT;
2652 }
2653
2654 static int ext4_delete_entry(handle_t *handle,
2655                              struct inode *dir,
2656                              struct ext4_dir_entry_2 *de_del,
2657                              struct buffer_head *bh)
2658 {
2659         int err, csum_size = 0;
2660
2661         if (ext4_has_inline_data(dir)) {
2662                 int has_inline_data = 1;
2663                 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2664                                                &has_inline_data);
2665                 if (has_inline_data)
2666                         return err;
2667         }
2668
2669         if (ext4_has_metadata_csum(dir->i_sb))
2670                 csum_size = sizeof(struct ext4_dir_entry_tail);
2671
2672         BUFFER_TRACE(bh, "get_write_access");
2673         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2674                                             EXT4_JTR_NONE);
2675         if (unlikely(err))
2676                 goto out;
2677
2678         err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2679                                         dir->i_sb->s_blocksize, csum_size);
2680         if (err)
2681                 goto out;
2682
2683         BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2684         err = ext4_handle_dirty_dirblock(handle, dir, bh);
2685         if (unlikely(err))
2686                 goto out;
2687
2688         return 0;
2689 out:
2690         if (err != -ENOENT)
2691                 ext4_std_error(dir->i_sb, err);
2692         return err;
2693 }
2694
2695 /*
2696  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2697  * since this indicates that nlinks count was previously 1 to avoid overflowing
2698  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2699  * that subdirectory link counts are not being maintained accurately.
2700  *
2701  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2702  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2703  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2704  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2705  */
2706 static void ext4_inc_count(struct inode *inode)
2707 {
2708         inc_nlink(inode);
2709         if (is_dx(inode) &&
2710             (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2711                 set_nlink(inode, 1);
2712 }
2713
2714 /*
2715  * If a directory had nlink == 1, then we should let it be 1. This indicates
2716  * directory has >EXT4_LINK_MAX subdirs.
2717  */
2718 static void ext4_dec_count(struct inode *inode)
2719 {
2720         if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2721                 drop_nlink(inode);
2722 }
2723
2724
2725 /*
2726  * Add non-directory inode to a directory. On success, the inode reference is
2727  * consumed by dentry is instantiation. This is also indicated by clearing of
2728  * *inodep pointer. On failure, the caller is responsible for dropping the
2729  * inode reference in the safe context.
2730  */
2731 static int ext4_add_nondir(handle_t *handle,
2732                 struct dentry *dentry, struct inode **inodep)
2733 {
2734         struct inode *dir = d_inode(dentry->d_parent);
2735         struct inode *inode = *inodep;
2736         int err = ext4_add_entry(handle, dentry, inode);
2737         if (!err) {
2738                 err = ext4_mark_inode_dirty(handle, inode);
2739                 if (IS_DIRSYNC(dir))
2740                         ext4_handle_sync(handle);
2741                 d_instantiate_new(dentry, inode);
2742                 *inodep = NULL;
2743                 return err;
2744         }
2745         drop_nlink(inode);
2746         ext4_orphan_add(handle, inode);
2747         unlock_new_inode(inode);
2748         return err;
2749 }
2750
2751 /*
2752  * By the time this is called, we already have created
2753  * the directory cache entry for the new file, but it
2754  * is so far negative - it has no inode.
2755  *
2756  * If the create succeeds, we fill in the inode information
2757  * with d_instantiate().
2758  */
2759 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2760                        struct dentry *dentry, umode_t mode, bool excl)
2761 {
2762         handle_t *handle;
2763         struct inode *inode;
2764         int err, credits, retries = 0;
2765
2766         err = dquot_initialize(dir);
2767         if (err)
2768                 return err;
2769
2770         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2771                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2772 retry:
2773         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2774                                             0, NULL, EXT4_HT_DIR, credits);
2775         handle = ext4_journal_current_handle();
2776         err = PTR_ERR(inode);
2777         if (!IS_ERR(inode)) {
2778                 inode->i_op = &ext4_file_inode_operations;
2779                 inode->i_fop = &ext4_file_operations;
2780                 ext4_set_aops(inode);
2781                 err = ext4_add_nondir(handle, dentry, &inode);
2782                 if (!err)
2783                         ext4_fc_track_create(handle, dentry);
2784         }
2785         if (handle)
2786                 ext4_journal_stop(handle);
2787         if (!IS_ERR_OR_NULL(inode))
2788                 iput(inode);
2789         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2790                 goto retry;
2791         return err;
2792 }
2793
2794 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2795                       struct dentry *dentry, umode_t mode, dev_t rdev)
2796 {
2797         handle_t *handle;
2798         struct inode *inode;
2799         int err, credits, retries = 0;
2800
2801         err = dquot_initialize(dir);
2802         if (err)
2803                 return err;
2804
2805         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2806                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2807 retry:
2808         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2809                                             0, NULL, EXT4_HT_DIR, credits);
2810         handle = ext4_journal_current_handle();
2811         err = PTR_ERR(inode);
2812         if (!IS_ERR(inode)) {
2813                 init_special_inode(inode, inode->i_mode, rdev);
2814                 inode->i_op = &ext4_special_inode_operations;
2815                 err = ext4_add_nondir(handle, dentry, &inode);
2816                 if (!err)
2817                         ext4_fc_track_create(handle, dentry);
2818         }
2819         if (handle)
2820                 ext4_journal_stop(handle);
2821         if (!IS_ERR_OR_NULL(inode))
2822                 iput(inode);
2823         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2824                 goto retry;
2825         return err;
2826 }
2827
2828 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2829                         struct dentry *dentry, umode_t mode)
2830 {
2831         handle_t *handle;
2832         struct inode *inode;
2833         int err, retries = 0;
2834
2835         err = dquot_initialize(dir);
2836         if (err)
2837                 return err;
2838
2839 retry:
2840         inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2841                                             NULL, 0, NULL,
2842                                             EXT4_HT_DIR,
2843                         EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2844                           4 + EXT4_XATTR_TRANS_BLOCKS);
2845         handle = ext4_journal_current_handle();
2846         err = PTR_ERR(inode);
2847         if (!IS_ERR(inode)) {
2848                 inode->i_op = &ext4_file_inode_operations;
2849                 inode->i_fop = &ext4_file_operations;
2850                 ext4_set_aops(inode);
2851                 d_tmpfile(dentry, inode);
2852                 err = ext4_orphan_add(handle, inode);
2853                 if (err)
2854                         goto err_unlock_inode;
2855                 mark_inode_dirty(inode);
2856                 unlock_new_inode(inode);
2857         }
2858         if (handle)
2859                 ext4_journal_stop(handle);
2860         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2861                 goto retry;
2862         return err;
2863 err_unlock_inode:
2864         ext4_journal_stop(handle);
2865         unlock_new_inode(inode);
2866         return err;
2867 }
2868
2869 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2870                           struct ext4_dir_entry_2 *de,
2871                           int blocksize, int csum_size,
2872                           unsigned int parent_ino, int dotdot_real_len)
2873 {
2874         de->inode = cpu_to_le32(inode->i_ino);
2875         de->name_len = 1;
2876         de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2877                                            blocksize);
2878         strcpy(de->name, ".");
2879         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2880
2881         de = ext4_next_entry(de, blocksize);
2882         de->inode = cpu_to_le32(parent_ino);
2883         de->name_len = 2;
2884         if (!dotdot_real_len)
2885                 de->rec_len = ext4_rec_len_to_disk(blocksize -
2886                                         (csum_size + ext4_dir_rec_len(1, NULL)),
2887                                         blocksize);
2888         else
2889                 de->rec_len = ext4_rec_len_to_disk(
2890                                         ext4_dir_rec_len(de->name_len, NULL),
2891                                         blocksize);
2892         strcpy(de->name, "..");
2893         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2894
2895         return ext4_next_entry(de, blocksize);
2896 }
2897
2898 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2899                              struct inode *inode)
2900 {
2901         struct buffer_head *dir_block = NULL;
2902         struct ext4_dir_entry_2 *de;
2903         ext4_lblk_t block = 0;
2904         unsigned int blocksize = dir->i_sb->s_blocksize;
2905         int csum_size = 0;
2906         int err;
2907
2908         if (ext4_has_metadata_csum(dir->i_sb))
2909                 csum_size = sizeof(struct ext4_dir_entry_tail);
2910
2911         if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2912                 err = ext4_try_create_inline_dir(handle, dir, inode);
2913                 if (err < 0 && err != -ENOSPC)
2914                         goto out;
2915                 if (!err)
2916                         goto out;
2917         }
2918
2919         inode->i_size = 0;
2920         dir_block = ext4_append(handle, inode, &block);
2921         if (IS_ERR(dir_block))
2922                 return PTR_ERR(dir_block);
2923         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2924         ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2925         set_nlink(inode, 2);
2926         if (csum_size)
2927                 ext4_initialize_dirent_tail(dir_block, blocksize);
2928
2929         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2930         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2931         if (err)
2932                 goto out;
2933         set_buffer_verified(dir_block);
2934 out:
2935         brelse(dir_block);
2936         return err;
2937 }
2938
2939 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2940                       struct dentry *dentry, umode_t mode)
2941 {
2942         handle_t *handle;
2943         struct inode *inode;
2944         int err, err2 = 0, credits, retries = 0;
2945
2946         if (EXT4_DIR_LINK_MAX(dir))
2947                 return -EMLINK;
2948
2949         err = dquot_initialize(dir);
2950         if (err)
2951                 return err;
2952
2953         credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2954                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2955 retry:
2956         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2957                                             &dentry->d_name,
2958                                             0, NULL, EXT4_HT_DIR, credits);
2959         handle = ext4_journal_current_handle();
2960         err = PTR_ERR(inode);
2961         if (IS_ERR(inode))
2962                 goto out_stop;
2963
2964         inode->i_op = &ext4_dir_inode_operations;
2965         inode->i_fop = &ext4_dir_operations;
2966         err = ext4_init_new_dir(handle, dir, inode);
2967         if (err)
2968                 goto out_clear_inode;
2969         err = ext4_mark_inode_dirty(handle, inode);
2970         if (!err)
2971                 err = ext4_add_entry(handle, dentry, inode);
2972         if (err) {
2973 out_clear_inode:
2974                 clear_nlink(inode);
2975                 ext4_orphan_add(handle, inode);
2976                 unlock_new_inode(inode);
2977                 err2 = ext4_mark_inode_dirty(handle, inode);
2978                 if (unlikely(err2))
2979                         err = err2;
2980                 ext4_journal_stop(handle);
2981                 iput(inode);
2982                 goto out_retry;
2983         }
2984         ext4_inc_count(dir);
2985
2986         ext4_update_dx_flag(dir);
2987         err = ext4_mark_inode_dirty(handle, dir);
2988         if (err)
2989                 goto out_clear_inode;
2990         d_instantiate_new(dentry, inode);
2991         ext4_fc_track_create(handle, dentry);
2992         if (IS_DIRSYNC(dir))
2993                 ext4_handle_sync(handle);
2994
2995 out_stop:
2996         if (handle)
2997                 ext4_journal_stop(handle);
2998 out_retry:
2999         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3000                 goto retry;
3001         return err;
3002 }
3003
3004 /*
3005  * routine to check that the specified directory is empty (for rmdir)
3006  */
3007 bool ext4_empty_dir(struct inode *inode)
3008 {
3009         unsigned int offset;
3010         struct buffer_head *bh;
3011         struct ext4_dir_entry_2 *de;
3012         struct super_block *sb;
3013
3014         if (ext4_has_inline_data(inode)) {
3015                 int has_inline_data = 1;
3016                 int ret;
3017
3018                 ret = empty_inline_dir(inode, &has_inline_data);
3019                 if (has_inline_data)
3020                         return ret;
3021         }
3022
3023         sb = inode->i_sb;
3024         if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3025                                         ext4_dir_rec_len(2, NULL)) {
3026                 EXT4_ERROR_INODE(inode, "invalid size");
3027                 return false;
3028         }
3029         /* The first directory block must not be a hole,
3030          * so treat it as DIRENT_HTREE
3031          */
3032         bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3033         if (IS_ERR(bh))
3034                 return false;
3035
3036         de = (struct ext4_dir_entry_2 *) bh->b_data;
3037         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3038                                  0) ||
3039             le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3040                 ext4_warning_inode(inode, "directory missing '.'");
3041                 brelse(bh);
3042                 return false;
3043         }
3044         offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3045         de = ext4_next_entry(de, sb->s_blocksize);
3046         if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3047                                  offset) ||
3048             le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3049                 ext4_warning_inode(inode, "directory missing '..'");
3050                 brelse(bh);
3051                 return false;
3052         }
3053         offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3054         while (offset < inode->i_size) {
3055                 if (!(offset & (sb->s_blocksize - 1))) {
3056                         unsigned int lblock;
3057                         brelse(bh);
3058                         lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3059                         bh = ext4_read_dirblock(inode, lblock, EITHER);
3060                         if (bh == NULL) {
3061                                 offset += sb->s_blocksize;
3062                                 continue;
3063                         }
3064                         if (IS_ERR(bh))
3065                                 return false;
3066                 }
3067                 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3068                                         (offset & (sb->s_blocksize - 1)));
3069                 if (ext4_check_dir_entry(inode, NULL, de, bh,
3070                                          bh->b_data, bh->b_size, offset)) {
3071                         offset = (offset | (sb->s_blocksize - 1)) + 1;
3072                         continue;
3073                 }
3074                 if (le32_to_cpu(de->inode)) {
3075                         brelse(bh);
3076                         return false;
3077                 }
3078                 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3079         }
3080         brelse(bh);
3081         return true;
3082 }
3083
3084 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3085 {
3086         int retval;
3087         struct inode *inode;
3088         struct buffer_head *bh;
3089         struct ext4_dir_entry_2 *de;
3090         handle_t *handle = NULL;
3091
3092         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3093                 return -EIO;
3094
3095         /* Initialize quotas before so that eventual writes go in
3096          * separate transaction */
3097         retval = dquot_initialize(dir);
3098         if (retval)
3099                 return retval;
3100         retval = dquot_initialize(d_inode(dentry));
3101         if (retval)
3102                 return retval;
3103
3104         retval = -ENOENT;
3105         bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3106         if (IS_ERR(bh))
3107                 return PTR_ERR(bh);
3108         if (!bh)
3109                 goto end_rmdir;
3110
3111         inode = d_inode(dentry);
3112
3113         retval = -EFSCORRUPTED;
3114         if (le32_to_cpu(de->inode) != inode->i_ino)
3115                 goto end_rmdir;
3116
3117         retval = -ENOTEMPTY;
3118         if (!ext4_empty_dir(inode))
3119                 goto end_rmdir;
3120
3121         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3122                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3123         if (IS_ERR(handle)) {
3124                 retval = PTR_ERR(handle);
3125                 handle = NULL;
3126                 goto end_rmdir;
3127         }
3128
3129         if (IS_DIRSYNC(dir))
3130                 ext4_handle_sync(handle);
3131
3132         retval = ext4_delete_entry(handle, dir, de, bh);
3133         if (retval)
3134                 goto end_rmdir;
3135         if (!EXT4_DIR_LINK_EMPTY(inode))
3136                 ext4_warning_inode(inode,
3137                              "empty directory '%.*s' has too many links (%u)",
3138                              dentry->d_name.len, dentry->d_name.name,
3139                              inode->i_nlink);
3140         inode_inc_iversion(inode);
3141         clear_nlink(inode);
3142         /* There's no need to set i_disksize: the fact that i_nlink is
3143          * zero will ensure that the right thing happens during any
3144          * recovery. */
3145         inode->i_size = 0;
3146         ext4_orphan_add(handle, inode);
3147         inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3148         retval = ext4_mark_inode_dirty(handle, inode);
3149         if (retval)
3150                 goto end_rmdir;
3151         ext4_dec_count(dir);
3152         ext4_update_dx_flag(dir);
3153         ext4_fc_track_unlink(handle, dentry);
3154         retval = ext4_mark_inode_dirty(handle, dir);
3155
3156 #if IS_ENABLED(CONFIG_UNICODE)
3157         /* VFS negative dentries are incompatible with Encoding and
3158          * Case-insensitiveness. Eventually we'll want avoid
3159          * invalidating the dentries here, alongside with returning the
3160          * negative dentries at ext4_lookup(), when it is better
3161          * supported by the VFS for the CI case.
3162          */
3163         if (IS_CASEFOLDED(dir))
3164                 d_invalidate(dentry);
3165 #endif
3166
3167 end_rmdir:
3168         brelse(bh);
3169         if (handle)
3170                 ext4_journal_stop(handle);
3171         return retval;
3172 }
3173
3174 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3175                   struct inode *inode)
3176 {
3177         int retval = -ENOENT;
3178         struct buffer_head *bh;
3179         struct ext4_dir_entry_2 *de;
3180         int skip_remove_dentry = 0;
3181
3182         bh = ext4_find_entry(dir, d_name, &de, NULL);
3183         if (IS_ERR(bh))
3184                 return PTR_ERR(bh);
3185
3186         if (!bh)
3187                 return -ENOENT;
3188
3189         if (le32_to_cpu(de->inode) != inode->i_ino) {
3190                 /*
3191                  * It's okay if we find dont find dentry which matches
3192                  * the inode. That's because it might have gotten
3193                  * renamed to a different inode number
3194                  */
3195                 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3196                         skip_remove_dentry = 1;
3197                 else
3198                         goto out;
3199         }
3200
3201         if (IS_DIRSYNC(dir))
3202                 ext4_handle_sync(handle);
3203
3204         if (!skip_remove_dentry) {
3205                 retval = ext4_delete_entry(handle, dir, de, bh);
3206                 if (retval)
3207                         goto out;
3208                 dir->i_ctime = dir->i_mtime = current_time(dir);
3209                 ext4_update_dx_flag(dir);
3210                 retval = ext4_mark_inode_dirty(handle, dir);
3211                 if (retval)
3212                         goto out;
3213         } else {
3214                 retval = 0;
3215         }
3216         if (inode->i_nlink == 0)
3217                 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3218                                    d_name->len, d_name->name);
3219         else
3220                 drop_nlink(inode);
3221         if (!inode->i_nlink)
3222                 ext4_orphan_add(handle, inode);
3223         inode->i_ctime = current_time(inode);
3224         retval = ext4_mark_inode_dirty(handle, inode);
3225
3226 out:
3227         brelse(bh);
3228         return retval;
3229 }
3230
3231 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3232 {
3233         handle_t *handle;
3234         int retval;
3235
3236         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3237                 return -EIO;
3238
3239         trace_ext4_unlink_enter(dir, dentry);
3240         /*
3241          * Initialize quotas before so that eventual writes go
3242          * in separate transaction
3243          */
3244         retval = dquot_initialize(dir);
3245         if (retval)
3246                 goto out_trace;
3247         retval = dquot_initialize(d_inode(dentry));
3248         if (retval)
3249                 goto out_trace;
3250
3251         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3252                                     EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3253         if (IS_ERR(handle)) {
3254                 retval = PTR_ERR(handle);
3255                 goto out_trace;
3256         }
3257
3258         retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3259         if (!retval)
3260                 ext4_fc_track_unlink(handle, dentry);
3261 #if IS_ENABLED(CONFIG_UNICODE)
3262         /* VFS negative dentries are incompatible with Encoding and
3263          * Case-insensitiveness. Eventually we'll want avoid
3264          * invalidating the dentries here, alongside with returning the
3265          * negative dentries at ext4_lookup(), when it is  better
3266          * supported by the VFS for the CI case.
3267          */
3268         if (IS_CASEFOLDED(dir))
3269                 d_invalidate(dentry);
3270 #endif
3271         if (handle)
3272                 ext4_journal_stop(handle);
3273
3274 out_trace:
3275         trace_ext4_unlink_exit(dentry, retval);
3276         return retval;
3277 }
3278
3279 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3280                                    struct fscrypt_str *disk_link)
3281 {
3282         struct buffer_head *bh;
3283         char *kaddr;
3284         int err = 0;
3285
3286         bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3287         if (IS_ERR(bh))
3288                 return PTR_ERR(bh);
3289
3290         BUFFER_TRACE(bh, "get_write_access");
3291         err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3292         if (err)
3293                 goto out;
3294
3295         kaddr = (char *)bh->b_data;
3296         memcpy(kaddr, disk_link->name, disk_link->len);
3297         inode->i_size = disk_link->len - 1;
3298         EXT4_I(inode)->i_disksize = inode->i_size;
3299         err = ext4_handle_dirty_metadata(handle, inode, bh);
3300 out:
3301         brelse(bh);
3302         return err;
3303 }
3304
3305 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3306                         struct dentry *dentry, const char *symname)
3307 {
3308         handle_t *handle;
3309         struct inode *inode;
3310         int err, len = strlen(symname);
3311         int credits;
3312         struct fscrypt_str disk_link;
3313         int retries = 0;
3314
3315         if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3316                 return -EIO;
3317
3318         err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3319                                       &disk_link);
3320         if (err)
3321                 return err;
3322
3323         err = dquot_initialize(dir);
3324         if (err)
3325                 return err;
3326
3327         /*
3328          * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3329          * directory. +3 for inode, inode bitmap, group descriptor allocation.
3330          * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3331          * modification.
3332          */
3333         credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3334                   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3335 retry:
3336         inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3337                                             &dentry->d_name, 0, NULL,
3338                                             EXT4_HT_DIR, credits);
3339         handle = ext4_journal_current_handle();
3340         if (IS_ERR(inode)) {
3341                 if (handle)
3342                         ext4_journal_stop(handle);
3343                 err = PTR_ERR(inode);
3344                 goto out_retry;
3345         }
3346
3347         if (IS_ENCRYPTED(inode)) {
3348                 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3349                 if (err)
3350                         goto err_drop_inode;
3351                 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3352         } else {
3353                 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3354                         inode->i_op = &ext4_symlink_inode_operations;
3355                 } else {
3356                         inode->i_op = &ext4_fast_symlink_inode_operations;
3357                         inode->i_link = (char *)&EXT4_I(inode)->i_data;
3358                 }
3359         }
3360
3361         if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3362                 /* alloc symlink block and fill it */
3363                 err = ext4_init_symlink_block(handle, inode, &disk_link);
3364                 if (err)
3365                         goto err_drop_inode;
3366         } else {
3367                 /* clear the extent format for fast symlink */
3368                 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3369                 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3370                        disk_link.len);
3371                 inode->i_size = disk_link.len - 1;
3372                 EXT4_I(inode)->i_disksize = inode->i_size;
3373         }
3374         err = ext4_add_nondir(handle, dentry, &inode);
3375         if (handle)
3376                 ext4_journal_stop(handle);
3377         iput(inode);
3378         goto out_retry;
3379
3380 err_drop_inode:
3381         clear_nlink(inode);
3382         ext4_orphan_add(handle, inode);
3383         unlock_new_inode(inode);
3384         if (handle)
3385                 ext4_journal_stop(handle);
3386         iput(inode);
3387 out_retry:
3388         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3389                 goto retry;
3390         if (disk_link.name != (unsigned char *)symname)
3391                 kfree(disk_link.name);
3392         return err;
3393 }
3394
3395 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3396 {
3397         handle_t *handle;
3398         int err, retries = 0;
3399 retry:
3400         handle = ext4_journal_start(dir, EXT4_HT_DIR,
3401                 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3402                  EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3403         if (IS_ERR(handle))
3404                 return PTR_ERR(handle);
3405
3406         if (IS_DIRSYNC(dir))
3407                 ext4_handle_sync(handle);
3408
3409         inode->i_ctime = current_time(inode);
3410         ext4_inc_count(inode);
3411         ihold(inode);
3412
3413         err = ext4_add_entry(handle, dentry, inode);
3414         if (!err) {
3415                 err = ext4_mark_inode_dirty(handle, inode);
3416                 /* this can happen only for tmpfile being
3417                  * linked the first time
3418                  */
3419                 if (inode->i_nlink == 1)
3420                         ext4_orphan_del(handle, inode);
3421                 d_instantiate(dentry, inode);
3422                 ext4_fc_track_link(handle, dentry);
3423         } else {
3424                 drop_nlink(inode);
3425                 iput(inode);
3426         }
3427         ext4_journal_stop(handle);
3428         if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3429                 goto retry;
3430         return err;
3431 }
3432
3433 static int ext4_link(struct dentry *old_dentry,
3434                      struct inode *dir, struct dentry *dentry)
3435 {
3436         struct inode *inode = d_inode(old_dentry);
3437         int err;
3438
3439         if (inode->i_nlink >= EXT4_LINK_MAX)
3440                 return -EMLINK;
3441
3442         err = fscrypt_prepare_link(old_dentry, dir, dentry);
3443         if (err)
3444                 return err;
3445
3446         if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3447             (!projid_eq(EXT4_I(dir)->i_projid,
3448                         EXT4_I(old_dentry->d_inode)->i_projid)))
3449                 return -EXDEV;
3450
3451         err = dquot_initialize(dir);
3452         if (err)
3453                 return err;
3454         return __ext4_link(dir, inode, dentry);
3455 }
3456
3457 /*
3458  * Try to find buffer head where contains the parent block.
3459  * It should be the inode block if it is inlined or the 1st block
3460  * if it is a normal dir.
3461  */
3462 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3463                                         struct inode *inode,
3464                                         int *retval,
3465                                         struct ext4_dir_entry_2 **parent_de,
3466                                         int *inlined)
3467 {
3468         struct buffer_head *bh;
3469
3470         if (!ext4_has_inline_data(inode)) {
3471                 struct ext4_dir_entry_2 *de;
3472                 unsigned int offset;
3473
3474                 /* The first directory block must not be a hole, so
3475                  * treat it as DIRENT_HTREE
3476                  */
3477                 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3478                 if (IS_ERR(bh)) {
3479                         *retval = PTR_ERR(bh);
3480                         return NULL;
3481                 }
3482
3483                 de = (struct ext4_dir_entry_2 *) bh->b_data;
3484                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3485                                          bh->b_size, 0) ||
3486                     le32_to_cpu(de->inode) != inode->i_ino ||
3487                     strcmp(".", de->name)) {
3488                         EXT4_ERROR_INODE(inode, "directory missing '.'");
3489                         brelse(bh);
3490                         *retval = -EFSCORRUPTED;
3491                         return NULL;
3492                 }
3493                 offset = ext4_rec_len_from_disk(de->rec_len,
3494                                                 inode->i_sb->s_blocksize);
3495                 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3496                 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3497                                          bh->b_size, offset) ||
3498                     le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3499                         EXT4_ERROR_INODE(inode, "directory missing '..'");
3500                         brelse(bh);
3501                         *retval = -EFSCORRUPTED;
3502                         return NULL;
3503                 }
3504                 *parent_de = de;
3505
3506                 return bh;
3507         }
3508
3509         *inlined = 1;
3510         return ext4_get_first_inline_block(inode, parent_de, retval);
3511 }
3512
3513 struct ext4_renament {
3514         struct inode *dir;
3515         struct dentry *dentry;
3516         struct inode *inode;
3517         bool is_dir;
3518         int dir_nlink_delta;
3519
3520         /* entry for "dentry" */
3521         struct buffer_head *bh;
3522         struct ext4_dir_entry_2 *de;
3523         int inlined;
3524
3525         /* entry for ".." in inode if it's a directory */
3526         struct buffer_head *dir_bh;
3527         struct ext4_dir_entry_2 *parent_de;
3528         int dir_inlined;
3529 };
3530
3531 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3532 {
3533         int retval;
3534
3535         ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3536                                               &retval, &ent->parent_de,
3537                                               &ent->dir_inlined);
3538         if (!ent->dir_bh)
3539                 return retval;
3540         if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3541                 return -EFSCORRUPTED;
3542         BUFFER_TRACE(ent->dir_bh, "get_write_access");
3543         return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3544                                              ent->dir_bh, EXT4_JTR_NONE);
3545 }
3546
3547 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3548                                   unsigned dir_ino)
3549 {
3550         int retval;
3551
3552         ent->parent_de->inode = cpu_to_le32(dir_ino);
3553         BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3554         if (!ent->dir_inlined) {
3555                 if (is_dx(ent->inode)) {
3556                         retval = ext4_handle_dirty_dx_node(handle,
3557                                                            ent->inode,
3558                                                            ent->dir_bh);
3559                 } else {
3560                         retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3561                                                             ent->dir_bh);
3562                 }
3563         } else {
3564                 retval = ext4_mark_inode_dirty(handle, ent->inode);
3565         }
3566         if (retval) {
3567                 ext4_std_error(ent->dir->i_sb, retval);
3568                 return retval;
3569         }
3570         return 0;
3571 }
3572
3573 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3574                        unsigned ino, unsigned file_type)
3575 {
3576         int retval, retval2;
3577
3578         BUFFER_TRACE(ent->bh, "get write access");
3579         retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3580                                                EXT4_JTR_NONE);
3581         if (retval)
3582                 return retval;
3583         ent->de->inode = cpu_to_le32(ino);
3584         if (ext4_has_feature_filetype(ent->dir->i_sb))
3585                 ent->de->file_type = file_type;
3586         inode_inc_iversion(ent->dir);
3587         ent->dir->i_ctime = ent->dir->i_mtime =
3588                 current_time(ent->dir);
3589         retval = ext4_mark_inode_dirty(handle, ent->dir);
3590         BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3591         if (!ent->inlined) {
3592                 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3593                 if (unlikely(retval2)) {
3594                         ext4_std_error(ent->dir->i_sb, retval2);
3595                         return retval2;
3596                 }
3597         }
3598         return retval;
3599 }
3600
3601 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3602                           unsigned ino, unsigned file_type)
3603 {
3604         struct ext4_renament old = *ent;
3605         int retval = 0;
3606
3607         /*
3608          * old->de could have moved from under us during make indexed dir,
3609          * so the old->de may no longer valid and need to find it again
3610          * before reset old inode info.
3611          */
3612         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3613         if (IS_ERR(old.bh))
3614                 retval = PTR_ERR(old.bh);
3615         if (!old.bh)
3616                 retval = -ENOENT;
3617         if (retval) {
3618                 ext4_std_error(old.dir->i_sb, retval);
3619                 return;
3620         }
3621
3622         ext4_setent(handle, &old, ino, file_type);
3623         brelse(old.bh);
3624 }
3625
3626 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3627                                   const struct qstr *d_name)
3628 {
3629         int retval = -ENOENT;
3630         struct buffer_head *bh;
3631         struct ext4_dir_entry_2 *de;
3632
3633         bh = ext4_find_entry(dir, d_name, &de, NULL);
3634         if (IS_ERR(bh))
3635                 return PTR_ERR(bh);
3636         if (bh) {
3637                 retval = ext4_delete_entry(handle, dir, de, bh);
3638                 brelse(bh);
3639         }
3640         return retval;
3641 }
3642
3643 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3644                                int force_reread)
3645 {
3646         int retval;
3647         /*
3648          * ent->de could have moved from under us during htree split, so make
3649          * sure that we are deleting the right entry.  We might also be pointing
3650          * to a stale entry in the unused part of ent->bh so just checking inum
3651          * and the name isn't enough.
3652          */
3653         if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3654             ent->de->name_len != ent->dentry->d_name.len ||
3655             strncmp(ent->de->name, ent->dentry->d_name.name,
3656                     ent->de->name_len) ||
3657             force_reread) {
3658                 retval = ext4_find_delete_entry(handle, ent->dir,
3659                                                 &ent->dentry->d_name);
3660         } else {
3661                 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3662                 if (retval == -ENOENT) {
3663                         retval = ext4_find_delete_entry(handle, ent->dir,
3664                                                         &ent->dentry->d_name);
3665                 }
3666         }
3667
3668         if (retval) {
3669                 ext4_warning_inode(ent->dir,
3670                                    "Deleting old file: nlink %d, error=%d",
3671                                    ent->dir->i_nlink, retval);
3672         }
3673 }
3674
3675 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3676 {
3677         if (ent->dir_nlink_delta) {
3678                 if (ent->dir_nlink_delta == -1)
3679                         ext4_dec_count(ent->dir);
3680                 else
3681                         ext4_inc_count(ent->dir);
3682                 ext4_mark_inode_dirty(handle, ent->dir);
3683         }
3684 }
3685
3686 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3687                                               struct ext4_renament *ent,
3688                                               int credits, handle_t **h)
3689 {
3690         struct inode *wh;
3691         handle_t *handle;
3692         int retries = 0;
3693
3694         /*
3695          * for inode block, sb block, group summaries,
3696          * and inode bitmap
3697          */
3698         credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3699                     EXT4_XATTR_TRANS_BLOCKS + 4);
3700 retry:
3701         wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3702                                          S_IFCHR | WHITEOUT_MODE,
3703                                          &ent->dentry->d_name, 0, NULL,
3704                                          EXT4_HT_DIR, credits);
3705
3706         handle = ext4_journal_current_handle();
3707         if (IS_ERR(wh)) {
3708                 if (handle)
3709                         ext4_journal_stop(handle);
3710                 if (PTR_ERR(wh) == -ENOSPC &&
3711                     ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3712                         goto retry;
3713         } else {
3714                 *h = handle;
3715                 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3716                 wh->i_op = &ext4_special_inode_operations;
3717         }
3718         return wh;
3719 }
3720
3721 /*
3722  * Anybody can rename anything with this: the permission checks are left to the
3723  * higher-level routines.
3724  *
3725  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3726  * while new_{dentry,inode) refers to the destination dentry/inode
3727  * This comes from rename(const char *oldpath, const char *newpath)
3728  */
3729 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3730                        struct dentry *old_dentry, struct inode *new_dir,
3731                        struct dentry *new_dentry, unsigned int flags)
3732 {
3733         handle_t *handle = NULL;
3734         struct ext4_renament old = {
3735                 .dir = old_dir,
3736                 .dentry = old_dentry,
3737                 .inode = d_inode(old_dentry),
3738         };
3739         struct ext4_renament new = {
3740                 .dir = new_dir,
3741                 .dentry = new_dentry,
3742                 .inode = d_inode(new_dentry),
3743         };
3744         int force_reread;
3745         int retval;
3746         struct inode *whiteout = NULL;
3747         int credits;
3748         u8 old_file_type;
3749
3750         if (new.inode && new.inode->i_nlink == 0) {
3751                 EXT4_ERROR_INODE(new.inode,
3752                                  "target of rename is already freed");
3753                 return -EFSCORRUPTED;
3754         }
3755
3756         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3757             (!projid_eq(EXT4_I(new_dir)->i_projid,
3758                         EXT4_I(old_dentry->d_inode)->i_projid)))
3759                 return -EXDEV;
3760
3761         retval = dquot_initialize(old.dir);
3762         if (retval)
3763                 return retval;
3764         retval = dquot_initialize(new.dir);
3765         if (retval)
3766                 return retval;
3767
3768         /* Initialize quotas before so that eventual writes go
3769          * in separate transaction */
3770         if (new.inode) {
3771                 retval = dquot_initialize(new.inode);
3772                 if (retval)
3773                         return retval;
3774         }
3775
3776         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3777         if (IS_ERR(old.bh))
3778                 return PTR_ERR(old.bh);
3779         /*
3780          *  Check for inode number is _not_ due to possible IO errors.
3781          *  We might rmdir the source, keep it as pwd of some process
3782          *  and merrily kill the link to whatever was created under the
3783          *  same name. Goodbye sticky bit ;-<
3784          */
3785         retval = -ENOENT;
3786         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3787                 goto release_bh;
3788
3789         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3790                                  &new.de, &new.inlined);
3791         if (IS_ERR(new.bh)) {
3792                 retval = PTR_ERR(new.bh);
3793                 new.bh = NULL;
3794                 goto release_bh;
3795         }
3796         if (new.bh) {
3797                 if (!new.inode) {
3798                         brelse(new.bh);
3799                         new.bh = NULL;
3800                 }
3801         }
3802         if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3803                 ext4_alloc_da_blocks(old.inode);
3804
3805         credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3806                    EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3807         if (!(flags & RENAME_WHITEOUT)) {
3808                 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3809                 if (IS_ERR(handle)) {
3810                         retval = PTR_ERR(handle);
3811                         goto release_bh;
3812                 }
3813         } else {
3814                 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3815                 if (IS_ERR(whiteout)) {
3816                         retval = PTR_ERR(whiteout);
3817                         goto release_bh;
3818                 }
3819         }
3820
3821         old_file_type = old.de->file_type;
3822         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3823                 ext4_handle_sync(handle);
3824
3825         if (S_ISDIR(old.inode->i_mode)) {
3826                 if (new.inode) {
3827                         retval = -ENOTEMPTY;
3828                         if (!ext4_empty_dir(new.inode))
3829                                 goto end_rename;
3830                 } else {
3831                         retval = -EMLINK;
3832                         if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3833                                 goto end_rename;
3834                 }
3835                 retval = ext4_rename_dir_prepare(handle, &old);
3836                 if (retval)
3837                         goto end_rename;
3838         }
3839         /*
3840          * If we're renaming a file within an inline_data dir and adding or
3841          * setting the new dirent causes a conversion from inline_data to
3842          * extents/blockmap, we need to force the dirent delete code to
3843          * re-read the directory, or else we end up trying to delete a dirent
3844          * from what is now the extent tree root (or a block map).
3845          */
3846         force_reread = (new.dir->i_ino == old.dir->i_ino &&
3847                         ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3848
3849         if (whiteout) {
3850                 /*
3851                  * Do this before adding a new entry, so the old entry is sure
3852                  * to be still pointing to the valid old entry.
3853                  */
3854                 retval = ext4_setent(handle, &old, whiteout->i_ino,
3855                                      EXT4_FT_CHRDEV);
3856                 if (retval)
3857                         goto end_rename;
3858                 retval = ext4_mark_inode_dirty(handle, whiteout);
3859                 if (unlikely(retval))
3860                         goto end_rename;
3861
3862         }
3863         if (!new.bh) {
3864                 retval = ext4_add_entry(handle, new.dentry, old.inode);
3865                 if (retval)
3866                         goto end_rename;
3867         } else {
3868                 retval = ext4_setent(handle, &new,
3869                                      old.inode->i_ino, old_file_type);
3870                 if (retval)
3871                         goto end_rename;
3872         }
3873         if (force_reread)
3874                 force_reread = !ext4_test_inode_flag(new.dir,
3875                                                      EXT4_INODE_INLINE_DATA);
3876
3877         /*
3878          * Like most other Unix systems, set the ctime for inodes on a
3879          * rename.
3880          */
3881         old.inode->i_ctime = current_time(old.inode);
3882         retval = ext4_mark_inode_dirty(handle, old.inode);
3883         if (unlikely(retval))
3884                 goto end_rename;
3885
3886         if (!whiteout) {
3887                 /*
3888                  * ok, that's it
3889                  */
3890                 ext4_rename_delete(handle, &old, force_reread);
3891         }
3892
3893         if (new.inode) {
3894                 ext4_dec_count(new.inode);
3895                 new.inode->i_ctime = current_time(new.inode);
3896         }
3897         old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3898         ext4_update_dx_flag(old.dir);
3899         if (old.dir_bh) {
3900                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3901                 if (retval)
3902                         goto end_rename;
3903
3904                 ext4_dec_count(old.dir);
3905                 if (new.inode) {
3906                         /* checked ext4_empty_dir above, can't have another
3907                          * parent, ext4_dec_count() won't work for many-linked
3908                          * dirs */
3909                         clear_nlink(new.inode);
3910                 } else {
3911                         ext4_inc_count(new.dir);
3912                         ext4_update_dx_flag(new.dir);
3913                         retval = ext4_mark_inode_dirty(handle, new.dir);
3914                         if (unlikely(retval))
3915                                 goto end_rename;
3916                 }
3917         }
3918         retval = ext4_mark_inode_dirty(handle, old.dir);
3919         if (unlikely(retval))
3920                 goto end_rename;
3921
3922         if (S_ISDIR(old.inode->i_mode)) {
3923                 /*
3924                  * We disable fast commits here that's because the
3925                  * replay code is not yet capable of changing dot dot
3926                  * dirents in directories.
3927                  */
3928                 ext4_fc_mark_ineligible(old.inode->i_sb,
3929                         EXT4_FC_REASON_RENAME_DIR, handle);
3930         } else {
3931                 struct super_block *sb = old.inode->i_sb;
3932
3933                 if (new.inode)
3934                         ext4_fc_track_unlink(handle, new.dentry);
3935                 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
3936                     !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
3937                     !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
3938                         __ext4_fc_track_link(handle, old.inode, new.dentry);
3939                         __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3940                         if (whiteout)
3941                                 __ext4_fc_track_create(handle, whiteout,
3942                                                        old.dentry);
3943                 }
3944         }
3945
3946         if (new.inode) {
3947                 retval = ext4_mark_inode_dirty(handle, new.inode);
3948                 if (unlikely(retval))
3949                         goto end_rename;
3950                 if (!new.inode->i_nlink)
3951                         ext4_orphan_add(handle, new.inode);
3952         }
3953         retval = 0;
3954
3955 end_rename:
3956         if (whiteout) {
3957                 if (retval) {
3958                         ext4_resetent(handle, &old,
3959                                       old.inode->i_ino, old_file_type);
3960                         drop_nlink(whiteout);
3961                         ext4_orphan_add(handle, whiteout);
3962                 }
3963                 unlock_new_inode(whiteout);
3964                 ext4_journal_stop(handle);
3965                 iput(whiteout);
3966         } else {
3967                 ext4_journal_stop(handle);
3968         }
3969 release_bh:
3970         brelse(old.dir_bh);
3971         brelse(old.bh);
3972         brelse(new.bh);
3973         return retval;
3974 }
3975
3976 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3977                              struct inode *new_dir, struct dentry *new_dentry)
3978 {
3979         handle_t *handle = NULL;
3980         struct ext4_renament old = {
3981                 .dir = old_dir,
3982                 .dentry = old_dentry,
3983                 .inode = d_inode(old_dentry),
3984         };
3985         struct ext4_renament new = {
3986                 .dir = new_dir,
3987                 .dentry = new_dentry,
3988                 .inode = d_inode(new_dentry),
3989         };
3990         u8 new_file_type;
3991         int retval;
3992         struct timespec64 ctime;
3993
3994         if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3995              !projid_eq(EXT4_I(new_dir)->i_projid,
3996                         EXT4_I(old_dentry->d_inode)->i_projid)) ||
3997             (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3998              !projid_eq(EXT4_I(old_dir)->i_projid,
3999                         EXT4_I(new_dentry->d_inode)->i_projid)))
4000                 return -EXDEV;
4001
4002         retval = dquot_initialize(old.dir);
4003         if (retval)
4004                 return retval;
4005         retval = dquot_initialize(new.dir);
4006         if (retval)
4007                 return retval;
4008
4009         old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4010                                  &old.de, &old.inlined);
4011         if (IS_ERR(old.bh))
4012                 return PTR_ERR(old.bh);
4013         /*
4014          *  Check for inode number is _not_ due to possible IO errors.
4015          *  We might rmdir the source, keep it as pwd of some process
4016          *  and merrily kill the link to whatever was created under the
4017          *  same name. Goodbye sticky bit ;-<
4018          */
4019         retval = -ENOENT;
4020         if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4021                 goto end_rename;
4022
4023         new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4024                                  &new.de, &new.inlined);
4025         if (IS_ERR(new.bh)) {
4026                 retval = PTR_ERR(new.bh);
4027                 new.bh = NULL;
4028                 goto end_rename;
4029         }
4030
4031         /* RENAME_EXCHANGE case: old *and* new must both exist */
4032         if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4033                 goto end_rename;
4034
4035         handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4036                 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4037                  2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4038         if (IS_ERR(handle)) {
4039                 retval = PTR_ERR(handle);
4040                 handle = NULL;
4041                 goto end_rename;
4042         }
4043
4044         if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4045                 ext4_handle_sync(handle);
4046
4047         if (S_ISDIR(old.inode->i_mode)) {
4048                 old.is_dir = true;
4049                 retval = ext4_rename_dir_prepare(handle, &old);
4050                 if (retval)
4051                         goto end_rename;
4052         }
4053         if (S_ISDIR(new.inode->i_mode)) {
4054                 new.is_dir = true;
4055                 retval = ext4_rename_dir_prepare(handle, &new);
4056                 if (retval)
4057                         goto end_rename;
4058         }
4059
4060         /*
4061          * Other than the special case of overwriting a directory, parents'
4062          * nlink only needs to be modified if this is a cross directory rename.
4063          */
4064         if (old.dir != new.dir && old.is_dir != new.is_dir) {
4065                 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4066                 new.dir_nlink_delta = -old.dir_nlink_delta;
4067                 retval = -EMLINK;
4068                 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4069                     (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4070                         goto end_rename;
4071         }
4072
4073         new_file_type = new.de->file_type;
4074         retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4075         if (retval)
4076                 goto end_rename;
4077
4078         retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4079         if (retval)
4080                 goto end_rename;
4081
4082         /*
4083          * Like most other Unix systems, set the ctime for inodes on a
4084          * rename.
4085          */
4086         ctime = current_time(old.inode);
4087         old.inode->i_ctime = ctime;
4088         new.inode->i_ctime = ctime;
4089         retval = ext4_mark_inode_dirty(handle, old.inode);
4090         if (unlikely(retval))
4091                 goto end_rename;
4092         retval = ext4_mark_inode_dirty(handle, new.inode);
4093         if (unlikely(retval))
4094                 goto end_rename;
4095         ext4_fc_mark_ineligible(new.inode->i_sb,
4096                                 EXT4_FC_REASON_CROSS_RENAME, handle);
4097         if (old.dir_bh) {
4098                 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4099                 if (retval)
4100                         goto end_rename;
4101         }
4102         if (new.dir_bh) {
4103                 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4104                 if (retval)
4105                         goto end_rename;
4106         }
4107         ext4_update_dir_count(handle, &old);
4108         ext4_update_dir_count(handle, &new);
4109         retval = 0;
4110
4111 end_rename:
4112         brelse(old.dir_bh);
4113         brelse(new.dir_bh);
4114         brelse(old.bh);
4115         brelse(new.bh);
4116         if (handle)
4117                 ext4_journal_stop(handle);
4118         return retval;
4119 }
4120
4121 static int ext4_rename2(struct user_namespace *mnt_userns,
4122                         struct inode *old_dir, struct dentry *old_dentry,
4123                         struct inode *new_dir, struct dentry *new_dentry,
4124                         unsigned int flags)
4125 {
4126         int err;
4127
4128         if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4129                 return -EIO;
4130
4131         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4132                 return -EINVAL;
4133
4134         err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4135                                      flags);
4136         if (err)
4137                 return err;
4138
4139         if (flags & RENAME_EXCHANGE) {
4140                 return ext4_cross_rename(old_dir, old_dentry,
4141                                          new_dir, new_dentry);
4142         }
4143
4144         return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4145 }
4146
4147 /*
4148  * directories can handle most operations...
4149  */
4150 const struct inode_operations ext4_dir_inode_operations = {
4151         .create         = ext4_create,
4152         .lookup         = ext4_lookup,
4153         .link           = ext4_link,
4154         .unlink         = ext4_unlink,
4155         .symlink        = ext4_symlink,
4156         .mkdir          = ext4_mkdir,
4157         .rmdir          = ext4_rmdir,
4158         .mknod          = ext4_mknod,
4159         .tmpfile        = ext4_tmpfile,
4160         .rename         = ext4_rename2,
4161         .setattr        = ext4_setattr,
4162         .getattr        = ext4_getattr,
4163         .listxattr      = ext4_listxattr,
4164         .get_acl        = ext4_get_acl,
4165         .set_acl        = ext4_set_acl,
4166         .fiemap         = ext4_fiemap,
4167         .fileattr_get   = ext4_fileattr_get,
4168         .fileattr_set   = ext4_fileattr_set,
4169 };
4170
4171 const struct inode_operations ext4_special_inode_operations = {
4172         .setattr        = ext4_setattr,
4173         .getattr        = ext4_getattr,
4174         .listxattr      = ext4_listxattr,
4175         .get_acl        = ext4_get_acl,
4176         .set_acl        = ext4_set_acl,
4177 };