fs/ext4/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/fs/ext4/xattr.c
   4  *
   5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6  *
   7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9  * Extended attributes for symlinks and special files added per
  10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12  *  Red Hat Inc.
  13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14  *  and Andreas Gruenbacher <agruen@suse.de>.
  15  */
  16
  17 /*
  18  * Extended attributes are stored directly in inodes (on file systems with
  19  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20  * field contains the block number if an inode uses an additional block. All
  21  * attributes must fit in the inode and one additional block. Blocks that
  22  * contain the identical set of attributes may be shared among several inodes.
  23  * Identical blocks are detected by keeping a cache of blocks that have
  24  * recently been accessed.
  25  *
  26  * The attributes in inodes and on blocks have a different header; the entries
  27  * are stored in the same format:
  28  *
  29  *   +------------------+
  30  *   | header           |
  31  *   | entry 1          | |
  32  *   | entry 2          | | growing downwards
  33  *   | entry 3          | v
  34  *   | four null bytes  |
  35  *   | . . .            |
  36  *   | value 1          | ^
  37  *   | value 3          | | growing upwards
  38  *   | value 2          | |
  39  *   +------------------+
  40  *
  41  * The header is followed by multiple entry descriptors. In disk blocks, the
  42  * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43  * attribute values are aligned to the end of the block in no specific order.
  44  *
  45  * Locking strategy
  46  * ----------------
  47  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48  * EA blocks are only changed if they are exclusive to an inode, so
  49  * holding xattr_sem also means that nothing but the EA block's reference
  50  * count can change. Multiple writers to the same block are synchronized
  51  * by the buffer lock.
  52  */
  53
  54 #include <linux/init.h>
  55 #include <linux/fs.h>
  56 #include <linux/slab.h>
  57 #include <linux/mbcache.h>
  58 #include <linux/quotaops.h>
  59 #include <linux/iversion.h>
  60 #include "ext4_jbd2.h"
  61 #include "ext4.h"
  62 #include "xattr.h"
  63 #include "acl.h"
  64
  65 #ifdef EXT4_XATTR_DEBUG
  66 # define ea_idebug(inode, fmt, ...)                                     \
  67         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  68                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69 # define ea_bdebug(bh, fmt, ...)                                        \
  70         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  71                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72 #else
  73 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  74 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  75 #endif
  76
  77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78                                           struct buffer_head *);
  79 static struct buffer_head *
  80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81                             struct mb_cache_entry **);
  82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83                                     size_t value_count);
  84 static void ext4_xattr_rehash(struct ext4_xattr_header *);
  85
  86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  87         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  89         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  90         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  91 #endif
  92         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  93 #ifdef CONFIG_EXT4_FS_SECURITY
  94         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  95 #endif
  96 };
  97
  98 const struct xattr_handler *ext4_xattr_handlers[] = {
  99         &ext4_xattr_user_handler,
 100         &ext4_xattr_trusted_handler,
 101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 102         &posix_acl_access_xattr_handler,
 103         &posix_acl_default_xattr_handler,
 104 #endif
 105 #ifdef CONFIG_EXT4_FS_SECURITY
 106         &ext4_xattr_security_handler,
 107 #endif
 108         NULL
 109 };
 110
 111 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 112                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 115                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117 static int
 118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119                         struct inode *inode);
 120
 121 #ifdef CONFIG_LOCKDEP
 122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123 {
 124         struct ext4_inode_info *ei = EXT4_I(ea_inode);
 125
 126         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 127         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
 128         lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
 129 }
 130 #endif
 131
 132 static __le32 ext4_xattr_block_csum(struct inode *inode,
 133                                     sector_t block_nr,
 134                                     struct ext4_xattr_header *hdr)
 135 {
 136         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 137         __u32 csum;
 138         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 139         __u32 dummy_csum = 0;
 140         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 141
 142         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 143                            sizeof(dsk_block_nr));
 144         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 145         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 146         offset += sizeof(dummy_csum);
 147         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 148                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 149
 150         return cpu_to_le32(csum);
 151 }
 152
 153 static int ext4_xattr_block_csum_verify(struct inode *inode,
 154                                         struct buffer_head *bh)
 155 {
 156         struct ext4_xattr_header *hdr = BHDR(bh);
 157         int ret = 1;
 158
 159         if (ext4_has_metadata_csum(inode->i_sb)) {
 160                 lock_buffer(bh);
 161                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 162                                                         bh->b_blocknr, hdr));
 163                 unlock_buffer(bh);
 164         }
 165         return ret;
 166 }
 167
 168 static void ext4_xattr_block_csum_set(struct inode *inode,
 169                                       struct buffer_head *bh)
 170 {
 171         if (ext4_has_metadata_csum(inode->i_sb))
 172                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 173                                                 bh->b_blocknr, BHDR(bh));
 174 }
 175
 176 static inline const struct xattr_handler *
 177 ext4_xattr_handler(int name_index)
 178 {
 179         const struct xattr_handler *handler = NULL;
 180
 181         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 182                 handler = ext4_xattr_handler_map[name_index];
 183         return handler;
 184 }
 185
 186 static int
 187 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 188                          void *value_start)
 189 {
 190         struct ext4_xattr_entry *e = entry;
 191
 192         /* Find the end of the names list */
 193         while (!IS_LAST_ENTRY(e)) {
 194                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 195                 if ((void *)next >= end)
 196                         return -EFSCORRUPTED;
 197                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 198                         return -EFSCORRUPTED;
 199                 e = next;
 200         }
 201
 202         /* Check the values */
 203         while (!IS_LAST_ENTRY(entry)) {
 204                 u32 size = le32_to_cpu(entry->e_value_size);
 205
 206                 if (size > EXT4_XATTR_SIZE_MAX)
 207                         return -EFSCORRUPTED;
 208
 209                 if (size != 0 && entry->e_value_inum == 0) {
 210                         u16 offs = le16_to_cpu(entry->e_value_offs);
 211                         void *value;
 212
 213                         /*
 214                          * The value cannot overlap the names, and the value
 215                          * with padding cannot extend beyond 'end'.  Check both
 216                          * the padded and unpadded sizes, since the size may
 217                          * overflow to 0 when adding padding.
 218                          */
 219                         if (offs > end - value_start)
 220                                 return -EFSCORRUPTED;
 221                         value = value_start + offs;
 222                         if (value < (void *)e + sizeof(u32) ||
 223                             size > end - value ||
 224                             EXT4_XATTR_SIZE(size) > end - value)
 225                                 return -EFSCORRUPTED;
 226                 }
 227                 entry = EXT4_XATTR_NEXT(entry);
 228         }
 229
 230         return 0;
 231 }
 232
 233 static inline int
 234 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 235                          const char *function, unsigned int line)
 236 {
 237         int error = -EFSCORRUPTED;
 238
 239         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 240             BHDR(bh)->h_blocks != cpu_to_le32(1))
 241                 goto errout;
 242         if (buffer_verified(bh))
 243                 return 0;
 244
 245         error = -EFSBADCRC;
 246         if (!ext4_xattr_block_csum_verify(inode, bh))
 247                 goto errout;
 248         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 249                                          bh->b_data);
 250 errout:
 251         if (error)
 252                 __ext4_error_inode(inode, function, line, 0,
 253                                    "corrupted xattr block %llu",
 254                                    (unsigned long long) bh->b_blocknr);
 255         else
 256                 set_buffer_verified(bh);
 257         return error;
 258 }
 259
 260 #define ext4_xattr_check_block(inode, bh) \
 261         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 262
 263
 264 static int
 265 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 266                          void *end, const char *function, unsigned int line)
 267 {
 268         int error = -EFSCORRUPTED;
 269
 270         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 271             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 272                 goto errout;
 273         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 274 errout:
 275         if (error)
 276                 __ext4_error_inode(inode, function, line, 0,
 277                                    "corrupted in-inode xattr");
 278         return error;
 279 }
 280
 281 #define xattr_check_inode(inode, header, end) \
 282         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 283
 284 static int
 285 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 286                  void *end, int name_index, const char *name, int sorted)
 287 {
 288         struct ext4_xattr_entry *entry, *next;
 289         size_t name_len;
 290         int cmp = 1;
 291
 292         if (name == NULL)
 293                 return -EINVAL;
 294         name_len = strlen(name);
 295         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 296                 next = EXT4_XATTR_NEXT(entry);
 297                 if ((void *) next >= end) {
 298                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 299                         return -EFSCORRUPTED;
 300                 }
 301                 cmp = name_index - entry->e_name_index;
 302                 if (!cmp)
 303                         cmp = name_len - entry->e_name_len;
 304                 if (!cmp)
 305                         cmp = memcmp(name, entry->e_name, name_len);
 306                 if (cmp <= 0 && (sorted || cmp == 0))
 307                         break;
 308         }
 309         *pentry = entry;
 310         return cmp ? -ENODATA : 0;
 311 }
 312
 313 static u32
 314 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 315 {
 316         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 317 }
 318
 319 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 320 {
 321         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 322                 (u32) inode_peek_iversion_raw(ea_inode);
 323 }
 324
 325 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 326 {
 327         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 328         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 329 }
 330
 331 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 332 {
 333         return (u32)ea_inode->i_atime.tv_sec;
 334 }
 335
 336 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 337 {
 338         ea_inode->i_atime.tv_sec = hash;
 339 }
 340
 341 /*
 342  * Read the EA value from an inode.
 343  */
 344 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 345 {
 346         int blocksize = 1 << ea_inode->i_blkbits;
 347         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 348         int tail_size = (size % blocksize) ?: blocksize;
 349         struct buffer_head *bhs_inline[8];
 350         struct buffer_head **bhs = bhs_inline;
 351         int i, ret;
 352
 353         if (bh_count > ARRAY_SIZE(bhs_inline)) {
 354                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 355                 if (!bhs)
 356                         return -ENOMEM;
 357         }
 358
 359         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 360                                true /* wait */, bhs);
 361         if (ret)
 362                 goto free_bhs;
 363
 364         for (i = 0; i < bh_count; i++) {
 365                 /* There shouldn't be any holes in ea_inode. */
 366                 if (!bhs[i]) {
 367                         ret = -EFSCORRUPTED;
 368                         goto put_bhs;
 369                 }
 370                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 371                        i < bh_count - 1 ? blocksize : tail_size);
 372         }
 373         ret = 0;
 374 put_bhs:
 375         for (i = 0; i < bh_count; i++)
 376                 brelse(bhs[i]);
 377 free_bhs:
 378         if (bhs != bhs_inline)
 379                 kfree(bhs);
 380         return ret;
 381 }
 382
 383 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 384
 385 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 386                                  u32 ea_inode_hash, struct inode **ea_inode)
 387 {
 388         struct inode *inode;
 389         int err;
 390
 391         /*
 392          * We have to check for this corruption early as otherwise
 393          * iget_locked() could wait indefinitely for the state of our
 394          * parent inode.
 395          */
 396         if (parent->i_ino == ea_ino) {
 397                 ext4_error(parent->i_sb,
 398                            "Parent and EA inode have the same ino %lu", ea_ino);
 399                 return -EFSCORRUPTED;
 400         }
 401
 402         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
 403         if (IS_ERR(inode)) {
 404                 err = PTR_ERR(inode);
 405                 ext4_error(parent->i_sb,
 406                            "error while reading EA inode %lu err=%d", ea_ino,
 407                            err);
 408                 return err;
 409         }
 410         ext4_xattr_inode_set_class(inode);
 411
 412         /*
 413          * Check whether this is an old Lustre-style xattr inode. Lustre
 414          * implementation does not have hash validation, rather it has a
 415          * backpointer from ea_inode to the parent inode.
 416          */
 417         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 418             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 419             inode->i_generation == parent->i_generation) {
 420                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 421                 ext4_xattr_inode_set_ref(inode, 1);
 422         } else {
 423                 inode_lock(inode);
 424                 inode->i_flags |= S_NOQUOTA;
 425                 inode_unlock(inode);
 426         }
 427
 428         *ea_inode = inode;
 429         return 0;
 430 }
 431
 432 /* Remove entry from mbcache when EA inode is getting evicted */
 433 void ext4_evict_ea_inode(struct inode *inode)
 434 {
 435         struct mb_cache_entry *oe;
 436
 437         if (!EA_INODE_CACHE(inode))
 438                 return;
 439         /* Wait for entry to get unused so that we can remove it */
 440         while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
 441                         ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
 442                 mb_cache_entry_wait_unused(oe);
 443                 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
 444         }
 445 }
 446
 447 static int
 448 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 449                                struct ext4_xattr_entry *entry, void *buffer,
 450                                size_t size)
 451 {
 452         u32 hash;
 453
 454         /* Verify stored hash matches calculated hash. */
 455         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 456         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 457                 return -EFSCORRUPTED;
 458
 459         if (entry) {
 460                 __le32 e_hash, tmp_data;
 461
 462                 /* Verify entry hash. */
 463                 tmp_data = cpu_to_le32(hash);
 464                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 465                                                &tmp_data, 1);
 466                 if (e_hash != entry->e_hash)
 467                         return -EFSCORRUPTED;
 468         }
 469         return 0;
 470 }
 471
 472 /*
 473  * Read xattr value from the EA inode.
 474  */
 475 static int
 476 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 477                      void *buffer, size_t size)
 478 {
 479         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 480         struct inode *ea_inode;
 481         int err;
 482
 483         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 484                                     le32_to_cpu(entry->e_hash), &ea_inode);
 485         if (err) {
 486                 ea_inode = NULL;
 487                 goto out;
 488         }
 489
 490         if (i_size_read(ea_inode) != size) {
 491                 ext4_warning_inode(ea_inode,
 492                                    "ea_inode file size=%llu entry size=%zu",
 493                                    i_size_read(ea_inode), size);
 494                 err = -EFSCORRUPTED;
 495                 goto out;
 496         }
 497
 498         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 499         if (err)
 500                 goto out;
 501
 502         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 503                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 504                                                      size);
 505                 if (err) {
 506                         ext4_warning_inode(ea_inode,
 507                                            "EA inode hash validation failed");
 508                         goto out;
 509                 }
 510
 511                 if (ea_inode_cache)
 512                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 513                                         ext4_xattr_inode_get_hash(ea_inode),
 514                                         ea_inode->i_ino, true /* reusable */);
 515         }
 516 out:
 517         iput(ea_inode);
 518         return err;
 519 }
 520
 521 static int
 522 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 523                      void *buffer, size_t buffer_size)
 524 {
 525         struct buffer_head *bh = NULL;
 526         struct ext4_xattr_entry *entry;
 527         size_t size;
 528         void *end;
 529         int error;
 530         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 531
 532         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 533                   name_index, name, buffer, (long)buffer_size);
 534
 535         if (!EXT4_I(inode)->i_file_acl)
 536                 return -ENODATA;
 537         ea_idebug(inode, "reading block %llu",
 538                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 539         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 540         if (IS_ERR(bh))
 541                 return PTR_ERR(bh);
 542         ea_bdebug(bh, "b_count=%d, refcount=%d",
 543                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 544         error = ext4_xattr_check_block(inode, bh);
 545         if (error)
 546                 goto cleanup;
 547         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 548         entry = BFIRST(bh);
 549         end = bh->b_data + bh->b_size;
 550         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 551         if (error)
 552                 goto cleanup;
 553         size = le32_to_cpu(entry->e_value_size);
 554         error = -ERANGE;
 555         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 556                 goto cleanup;
 557         if (buffer) {
 558                 if (size > buffer_size)
 559                         goto cleanup;
 560                 if (entry->e_value_inum) {
 561                         error = ext4_xattr_inode_get(inode, entry, buffer,
 562                                                      size);
 563                         if (error)
 564                                 goto cleanup;
 565                 } else {
 566                         u16 offset = le16_to_cpu(entry->e_value_offs);
 567                         void *p = bh->b_data + offset;
 568
 569                         if (unlikely(p + size > end))
 570                                 goto cleanup;
 571                         memcpy(buffer, p, size);
 572                 }
 573         }
 574         error = size;
 575
 576 cleanup:
 577         brelse(bh);
 578         return error;
 579 }
 580
 581 int
 582 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 583                      void *buffer, size_t buffer_size)
 584 {
 585         struct ext4_xattr_ibody_header *header;
 586         struct ext4_xattr_entry *entry;
 587         struct ext4_inode *raw_inode;
 588         struct ext4_iloc iloc;
 589         size_t size;
 590         void *end;
 591         int error;
 592
 593         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 594                 return -ENODATA;
 595         error = ext4_get_inode_loc(inode, &iloc);
 596         if (error)
 597                 return error;
 598         raw_inode = ext4_raw_inode(&iloc);
 599         header = IHDR(inode, raw_inode);
 600         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 601         error = xattr_check_inode(inode, header, end);
 602         if (error)
 603                 goto cleanup;
 604         entry = IFIRST(header);
 605         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 606         if (error)
 607                 goto cleanup;
 608         size = le32_to_cpu(entry->e_value_size);
 609         error = -ERANGE;
 610         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 611                 goto cleanup;
 612         if (buffer) {
 613                 if (size > buffer_size)
 614                         goto cleanup;
 615                 if (entry->e_value_inum) {
 616                         error = ext4_xattr_inode_get(inode, entry, buffer,
 617                                                      size);
 618                         if (error)
 619                                 goto cleanup;
 620                 } else {
 621                         u16 offset = le16_to_cpu(entry->e_value_offs);
 622                         void *p = (void *)IFIRST(header) + offset;
 623
 624                         if (unlikely(p + size > end))
 625                                 goto cleanup;
 626                         memcpy(buffer, p, size);
 627                 }
 628         }
 629         error = size;
 630
 631 cleanup:
 632         brelse(iloc.bh);
 633         return error;
 634 }
 635
 636 /*
 637  * ext4_xattr_get()
 638  *
 639  * Copy an extended attribute into the buffer
 640  * provided, or compute the buffer size required.
 641  * Buffer is NULL to compute the size of the buffer required.
 642  *
 643  * Returns a negative error number on failure, or the number of bytes
 644  * used / required on success.
 645  */
 646 int
 647 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 648                void *buffer, size_t buffer_size)
 649 {
 650         int error;
 651
 652         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 653                 return -EIO;
 654
 655         if (strlen(name) > 255)
 656                 return -ERANGE;
 657
 658         down_read(&EXT4_I(inode)->xattr_sem);
 659         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 660                                      buffer_size);
 661         if (error == -ENODATA)
 662                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 663                                              buffer_size);
 664         up_read(&EXT4_I(inode)->xattr_sem);
 665         return error;
 666 }
 667
 668 static int
 669 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 670                         char *buffer, size_t buffer_size)
 671 {
 672         size_t rest = buffer_size;
 673
 674         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 675                 const struct xattr_handler *handler =
 676                         ext4_xattr_handler(entry->e_name_index);
 677
 678                 if (handler && (!handler->list || handler->list(dentry))) {
 679                         const char *prefix = handler->prefix ?: handler->name;
 680                         size_t prefix_len = strlen(prefix);
 681                         size_t size = prefix_len + entry->e_name_len + 1;
 682
 683                         if (buffer) {
 684                                 if (size > rest)
 685                                         return -ERANGE;
 686                                 memcpy(buffer, prefix, prefix_len);
 687                                 buffer += prefix_len;
 688                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 689                                 buffer += entry->e_name_len;
 690                                 *buffer++ = 0;
 691                         }
 692                         rest -= size;
 693                 }
 694         }
 695         return buffer_size - rest;  /* total size */
 696 }
 697
 698 static int
 699 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 700 {
 701         struct inode *inode = d_inode(dentry);
 702         struct buffer_head *bh = NULL;
 703         int error;
 704
 705         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 706                   buffer, (long)buffer_size);
 707
 708         if (!EXT4_I(inode)->i_file_acl)
 709                 return 0;
 710         ea_idebug(inode, "reading block %llu",
 711                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 712         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 713         if (IS_ERR(bh))
 714                 return PTR_ERR(bh);
 715         ea_bdebug(bh, "b_count=%d, refcount=%d",
 716                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 717         error = ext4_xattr_check_block(inode, bh);
 718         if (error)
 719                 goto cleanup;
 720         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 721         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 722                                         buffer_size);
 723 cleanup:
 724         brelse(bh);
 725         return error;
 726 }
 727
 728 static int
 729 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 730 {
 731         struct inode *inode = d_inode(dentry);
 732         struct ext4_xattr_ibody_header *header;
 733         struct ext4_inode *raw_inode;
 734         struct ext4_iloc iloc;
 735         void *end;
 736         int error;
 737
 738         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 739                 return 0;
 740         error = ext4_get_inode_loc(inode, &iloc);
 741         if (error)
 742                 return error;
 743         raw_inode = ext4_raw_inode(&iloc);
 744         header = IHDR(inode, raw_inode);
 745         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 746         error = xattr_check_inode(inode, header, end);
 747         if (error)
 748                 goto cleanup;
 749         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 750                                         buffer, buffer_size);
 751
 752 cleanup:
 753         brelse(iloc.bh);
 754         return error;
 755 }
 756
 757 /*
 758  * Inode operation listxattr()
 759  *
 760  * d_inode(dentry)->i_rwsem: don't care
 761  *
 762  * Copy a list of attribute names into the buffer
 763  * provided, or compute the buffer size required.
 764  * Buffer is NULL to compute the size of the buffer required.
 765  *
 766  * Returns a negative error number on failure, or the number of bytes
 767  * used / required on success.
 768  */
 769 ssize_t
 770 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 771 {
 772         int ret, ret2;
 773
 774         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 775         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 776         if (ret < 0)
 777                 goto errout;
 778         if (buffer) {
 779                 buffer += ret;
 780                 buffer_size -= ret;
 781         }
 782         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 783         if (ret < 0)
 784                 goto errout;
 785         ret += ret2;
 786 errout:
 787         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 788         return ret;
 789 }
 790
 791 /*
 792  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 793  * not set, set it.
 794  */
 795 static void ext4_xattr_update_super_block(handle_t *handle,
 796                                           struct super_block *sb)
 797 {
 798         if (ext4_has_feature_xattr(sb))
 799                 return;
 800
 801         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 802         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 803                 ext4_set_feature_xattr(sb);
 804                 ext4_handle_dirty_super(handle, sb);
 805         }
 806 }
 807
 808 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 809 {
 810         struct ext4_iloc iloc = { .bh = NULL };
 811         struct buffer_head *bh = NULL;
 812         struct ext4_inode *raw_inode;
 813         struct ext4_xattr_ibody_header *header;
 814         struct ext4_xattr_entry *entry;
 815         qsize_t ea_inode_refs = 0;
 816         void *end;
 817         int ret;
 818
 819         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 820
 821         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 822                 ret = ext4_get_inode_loc(inode, &iloc);
 823                 if (ret)
 824                         goto out;
 825                 raw_inode = ext4_raw_inode(&iloc);
 826                 header = IHDR(inode, raw_inode);
 827                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 828                 ret = xattr_check_inode(inode, header, end);
 829                 if (ret)
 830                         goto out;
 831
 832                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 833                      entry = EXT4_XATTR_NEXT(entry))
 834                         if (entry->e_value_inum)
 835                                 ea_inode_refs++;
 836         }
 837
 838         if (EXT4_I(inode)->i_file_acl) {
 839                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 840                 if (IS_ERR(bh)) {
 841                         ret = PTR_ERR(bh);
 842                         bh = NULL;
 843                         goto out;
 844                 }
 845
 846                 ret = ext4_xattr_check_block(inode, bh);
 847                 if (ret)
 848                         goto out;
 849
 850                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 851                      entry = EXT4_XATTR_NEXT(entry))
 852                         if (entry->e_value_inum)
 853                                 ea_inode_refs++;
 854         }
 855         *usage = ea_inode_refs + 1;
 856         ret = 0;
 857 out:
 858         brelse(iloc.bh);
 859         brelse(bh);
 860         return ret;
 861 }
 862
 863 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 864 {
 865         struct super_block *sb = inode->i_sb;
 866         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 867                                     inode->i_blkbits);
 868         size_t mask = ~(cluster_size - 1);
 869
 870         return (length + cluster_size - 1) & mask;
 871 }
 872
 873 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 874 {
 875         int err;
 876
 877         err = dquot_alloc_inode(inode);
 878         if (err)
 879                 return err;
 880         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 881         if (err)
 882                 dquot_free_inode(inode);
 883         return err;
 884 }
 885
 886 static void ext4_xattr_inode_free_quota(struct inode *parent,
 887                                         struct inode *ea_inode,
 888                                         size_t len)
 889 {
 890         if (ea_inode &&
 891             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 892                 return;
 893         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 894         dquot_free_inode(parent);
 895 }
 896
 897 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 898                              struct buffer_head *block_bh, size_t value_len,
 899                              bool is_create)
 900 {
 901         int credits;
 902         int blocks;
 903
 904         /*
 905          * 1) Owner inode update
 906          * 2) Ref count update on old xattr block
 907          * 3) new xattr block
 908          * 4) block bitmap update for new xattr block
 909          * 5) group descriptor for new xattr block
 910          * 6) block bitmap update for old xattr block
 911          * 7) group descriptor for old block
 912          *
 913          * 6 & 7 can happen if we have two racing threads T_a and T_b
 914          * which are each trying to set an xattr on inodes I_a and I_b
 915          * which were both initially sharing an xattr block.
 916          */
 917         credits = 7;
 918
 919         /* Quota updates. */
 920         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 921
 922         /*
 923          * In case of inline data, we may push out the data to a block,
 924          * so we need to reserve credits for this eventuality
 925          */
 926         if (inode && ext4_has_inline_data(inode))
 927                 credits += ext4_writepage_trans_blocks(inode) + 1;
 928
 929         /* We are done if ea_inode feature is not enabled. */
 930         if (!ext4_has_feature_ea_inode(sb))
 931                 return credits;
 932
 933         /* New ea_inode, inode map, block bitmap, group descriptor. */
 934         credits += 4;
 935
 936         /* Data blocks. */
 937         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 938
 939         /* Indirection block or one level of extent tree. */
 940         blocks += 1;
 941
 942         /* Block bitmap and group descriptor updates for each block. */
 943         credits += blocks * 2;
 944
 945         /* Blocks themselves. */
 946         credits += blocks;
 947
 948         if (!is_create) {
 949                 /* Dereference ea_inode holding old xattr value.
 950                  * Old ea_inode, inode map, block bitmap, group descriptor.
 951                  */
 952                 credits += 4;
 953
 954                 /* Data blocks for old ea_inode. */
 955                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 956
 957                 /* Indirection block or one level of extent tree for old
 958                  * ea_inode.
 959                  */
 960                 blocks += 1;
 961
 962                 /* Block bitmap and group descriptor updates for each block. */
 963                 credits += blocks * 2;
 964         }
 965
 966         /* We may need to clone the existing xattr block in which case we need
 967          * to increment ref counts for existing ea_inodes referenced by it.
 968          */
 969         if (block_bh) {
 970                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
 971
 972                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 973                         if (entry->e_value_inum)
 974                                 /* Ref count update on ea_inode. */
 975                                 credits += 1;
 976         }
 977         return credits;
 978 }
 979
 980 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 981                                      int credits, struct buffer_head *bh,
 982                                      bool dirty, bool block_csum)
 983 {
 984         int error;
 985
 986         if (!ext4_handle_valid(handle))
 987                 return 0;
 988
 989         if (handle->h_buffer_credits >= credits)
 990                 return 0;
 991
 992         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 993         if (!error)
 994                 return 0;
 995         if (error < 0) {
 996                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 997                 return error;
 998         }
 999
1000         if (bh && dirty) {
1001                 if (block_csum)
1002                         ext4_xattr_block_csum_set(inode, bh);
1003                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1004                 if (error) {
1005                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1006                                      error);
1007                         return error;
1008                 }
1009         }
1010
1011         error = ext4_journal_restart(handle, credits);
1012         if (error) {
1013                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1014                 return error;
1015         }
1016
1017         if (bh) {
1018                 error = ext4_journal_get_write_access(handle, bh);
1019                 if (error) {
1020                         ext4_warning(inode->i_sb,
1021                                      "Get write access failed (error %d)",
1022                                      error);
1023                         return error;
1024                 }
1025         }
1026         return 0;
1027 }
1028
1029 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1030                                        int ref_change)
1031 {
1032         struct ext4_iloc iloc;
1033         s64 ref_count;
1034         int ret;
1035
1036         inode_lock(ea_inode);
1037
1038         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1039         if (ret)
1040                 goto out;
1041
1042         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1043         ref_count += ref_change;
1044         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1045
1046         if (ref_change > 0) {
1047                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1048                           ea_inode->i_ino, ref_count);
1049
1050                 if (ref_count == 1) {
1051                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1052                                   ea_inode->i_ino, ea_inode->i_nlink);
1053
1054                         set_nlink(ea_inode, 1);
1055                         ext4_orphan_del(handle, ea_inode);
1056                 }
1057         } else {
1058                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1059                           ea_inode->i_ino, ref_count);
1060
1061                 if (ref_count == 0) {
1062                         WARN_ONCE(ea_inode->i_nlink != 1,
1063                                   "EA inode %lu i_nlink=%u",
1064                                   ea_inode->i_ino, ea_inode->i_nlink);
1065
1066                         clear_nlink(ea_inode);
1067                         ext4_orphan_add(handle, ea_inode);
1068                 }
1069         }
1070
1071         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1072         if (ret)
1073                 ext4_warning_inode(ea_inode,
1074                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1075 out:
1076         inode_unlock(ea_inode);
1077         return ret;
1078 }
1079
1080 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1081 {
1082         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1083 }
1084
1085 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1086 {
1087         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1088 }
1089
1090 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1091                                         struct ext4_xattr_entry *first)
1092 {
1093         struct inode *ea_inode;
1094         struct ext4_xattr_entry *entry;
1095         struct ext4_xattr_entry *failed_entry;
1096         unsigned int ea_ino;
1097         int err, saved_err;
1098
1099         for (entry = first; !IS_LAST_ENTRY(entry);
1100              entry = EXT4_XATTR_NEXT(entry)) {
1101                 if (!entry->e_value_inum)
1102                         continue;
1103                 ea_ino = le32_to_cpu(entry->e_value_inum);
1104                 err = ext4_xattr_inode_iget(parent, ea_ino,
1105                                             le32_to_cpu(entry->e_hash),
1106                                             &ea_inode);
1107                 if (err)
1108                         goto cleanup;
1109                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1110                 if (err) {
1111                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1112                         iput(ea_inode);
1113                         goto cleanup;
1114                 }
1115                 iput(ea_inode);
1116         }
1117         return 0;
1118
1119 cleanup:
1120         saved_err = err;
1121         failed_entry = entry;
1122
1123         for (entry = first; entry != failed_entry;
1124              entry = EXT4_XATTR_NEXT(entry)) {
1125                 if (!entry->e_value_inum)
1126                         continue;
1127                 ea_ino = le32_to_cpu(entry->e_value_inum);
1128                 err = ext4_xattr_inode_iget(parent, ea_ino,
1129                                             le32_to_cpu(entry->e_hash),
1130                                             &ea_inode);
1131                 if (err) {
1132                         ext4_warning(parent->i_sb,
1133                                      "cleanup ea_ino %u iget error %d", ea_ino,
1134                                      err);
1135                         continue;
1136                 }
1137                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1138                 if (err)
1139                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1140                                            err);
1141                 iput(ea_inode);
1142         }
1143         return saved_err;
1144 }
1145
1146 static void
1147 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1148                              struct buffer_head *bh,
1149                              struct ext4_xattr_entry *first, bool block_csum,
1150                              struct ext4_xattr_inode_array **ea_inode_array,
1151                              int extra_credits, bool skip_quota)
1152 {
1153         struct inode *ea_inode;
1154         struct ext4_xattr_entry *entry;
1155         bool dirty = false;
1156         unsigned int ea_ino;
1157         int err;
1158         int credits;
1159
1160         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1161         credits = 2 + extra_credits;
1162
1163         for (entry = first; !IS_LAST_ENTRY(entry);
1164              entry = EXT4_XATTR_NEXT(entry)) {
1165                 if (!entry->e_value_inum)
1166                         continue;
1167                 ea_ino = le32_to_cpu(entry->e_value_inum);
1168                 err = ext4_xattr_inode_iget(parent, ea_ino,
1169                                             le32_to_cpu(entry->e_hash),
1170                                             &ea_inode);
1171                 if (err)
1172                         continue;
1173
1174                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1175                 if (err) {
1176                         ext4_warning_inode(ea_inode,
1177                                            "Expand inode array err=%d", err);
1178                         iput(ea_inode);
1179                         continue;
1180                 }
1181
1182                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1183                                                 dirty, block_csum);
1184                 if (err) {
1185                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1186                                            err);
1187                         continue;
1188                 }
1189
1190                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1191                 if (err) {
1192                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1193                                            err);
1194                         continue;
1195                 }
1196
1197                 if (!skip_quota)
1198                         ext4_xattr_inode_free_quota(parent, ea_inode,
1199                                               le32_to_cpu(entry->e_value_size));
1200
1201                 /*
1202                  * Forget about ea_inode within the same transaction that
1203                  * decrements the ref count. This avoids duplicate decrements in
1204                  * case the rest of the work spills over to subsequent
1205                  * transactions.
1206                  */
1207                 entry->e_value_inum = 0;
1208                 entry->e_value_size = 0;
1209
1210                 dirty = true;
1211         }
1212
1213         if (dirty) {
1214                 /*
1215                  * Note that we are deliberately skipping csum calculation for
1216                  * the final update because we do not expect any journal
1217                  * restarts until xattr block is freed.
1218                  */
1219
1220                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1221                 if (err)
1222                         ext4_warning_inode(parent,
1223                                            "handle dirty metadata err=%d", err);
1224         }
1225 }
1226
1227 /*
1228  * Release the xattr block BH: If the reference count is > 1, decrement it;
1229  * otherwise free the block.
1230  */
1231 static void
1232 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1233                          struct buffer_head *bh,
1234                          struct ext4_xattr_inode_array **ea_inode_array,
1235                          int extra_credits)
1236 {
1237         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1238         u32 hash, ref;
1239         int error = 0;
1240
1241         BUFFER_TRACE(bh, "get_write_access");
1242         error = ext4_journal_get_write_access(handle, bh);
1243         if (error)
1244                 goto out;
1245
1246 retry_ref:
1247         lock_buffer(bh);
1248         hash = le32_to_cpu(BHDR(bh)->h_hash);
1249         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1250         if (ref == 1) {
1251                 ea_bdebug(bh, "refcount now=0; freeing");
1252                 /*
1253                  * This must happen under buffer lock for
1254                  * ext4_xattr_block_set() to reliably detect freed block
1255                  */
1256                 if (ea_block_cache) {
1257                         struct mb_cache_entry *oe;
1258
1259                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1260                                                           bh->b_blocknr);
1261                         if (oe) {
1262                                 unlock_buffer(bh);
1263                                 mb_cache_entry_wait_unused(oe);
1264                                 mb_cache_entry_put(ea_block_cache, oe);
1265                                 goto retry_ref;
1266                         }
1267                 }
1268                 get_bh(bh);
1269                 unlock_buffer(bh);
1270
1271                 if (ext4_has_feature_ea_inode(inode->i_sb))
1272                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1273                                                      BFIRST(bh),
1274                                                      true /* block_csum */,
1275                                                      ea_inode_array,
1276                                                      extra_credits,
1277                                                      true /* skip_quota */);
1278                 ext4_free_blocks(handle, inode, bh, 0, 1,
1279                                  EXT4_FREE_BLOCKS_METADATA |
1280                                  EXT4_FREE_BLOCKS_FORGET);
1281         } else {
1282                 ref--;
1283                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1284                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1285                         struct mb_cache_entry *ce;
1286
1287                         if (ea_block_cache) {
1288                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1289                                                         bh->b_blocknr);
1290                                 if (ce) {
1291                                         set_bit(MBE_REUSABLE_B, &ce->e_flags);
1292                                         mb_cache_entry_put(ea_block_cache, ce);
1293                                 }
1294                         }
1295                 }
1296
1297                 ext4_xattr_block_csum_set(inode, bh);
1298                 /*
1299                  * Beware of this ugliness: Releasing of xattr block references
1300                  * from different inodes can race and so we have to protect
1301                  * from a race where someone else frees the block (and releases
1302                  * its journal_head) before we are done dirtying the buffer. In
1303                  * nojournal mode this race is harmless and we actually cannot
1304                  * call ext4_handle_dirty_metadata() with locked buffer as
1305                  * that function can call sync_dirty_buffer() so for that case
1306                  * we handle the dirtying after unlocking the buffer.
1307                  */
1308                 if (ext4_handle_valid(handle))
1309                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1310                 unlock_buffer(bh);
1311                 if (!ext4_handle_valid(handle))
1312                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1313                 if (IS_SYNC(inode))
1314                         ext4_handle_sync(handle);
1315                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1316                 ea_bdebug(bh, "refcount now=%d; releasing",
1317                           le32_to_cpu(BHDR(bh)->h_refcount));
1318         }
1319 out:
1320         ext4_std_error(inode->i_sb, error);
1321         return;
1322 }
1323
1324 /*
1325  * Find the available free space for EAs. This also returns the total number of
1326  * bytes used by EA entries.
1327  */
1328 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1329                                     size_t *min_offs, void *base, int *total)
1330 {
1331         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1332                 if (!last->e_value_inum && last->e_value_size) {
1333                         size_t offs = le16_to_cpu(last->e_value_offs);
1334                         if (offs < *min_offs)
1335                                 *min_offs = offs;
1336                 }
1337                 if (total)
1338                         *total += EXT4_XATTR_LEN(last->e_name_len);
1339         }
1340         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1341 }
1342
1343 /*
1344  * Write the value of the EA in an inode.
1345  */
1346 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1347                                   const void *buf, int bufsize)
1348 {
1349         struct buffer_head *bh = NULL;
1350         unsigned long block = 0;
1351         int blocksize = ea_inode->i_sb->s_blocksize;
1352         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1353         int csize, wsize = 0;
1354         int ret = 0;
1355         int retries = 0;
1356
1357 retry:
1358         while (ret >= 0 && ret < max_blocks) {
1359                 struct ext4_map_blocks map;
1360                 map.m_lblk = block += ret;
1361                 map.m_len = max_blocks -= ret;
1362
1363                 ret = ext4_map_blocks(handle, ea_inode, &map,
1364                                       EXT4_GET_BLOCKS_CREATE);
1365                 if (ret <= 0) {
1366                         ext4_mark_inode_dirty(handle, ea_inode);
1367                         if (ret == -ENOSPC &&
1368                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1369                                 ret = 0;
1370                                 goto retry;
1371                         }
1372                         break;
1373                 }
1374         }
1375
1376         if (ret < 0)
1377                 return ret;
1378
1379         block = 0;
1380         while (wsize < bufsize) {
1381                 if (bh != NULL)
1382                         brelse(bh);
1383                 csize = (bufsize - wsize) > blocksize ? blocksize :
1384                                                                 bufsize - wsize;
1385                 bh = ext4_getblk(handle, ea_inode, block, 0);
1386                 if (IS_ERR(bh))
1387                         return PTR_ERR(bh);
1388                 if (!bh) {
1389                         WARN_ON_ONCE(1);
1390                         EXT4_ERROR_INODE(ea_inode,
1391                                          "ext4_getblk() return bh = NULL");
1392                         return -EFSCORRUPTED;
1393                 }
1394                 ret = ext4_journal_get_write_access(handle, bh);
1395                 if (ret)
1396                         goto out;
1397
1398                 memcpy(bh->b_data, buf, csize);
1399                 set_buffer_uptodate(bh);
1400                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1401
1402                 buf += csize;
1403                 wsize += csize;
1404                 block += 1;
1405         }
1406
1407         inode_lock(ea_inode);
1408         i_size_write(ea_inode, wsize);
1409         ext4_update_i_disksize(ea_inode, wsize);
1410         inode_unlock(ea_inode);
1411
1412         ext4_mark_inode_dirty(handle, ea_inode);
1413
1414 out:
1415         brelse(bh);
1416
1417         return ret;
1418 }
1419
1420 /*
1421  * Create an inode to store the value of a large EA.
1422  */
1423 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1424                                              struct inode *inode, u32 hash)
1425 {
1426         struct inode *ea_inode = NULL;
1427         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1428         int err;
1429
1430         if (inode->i_sb->s_root == NULL) {
1431                 ext4_warning(inode->i_sb,
1432                              "refuse to create EA inode when umounting");
1433                 WARN_ON(1);
1434                 return ERR_PTR(-EINVAL);
1435         }
1436
1437         /*
1438          * Let the next inode be the goal, so we try and allocate the EA inode
1439          * in the same group, or nearby one.
1440          */
1441         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1442                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1443                                   EXT4_EA_INODE_FL);
1444         if (!IS_ERR(ea_inode)) {
1445                 ea_inode->i_op = &ext4_file_inode_operations;
1446                 ea_inode->i_fop = &ext4_file_operations;
1447                 ext4_set_aops(ea_inode);
1448                 ext4_xattr_inode_set_class(ea_inode);
1449                 unlock_new_inode(ea_inode);
1450                 ext4_xattr_inode_set_ref(ea_inode, 1);
1451                 ext4_xattr_inode_set_hash(ea_inode, hash);
1452                 err = ext4_mark_inode_dirty(handle, ea_inode);
1453                 if (!err)
1454                         err = ext4_inode_attach_jinode(ea_inode);
1455                 if (err) {
1456                         if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1457                                 ext4_warning_inode(ea_inode,
1458                                         "cleanup dec ref error %d", err);
1459                         iput(ea_inode);
1460                         return ERR_PTR(err);
1461                 }
1462
1463                 /*
1464                  * Xattr inodes are shared therefore quota charging is performed
1465                  * at a higher level.
1466                  */
1467                 dquot_free_inode(ea_inode);
1468                 dquot_drop(ea_inode);
1469                 inode_lock(ea_inode);
1470                 ea_inode->i_flags |= S_NOQUOTA;
1471                 inode_unlock(ea_inode);
1472         }
1473
1474         return ea_inode;
1475 }
1476
1477 static struct inode *
1478 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1479                             size_t value_len, u32 hash)
1480 {
1481         struct inode *ea_inode;
1482         struct mb_cache_entry *ce;
1483         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1484         void *ea_data;
1485
1486         if (!ea_inode_cache)
1487                 return NULL;
1488
1489         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1490         if (!ce)
1491                 return NULL;
1492
1493         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1494                      !(current->flags & PF_MEMALLOC_NOFS));
1495
1496         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1497         if (!ea_data) {
1498                 mb_cache_entry_put(ea_inode_cache, ce);
1499                 return NULL;
1500         }
1501
1502         while (ce) {
1503                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1504                                      EXT4_IGET_EA_INODE);
1505                 if (IS_ERR(ea_inode))
1506                         goto next_entry;
1507                 ext4_xattr_inode_set_class(ea_inode);
1508                 if (i_size_read(ea_inode) == value_len &&
1509                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1510                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1511                                                     value_len) &&
1512                     !memcmp(value, ea_data, value_len)) {
1513                         mb_cache_entry_touch(ea_inode_cache, ce);
1514                         mb_cache_entry_put(ea_inode_cache, ce);
1515                         kvfree(ea_data);
1516                         return ea_inode;
1517                 }
1518                 iput(ea_inode);
1519         next_entry:
1520                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1521         }
1522         kvfree(ea_data);
1523         return NULL;
1524 }
1525
1526 /*
1527  * Add value of the EA in an inode.
1528  */
1529 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1530                                           const void *value, size_t value_len,
1531                                           struct inode **ret_inode)
1532 {
1533         struct inode *ea_inode;
1534         u32 hash;
1535         int err;
1536
1537         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1538         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1539         if (ea_inode) {
1540                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1541                 if (err) {
1542                         iput(ea_inode);
1543                         return err;
1544                 }
1545
1546                 *ret_inode = ea_inode;
1547                 return 0;
1548         }
1549
1550         /* Create an inode for the EA value */
1551         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1552         if (IS_ERR(ea_inode))
1553                 return PTR_ERR(ea_inode);
1554
1555         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1556         if (err) {
1557                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1558                 iput(ea_inode);
1559                 return err;
1560         }
1561
1562         if (EA_INODE_CACHE(inode))
1563                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1564                                       ea_inode->i_ino, true /* reusable */);
1565
1566         *ret_inode = ea_inode;
1567         return 0;
1568 }
1569
1570 /*
1571  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1572  * feature is enabled.
1573  */
1574 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1575
1576 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1577                                 struct ext4_xattr_search *s,
1578                                 handle_t *handle, struct inode *inode,
1579                                 bool is_block)
1580 {
1581         struct ext4_xattr_entry *last, *next;
1582         struct ext4_xattr_entry *here = s->here;
1583         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1584         int in_inode = i->in_inode;
1585         struct inode *old_ea_inode = NULL;
1586         struct inode *new_ea_inode = NULL;
1587         size_t old_size, new_size;
1588         int ret;
1589
1590         /* Space used by old and new values. */
1591         old_size = (!s->not_found && !here->e_value_inum) ?
1592                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1593         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1594
1595         /*
1596          * Optimization for the simple case when old and new values have the
1597          * same padded sizes. Not applicable if external inodes are involved.
1598          */
1599         if (new_size && new_size == old_size) {
1600                 size_t offs = le16_to_cpu(here->e_value_offs);
1601                 void *val = s->base + offs;
1602
1603                 here->e_value_size = cpu_to_le32(i->value_len);
1604                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1605                         memset(val, 0, new_size);
1606                 } else {
1607                         memcpy(val, i->value, i->value_len);
1608                         /* Clear padding bytes. */
1609                         memset(val + i->value_len, 0, new_size - i->value_len);
1610                 }
1611                 goto update_hash;
1612         }
1613
1614         /* Compute min_offs and last. */
1615         last = s->first;
1616         for (; !IS_LAST_ENTRY(last); last = next) {
1617                 next = EXT4_XATTR_NEXT(last);
1618                 if ((void *)next >= s->end) {
1619                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1620                         ret = -EFSCORRUPTED;
1621                         goto out;
1622                 }
1623                 if (!last->e_value_inum && last->e_value_size) {
1624                         size_t offs = le16_to_cpu(last->e_value_offs);
1625                         if (offs < min_offs)
1626                                 min_offs = offs;
1627                 }
1628         }
1629
1630         /* Check whether we have enough space. */
1631         if (i->value) {
1632                 size_t free;
1633
1634                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1635                 if (!s->not_found)
1636                         free += EXT4_XATTR_LEN(name_len) + old_size;
1637
1638                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1639                         ret = -ENOSPC;
1640                         goto out;
1641                 }
1642
1643                 /*
1644                  * If storing the value in an external inode is an option,
1645                  * reserve space for xattr entries/names in the external
1646                  * attribute block so that a long value does not occupy the
1647                  * whole space and prevent futher entries being added.
1648                  */
1649                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1650                     new_size && is_block &&
1651                     (min_offs + old_size - new_size) <
1652                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1653                         ret = -ENOSPC;
1654                         goto out;
1655                 }
1656         }
1657
1658         /*
1659          * Getting access to old and new ea inodes is subject to failures.
1660          * Finish that work before doing any modifications to the xattr data.
1661          */
1662         if (!s->not_found && here->e_value_inum) {
1663                 ret = ext4_xattr_inode_iget(inode,
1664                                             le32_to_cpu(here->e_value_inum),
1665                                             le32_to_cpu(here->e_hash),
1666                                             &old_ea_inode);
1667                 if (ret) {
1668                         old_ea_inode = NULL;
1669                         goto out;
1670                 }
1671         }
1672         if (i->value && in_inode) {
1673                 WARN_ON_ONCE(!i->value_len);
1674
1675                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1676                 if (ret)
1677                         goto out;
1678
1679                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1680                                                      i->value_len,
1681                                                      &new_ea_inode);
1682                 if (ret) {
1683                         new_ea_inode = NULL;
1684                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1685                         goto out;
1686                 }
1687         }
1688
1689         if (old_ea_inode) {
1690                 /* We are ready to release ref count on the old_ea_inode. */
1691                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1692                 if (ret) {
1693                         /* Release newly required ref count on new_ea_inode. */
1694                         if (new_ea_inode) {
1695                                 int err;
1696
1697                                 err = ext4_xattr_inode_dec_ref(handle,
1698                                                                new_ea_inode);
1699                                 if (err)
1700                                         ext4_warning_inode(new_ea_inode,
1701                                                   "dec ref new_ea_inode err=%d",
1702                                                   err);
1703                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1704                                                             i->value_len);
1705                         }
1706                         goto out;
1707                 }
1708
1709                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1710                                             le32_to_cpu(here->e_value_size));
1711         }
1712
1713         /* No failures allowed past this point. */
1714
1715         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1716                 /* Remove the old value. */
1717                 void *first_val = s->base + min_offs;
1718                 size_t offs = le16_to_cpu(here->e_value_offs);
1719                 void *val = s->base + offs;
1720
1721                 memmove(first_val + old_size, first_val, val - first_val);
1722                 memset(first_val, 0, old_size);
1723                 min_offs += old_size;
1724
1725                 /* Adjust all value offsets. */
1726                 last = s->first;
1727                 while (!IS_LAST_ENTRY(last)) {
1728                         size_t o = le16_to_cpu(last->e_value_offs);
1729
1730                         if (!last->e_value_inum &&
1731                             last->e_value_size && o < offs)
1732                                 last->e_value_offs = cpu_to_le16(o + old_size);
1733                         last = EXT4_XATTR_NEXT(last);
1734                 }
1735         }
1736
1737         if (!i->value) {
1738                 /* Remove old name. */
1739                 size_t size = EXT4_XATTR_LEN(name_len);
1740
1741                 last = ENTRY((void *)last - size);
1742                 memmove(here, (void *)here + size,
1743                         (void *)last - (void *)here + sizeof(__u32));
1744                 memset(last, 0, size);
1745
1746                 /*
1747                  * Update i_inline_off - moved ibody region might contain
1748                  * system.data attribute.  Handling a failure here won't
1749                  * cause other complications for setting an xattr.
1750                  */
1751                 if (!is_block && ext4_has_inline_data(inode)) {
1752                         ret = ext4_find_inline_data_nolock(inode);
1753                         if (ret) {
1754                                 ext4_warning_inode(inode,
1755                                         "unable to update i_inline_off");
1756                                 goto out;
1757                         }
1758                 }
1759         } else if (s->not_found) {
1760                 /* Insert new name. */
1761                 size_t size = EXT4_XATTR_LEN(name_len);
1762                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1763
1764                 memmove((void *)here + size, here, rest);
1765                 memset(here, 0, size);
1766                 here->e_name_index = i->name_index;
1767                 here->e_name_len = name_len;
1768                 memcpy(here->e_name, i->name, name_len);
1769         } else {
1770                 /* This is an update, reset value info. */
1771                 here->e_value_inum = 0;
1772                 here->e_value_offs = 0;
1773                 here->e_value_size = 0;
1774         }
1775
1776         if (i->value) {
1777                 /* Insert new value. */
1778                 if (in_inode) {
1779                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1780                 } else if (i->value_len) {
1781                         void *val = s->base + min_offs - new_size;
1782
1783                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1784                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1785                                 memset(val, 0, new_size);
1786                         } else {
1787                                 memcpy(val, i->value, i->value_len);
1788                                 /* Clear padding bytes. */
1789                                 memset(val + i->value_len, 0,
1790                                        new_size - i->value_len);
1791                         }
1792                 }
1793                 here->e_value_size = cpu_to_le32(i->value_len);
1794         }
1795
1796 update_hash:
1797         if (i->value) {
1798                 __le32 hash = 0;
1799
1800                 /* Entry hash calculation. */
1801                 if (in_inode) {
1802                         __le32 crc32c_hash;
1803
1804                         /*
1805                          * Feed crc32c hash instead of the raw value for entry
1806                          * hash calculation. This is to avoid walking
1807                          * potentially long value buffer again.
1808                          */
1809                         crc32c_hash = cpu_to_le32(
1810                                        ext4_xattr_inode_get_hash(new_ea_inode));
1811                         hash = ext4_xattr_hash_entry(here->e_name,
1812                                                      here->e_name_len,
1813                                                      &crc32c_hash, 1);
1814                 } else if (is_block) {
1815                         __le32 *value = s->base + le16_to_cpu(
1816                                                         here->e_value_offs);
1817
1818                         hash = ext4_xattr_hash_entry(here->e_name,
1819                                                      here->e_name_len, value,
1820                                                      new_size >> 2);
1821                 }
1822                 here->e_hash = hash;
1823         }
1824
1825         if (is_block)
1826                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1827
1828         ret = 0;
1829 out:
1830         iput(old_ea_inode);
1831         iput(new_ea_inode);
1832         return ret;
1833 }
1834
1835 struct ext4_xattr_block_find {
1836         struct ext4_xattr_search s;
1837         struct buffer_head *bh;
1838 };
1839
1840 static int
1841 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1842                       struct ext4_xattr_block_find *bs)
1843 {
1844         struct super_block *sb = inode->i_sb;
1845         int error;
1846
1847         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1848                   i->name_index, i->name, i->value, (long)i->value_len);
1849
1850         if (EXT4_I(inode)->i_file_acl) {
1851                 /* The inode already has an extended attribute block. */
1852                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1853                 if (IS_ERR(bs->bh)) {
1854                         error = PTR_ERR(bs->bh);
1855                         bs->bh = NULL;
1856                         return error;
1857                 }
1858                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1859                         atomic_read(&(bs->bh->b_count)),
1860                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1861                 error = ext4_xattr_check_block(inode, bs->bh);
1862                 if (error)
1863                         return error;
1864                 /* Find the named attribute. */
1865                 bs->s.base = BHDR(bs->bh);
1866                 bs->s.first = BFIRST(bs->bh);
1867                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1868                 bs->s.here = bs->s.first;
1869                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1870                                          i->name_index, i->name, 1);
1871                 if (error && error != -ENODATA)
1872                         return error;
1873                 bs->s.not_found = error;
1874         }
1875         return 0;
1876 }
1877
1878 static int
1879 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1880                      struct ext4_xattr_info *i,
1881                      struct ext4_xattr_block_find *bs)
1882 {
1883         struct super_block *sb = inode->i_sb;
1884         struct buffer_head *new_bh = NULL;
1885         struct ext4_xattr_search s_copy = bs->s;
1886         struct ext4_xattr_search *s = &s_copy;
1887         struct mb_cache_entry *ce = NULL;
1888         int error = 0;
1889         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1890         struct inode *ea_inode = NULL, *tmp_inode;
1891         size_t old_ea_inode_quota = 0;
1892         unsigned int ea_ino;
1893
1894
1895 #define header(x) ((struct ext4_xattr_header *)(x))
1896
1897         if (s->base) {
1898                 int offset = (char *)s->here - bs->bh->b_data;
1899
1900                 BUFFER_TRACE(bs->bh, "get_write_access");
1901                 error = ext4_journal_get_write_access(handle, bs->bh);
1902                 if (error)
1903                         goto cleanup;
1904                 lock_buffer(bs->bh);
1905
1906                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1907                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1908
1909                         /*
1910                          * This must happen under buffer lock for
1911                          * ext4_xattr_block_set() to reliably detect modified
1912                          * block
1913                          */
1914                         if (ea_block_cache) {
1915                                 struct mb_cache_entry *oe;
1916
1917                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1918                                         hash, bs->bh->b_blocknr);
1919                                 if (oe) {
1920                                         /*
1921                                          * Xattr block is getting reused. Leave
1922                                          * it alone.
1923                                          */
1924                                         mb_cache_entry_put(ea_block_cache, oe);
1925                                         goto clone_block;
1926                                 }
1927                         }
1928                         ea_bdebug(bs->bh, "modifying in-place");
1929                         error = ext4_xattr_set_entry(i, s, handle, inode,
1930                                                      true /* is_block */);
1931                         ext4_xattr_block_csum_set(inode, bs->bh);
1932                         unlock_buffer(bs->bh);
1933                         if (error == -EFSCORRUPTED)
1934                                 goto bad_block;
1935                         if (!error)
1936                                 error = ext4_handle_dirty_metadata(handle,
1937                                                                    inode,
1938                                                                    bs->bh);
1939                         if (error)
1940                                 goto cleanup;
1941                         goto inserted;
1942                 }
1943 clone_block:
1944                 unlock_buffer(bs->bh);
1945                 ea_bdebug(bs->bh, "cloning");
1946                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1947                 error = -ENOMEM;
1948                 if (s->base == NULL)
1949                         goto cleanup;
1950                 s->first = ENTRY(header(s->base)+1);
1951                 header(s->base)->h_refcount = cpu_to_le32(1);
1952                 s->here = ENTRY(s->base + offset);
1953                 s->end = s->base + bs->bh->b_size;
1954
1955                 /*
1956                  * If existing entry points to an xattr inode, we need
1957                  * to prevent ext4_xattr_set_entry() from decrementing
1958                  * ref count on it because the reference belongs to the
1959                  * original block. In this case, make the entry look
1960                  * like it has an empty value.
1961                  */
1962                 if (!s->not_found && s->here->e_value_inum) {
1963                         ea_ino = le32_to_cpu(s->here->e_value_inum);
1964                         error = ext4_xattr_inode_iget(inode, ea_ino,
1965                                       le32_to_cpu(s->here->e_hash),
1966                                       &tmp_inode);
1967                         if (error)
1968                                 goto cleanup;
1969
1970                         if (!ext4_test_inode_state(tmp_inode,
1971                                         EXT4_STATE_LUSTRE_EA_INODE)) {
1972                                 /*
1973                                  * Defer quota free call for previous
1974                                  * inode until success is guaranteed.
1975                                  */
1976                                 old_ea_inode_quota = le32_to_cpu(
1977                                                 s->here->e_value_size);
1978                         }
1979                         iput(tmp_inode);
1980
1981                         s->here->e_value_inum = 0;
1982                         s->here->e_value_size = 0;
1983                 }
1984         } else {
1985                 /* Allocate a buffer where we construct the new block. */
1986                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1987                 /* assert(header == s->base) */
1988                 error = -ENOMEM;
1989                 if (s->base == NULL)
1990                         goto cleanup;
1991                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1992                 header(s->base)->h_blocks = cpu_to_le32(1);
1993                 header(s->base)->h_refcount = cpu_to_le32(1);
1994                 s->first = ENTRY(header(s->base)+1);
1995                 s->here = ENTRY(header(s->base)+1);
1996                 s->end = s->base + sb->s_blocksize;
1997         }
1998
1999         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2000         if (error == -EFSCORRUPTED)
2001                 goto bad_block;
2002         if (error)
2003                 goto cleanup;
2004
2005         if (i->value && s->here->e_value_inum) {
2006                 /*
2007                  * A ref count on ea_inode has been taken as part of the call to
2008                  * ext4_xattr_set_entry() above. We would like to drop this
2009                  * extra ref but we have to wait until the xattr block is
2010                  * initialized and has its own ref count on the ea_inode.
2011                  */
2012                 ea_ino = le32_to_cpu(s->here->e_value_inum);
2013                 error = ext4_xattr_inode_iget(inode, ea_ino,
2014                                               le32_to_cpu(s->here->e_hash),
2015                                               &ea_inode);
2016                 if (error) {
2017                         ea_inode = NULL;
2018                         goto cleanup;
2019                 }
2020         }
2021
2022 inserted:
2023         if (!IS_LAST_ENTRY(s->first)) {
2024                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2025                                                      &ce);
2026                 if (new_bh) {
2027                         /* We found an identical block in the cache. */
2028                         if (new_bh == bs->bh)
2029                                 ea_bdebug(new_bh, "keeping");
2030                         else {
2031                                 u32 ref;
2032
2033 #ifdef EXT4_XATTR_DEBUG
2034                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2035 #endif
2036                                 /* The old block is released after updating
2037                                    the inode. */
2038                                 error = dquot_alloc_block(inode,
2039                                                 EXT4_C2B(EXT4_SB(sb), 1));
2040                                 if (error)
2041                                         goto cleanup;
2042                                 BUFFER_TRACE(new_bh, "get_write_access");
2043                                 error = ext4_journal_get_write_access(handle,
2044                                                                       new_bh);
2045                                 if (error)
2046                                         goto cleanup_dquot;
2047                                 lock_buffer(new_bh);
2048                                 /*
2049                                  * We have to be careful about races with
2050                                  * adding references to xattr block. Once we
2051                                  * hold buffer lock xattr block's state is
2052                                  * stable so we can check the additional
2053                                  * reference fits.
2054                                  */
2055                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2056                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2057                                         /*
2058                                          * Undo everything and check mbcache
2059                                          * again.
2060                                          */
2061                                         unlock_buffer(new_bh);
2062                                         dquot_free_block(inode,
2063                                                          EXT4_C2B(EXT4_SB(sb),
2064                                                                   1));
2065                                         brelse(new_bh);
2066                                         mb_cache_entry_put(ea_block_cache, ce);
2067                                         ce = NULL;
2068                                         new_bh = NULL;
2069                                         goto inserted;
2070                                 }
2071                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2072                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2073                                         clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2074                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2075                                           ref);
2076                                 ext4_xattr_block_csum_set(inode, new_bh);
2077                                 unlock_buffer(new_bh);
2078                                 error = ext4_handle_dirty_metadata(handle,
2079                                                                    inode,
2080                                                                    new_bh);
2081                                 if (error)
2082                                         goto cleanup_dquot;
2083                         }
2084                         mb_cache_entry_touch(ea_block_cache, ce);
2085                         mb_cache_entry_put(ea_block_cache, ce);
2086                         ce = NULL;
2087                 } else if (bs->bh && s->base == bs->bh->b_data) {
2088                         /* We were modifying this block in-place. */
2089                         ea_bdebug(bs->bh, "keeping this block");
2090                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2091                         new_bh = bs->bh;
2092                         get_bh(new_bh);
2093                 } else {
2094                         /* We need to allocate a new block */
2095                         ext4_fsblk_t goal, block;
2096
2097 #ifdef EXT4_XATTR_DEBUG
2098                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2099 #endif
2100                         goal = ext4_group_first_block_no(sb,
2101                                                 EXT4_I(inode)->i_block_group);
2102                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2103                                                      NULL, &error);
2104                         if (error)
2105                                 goto cleanup;
2106
2107                         ea_idebug(inode, "creating block %llu",
2108                                   (unsigned long long)block);
2109
2110                         new_bh = sb_getblk(sb, block);
2111                         if (unlikely(!new_bh)) {
2112                                 error = -ENOMEM;
2113 getblk_failed:
2114                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2115                                                  EXT4_FREE_BLOCKS_METADATA);
2116                                 goto cleanup;
2117                         }
2118                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2119                                                       ENTRY(header(s->base)+1));
2120                         if (error)
2121                                 goto getblk_failed;
2122                         if (ea_inode) {
2123                                 /* Drop the extra ref on ea_inode. */
2124                                 error = ext4_xattr_inode_dec_ref(handle,
2125                                                                  ea_inode);
2126                                 if (error)
2127                                         ext4_warning_inode(ea_inode,
2128                                                            "dec ref error=%d",
2129                                                            error);
2130                                 iput(ea_inode);
2131                                 ea_inode = NULL;
2132                         }
2133
2134                         lock_buffer(new_bh);
2135                         error = ext4_journal_get_create_access(handle, new_bh);
2136                         if (error) {
2137                                 unlock_buffer(new_bh);
2138                                 error = -EIO;
2139                                 goto getblk_failed;
2140                         }
2141                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2142                         ext4_xattr_block_csum_set(inode, new_bh);
2143                         set_buffer_uptodate(new_bh);
2144                         unlock_buffer(new_bh);
2145                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2146                         error = ext4_handle_dirty_metadata(handle, inode,
2147                                                            new_bh);
2148                         if (error)
2149                                 goto cleanup;
2150                 }
2151         }
2152
2153         if (old_ea_inode_quota)
2154                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2155
2156         /* Update the inode. */
2157         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2158
2159         /* Drop the previous xattr block. */
2160         if (bs->bh && bs->bh != new_bh) {
2161                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2162
2163                 ext4_xattr_release_block(handle, inode, bs->bh,
2164                                          &ea_inode_array,
2165                                          0 /* extra_credits */);
2166                 ext4_xattr_inode_array_free(ea_inode_array);
2167         }
2168         error = 0;
2169
2170 cleanup:
2171         if (ea_inode) {
2172                 int error2;
2173
2174                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2175                 if (error2)
2176                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2177                                            error2);
2178
2179                 /* If there was an error, revert the quota charge. */
2180                 if (error)
2181                         ext4_xattr_inode_free_quota(inode, ea_inode,
2182                                                     i_size_read(ea_inode));
2183                 iput(ea_inode);
2184         }
2185         if (ce)
2186                 mb_cache_entry_put(ea_block_cache, ce);
2187         brelse(new_bh);
2188         if (!(bs->bh && s->base == bs->bh->b_data))
2189                 kfree(s->base);
2190
2191         return error;
2192
2193 cleanup_dquot:
2194         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2195         goto cleanup;
2196
2197 bad_block:
2198         EXT4_ERROR_INODE(inode, "bad block %llu",
2199                          EXT4_I(inode)->i_file_acl);
2200         goto cleanup;
2201
2202 #undef header
2203 }
2204
2205 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2206                           struct ext4_xattr_ibody_find *is)
2207 {
2208         struct ext4_xattr_ibody_header *header;
2209         struct ext4_inode *raw_inode;
2210         int error;
2211
2212         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2213                 return 0;
2214
2215         raw_inode = ext4_raw_inode(&is->iloc);
2216         header = IHDR(inode, raw_inode);
2217         is->s.base = is->s.first = IFIRST(header);
2218         is->s.here = is->s.first;
2219         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2220         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2221                 error = xattr_check_inode(inode, header, is->s.end);
2222                 if (error)
2223                         return error;
2224                 /* Find the named attribute. */
2225                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2226                                          i->name_index, i->name, 0);
2227                 if (error && error != -ENODATA)
2228                         return error;
2229                 is->s.not_found = error;
2230         }
2231         return 0;
2232 }
2233
2234 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2235                                 struct ext4_xattr_info *i,
2236                                 struct ext4_xattr_ibody_find *is)
2237 {
2238         struct ext4_xattr_ibody_header *header;
2239         struct ext4_xattr_search *s = &is->s;
2240         int error;
2241
2242         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2243                 return -ENOSPC;
2244
2245         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2246         if (error)
2247                 return error;
2248         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2249         if (!IS_LAST_ENTRY(s->first)) {
2250                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2251                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2252         } else {
2253                 header->h_magic = cpu_to_le32(0);
2254                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2255         }
2256         return 0;
2257 }
2258
2259 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2260                                  struct ext4_xattr_info *i)
2261 {
2262         void *value;
2263
2264         /* When e_value_inum is set the value is stored externally. */
2265         if (s->here->e_value_inum)
2266                 return 0;
2267         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2268                 return 0;
2269         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2270         return !memcmp(value, i->value, i->value_len);
2271 }
2272
2273 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2274 {
2275         struct buffer_head *bh;
2276         int error;
2277
2278         if (!EXT4_I(inode)->i_file_acl)
2279                 return NULL;
2280         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2281         if (IS_ERR(bh))
2282                 return bh;
2283         error = ext4_xattr_check_block(inode, bh);
2284         if (error) {
2285                 brelse(bh);
2286                 return ERR_PTR(error);
2287         }
2288         return bh;
2289 }
2290
2291 /*
2292  * ext4_xattr_set_handle()
2293  *
2294  * Create, replace or remove an extended attribute for this inode.  Value
2295  * is NULL to remove an existing extended attribute, and non-NULL to
2296  * either replace an existing extended attribute, or create a new extended
2297  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2298  * specify that an extended attribute must exist and must not exist
2299  * previous to the call, respectively.
2300  *
2301  * Returns 0, or a negative error number on failure.
2302  */
2303 int
2304 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2305                       const char *name, const void *value, size_t value_len,
2306                       int flags)
2307 {
2308         struct ext4_xattr_info i = {
2309                 .name_index = name_index,
2310                 .name = name,
2311                 .value = value,
2312                 .value_len = value_len,
2313                 .in_inode = 0,
2314         };
2315         struct ext4_xattr_ibody_find is = {
2316                 .s = { .not_found = -ENODATA, },
2317         };
2318         struct ext4_xattr_block_find bs = {
2319                 .s = { .not_found = -ENODATA, },
2320         };
2321         int no_expand;
2322         int error;
2323
2324         if (!name)
2325                 return -EINVAL;
2326         if (strlen(name) > 255)
2327                 return -ERANGE;
2328
2329         ext4_write_lock_xattr(inode, &no_expand);
2330
2331         /* Check journal credits under write lock. */
2332         if (ext4_handle_valid(handle)) {
2333                 struct buffer_head *bh;
2334                 int credits;
2335
2336                 bh = ext4_xattr_get_block(inode);
2337                 if (IS_ERR(bh)) {
2338                         error = PTR_ERR(bh);
2339                         goto cleanup;
2340                 }
2341
2342                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2343                                                    value_len,
2344                                                    flags & XATTR_CREATE);
2345                 brelse(bh);
2346
2347                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2348                         error = -ENOSPC;
2349                         goto cleanup;
2350                 }
2351                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2352         }
2353
2354         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2355         if (error)
2356                 goto cleanup;
2357
2358         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2359                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2360                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2361                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2362         }
2363
2364         error = ext4_xattr_ibody_find(inode, &i, &is);
2365         if (error)
2366                 goto cleanup;
2367         if (is.s.not_found)
2368                 error = ext4_xattr_block_find(inode, &i, &bs);
2369         if (error)
2370                 goto cleanup;
2371         if (is.s.not_found && bs.s.not_found) {
2372                 error = -ENODATA;
2373                 if (flags & XATTR_REPLACE)
2374                         goto cleanup;
2375                 error = 0;
2376                 if (!value)
2377                         goto cleanup;
2378         } else {
2379                 error = -EEXIST;
2380                 if (flags & XATTR_CREATE)
2381                         goto cleanup;
2382         }
2383
2384         if (!value) {
2385                 if (!is.s.not_found)
2386                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2387                 else if (!bs.s.not_found)
2388                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2389         } else {
2390                 error = 0;
2391                 /* Xattr value did not change? Save us some work and bail out */
2392                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2393                         goto cleanup;
2394                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2395                         goto cleanup;
2396
2397                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2398                     (EXT4_XATTR_SIZE(i.value_len) >
2399                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2400                         i.in_inode = 1;
2401 retry_inode:
2402                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2403                 if (!error && !bs.s.not_found) {
2404                         i.value = NULL;
2405                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2406                 } else if (error == -ENOSPC) {
2407                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2408                                 brelse(bs.bh);
2409                                 bs.bh = NULL;
2410                                 error = ext4_xattr_block_find(inode, &i, &bs);
2411                                 if (error)
2412                                         goto cleanup;
2413                         }
2414                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2415                         if (!error && !is.s.not_found) {
2416                                 i.value = NULL;
2417                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2418                                                              &is);
2419                         } else if (error == -ENOSPC) {
2420                                 /*
2421                                  * Xattr does not fit in the block, store at
2422                                  * external inode if possible.
2423                                  */
2424                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2425                                     i.value_len && !i.in_inode) {
2426                                         i.in_inode = 1;
2427                                         goto retry_inode;
2428                                 }
2429                         }
2430                 }
2431         }
2432         if (!error) {
2433                 ext4_xattr_update_super_block(handle, inode->i_sb);
2434                 inode->i_ctime = current_time(inode);
2435                 if (!value)
2436                         no_expand = 0;
2437                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2438                 /*
2439                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2440                  * error != 0.
2441                  */
2442                 is.iloc.bh = NULL;
2443                 if (IS_SYNC(inode))
2444                         ext4_handle_sync(handle);
2445         }
2446
2447 cleanup:
2448         brelse(is.iloc.bh);
2449         brelse(bs.bh);
2450         ext4_write_unlock_xattr(inode, &no_expand);
2451         return error;
2452 }
2453
2454 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2455                            bool is_create, int *credits)
2456 {
2457         struct buffer_head *bh;
2458         int err;
2459
2460         *credits = 0;
2461
2462         if (!EXT4_SB(inode->i_sb)->s_journal)
2463                 return 0;
2464
2465         down_read(&EXT4_I(inode)->xattr_sem);
2466
2467         bh = ext4_xattr_get_block(inode);
2468         if (IS_ERR(bh)) {
2469                 err = PTR_ERR(bh);
2470         } else {
2471                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2472                                                     value_len, is_create);
2473                 brelse(bh);
2474                 err = 0;
2475         }
2476
2477         up_read(&EXT4_I(inode)->xattr_sem);
2478         return err;
2479 }
2480
2481 /*
2482  * ext4_xattr_set()
2483  *
2484  * Like ext4_xattr_set_handle, but start from an inode. This extended
2485  * attribute modification is a filesystem transaction by itself.
2486  *
2487  * Returns 0, or a negative error number on failure.
2488  */
2489 int
2490 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2491                const void *value, size_t value_len, int flags)
2492 {
2493         handle_t *handle;
2494         struct super_block *sb = inode->i_sb;
2495         int error, retries = 0;
2496         int credits;
2497
2498         error = dquot_initialize(inode);
2499         if (error)
2500                 return error;
2501
2502 retry:
2503         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2504                                        &credits);
2505         if (error)
2506                 return error;
2507
2508         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2509         if (IS_ERR(handle)) {
2510                 error = PTR_ERR(handle);
2511         } else {
2512                 int error2;
2513
2514                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2515                                               value, value_len, flags);
2516                 error2 = ext4_journal_stop(handle);
2517                 if (error == -ENOSPC &&
2518                     ext4_should_retry_alloc(sb, &retries))
2519                         goto retry;
2520                 if (error == 0)
2521                         error = error2;
2522         }
2523
2524         return error;
2525 }
2526
2527 /*
2528  * Shift the EA entries in the inode to create space for the increased
2529  * i_extra_isize.
2530  */
2531 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2532                                      int value_offs_shift, void *to,
2533                                      void *from, size_t n)
2534 {
2535         struct ext4_xattr_entry *last = entry;
2536         int new_offs;
2537
2538         /* We always shift xattr headers further thus offsets get lower */
2539         BUG_ON(value_offs_shift > 0);
2540
2541         /* Adjust the value offsets of the entries */
2542         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2543                 if (!last->e_value_inum && last->e_value_size) {
2544                         new_offs = le16_to_cpu(last->e_value_offs) +
2545                                                         value_offs_shift;
2546                         last->e_value_offs = cpu_to_le16(new_offs);
2547                 }
2548         }
2549         /* Shift the entries by n bytes */
2550         memmove(to, from, n);
2551 }
2552
2553 /*
2554  * Move xattr pointed to by 'entry' from inode into external xattr block
2555  */
2556 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2557                                     struct ext4_inode *raw_inode,
2558                                     struct ext4_xattr_entry *entry)
2559 {
2560         struct ext4_xattr_ibody_find *is = NULL;
2561         struct ext4_xattr_block_find *bs = NULL;
2562         char *buffer = NULL, *b_entry_name = NULL;
2563         size_t value_size = le32_to_cpu(entry->e_value_size);
2564         struct ext4_xattr_info i = {
2565                 .value = NULL,
2566                 .value_len = 0,
2567                 .name_index = entry->e_name_index,
2568                 .in_inode = !!entry->e_value_inum,
2569         };
2570         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2571         int needs_kvfree = 0;
2572         int error;
2573
2574         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2575         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2576         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2577         if (!is || !bs || !b_entry_name) {
2578                 error = -ENOMEM;
2579                 goto out;
2580         }
2581
2582         is->s.not_found = -ENODATA;
2583         bs->s.not_found = -ENODATA;
2584         is->iloc.bh = NULL;
2585         bs->bh = NULL;
2586
2587         /* Save the entry name and the entry value */
2588         if (entry->e_value_inum) {
2589                 buffer = kvmalloc(value_size, GFP_NOFS);
2590                 if (!buffer) {
2591                         error = -ENOMEM;
2592                         goto out;
2593                 }
2594                 needs_kvfree = 1;
2595                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2596                 if (error)
2597                         goto out;
2598         } else {
2599                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2600                 buffer = (void *)IFIRST(header) + value_offs;
2601         }
2602
2603         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2604         b_entry_name[entry->e_name_len] = '\0';
2605         i.name = b_entry_name;
2606
2607         error = ext4_get_inode_loc(inode, &is->iloc);
2608         if (error)
2609                 goto out;
2610
2611         error = ext4_xattr_ibody_find(inode, &i, is);
2612         if (error)
2613                 goto out;
2614
2615         i.value = buffer;
2616         i.value_len = value_size;
2617         error = ext4_xattr_block_find(inode, &i, bs);
2618         if (error)
2619                 goto out;
2620
2621         /* Move ea entry from the inode into the block */
2622         error = ext4_xattr_block_set(handle, inode, &i, bs);
2623         if (error)
2624                 goto out;
2625
2626         /* Remove the chosen entry from the inode */
2627         i.value = NULL;
2628         i.value_len = 0;
2629         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2630
2631 out:
2632         kfree(b_entry_name);
2633         if (needs_kvfree && buffer)
2634                 kvfree(buffer);
2635         if (is)
2636                 brelse(is->iloc.bh);
2637         if (bs)
2638                 brelse(bs->bh);
2639         kfree(is);
2640         kfree(bs);
2641
2642         return error;
2643 }
2644
2645 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2646                                        struct ext4_inode *raw_inode,
2647                                        int isize_diff, size_t ifree,
2648                                        size_t bfree, int *total_ino)
2649 {
2650         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2651         struct ext4_xattr_entry *small_entry;
2652         struct ext4_xattr_entry *entry;
2653         struct ext4_xattr_entry *last;
2654         unsigned int entry_size;        /* EA entry size */
2655         unsigned int total_size;        /* EA entry size + value size */
2656         unsigned int min_total_size;
2657         int error;
2658
2659         while (isize_diff > ifree) {
2660                 entry = NULL;
2661                 small_entry = NULL;
2662                 min_total_size = ~0U;
2663                 last = IFIRST(header);
2664                 /* Find the entry best suited to be pushed into EA block */
2665                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2666                         /* never move system.data out of the inode */
2667                         if ((last->e_name_len == 4) &&
2668                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2669                             !memcmp(last->e_name, "data", 4))
2670                                 continue;
2671                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2672                         if (!last->e_value_inum)
2673                                 total_size += EXT4_XATTR_SIZE(
2674                                                le32_to_cpu(last->e_value_size));
2675                         if (total_size <= bfree &&
2676                             total_size < min_total_size) {
2677                                 if (total_size + ifree < isize_diff) {
2678                                         small_entry = last;
2679                                 } else {
2680                                         entry = last;
2681                                         min_total_size = total_size;
2682                                 }
2683                         }
2684                 }
2685
2686                 if (entry == NULL) {
2687                         if (small_entry == NULL)
2688                                 return -ENOSPC;
2689                         entry = small_entry;
2690                 }
2691
2692                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2693                 total_size = entry_size;
2694                 if (!entry->e_value_inum)
2695                         total_size += EXT4_XATTR_SIZE(
2696                                               le32_to_cpu(entry->e_value_size));
2697                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2698                                                  entry);
2699                 if (error)
2700                         return error;
2701
2702                 *total_ino -= entry_size;
2703                 ifree += total_size;
2704                 bfree -= total_size;
2705         }
2706
2707         return 0;
2708 }
2709
2710 /*
2711  * Expand an inode by new_extra_isize bytes when EAs are present.
2712  * Returns 0 on success or negative error number on failure.
2713  */
2714 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2715                                struct ext4_inode *raw_inode, handle_t *handle)
2716 {
2717         struct ext4_xattr_ibody_header *header;
2718         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2719         static unsigned int mnt_count;
2720         size_t min_offs;
2721         size_t ifree, bfree;
2722         int total_ino;
2723         void *base, *end;
2724         int error = 0, tried_min_extra_isize = 0;
2725         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2726         int isize_diff; /* How much do we need to grow i_extra_isize */
2727
2728 retry:
2729         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2730         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2731                 return 0;
2732
2733         header = IHDR(inode, raw_inode);
2734
2735         /*
2736          * Check if enough free space is available in the inode to shift the
2737          * entries ahead by new_extra_isize.
2738          */
2739
2740         base = IFIRST(header);
2741         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2742         min_offs = end - base;
2743         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2744
2745         error = xattr_check_inode(inode, header, end);
2746         if (error)
2747                 goto cleanup;
2748
2749         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2750         if (ifree >= isize_diff)
2751                 goto shift;
2752
2753         /*
2754          * Enough free space isn't available in the inode, check if
2755          * EA block can hold new_extra_isize bytes.
2756          */
2757         if (EXT4_I(inode)->i_file_acl) {
2758                 struct buffer_head *bh;
2759
2760                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2761                 if (IS_ERR(bh)) {
2762                         error = PTR_ERR(bh);
2763                         goto cleanup;
2764                 }
2765                 error = ext4_xattr_check_block(inode, bh);
2766                 if (error) {
2767                         brelse(bh);
2768                         goto cleanup;
2769                 }
2770                 base = BHDR(bh);
2771                 end = bh->b_data + bh->b_size;
2772                 min_offs = end - base;
2773                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2774                                               NULL);
2775                 brelse(bh);
2776                 if (bfree + ifree < isize_diff) {
2777                         if (!tried_min_extra_isize && s_min_extra_isize) {
2778                                 tried_min_extra_isize++;
2779                                 new_extra_isize = s_min_extra_isize;
2780                                 goto retry;
2781                         }
2782                         error = -ENOSPC;
2783                         goto cleanup;
2784                 }
2785         } else {
2786                 bfree = inode->i_sb->s_blocksize;
2787         }
2788
2789         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2790                                             isize_diff, ifree, bfree,
2791                                             &total_ino);
2792         if (error) {
2793                 if (error == -ENOSPC && !tried_min_extra_isize &&
2794                     s_min_extra_isize) {
2795                         tried_min_extra_isize++;
2796                         new_extra_isize = s_min_extra_isize;
2797                         goto retry;
2798                 }
2799                 goto cleanup;
2800         }
2801 shift:
2802         /* Adjust the offsets and shift the remaining entries ahead */
2803         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2804                         - new_extra_isize, (void *)raw_inode +
2805                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2806                         (void *)header, total_ino);
2807         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2808
2809         if (ext4_has_inline_data(inode))
2810                 error = ext4_find_inline_data_nolock(inode);
2811
2812 cleanup:
2813         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2814                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2815                              inode->i_ino);
2816                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2817         }
2818         return error;
2819 }
2820
2821 #define EIA_INCR 16 /* must be 2^n */
2822 #define EIA_MASK (EIA_INCR - 1)
2823
2824 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2825  * If @ea_inode_array is new or full it will be grown and the old
2826  * contents copied over.
2827  */
2828 static int
2829 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2830                         struct inode *inode)
2831 {
2832         if (*ea_inode_array == NULL) {
2833                 /*
2834                  * Start with 15 inodes, so it fits into a power-of-two size.
2835                  * If *ea_inode_array is NULL, this is essentially offsetof()
2836                  */
2837                 (*ea_inode_array) =
2838                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2839                                          inodes[EIA_MASK]),
2840                                 GFP_NOFS);
2841                 if (*ea_inode_array == NULL)
2842                         return -ENOMEM;
2843                 (*ea_inode_array)->count = 0;
2844         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2845                 /* expand the array once all 15 + n * 16 slots are full */
2846                 struct ext4_xattr_inode_array *new_array = NULL;
2847                 int count = (*ea_inode_array)->count;
2848
2849                 /* if new_array is NULL, this is essentially offsetof() */
2850                 new_array = kmalloc(
2851                                 offsetof(struct ext4_xattr_inode_array,
2852                                          inodes[count + EIA_INCR]),
2853                                 GFP_NOFS);
2854                 if (new_array == NULL)
2855                         return -ENOMEM;
2856                 memcpy(new_array, *ea_inode_array,
2857                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2858                 kfree(*ea_inode_array);
2859                 *ea_inode_array = new_array;
2860         }
2861         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2862         return 0;
2863 }
2864
2865 /*
2866  * ext4_xattr_delete_inode()
2867  *
2868  * Free extended attribute resources associated with this inode. Traverse
2869  * all entries and decrement reference on any xattr inodes associated with this
2870  * inode. This is called immediately before an inode is freed. We have exclusive
2871  * access to the inode. If an orphan inode is deleted it will also release its
2872  * references on xattr block and xattr inodes.
2873  */
2874 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2875                             struct ext4_xattr_inode_array **ea_inode_array,
2876                             int extra_credits)
2877 {
2878         struct buffer_head *bh = NULL;
2879         struct ext4_xattr_ibody_header *header;
2880         struct ext4_iloc iloc = { .bh = NULL };
2881         struct ext4_xattr_entry *entry;
2882         struct inode *ea_inode;
2883         int error;
2884
2885         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2886                                           NULL /* bh */,
2887                                           false /* dirty */,
2888                                           false /* block_csum */);
2889         if (error) {
2890                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2891                 goto cleanup;
2892         }
2893
2894         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2895             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2896
2897                 error = ext4_get_inode_loc(inode, &iloc);
2898                 if (error) {
2899                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2900                         goto cleanup;
2901                 }
2902
2903                 error = ext4_journal_get_write_access(handle, iloc.bh);
2904                 if (error) {
2905                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2906                                          error);
2907                         goto cleanup;
2908                 }
2909
2910                 header = IHDR(inode, ext4_raw_inode(&iloc));
2911                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2912                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2913                                                      IFIRST(header),
2914                                                      false /* block_csum */,
2915                                                      ea_inode_array,
2916                                                      extra_credits,
2917                                                      false /* skip_quota */);
2918         }
2919
2920         if (EXT4_I(inode)->i_file_acl) {
2921                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2922                 if (IS_ERR(bh)) {
2923                         error = PTR_ERR(bh);
2924                         if (error == -EIO)
2925                                 EXT4_ERROR_INODE(inode, "block %llu read error",
2926                                                  EXT4_I(inode)->i_file_acl);
2927                         bh = NULL;
2928                         goto cleanup;
2929                 }
2930                 error = ext4_xattr_check_block(inode, bh);
2931                 if (error)
2932                         goto cleanup;
2933
2934                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2935                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2936                              entry = EXT4_XATTR_NEXT(entry)) {
2937                                 if (!entry->e_value_inum)
2938                                         continue;
2939                                 error = ext4_xattr_inode_iget(inode,
2940                                               le32_to_cpu(entry->e_value_inum),
2941                                               le32_to_cpu(entry->e_hash),
2942                                               &ea_inode);
2943                                 if (error)
2944                                         continue;
2945                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2946                                               le32_to_cpu(entry->e_value_size));
2947                                 iput(ea_inode);
2948                         }
2949
2950                 }
2951
2952                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2953                                          extra_credits);
2954                 /*
2955                  * Update i_file_acl value in the same transaction that releases
2956                  * block.
2957                  */
2958                 EXT4_I(inode)->i_file_acl = 0;
2959                 error = ext4_mark_inode_dirty(handle, inode);
2960                 if (error) {
2961                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2962                                          error);
2963                         goto cleanup;
2964                 }
2965         }
2966         error = 0;
2967 cleanup:
2968         brelse(iloc.bh);
2969         brelse(bh);
2970         return error;
2971 }
2972
2973 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2974 {
2975         int idx;
2976
2977         if (ea_inode_array == NULL)
2978                 return;
2979
2980         for (idx = 0; idx < ea_inode_array->count; ++idx)
2981                 iput(ea_inode_array->inodes[idx]);
2982         kfree(ea_inode_array);
2983 }
2984
2985 /*
2986  * ext4_xattr_block_cache_insert()
2987  *
2988  * Create a new entry in the extended attribute block cache, and insert
2989  * it unless such an entry is already in the cache.
2990  *
2991  * Returns 0, or a negative error number on failure.
2992  */
2993 static void
2994 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2995                               struct buffer_head *bh)
2996 {
2997         struct ext4_xattr_header *header = BHDR(bh);
2998         __u32 hash = le32_to_cpu(header->h_hash);
2999         int reusable = le32_to_cpu(header->h_refcount) <
3000                        EXT4_XATTR_REFCOUNT_MAX;
3001         int error;
3002
3003         if (!ea_block_cache)
3004                 return;
3005         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3006                                       bh->b_blocknr, reusable);
3007         if (error) {
3008                 if (error == -EBUSY)
3009                         ea_bdebug(bh, "already in cache");
3010         } else
3011                 ea_bdebug(bh, "inserting [%x]", (int)hash);
3012 }
3013
3014 /*
3015  * ext4_xattr_cmp()
3016  *
3017  * Compare two extended attribute blocks for equality.
3018  *
3019  * Returns 0 if the blocks are equal, 1 if they differ, and
3020  * a negative error number on errors.
3021  */
3022 static int
3023 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3024                struct ext4_xattr_header *header2)
3025 {
3026         struct ext4_xattr_entry *entry1, *entry2;
3027
3028         entry1 = ENTRY(header1+1);
3029         entry2 = ENTRY(header2+1);
3030         while (!IS_LAST_ENTRY(entry1)) {
3031                 if (IS_LAST_ENTRY(entry2))
3032                         return 1;
3033                 if (entry1->e_hash != entry2->e_hash ||
3034                     entry1->e_name_index != entry2->e_name_index ||
3035                     entry1->e_name_len != entry2->e_name_len ||
3036                     entry1->e_value_size != entry2->e_value_size ||
3037                     entry1->e_value_inum != entry2->e_value_inum ||
3038                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3039                         return 1;
3040                 if (!entry1->e_value_inum &&
3041                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3042                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3043                            le32_to_cpu(entry1->e_value_size)))
3044                         return 1;
3045
3046                 entry1 = EXT4_XATTR_NEXT(entry1);
3047                 entry2 = EXT4_XATTR_NEXT(entry2);
3048         }
3049         if (!IS_LAST_ENTRY(entry2))
3050                 return 1;
3051         return 0;
3052 }
3053
3054 /*
3055  * ext4_xattr_block_cache_find()
3056  *
3057  * Find an identical extended attribute block.
3058  *
3059  * Returns a pointer to the block found, or NULL if such a block was
3060  * not found or an error occurred.
3061  */
3062 static struct buffer_head *
3063 ext4_xattr_block_cache_find(struct inode *inode,
3064                             struct ext4_xattr_header *header,
3065                             struct mb_cache_entry **pce)
3066 {
3067         __u32 hash = le32_to_cpu(header->h_hash);
3068         struct mb_cache_entry *ce;
3069         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3070
3071         if (!ea_block_cache)
3072                 return NULL;
3073         if (!header->h_hash)
3074                 return NULL;  /* never share */
3075         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3076         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3077         while (ce) {
3078                 struct buffer_head *bh;
3079
3080                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3081                 if (IS_ERR(bh)) {
3082                         if (PTR_ERR(bh) == -ENOMEM)
3083                                 return NULL;
3084                         bh = NULL;
3085                         EXT4_ERROR_INODE(inode, "block %lu read error",
3086                                          (unsigned long)ce->e_value);
3087                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3088                         *pce = ce;
3089                         return bh;
3090                 }
3091                 brelse(bh);
3092                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3093         }
3094         return NULL;
3095 }
3096
3097 #define NAME_HASH_SHIFT 5
3098 #define VALUE_HASH_SHIFT 16
3099
3100 /*
3101  * ext4_xattr_hash_entry()
3102  *
3103  * Compute the hash of an extended attribute.
3104  */
3105 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3106                                     size_t value_count)
3107 {
3108         __u32 hash = 0;
3109
3110         while (name_len--) {
3111                 hash = (hash << NAME_HASH_SHIFT) ^
3112                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3113                        *name++;
3114         }
3115         while (value_count--) {
3116                 hash = (hash << VALUE_HASH_SHIFT) ^
3117                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3118                        le32_to_cpu(*value++);
3119         }
3120         return cpu_to_le32(hash);
3121 }
3122
3123 #undef NAME_HASH_SHIFT
3124 #undef VALUE_HASH_SHIFT
3125
3126 #define BLOCK_HASH_SHIFT 16
3127
3128 /*
3129  * ext4_xattr_rehash()
3130  *
3131  * Re-compute the extended attribute hash value after an entry has changed.
3132  */
3133 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3134 {
3135         struct ext4_xattr_entry *here;
3136         __u32 hash = 0;
3137
3138         here = ENTRY(header+1);
3139         while (!IS_LAST_ENTRY(here)) {
3140                 if (!here->e_hash) {
3141                         /* Block is not shared if an entry's hash value == 0 */
3142                         hash = 0;
3143                         break;
3144                 }
3145                 hash = (hash << BLOCK_HASH_SHIFT) ^
3146                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3147                        le32_to_cpu(here->e_hash);
3148                 here = EXT4_XATTR_NEXT(here);
3149         }
3150         header->h_hash = cpu_to_le32(hash);
3151 }
3152
3153 #undef BLOCK_HASH_SHIFT
3154
3155 #define HASH_BUCKET_BITS        10
3156
3157 struct mb_cache *
3158 ext4_xattr_create_cache(void)
3159 {
3160         return mb_cache_create(HASH_BUCKET_BITS);
3161 }
3162
3163 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3164 {
3165         if (cache)
3166                 mb_cache_destroy(cache);
3167 }
3168