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         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 125 }
 126 #endif
 127
 128 static __le32 ext4_xattr_block_csum(struct inode *inode,
 129                                     sector_t block_nr,
 130                                     struct ext4_xattr_header *hdr)
 131 {
 132         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133         __u32 csum;
 134         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 135         __u32 dummy_csum = 0;
 136         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 137
 138         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 139                            sizeof(dsk_block_nr));
 140         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 141         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 142         offset += sizeof(dummy_csum);
 143         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 144                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 145
 146         return cpu_to_le32(csum);
 147 }
 148
 149 static int ext4_xattr_block_csum_verify(struct inode *inode,
 150                                         struct buffer_head *bh)
 151 {
 152         struct ext4_xattr_header *hdr = BHDR(bh);
 153         int ret = 1;
 154
 155         if (ext4_has_metadata_csum(inode->i_sb)) {
 156                 lock_buffer(bh);
 157                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 158                                                         bh->b_blocknr, hdr));
 159                 unlock_buffer(bh);
 160         }
 161         return ret;
 162 }
 163
 164 static void ext4_xattr_block_csum_set(struct inode *inode,
 165                                       struct buffer_head *bh)
 166 {
 167         if (ext4_has_metadata_csum(inode->i_sb))
 168                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 169                                                 bh->b_blocknr, BHDR(bh));
 170 }
 171
 172 static inline const struct xattr_handler *
 173 ext4_xattr_handler(int name_index)
 174 {
 175         const struct xattr_handler *handler = NULL;
 176
 177         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 178                 handler = ext4_xattr_handler_map[name_index];
 179         return handler;
 180 }
 181
 182 static int
 183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 184                          void *value_start)
 185 {
 186         struct ext4_xattr_entry *e = entry;
 187
 188         /* Find the end of the names list */
 189         while (!IS_LAST_ENTRY(e)) {
 190                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 191                 if ((void *)next >= end)
 192                         return -EFSCORRUPTED;
 193                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 194                         return -EFSCORRUPTED;
 195                 e = next;
 196         }
 197
 198         /* Check the values */
 199         while (!IS_LAST_ENTRY(entry)) {
 200                 u32 size = le32_to_cpu(entry->e_value_size);
 201
 202                 if (size > EXT4_XATTR_SIZE_MAX)
 203                         return -EFSCORRUPTED;
 204
 205                 if (size != 0 && entry->e_value_inum == 0) {
 206                         u16 offs = le16_to_cpu(entry->e_value_offs);
 207                         void *value;
 208
 209                         /*
 210                          * The value cannot overlap the names, and the value
 211                          * with padding cannot extend beyond 'end'.  Check both
 212                          * the padded and unpadded sizes, since the size may
 213                          * overflow to 0 when adding padding.
 214                          */
 215                         if (offs > end - value_start)
 216                                 return -EFSCORRUPTED;
 217                         value = value_start + offs;
 218                         if (value < (void *)e + sizeof(u32) ||
 219                             size > end - value ||
 220                             EXT4_XATTR_SIZE(size) > end - value)
 221                                 return -EFSCORRUPTED;
 222                 }
 223                 entry = EXT4_XATTR_NEXT(entry);
 224         }
 225
 226         return 0;
 227 }
 228
 229 static inline int
 230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 231                          const char *function, unsigned int line)
 232 {
 233         int error = -EFSCORRUPTED;
 234
 235         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 236             BHDR(bh)->h_blocks != cpu_to_le32(1))
 237                 goto errout;
 238         if (buffer_verified(bh))
 239                 return 0;
 240
 241         error = -EFSBADCRC;
 242         if (!ext4_xattr_block_csum_verify(inode, bh))
 243                 goto errout;
 244         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 245                                          bh->b_data);
 246 errout:
 247         if (error)
 248                 __ext4_error_inode(inode, function, line, 0,
 249                                    "corrupted xattr block %llu",
 250                                    (unsigned long long) bh->b_blocknr);
 251         else
 252                 set_buffer_verified(bh);
 253         return error;
 254 }
 255
 256 #define ext4_xattr_check_block(inode, bh) \
 257         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 258
 259
 260 static int
 261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 262                          void *end, const char *function, unsigned int line)
 263 {
 264         int error = -EFSCORRUPTED;
 265
 266         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 267             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 268                 goto errout;
 269         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 270 errout:
 271         if (error)
 272                 __ext4_error_inode(inode, function, line, 0,
 273                                    "corrupted in-inode xattr");
 274         return error;
 275 }
 276
 277 #define xattr_check_inode(inode, header, end) \
 278         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 279
 280 static int
 281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 282                  void *end, int name_index, const char *name, int sorted)
 283 {
 284         struct ext4_xattr_entry *entry, *next;
 285         size_t name_len;
 286         int cmp = 1;
 287
 288         if (name == NULL)
 289                 return -EINVAL;
 290         name_len = strlen(name);
 291         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 292                 next = EXT4_XATTR_NEXT(entry);
 293                 if ((void *) next >= end) {
 294                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 295                         return -EFSCORRUPTED;
 296                 }
 297                 cmp = name_index - entry->e_name_index;
 298                 if (!cmp)
 299                         cmp = name_len - entry->e_name_len;
 300                 if (!cmp)
 301                         cmp = memcmp(name, entry->e_name, name_len);
 302                 if (cmp <= 0 && (sorted || cmp == 0))
 303                         break;
 304         }
 305         *pentry = entry;
 306         return cmp ? -ENODATA : 0;
 307 }
 308
 309 static u32
 310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 311 {
 312         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 313 }
 314
 315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 316 {
 317         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 318                 (u32) inode_peek_iversion_raw(ea_inode);
 319 }
 320
 321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 322 {
 323         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 324         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 325 }
 326
 327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 328 {
 329         return (u32)ea_inode->i_atime.tv_sec;
 330 }
 331
 332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 333 {
 334         ea_inode->i_atime.tv_sec = hash;
 335 }
 336
 337 /*
 338  * Read the EA value from an inode.
 339  */
 340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 341 {
 342         int blocksize = 1 << ea_inode->i_blkbits;
 343         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 344         int tail_size = (size % blocksize) ?: blocksize;
 345         struct buffer_head *bhs_inline[8];
 346         struct buffer_head **bhs = bhs_inline;
 347         int i, ret;
 348
 349         if (bh_count > ARRAY_SIZE(bhs_inline)) {
 350                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 351                 if (!bhs)
 352                         return -ENOMEM;
 353         }
 354
 355         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 356                                true /* wait */, bhs);
 357         if (ret)
 358                 goto free_bhs;
 359
 360         for (i = 0; i < bh_count; i++) {
 361                 /* There shouldn't be any holes in ea_inode. */
 362                 if (!bhs[i]) {
 363                         ret = -EFSCORRUPTED;
 364                         goto put_bhs;
 365                 }
 366                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 367                        i < bh_count - 1 ? blocksize : tail_size);
 368         }
 369         ret = 0;
 370 put_bhs:
 371         for (i = 0; i < bh_count; i++)
 372                 brelse(bhs[i]);
 373 free_bhs:
 374         if (bhs != bhs_inline)
 375                 kfree(bhs);
 376         return ret;
 377 }
 378
 379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 380
 381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 382                                  u32 ea_inode_hash, struct inode **ea_inode)
 383 {
 384         struct inode *inode;
 385         int err;
 386
 387         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 388         if (IS_ERR(inode)) {
 389                 err = PTR_ERR(inode);
 390                 ext4_error(parent->i_sb,
 391                            "error while reading EA inode %lu err=%d", ea_ino,
 392                            err);
 393                 return err;
 394         }
 395
 396         if (is_bad_inode(inode)) {
 397                 ext4_error(parent->i_sb,
 398                            "error while reading EA inode %lu is_bad_inode",
 399                            ea_ino);
 400                 err = -EIO;
 401                 goto error;
 402         }
 403
 404         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 405                 ext4_error(parent->i_sb,
 406                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 407                             ea_ino);
 408                 err = -EINVAL;
 409                 goto error;
 410         }
 411
 412         ext4_xattr_inode_set_class(inode);
 413
 414         /*
 415          * Check whether this is an old Lustre-style xattr inode. Lustre
 416          * implementation does not have hash validation, rather it has a
 417          * backpointer from ea_inode to the parent inode.
 418          */
 419         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 420             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 421             inode->i_generation == parent->i_generation) {
 422                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 423                 ext4_xattr_inode_set_ref(inode, 1);
 424         } else {
 425                 inode_lock(inode);
 426                 inode->i_flags |= S_NOQUOTA;
 427                 inode_unlock(inode);
 428         }
 429
 430         *ea_inode = inode;
 431         return 0;
 432 error:
 433         iput(inode);
 434         return err;
 435 }
 436
 437 static int
 438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 439                                struct ext4_xattr_entry *entry, void *buffer,
 440                                size_t size)
 441 {
 442         u32 hash;
 443
 444         /* Verify stored hash matches calculated hash. */
 445         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 446         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 447                 return -EFSCORRUPTED;
 448
 449         if (entry) {
 450                 __le32 e_hash, tmp_data;
 451
 452                 /* Verify entry hash. */
 453                 tmp_data = cpu_to_le32(hash);
 454                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 455                                                &tmp_data, 1);
 456                 if (e_hash != entry->e_hash)
 457                         return -EFSCORRUPTED;
 458         }
 459         return 0;
 460 }
 461
 462 /*
 463  * Read xattr value from the EA inode.
 464  */
 465 static int
 466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 467                      void *buffer, size_t size)
 468 {
 469         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 470         struct inode *ea_inode;
 471         int err;
 472
 473         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 474                                     le32_to_cpu(entry->e_hash), &ea_inode);
 475         if (err) {
 476                 ea_inode = NULL;
 477                 goto out;
 478         }
 479
 480         if (i_size_read(ea_inode) != size) {
 481                 ext4_warning_inode(ea_inode,
 482                                    "ea_inode file size=%llu entry size=%zu",
 483                                    i_size_read(ea_inode), size);
 484                 err = -EFSCORRUPTED;
 485                 goto out;
 486         }
 487
 488         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 489         if (err)
 490                 goto out;
 491
 492         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 493                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 494                                                      size);
 495                 if (err) {
 496                         ext4_warning_inode(ea_inode,
 497                                            "EA inode hash validation failed");
 498                         goto out;
 499                 }
 500
 501                 if (ea_inode_cache)
 502                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 503                                         ext4_xattr_inode_get_hash(ea_inode),
 504                                         ea_inode->i_ino, true /* reusable */);
 505         }
 506 out:
 507         iput(ea_inode);
 508         return err;
 509 }
 510
 511 static int
 512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 513                      void *buffer, size_t buffer_size)
 514 {
 515         struct buffer_head *bh = NULL;
 516         struct ext4_xattr_entry *entry;
 517         size_t size;
 518         void *end;
 519         int error;
 520         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 521
 522         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 523                   name_index, name, buffer, (long)buffer_size);
 524
 525         if (!EXT4_I(inode)->i_file_acl)
 526                 return -ENODATA;
 527         ea_idebug(inode, "reading block %llu",
 528                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 529         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 530         if (IS_ERR(bh))
 531                 return PTR_ERR(bh);
 532         ea_bdebug(bh, "b_count=%d, refcount=%d",
 533                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 534         error = ext4_xattr_check_block(inode, bh);
 535         if (error)
 536                 goto cleanup;
 537         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 538         entry = BFIRST(bh);
 539         end = bh->b_data + bh->b_size;
 540         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 541         if (error)
 542                 goto cleanup;
 543         size = le32_to_cpu(entry->e_value_size);
 544         error = -ERANGE;
 545         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 546                 goto cleanup;
 547         if (buffer) {
 548                 if (size > buffer_size)
 549                         goto cleanup;
 550                 if (entry->e_value_inum) {
 551                         error = ext4_xattr_inode_get(inode, entry, buffer,
 552                                                      size);
 553                         if (error)
 554                                 goto cleanup;
 555                 } else {
 556                         u16 offset = le16_to_cpu(entry->e_value_offs);
 557                         void *p = bh->b_data + offset;
 558
 559                         if (unlikely(p + size > end))
 560                                 goto cleanup;
 561                         memcpy(buffer, p, size);
 562                 }
 563         }
 564         error = size;
 565
 566 cleanup:
 567         brelse(bh);
 568         return error;
 569 }
 570
 571 int
 572 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 573                      void *buffer, size_t buffer_size)
 574 {
 575         struct ext4_xattr_ibody_header *header;
 576         struct ext4_xattr_entry *entry;
 577         struct ext4_inode *raw_inode;
 578         struct ext4_iloc iloc;
 579         size_t size;
 580         void *end;
 581         int error;
 582
 583         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 584                 return -ENODATA;
 585         error = ext4_get_inode_loc(inode, &iloc);
 586         if (error)
 587                 return error;
 588         raw_inode = ext4_raw_inode(&iloc);
 589         header = IHDR(inode, raw_inode);
 590         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 591         error = xattr_check_inode(inode, header, end);
 592         if (error)
 593                 goto cleanup;
 594         entry = IFIRST(header);
 595         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 596         if (error)
 597                 goto cleanup;
 598         size = le32_to_cpu(entry->e_value_size);
 599         error = -ERANGE;
 600         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 601                 goto cleanup;
 602         if (buffer) {
 603                 if (size > buffer_size)
 604                         goto cleanup;
 605                 if (entry->e_value_inum) {
 606                         error = ext4_xattr_inode_get(inode, entry, buffer,
 607                                                      size);
 608                         if (error)
 609                                 goto cleanup;
 610                 } else {
 611                         u16 offset = le16_to_cpu(entry->e_value_offs);
 612                         void *p = (void *)IFIRST(header) + offset;
 613
 614                         if (unlikely(p + size > end))
 615                                 goto cleanup;
 616                         memcpy(buffer, p, size);
 617                 }
 618         }
 619         error = size;
 620
 621 cleanup:
 622         brelse(iloc.bh);
 623         return error;
 624 }
 625
 626 /*
 627  * ext4_xattr_get()
 628  *
 629  * Copy an extended attribute into the buffer
 630  * provided, or compute the buffer size required.
 631  * Buffer is NULL to compute the size of the buffer required.
 632  *
 633  * Returns a negative error number on failure, or the number of bytes
 634  * used / required on success.
 635  */
 636 int
 637 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 638                void *buffer, size_t buffer_size)
 639 {
 640         int error;
 641
 642         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 643                 return -EIO;
 644
 645         if (strlen(name) > 255)
 646                 return -ERANGE;
 647
 648         down_read(&EXT4_I(inode)->xattr_sem);
 649         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 650                                      buffer_size);
 651         if (error == -ENODATA)
 652                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 653                                              buffer_size);
 654         up_read(&EXT4_I(inode)->xattr_sem);
 655         return error;
 656 }
 657
 658 static int
 659 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 660                         char *buffer, size_t buffer_size)
 661 {
 662         size_t rest = buffer_size;
 663
 664         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 665                 const struct xattr_handler *handler =
 666                         ext4_xattr_handler(entry->e_name_index);
 667
 668                 if (handler && (!handler->list || handler->list(dentry))) {
 669                         const char *prefix = handler->prefix ?: handler->name;
 670                         size_t prefix_len = strlen(prefix);
 671                         size_t size = prefix_len + entry->e_name_len + 1;
 672
 673                         if (buffer) {
 674                                 if (size > rest)
 675                                         return -ERANGE;
 676                                 memcpy(buffer, prefix, prefix_len);
 677                                 buffer += prefix_len;
 678                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 679                                 buffer += entry->e_name_len;
 680                                 *buffer++ = 0;
 681                         }
 682                         rest -= size;
 683                 }
 684         }
 685         return buffer_size - rest;  /* total size */
 686 }
 687
 688 static int
 689 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 690 {
 691         struct inode *inode = d_inode(dentry);
 692         struct buffer_head *bh = NULL;
 693         int error;
 694
 695         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 696                   buffer, (long)buffer_size);
 697
 698         if (!EXT4_I(inode)->i_file_acl)
 699                 return 0;
 700         ea_idebug(inode, "reading block %llu",
 701                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 702         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 703         if (IS_ERR(bh))
 704                 return PTR_ERR(bh);
 705         ea_bdebug(bh, "b_count=%d, refcount=%d",
 706                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 707         error = ext4_xattr_check_block(inode, bh);
 708         if (error)
 709                 goto cleanup;
 710         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 711         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 712                                         buffer_size);
 713 cleanup:
 714         brelse(bh);
 715         return error;
 716 }
 717
 718 static int
 719 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 720 {
 721         struct inode *inode = d_inode(dentry);
 722         struct ext4_xattr_ibody_header *header;
 723         struct ext4_inode *raw_inode;
 724         struct ext4_iloc iloc;
 725         void *end;
 726         int error;
 727
 728         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 729                 return 0;
 730         error = ext4_get_inode_loc(inode, &iloc);
 731         if (error)
 732                 return error;
 733         raw_inode = ext4_raw_inode(&iloc);
 734         header = IHDR(inode, raw_inode);
 735         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 736         error = xattr_check_inode(inode, header, end);
 737         if (error)
 738                 goto cleanup;
 739         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 740                                         buffer, buffer_size);
 741
 742 cleanup:
 743         brelse(iloc.bh);
 744         return error;
 745 }
 746
 747 /*
 748  * Inode operation listxattr()
 749  *
 750  * d_inode(dentry)->i_rwsem: don't care
 751  *
 752  * Copy a list of attribute names into the buffer
 753  * provided, or compute the buffer size required.
 754  * Buffer is NULL to compute the size of the buffer required.
 755  *
 756  * Returns a negative error number on failure, or the number of bytes
 757  * used / required on success.
 758  */
 759 ssize_t
 760 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 761 {
 762         int ret, ret2;
 763
 764         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 765         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 766         if (ret < 0)
 767                 goto errout;
 768         if (buffer) {
 769                 buffer += ret;
 770                 buffer_size -= ret;
 771         }
 772         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 773         if (ret < 0)
 774                 goto errout;
 775         ret += ret2;
 776 errout:
 777         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 778         return ret;
 779 }
 780
 781 /*
 782  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 783  * not set, set it.
 784  */
 785 static void ext4_xattr_update_super_block(handle_t *handle,
 786                                           struct super_block *sb)
 787 {
 788         if (ext4_has_feature_xattr(sb))
 789                 return;
 790
 791         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 792         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 793                 ext4_set_feature_xattr(sb);
 794                 ext4_handle_dirty_super(handle, sb);
 795         }
 796 }
 797
 798 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 799 {
 800         struct ext4_iloc iloc = { .bh = NULL };
 801         struct buffer_head *bh = NULL;
 802         struct ext4_inode *raw_inode;
 803         struct ext4_xattr_ibody_header *header;
 804         struct ext4_xattr_entry *entry;
 805         qsize_t ea_inode_refs = 0;
 806         void *end;
 807         int ret;
 808
 809         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 810
 811         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 812                 ret = ext4_get_inode_loc(inode, &iloc);
 813                 if (ret)
 814                         goto out;
 815                 raw_inode = ext4_raw_inode(&iloc);
 816                 header = IHDR(inode, raw_inode);
 817                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 818                 ret = xattr_check_inode(inode, header, end);
 819                 if (ret)
 820                         goto out;
 821
 822                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 823                      entry = EXT4_XATTR_NEXT(entry))
 824                         if (entry->e_value_inum)
 825                                 ea_inode_refs++;
 826         }
 827
 828         if (EXT4_I(inode)->i_file_acl) {
 829                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 830                 if (IS_ERR(bh)) {
 831                         ret = PTR_ERR(bh);
 832                         bh = NULL;
 833                         goto out;
 834                 }
 835
 836                 ret = ext4_xattr_check_block(inode, bh);
 837                 if (ret)
 838                         goto out;
 839
 840                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 841                      entry = EXT4_XATTR_NEXT(entry))
 842                         if (entry->e_value_inum)
 843                                 ea_inode_refs++;
 844         }
 845         *usage = ea_inode_refs + 1;
 846         ret = 0;
 847 out:
 848         brelse(iloc.bh);
 849         brelse(bh);
 850         return ret;
 851 }
 852
 853 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 854 {
 855         struct super_block *sb = inode->i_sb;
 856         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 857                                     inode->i_blkbits);
 858         size_t mask = ~(cluster_size - 1);
 859
 860         return (length + cluster_size - 1) & mask;
 861 }
 862
 863 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 864 {
 865         int err;
 866
 867         err = dquot_alloc_inode(inode);
 868         if (err)
 869                 return err;
 870         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 871         if (err)
 872                 dquot_free_inode(inode);
 873         return err;
 874 }
 875
 876 static void ext4_xattr_inode_free_quota(struct inode *parent,
 877                                         struct inode *ea_inode,
 878                                         size_t len)
 879 {
 880         if (ea_inode &&
 881             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 882                 return;
 883         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 884         dquot_free_inode(parent);
 885 }
 886
 887 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 888                              struct buffer_head *block_bh, size_t value_len,
 889                              bool is_create)
 890 {
 891         int credits;
 892         int blocks;
 893
 894         /*
 895          * 1) Owner inode update
 896          * 2) Ref count update on old xattr block
 897          * 3) new xattr block
 898          * 4) block bitmap update for new xattr block
 899          * 5) group descriptor for new xattr block
 900          * 6) block bitmap update for old xattr block
 901          * 7) group descriptor for old block
 902          *
 903          * 6 & 7 can happen if we have two racing threads T_a and T_b
 904          * which are each trying to set an xattr on inodes I_a and I_b
 905          * which were both initially sharing an xattr block.
 906          */
 907         credits = 7;
 908
 909         /* Quota updates. */
 910         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 911
 912         /*
 913          * In case of inline data, we may push out the data to a block,
 914          * so we need to reserve credits for this eventuality
 915          */
 916         if (inode && ext4_has_inline_data(inode))
 917                 credits += ext4_writepage_trans_blocks(inode) + 1;
 918
 919         /* We are done if ea_inode feature is not enabled. */
 920         if (!ext4_has_feature_ea_inode(sb))
 921                 return credits;
 922
 923         /* New ea_inode, inode map, block bitmap, group descriptor. */
 924         credits += 4;
 925
 926         /* Data blocks. */
 927         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 928
 929         /* Indirection block or one level of extent tree. */
 930         blocks += 1;
 931
 932         /* Block bitmap and group descriptor updates for each block. */
 933         credits += blocks * 2;
 934
 935         /* Blocks themselves. */
 936         credits += blocks;
 937
 938         if (!is_create) {
 939                 /* Dereference ea_inode holding old xattr value.
 940                  * Old ea_inode, inode map, block bitmap, group descriptor.
 941                  */
 942                 credits += 4;
 943
 944                 /* Data blocks for old ea_inode. */
 945                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 946
 947                 /* Indirection block or one level of extent tree for old
 948                  * ea_inode.
 949                  */
 950                 blocks += 1;
 951
 952                 /* Block bitmap and group descriptor updates for each block. */
 953                 credits += blocks * 2;
 954         }
 955
 956         /* We may need to clone the existing xattr block in which case we need
 957          * to increment ref counts for existing ea_inodes referenced by it.
 958          */
 959         if (block_bh) {
 960                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
 961
 962                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 963                         if (entry->e_value_inum)
 964                                 /* Ref count update on ea_inode. */
 965                                 credits += 1;
 966         }
 967         return credits;
 968 }
 969
 970 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 971                                      int credits, struct buffer_head *bh,
 972                                      bool dirty, bool block_csum)
 973 {
 974         int error;
 975
 976         if (!ext4_handle_valid(handle))
 977                 return 0;
 978
 979         if (handle->h_buffer_credits >= credits)
 980                 return 0;
 981
 982         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 983         if (!error)
 984                 return 0;
 985         if (error < 0) {
 986                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 987                 return error;
 988         }
 989
 990         if (bh && dirty) {
 991                 if (block_csum)
 992                         ext4_xattr_block_csum_set(inode, bh);
 993                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
 994                 if (error) {
 995                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 996                                      error);
 997                         return error;
 998                 }
 999         }
1000
1001         error = ext4_journal_restart(handle, credits);
1002         if (error) {
1003                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1004                 return error;
1005         }
1006
1007         if (bh) {
1008                 error = ext4_journal_get_write_access(handle, bh);
1009                 if (error) {
1010                         ext4_warning(inode->i_sb,
1011                                      "Get write access failed (error %d)",
1012                                      error);
1013                         return error;
1014                 }
1015         }
1016         return 0;
1017 }
1018
1019 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1020                                        int ref_change)
1021 {
1022         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1023         struct ext4_iloc iloc;
1024         s64 ref_count;
1025         u32 hash;
1026         int ret;
1027
1028         inode_lock(ea_inode);
1029
1030         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1031         if (ret) {
1032                 iloc.bh = NULL;
1033                 goto out;
1034         }
1035
1036         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1037         ref_count += ref_change;
1038         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1039
1040         if (ref_change > 0) {
1041                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1042                           ea_inode->i_ino, ref_count);
1043
1044                 if (ref_count == 1) {
1045                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1046                                   ea_inode->i_ino, ea_inode->i_nlink);
1047
1048                         set_nlink(ea_inode, 1);
1049                         ext4_orphan_del(handle, ea_inode);
1050
1051                         if (ea_inode_cache) {
1052                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1053                                 mb_cache_entry_create(ea_inode_cache,
1054                                                       GFP_NOFS, hash,
1055                                                       ea_inode->i_ino,
1056                                                       true /* reusable */);
1057                         }
1058                 }
1059         } else {
1060                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1061                           ea_inode->i_ino, ref_count);
1062
1063                 if (ref_count == 0) {
1064                         WARN_ONCE(ea_inode->i_nlink != 1,
1065                                   "EA inode %lu i_nlink=%u",
1066                                   ea_inode->i_ino, ea_inode->i_nlink);
1067
1068                         clear_nlink(ea_inode);
1069                         ext4_orphan_add(handle, ea_inode);
1070
1071                         if (ea_inode_cache) {
1072                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1073                                 mb_cache_entry_delete(ea_inode_cache, hash,
1074                                                       ea_inode->i_ino);
1075                         }
1076                 }
1077         }
1078
1079         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1080         iloc.bh = NULL;
1081         if (ret)
1082                 ext4_warning_inode(ea_inode,
1083                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1084 out:
1085         brelse(iloc.bh);
1086         inode_unlock(ea_inode);
1087         return ret;
1088 }
1089
1090 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1091 {
1092         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1093 }
1094
1095 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1096 {
1097         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1098 }
1099
1100 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1101                                         struct ext4_xattr_entry *first)
1102 {
1103         struct inode *ea_inode;
1104         struct ext4_xattr_entry *entry;
1105         struct ext4_xattr_entry *failed_entry;
1106         unsigned int ea_ino;
1107         int err, saved_err;
1108
1109         for (entry = first; !IS_LAST_ENTRY(entry);
1110              entry = EXT4_XATTR_NEXT(entry)) {
1111                 if (!entry->e_value_inum)
1112                         continue;
1113                 ea_ino = le32_to_cpu(entry->e_value_inum);
1114                 err = ext4_xattr_inode_iget(parent, ea_ino,
1115                                             le32_to_cpu(entry->e_hash),
1116                                             &ea_inode);
1117                 if (err)
1118                         goto cleanup;
1119                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1120                 if (err) {
1121                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1122                         iput(ea_inode);
1123                         goto cleanup;
1124                 }
1125                 iput(ea_inode);
1126         }
1127         return 0;
1128
1129 cleanup:
1130         saved_err = err;
1131         failed_entry = entry;
1132
1133         for (entry = first; entry != failed_entry;
1134              entry = EXT4_XATTR_NEXT(entry)) {
1135                 if (!entry->e_value_inum)
1136                         continue;
1137                 ea_ino = le32_to_cpu(entry->e_value_inum);
1138                 err = ext4_xattr_inode_iget(parent, ea_ino,
1139                                             le32_to_cpu(entry->e_hash),
1140                                             &ea_inode);
1141                 if (err) {
1142                         ext4_warning(parent->i_sb,
1143                                      "cleanup ea_ino %u iget error %d", ea_ino,
1144                                      err);
1145                         continue;
1146                 }
1147                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1148                 if (err)
1149                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1150                                            err);
1151                 iput(ea_inode);
1152         }
1153         return saved_err;
1154 }
1155
1156 static void
1157 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1158                              struct buffer_head *bh,
1159                              struct ext4_xattr_entry *first, bool block_csum,
1160                              struct ext4_xattr_inode_array **ea_inode_array,
1161                              int extra_credits, bool skip_quota)
1162 {
1163         struct inode *ea_inode;
1164         struct ext4_xattr_entry *entry;
1165         bool dirty = false;
1166         unsigned int ea_ino;
1167         int err;
1168         int credits;
1169
1170         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1171         credits = 2 + extra_credits;
1172
1173         for (entry = first; !IS_LAST_ENTRY(entry);
1174              entry = EXT4_XATTR_NEXT(entry)) {
1175                 if (!entry->e_value_inum)
1176                         continue;
1177                 ea_ino = le32_to_cpu(entry->e_value_inum);
1178                 err = ext4_xattr_inode_iget(parent, ea_ino,
1179                                             le32_to_cpu(entry->e_hash),
1180                                             &ea_inode);
1181                 if (err)
1182                         continue;
1183
1184                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1185                 if (err) {
1186                         ext4_warning_inode(ea_inode,
1187                                            "Expand inode array err=%d", err);
1188                         iput(ea_inode);
1189                         continue;
1190                 }
1191
1192                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1193                                                 dirty, block_csum);
1194                 if (err) {
1195                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1196                                            err);
1197                         continue;
1198                 }
1199
1200                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1201                 if (err) {
1202                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1203                                            err);
1204                         continue;
1205                 }
1206
1207                 if (!skip_quota)
1208                         ext4_xattr_inode_free_quota(parent, ea_inode,
1209                                               le32_to_cpu(entry->e_value_size));
1210
1211                 /*
1212                  * Forget about ea_inode within the same transaction that
1213                  * decrements the ref count. This avoids duplicate decrements in
1214                  * case the rest of the work spills over to subsequent
1215                  * transactions.
1216                  */
1217                 entry->e_value_inum = 0;
1218                 entry->e_value_size = 0;
1219
1220                 dirty = true;
1221         }
1222
1223         if (dirty) {
1224                 /*
1225                  * Note that we are deliberately skipping csum calculation for
1226                  * the final update because we do not expect any journal
1227                  * restarts until xattr block is freed.
1228                  */
1229
1230                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1231                 if (err)
1232                         ext4_warning_inode(parent,
1233                                            "handle dirty metadata err=%d", err);
1234         }
1235 }
1236
1237 /*
1238  * Release the xattr block BH: If the reference count is > 1, decrement it;
1239  * otherwise free the block.
1240  */
1241 static void
1242 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1243                          struct buffer_head *bh,
1244                          struct ext4_xattr_inode_array **ea_inode_array,
1245                          int extra_credits)
1246 {
1247         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1248         u32 hash, ref;
1249         int error = 0;
1250
1251         BUFFER_TRACE(bh, "get_write_access");
1252         error = ext4_journal_get_write_access(handle, bh);
1253         if (error)
1254                 goto out;
1255
1256         lock_buffer(bh);
1257         hash = le32_to_cpu(BHDR(bh)->h_hash);
1258         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1259         if (ref == 1) {
1260                 ea_bdebug(bh, "refcount now=0; freeing");
1261                 /*
1262                  * This must happen under buffer lock for
1263                  * ext4_xattr_block_set() to reliably detect freed block
1264                  */
1265                 if (ea_block_cache)
1266                         mb_cache_entry_delete(ea_block_cache, hash,
1267                                               bh->b_blocknr);
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                                         ce->e_reusable = 1;
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         /*
1431          * Let the next inode be the goal, so we try and allocate the EA inode
1432          * in the same group, or nearby one.
1433          */
1434         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1435                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1436                                   EXT4_EA_INODE_FL);
1437         if (!IS_ERR(ea_inode)) {
1438                 ea_inode->i_op = &ext4_file_inode_operations;
1439                 ea_inode->i_fop = &ext4_file_operations;
1440                 ext4_set_aops(ea_inode);
1441                 ext4_xattr_inode_set_class(ea_inode);
1442                 unlock_new_inode(ea_inode);
1443                 ext4_xattr_inode_set_ref(ea_inode, 1);
1444                 ext4_xattr_inode_set_hash(ea_inode, hash);
1445                 err = ext4_mark_inode_dirty(handle, ea_inode);
1446                 if (!err)
1447                         err = ext4_inode_attach_jinode(ea_inode);
1448                 if (err) {
1449                         iput(ea_inode);
1450                         return ERR_PTR(err);
1451                 }
1452
1453                 /*
1454                  * Xattr inodes are shared therefore quota charging is performed
1455                  * at a higher level.
1456                  */
1457                 dquot_free_inode(ea_inode);
1458                 dquot_drop(ea_inode);
1459                 inode_lock(ea_inode);
1460                 ea_inode->i_flags |= S_NOQUOTA;
1461                 inode_unlock(ea_inode);
1462         }
1463
1464         return ea_inode;
1465 }
1466
1467 static struct inode *
1468 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1469                             size_t value_len, u32 hash)
1470 {
1471         struct inode *ea_inode;
1472         struct mb_cache_entry *ce;
1473         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1474         void *ea_data;
1475
1476         if (!ea_inode_cache)
1477                 return NULL;
1478
1479         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1480         if (!ce)
1481                 return NULL;
1482
1483         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1484                      !(current->flags & PF_MEMALLOC_NOFS));
1485
1486         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1487         if (!ea_data) {
1488                 mb_cache_entry_put(ea_inode_cache, ce);
1489                 return NULL;
1490         }
1491
1492         while (ce) {
1493                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1494                                      EXT4_IGET_NORMAL);
1495                 if (!IS_ERR(ea_inode) &&
1496                     !is_bad_inode(ea_inode) &&
1497                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1498                     i_size_read(ea_inode) == value_len &&
1499                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1500                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1501                                                     value_len) &&
1502                     !memcmp(value, ea_data, value_len)) {
1503                         mb_cache_entry_touch(ea_inode_cache, ce);
1504                         mb_cache_entry_put(ea_inode_cache, ce);
1505                         kvfree(ea_data);
1506                         return ea_inode;
1507                 }
1508
1509                 if (!IS_ERR(ea_inode))
1510                         iput(ea_inode);
1511                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1512         }
1513         kvfree(ea_data);
1514         return NULL;
1515 }
1516
1517 /*
1518  * Add value of the EA in an inode.
1519  */
1520 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1521                                           const void *value, size_t value_len,
1522                                           struct inode **ret_inode)
1523 {
1524         struct inode *ea_inode;
1525         u32 hash;
1526         int err;
1527
1528         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1529         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1530         if (ea_inode) {
1531                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1532                 if (err) {
1533                         iput(ea_inode);
1534                         return err;
1535                 }
1536
1537                 *ret_inode = ea_inode;
1538                 return 0;
1539         }
1540
1541         /* Create an inode for the EA value */
1542         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1543         if (IS_ERR(ea_inode))
1544                 return PTR_ERR(ea_inode);
1545
1546         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1547         if (err) {
1548                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1549                 iput(ea_inode);
1550                 return err;
1551         }
1552
1553         if (EA_INODE_CACHE(inode))
1554                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1555                                       ea_inode->i_ino, true /* reusable */);
1556
1557         *ret_inode = ea_inode;
1558         return 0;
1559 }
1560
1561 /*
1562  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1563  * feature is enabled.
1564  */
1565 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1566
1567 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1568                                 struct ext4_xattr_search *s,
1569                                 handle_t *handle, struct inode *inode,
1570                                 bool is_block)
1571 {
1572         struct ext4_xattr_entry *last, *next;
1573         struct ext4_xattr_entry *here = s->here;
1574         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1575         int in_inode = i->in_inode;
1576         struct inode *old_ea_inode = NULL;
1577         struct inode *new_ea_inode = NULL;
1578         size_t old_size, new_size;
1579         int ret;
1580
1581         /* Space used by old and new values. */
1582         old_size = (!s->not_found && !here->e_value_inum) ?
1583                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1584         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1585
1586         /*
1587          * Optimization for the simple case when old and new values have the
1588          * same padded sizes. Not applicable if external inodes are involved.
1589          */
1590         if (new_size && new_size == old_size) {
1591                 size_t offs = le16_to_cpu(here->e_value_offs);
1592                 void *val = s->base + offs;
1593
1594                 here->e_value_size = cpu_to_le32(i->value_len);
1595                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1596                         memset(val, 0, new_size);
1597                 } else {
1598                         memcpy(val, i->value, i->value_len);
1599                         /* Clear padding bytes. */
1600                         memset(val + i->value_len, 0, new_size - i->value_len);
1601                 }
1602                 goto update_hash;
1603         }
1604
1605         /* Compute min_offs and last. */
1606         last = s->first;
1607         for (; !IS_LAST_ENTRY(last); last = next) {
1608                 next = EXT4_XATTR_NEXT(last);
1609                 if ((void *)next >= s->end) {
1610                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1611                         ret = -EFSCORRUPTED;
1612                         goto out;
1613                 }
1614                 if (!last->e_value_inum && last->e_value_size) {
1615                         size_t offs = le16_to_cpu(last->e_value_offs);
1616                         if (offs < min_offs)
1617                                 min_offs = offs;
1618                 }
1619         }
1620
1621         /* Check whether we have enough space. */
1622         if (i->value) {
1623                 size_t free;
1624
1625                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1626                 if (!s->not_found)
1627                         free += EXT4_XATTR_LEN(name_len) + old_size;
1628
1629                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1630                         ret = -ENOSPC;
1631                         goto out;
1632                 }
1633
1634                 /*
1635                  * If storing the value in an external inode is an option,
1636                  * reserve space for xattr entries/names in the external
1637                  * attribute block so that a long value does not occupy the
1638                  * whole space and prevent futher entries being added.
1639                  */
1640                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1641                     new_size && is_block &&
1642                     (min_offs + old_size - new_size) <
1643                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1644                         ret = -ENOSPC;
1645                         goto out;
1646                 }
1647         }
1648
1649         /*
1650          * Getting access to old and new ea inodes is subject to failures.
1651          * Finish that work before doing any modifications to the xattr data.
1652          */
1653         if (!s->not_found && here->e_value_inum) {
1654                 ret = ext4_xattr_inode_iget(inode,
1655                                             le32_to_cpu(here->e_value_inum),
1656                                             le32_to_cpu(here->e_hash),
1657                                             &old_ea_inode);
1658                 if (ret) {
1659                         old_ea_inode = NULL;
1660                         goto out;
1661                 }
1662         }
1663         if (i->value && in_inode) {
1664                 WARN_ON_ONCE(!i->value_len);
1665
1666                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1667                 if (ret)
1668                         goto out;
1669
1670                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1671                                                      i->value_len,
1672                                                      &new_ea_inode);
1673                 if (ret) {
1674                         new_ea_inode = NULL;
1675                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1676                         goto out;
1677                 }
1678         }
1679
1680         if (old_ea_inode) {
1681                 /* We are ready to release ref count on the old_ea_inode. */
1682                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1683                 if (ret) {
1684                         /* Release newly required ref count on new_ea_inode. */
1685                         if (new_ea_inode) {
1686                                 int err;
1687
1688                                 err = ext4_xattr_inode_dec_ref(handle,
1689                                                                new_ea_inode);
1690                                 if (err)
1691                                         ext4_warning_inode(new_ea_inode,
1692                                                   "dec ref new_ea_inode err=%d",
1693                                                   err);
1694                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1695                                                             i->value_len);
1696                         }
1697                         goto out;
1698                 }
1699
1700                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1701                                             le32_to_cpu(here->e_value_size));
1702         }
1703
1704         /* No failures allowed past this point. */
1705
1706         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1707                 /* Remove the old value. */
1708                 void *first_val = s->base + min_offs;
1709                 size_t offs = le16_to_cpu(here->e_value_offs);
1710                 void *val = s->base + offs;
1711
1712                 memmove(first_val + old_size, first_val, val - first_val);
1713                 memset(first_val, 0, old_size);
1714                 min_offs += old_size;
1715
1716                 /* Adjust all value offsets. */
1717                 last = s->first;
1718                 while (!IS_LAST_ENTRY(last)) {
1719                         size_t o = le16_to_cpu(last->e_value_offs);
1720
1721                         if (!last->e_value_inum &&
1722                             last->e_value_size && o < offs)
1723                                 last->e_value_offs = cpu_to_le16(o + old_size);
1724                         last = EXT4_XATTR_NEXT(last);
1725                 }
1726         }
1727
1728         if (!i->value) {
1729                 /* Remove old name. */
1730                 size_t size = EXT4_XATTR_LEN(name_len);
1731
1732                 last = ENTRY((void *)last - size);
1733                 memmove(here, (void *)here + size,
1734                         (void *)last - (void *)here + sizeof(__u32));
1735                 memset(last, 0, size);
1736         } else if (s->not_found) {
1737                 /* Insert new name. */
1738                 size_t size = EXT4_XATTR_LEN(name_len);
1739                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1740
1741                 memmove((void *)here + size, here, rest);
1742                 memset(here, 0, size);
1743                 here->e_name_index = i->name_index;
1744                 here->e_name_len = name_len;
1745                 memcpy(here->e_name, i->name, name_len);
1746         } else {
1747                 /* This is an update, reset value info. */
1748                 here->e_value_inum = 0;
1749                 here->e_value_offs = 0;
1750                 here->e_value_size = 0;
1751         }
1752
1753         if (i->value) {
1754                 /* Insert new value. */
1755                 if (in_inode) {
1756                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1757                 } else if (i->value_len) {
1758                         void *val = s->base + min_offs - new_size;
1759
1760                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1761                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1762                                 memset(val, 0, new_size);
1763                         } else {
1764                                 memcpy(val, i->value, i->value_len);
1765                                 /* Clear padding bytes. */
1766                                 memset(val + i->value_len, 0,
1767                                        new_size - i->value_len);
1768                         }
1769                 }
1770                 here->e_value_size = cpu_to_le32(i->value_len);
1771         }
1772
1773 update_hash:
1774         if (i->value) {
1775                 __le32 hash = 0;
1776
1777                 /* Entry hash calculation. */
1778                 if (in_inode) {
1779                         __le32 crc32c_hash;
1780
1781                         /*
1782                          * Feed crc32c hash instead of the raw value for entry
1783                          * hash calculation. This is to avoid walking
1784                          * potentially long value buffer again.
1785                          */
1786                         crc32c_hash = cpu_to_le32(
1787                                        ext4_xattr_inode_get_hash(new_ea_inode));
1788                         hash = ext4_xattr_hash_entry(here->e_name,
1789                                                      here->e_name_len,
1790                                                      &crc32c_hash, 1);
1791                 } else if (is_block) {
1792                         __le32 *value = s->base + le16_to_cpu(
1793                                                         here->e_value_offs);
1794
1795                         hash = ext4_xattr_hash_entry(here->e_name,
1796                                                      here->e_name_len, value,
1797                                                      new_size >> 2);
1798                 }
1799                 here->e_hash = hash;
1800         }
1801
1802         if (is_block)
1803                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1804
1805         ret = 0;
1806 out:
1807         iput(old_ea_inode);
1808         iput(new_ea_inode);
1809         return ret;
1810 }
1811
1812 struct ext4_xattr_block_find {
1813         struct ext4_xattr_search s;
1814         struct buffer_head *bh;
1815 };
1816
1817 static int
1818 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1819                       struct ext4_xattr_block_find *bs)
1820 {
1821         struct super_block *sb = inode->i_sb;
1822         int error;
1823
1824         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1825                   i->name_index, i->name, i->value, (long)i->value_len);
1826
1827         if (EXT4_I(inode)->i_file_acl) {
1828                 /* The inode already has an extended attribute block. */
1829                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1830                 if (IS_ERR(bs->bh)) {
1831                         error = PTR_ERR(bs->bh);
1832                         bs->bh = NULL;
1833                         return error;
1834                 }
1835                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1836                         atomic_read(&(bs->bh->b_count)),
1837                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1838                 error = ext4_xattr_check_block(inode, bs->bh);
1839                 if (error)
1840                         return error;
1841                 /* Find the named attribute. */
1842                 bs->s.base = BHDR(bs->bh);
1843                 bs->s.first = BFIRST(bs->bh);
1844                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1845                 bs->s.here = bs->s.first;
1846                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1847                                          i->name_index, i->name, 1);
1848                 if (error && error != -ENODATA)
1849                         return error;
1850                 bs->s.not_found = error;
1851         }
1852         return 0;
1853 }
1854
1855 static int
1856 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1857                      struct ext4_xattr_info *i,
1858                      struct ext4_xattr_block_find *bs)
1859 {
1860         struct super_block *sb = inode->i_sb;
1861         struct buffer_head *new_bh = NULL;
1862         struct ext4_xattr_search s_copy = bs->s;
1863         struct ext4_xattr_search *s = &s_copy;
1864         struct mb_cache_entry *ce = NULL;
1865         int error = 0;
1866         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1867         struct inode *ea_inode = NULL, *tmp_inode;
1868         size_t old_ea_inode_quota = 0;
1869         unsigned int ea_ino;
1870
1871
1872 #define header(x) ((struct ext4_xattr_header *)(x))
1873
1874         if (s->base) {
1875                 BUFFER_TRACE(bs->bh, "get_write_access");
1876                 error = ext4_journal_get_write_access(handle, bs->bh);
1877                 if (error)
1878                         goto cleanup;
1879                 lock_buffer(bs->bh);
1880
1881                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1882                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1883
1884                         /*
1885                          * This must happen under buffer lock for
1886                          * ext4_xattr_block_set() to reliably detect modified
1887                          * block
1888                          */
1889                         if (ea_block_cache)
1890                                 mb_cache_entry_delete(ea_block_cache, hash,
1891                                                       bs->bh->b_blocknr);
1892                         ea_bdebug(bs->bh, "modifying in-place");
1893                         error = ext4_xattr_set_entry(i, s, handle, inode,
1894                                                      true /* is_block */);
1895                         ext4_xattr_block_csum_set(inode, bs->bh);
1896                         unlock_buffer(bs->bh);
1897                         if (error == -EFSCORRUPTED)
1898                                 goto bad_block;
1899                         if (!error)
1900                                 error = ext4_handle_dirty_metadata(handle,
1901                                                                    inode,
1902                                                                    bs->bh);
1903                         if (error)
1904                                 goto cleanup;
1905                         goto inserted;
1906                 } else {
1907                         int offset = (char *)s->here - bs->bh->b_data;
1908
1909                         unlock_buffer(bs->bh);
1910                         ea_bdebug(bs->bh, "cloning");
1911                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1912                         error = -ENOMEM;
1913                         if (s->base == NULL)
1914                                 goto cleanup;
1915                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1916                         s->first = ENTRY(header(s->base)+1);
1917                         header(s->base)->h_refcount = cpu_to_le32(1);
1918                         s->here = ENTRY(s->base + offset);
1919                         s->end = s->base + bs->bh->b_size;
1920
1921                         /*
1922                          * If existing entry points to an xattr inode, we need
1923                          * to prevent ext4_xattr_set_entry() from decrementing
1924                          * ref count on it because the reference belongs to the
1925                          * original block. In this case, make the entry look
1926                          * like it has an empty value.
1927                          */
1928                         if (!s->not_found && s->here->e_value_inum) {
1929                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1930                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1931                                               le32_to_cpu(s->here->e_hash),
1932                                               &tmp_inode);
1933                                 if (error)
1934                                         goto cleanup;
1935
1936                                 if (!ext4_test_inode_state(tmp_inode,
1937                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1938                                         /*
1939                                          * Defer quota free call for previous
1940                                          * inode until success is guaranteed.
1941                                          */
1942                                         old_ea_inode_quota = le32_to_cpu(
1943                                                         s->here->e_value_size);
1944                                 }
1945                                 iput(tmp_inode);
1946
1947                                 s->here->e_value_inum = 0;
1948                                 s->here->e_value_size = 0;
1949                         }
1950                 }
1951         } else {
1952                 /* Allocate a buffer where we construct the new block. */
1953                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1954                 /* assert(header == s->base) */
1955                 error = -ENOMEM;
1956                 if (s->base == NULL)
1957                         goto cleanup;
1958                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1959                 header(s->base)->h_blocks = cpu_to_le32(1);
1960                 header(s->base)->h_refcount = cpu_to_le32(1);
1961                 s->first = ENTRY(header(s->base)+1);
1962                 s->here = ENTRY(header(s->base)+1);
1963                 s->end = s->base + sb->s_blocksize;
1964         }
1965
1966         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1967         if (error == -EFSCORRUPTED)
1968                 goto bad_block;
1969         if (error)
1970                 goto cleanup;
1971
1972         if (i->value && s->here->e_value_inum) {
1973                 /*
1974                  * A ref count on ea_inode has been taken as part of the call to
1975                  * ext4_xattr_set_entry() above. We would like to drop this
1976                  * extra ref but we have to wait until the xattr block is
1977                  * initialized and has its own ref count on the ea_inode.
1978                  */
1979                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1980                 error = ext4_xattr_inode_iget(inode, ea_ino,
1981                                               le32_to_cpu(s->here->e_hash),
1982                                               &ea_inode);
1983                 if (error) {
1984                         ea_inode = NULL;
1985                         goto cleanup;
1986                 }
1987         }
1988
1989 inserted:
1990         if (!IS_LAST_ENTRY(s->first)) {
1991                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1992                                                      &ce);
1993                 if (new_bh) {
1994                         /* We found an identical block in the cache. */
1995                         if (new_bh == bs->bh)
1996                                 ea_bdebug(new_bh, "keeping");
1997                         else {
1998                                 u32 ref;
1999
2000                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2001
2002                                 /* The old block is released after updating
2003                                    the inode. */
2004                                 error = dquot_alloc_block(inode,
2005                                                 EXT4_C2B(EXT4_SB(sb), 1));
2006                                 if (error)
2007                                         goto cleanup;
2008                                 BUFFER_TRACE(new_bh, "get_write_access");
2009                                 error = ext4_journal_get_write_access(handle,
2010                                                                       new_bh);
2011                                 if (error)
2012                                         goto cleanup_dquot;
2013                                 lock_buffer(new_bh);
2014                                 /*
2015                                  * We have to be careful about races with
2016                                  * freeing, rehashing or adding references to
2017                                  * xattr block. Once we hold buffer lock xattr
2018                                  * block's state is stable so we can check
2019                                  * whether the block got freed / rehashed or
2020                                  * not.  Since we unhash mbcache entry under
2021                                  * buffer lock when freeing / rehashing xattr
2022                                  * block, checking whether entry is still
2023                                  * hashed is reliable. Same rules hold for
2024                                  * e_reusable handling.
2025                                  */
2026                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2027                                     !ce->e_reusable) {
2028                                         /*
2029                                          * Undo everything and check mbcache
2030                                          * again.
2031                                          */
2032                                         unlock_buffer(new_bh);
2033                                         dquot_free_block(inode,
2034                                                          EXT4_C2B(EXT4_SB(sb),
2035                                                                   1));
2036                                         brelse(new_bh);
2037                                         mb_cache_entry_put(ea_block_cache, ce);
2038                                         ce = NULL;
2039                                         new_bh = NULL;
2040                                         goto inserted;
2041                                 }
2042                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2043                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2044                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2045                                         ce->e_reusable = 0;
2046                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2047                                           ref);
2048                                 ext4_xattr_block_csum_set(inode, new_bh);
2049                                 unlock_buffer(new_bh);
2050                                 error = ext4_handle_dirty_metadata(handle,
2051                                                                    inode,
2052                                                                    new_bh);
2053                                 if (error)
2054                                         goto cleanup_dquot;
2055                         }
2056                         mb_cache_entry_touch(ea_block_cache, ce);
2057                         mb_cache_entry_put(ea_block_cache, ce);
2058                         ce = NULL;
2059                 } else if (bs->bh && s->base == bs->bh->b_data) {
2060                         /* We were modifying this block in-place. */
2061                         ea_bdebug(bs->bh, "keeping this block");
2062                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2063                         new_bh = bs->bh;
2064                         get_bh(new_bh);
2065                 } else {
2066                         /* We need to allocate a new block */
2067                         ext4_fsblk_t goal, block;
2068
2069                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2070
2071                         goal = ext4_group_first_block_no(sb,
2072                                                 EXT4_I(inode)->i_block_group);
2073
2074                         /* non-extent files can't have physical blocks past 2^32 */
2075                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2076                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2077
2078                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2079                                                      NULL, &error);
2080                         if (error)
2081                                 goto cleanup;
2082
2083                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2084                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2085
2086                         ea_idebug(inode, "creating block %llu",
2087                                   (unsigned long long)block);
2088
2089                         new_bh = sb_getblk(sb, block);
2090                         if (unlikely(!new_bh)) {
2091                                 error = -ENOMEM;
2092 getblk_failed:
2093                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2094                                                  EXT4_FREE_BLOCKS_METADATA);
2095                                 goto cleanup;
2096                         }
2097                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2098                                                       ENTRY(header(s->base)+1));
2099                         if (error)
2100                                 goto getblk_failed;
2101                         if (ea_inode) {
2102                                 /* Drop the extra ref on ea_inode. */
2103                                 error = ext4_xattr_inode_dec_ref(handle,
2104                                                                  ea_inode);
2105                                 if (error)
2106                                         ext4_warning_inode(ea_inode,
2107                                                            "dec ref error=%d",
2108                                                            error);
2109                                 iput(ea_inode);
2110                                 ea_inode = NULL;
2111                         }
2112
2113                         lock_buffer(new_bh);
2114                         error = ext4_journal_get_create_access(handle, new_bh);
2115                         if (error) {
2116                                 unlock_buffer(new_bh);
2117                                 error = -EIO;
2118                                 goto getblk_failed;
2119                         }
2120                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2121                         ext4_xattr_block_csum_set(inode, new_bh);
2122                         set_buffer_uptodate(new_bh);
2123                         unlock_buffer(new_bh);
2124                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2125                         error = ext4_handle_dirty_metadata(handle, inode,
2126                                                            new_bh);
2127                         if (error)
2128                                 goto cleanup;
2129                 }
2130         }
2131
2132         if (old_ea_inode_quota)
2133                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2134
2135         /* Update the inode. */
2136         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2137
2138         /* Drop the previous xattr block. */
2139         if (bs->bh && bs->bh != new_bh) {
2140                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2141
2142                 ext4_xattr_release_block(handle, inode, bs->bh,
2143                                          &ea_inode_array,
2144                                          0 /* extra_credits */);
2145                 ext4_xattr_inode_array_free(ea_inode_array);
2146         }
2147         error = 0;
2148
2149 cleanup:
2150         if (ea_inode) {
2151                 int error2;
2152
2153                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2154                 if (error2)
2155                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2156                                            error2);
2157
2158                 /* If there was an error, revert the quota charge. */
2159                 if (error)
2160                         ext4_xattr_inode_free_quota(inode, ea_inode,
2161                                                     i_size_read(ea_inode));
2162                 iput(ea_inode);
2163         }
2164         if (ce)
2165                 mb_cache_entry_put(ea_block_cache, ce);
2166         brelse(new_bh);
2167         if (!(bs->bh && s->base == bs->bh->b_data))
2168                 kfree(s->base);
2169
2170         return error;
2171
2172 cleanup_dquot:
2173         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2174         goto cleanup;
2175
2176 bad_block:
2177         EXT4_ERROR_INODE(inode, "bad block %llu",
2178                          EXT4_I(inode)->i_file_acl);
2179         goto cleanup;
2180
2181 #undef header
2182 }
2183
2184 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2185                           struct ext4_xattr_ibody_find *is)
2186 {
2187         struct ext4_xattr_ibody_header *header;
2188         struct ext4_inode *raw_inode;
2189         int error;
2190
2191         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2192                 return 0;
2193
2194         raw_inode = ext4_raw_inode(&is->iloc);
2195         header = IHDR(inode, raw_inode);
2196         is->s.base = is->s.first = IFIRST(header);
2197         is->s.here = is->s.first;
2198         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2199         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2200                 error = xattr_check_inode(inode, header, is->s.end);
2201                 if (error)
2202                         return error;
2203                 /* Find the named attribute. */
2204                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2205                                          i->name_index, i->name, 0);
2206                 if (error && error != -ENODATA)
2207                         return error;
2208                 is->s.not_found = error;
2209         }
2210         return 0;
2211 }
2212
2213 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2214                                 struct ext4_xattr_info *i,
2215                                 struct ext4_xattr_ibody_find *is)
2216 {
2217         struct ext4_xattr_ibody_header *header;
2218         struct ext4_xattr_search *s = &is->s;
2219         int error;
2220
2221         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2222                 return -ENOSPC;
2223
2224         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2225         if (error)
2226                 return error;
2227         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2228         if (!IS_LAST_ENTRY(s->first)) {
2229                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2230                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2231         } else {
2232                 header->h_magic = cpu_to_le32(0);
2233                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2234         }
2235         return 0;
2236 }
2237
2238 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2239                                 struct ext4_xattr_info *i,
2240                                 struct ext4_xattr_ibody_find *is)
2241 {
2242         struct ext4_xattr_ibody_header *header;
2243         struct ext4_xattr_search *s = &is->s;
2244         int error;
2245
2246         if (EXT4_I(inode)->i_extra_isize == 0)
2247                 return -ENOSPC;
2248         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2249         if (error)
2250                 return error;
2251         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2252         if (!IS_LAST_ENTRY(s->first)) {
2253                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2254                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2255         } else {
2256                 header->h_magic = cpu_to_le32(0);
2257                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2258         }
2259         return 0;
2260 }
2261
2262 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2263                                  struct ext4_xattr_info *i)
2264 {
2265         void *value;
2266
2267         /* When e_value_inum is set the value is stored externally. */
2268         if (s->here->e_value_inum)
2269                 return 0;
2270         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2271                 return 0;
2272         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2273         return !memcmp(value, i->value, i->value_len);
2274 }
2275
2276 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2277 {
2278         struct buffer_head *bh;
2279         int error;
2280
2281         if (!EXT4_I(inode)->i_file_acl)
2282                 return NULL;
2283         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2284         if (IS_ERR(bh))
2285                 return bh;
2286         error = ext4_xattr_check_block(inode, bh);
2287         if (error) {
2288                 brelse(bh);
2289                 return ERR_PTR(error);
2290         }
2291         return bh;
2292 }
2293
2294 /*
2295  * ext4_xattr_set_handle()
2296  *
2297  * Create, replace or remove an extended attribute for this inode.  Value
2298  * is NULL to remove an existing extended attribute, and non-NULL to
2299  * either replace an existing extended attribute, or create a new extended
2300  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2301  * specify that an extended attribute must exist and must not exist
2302  * previous to the call, respectively.
2303  *
2304  * Returns 0, or a negative error number on failure.
2305  */
2306 int
2307 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2308                       const char *name, const void *value, size_t value_len,
2309                       int flags)
2310 {
2311         struct ext4_xattr_info i = {
2312                 .name_index = name_index,
2313                 .name = name,
2314                 .value = value,
2315                 .value_len = value_len,
2316                 .in_inode = 0,
2317         };
2318         struct ext4_xattr_ibody_find is = {
2319                 .s = { .not_found = -ENODATA, },
2320         };
2321         struct ext4_xattr_block_find bs = {
2322                 .s = { .not_found = -ENODATA, },
2323         };
2324         int no_expand;
2325         int error;
2326
2327         if (!name)
2328                 return -EINVAL;
2329         if (strlen(name) > 255)
2330                 return -ERANGE;
2331
2332         ext4_write_lock_xattr(inode, &no_expand);
2333
2334         /* Check journal credits under write lock. */
2335         if (ext4_handle_valid(handle)) {
2336                 struct buffer_head *bh;
2337                 int credits;
2338
2339                 bh = ext4_xattr_get_block(inode);
2340                 if (IS_ERR(bh)) {
2341                         error = PTR_ERR(bh);
2342                         goto cleanup;
2343                 }
2344
2345                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2346                                                    value_len,
2347                                                    flags & XATTR_CREATE);
2348                 brelse(bh);
2349
2350                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2351                         error = -ENOSPC;
2352                         goto cleanup;
2353                 }
2354                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2355         }
2356
2357         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2358         if (error)
2359                 goto cleanup;
2360
2361         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2362                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2363                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2364                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2365         }
2366
2367         error = ext4_xattr_ibody_find(inode, &i, &is);
2368         if (error)
2369                 goto cleanup;
2370         if (is.s.not_found)
2371                 error = ext4_xattr_block_find(inode, &i, &bs);
2372         if (error)
2373                 goto cleanup;
2374         if (is.s.not_found && bs.s.not_found) {
2375                 error = -ENODATA;
2376                 if (flags & XATTR_REPLACE)
2377                         goto cleanup;
2378                 error = 0;
2379                 if (!value)
2380                         goto cleanup;
2381         } else {
2382                 error = -EEXIST;
2383                 if (flags & XATTR_CREATE)
2384                         goto cleanup;
2385         }
2386
2387         if (!value) {
2388                 if (!is.s.not_found)
2389                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2390                 else if (!bs.s.not_found)
2391                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2392         } else {
2393                 error = 0;
2394                 /* Xattr value did not change? Save us some work and bail out */
2395                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2396                         goto cleanup;
2397                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2398                         goto cleanup;
2399
2400                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2401                     (EXT4_XATTR_SIZE(i.value_len) >
2402                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2403                         i.in_inode = 1;
2404 retry_inode:
2405                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2406                 if (!error && !bs.s.not_found) {
2407                         i.value = NULL;
2408                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2409                 } else if (error == -ENOSPC) {
2410                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2411                                 brelse(bs.bh);
2412                                 bs.bh = NULL;
2413                                 error = ext4_xattr_block_find(inode, &i, &bs);
2414                                 if (error)
2415                                         goto cleanup;
2416                         }
2417                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2418                         if (!error && !is.s.not_found) {
2419                                 i.value = NULL;
2420                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2421                                                              &is);
2422                         } else if (error == -ENOSPC) {
2423                                 /*
2424                                  * Xattr does not fit in the block, store at
2425                                  * external inode if possible.
2426                                  */
2427                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2428                                     i.value_len && !i.in_inode) {
2429                                         i.in_inode = 1;
2430                                         goto retry_inode;
2431                                 }
2432                         }
2433                 }
2434         }
2435         if (!error) {
2436                 ext4_xattr_update_super_block(handle, inode->i_sb);
2437                 inode->i_ctime = current_time(inode);
2438                 if (!value)
2439                         no_expand = 0;
2440                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2441                 /*
2442                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2443                  * error != 0.
2444                  */
2445                 is.iloc.bh = NULL;
2446                 if (IS_SYNC(inode))
2447                         ext4_handle_sync(handle);
2448         }
2449
2450 cleanup:
2451         brelse(is.iloc.bh);
2452         brelse(bs.bh);
2453         ext4_write_unlock_xattr(inode, &no_expand);
2454         return error;
2455 }
2456
2457 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2458                            bool is_create, int *credits)
2459 {
2460         struct buffer_head *bh;
2461         int err;
2462
2463         *credits = 0;
2464
2465         if (!EXT4_SB(inode->i_sb)->s_journal)
2466                 return 0;
2467
2468         down_read(&EXT4_I(inode)->xattr_sem);
2469
2470         bh = ext4_xattr_get_block(inode);
2471         if (IS_ERR(bh)) {
2472                 err = PTR_ERR(bh);
2473         } else {
2474                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2475                                                     value_len, is_create);
2476                 brelse(bh);
2477                 err = 0;
2478         }
2479
2480         up_read(&EXT4_I(inode)->xattr_sem);
2481         return err;
2482 }
2483
2484 /*
2485  * ext4_xattr_set()
2486  *
2487  * Like ext4_xattr_set_handle, but start from an inode. This extended
2488  * attribute modification is a filesystem transaction by itself.
2489  *
2490  * Returns 0, or a negative error number on failure.
2491  */
2492 int
2493 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2494                const void *value, size_t value_len, int flags)
2495 {
2496         handle_t *handle;
2497         struct super_block *sb = inode->i_sb;
2498         int error, retries = 0;
2499         int credits;
2500
2501         error = dquot_initialize(inode);
2502         if (error)
2503                 return error;
2504
2505 retry:
2506         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2507                                        &credits);
2508         if (error)
2509                 return error;
2510
2511         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2512         if (IS_ERR(handle)) {
2513                 error = PTR_ERR(handle);
2514         } else {
2515                 int error2;
2516
2517                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2518                                               value, value_len, flags);
2519                 error2 = ext4_journal_stop(handle);
2520                 if (error == -ENOSPC &&
2521                     ext4_should_retry_alloc(sb, &retries))
2522                         goto retry;
2523                 if (error == 0)
2524                         error = error2;
2525         }
2526
2527         return error;
2528 }
2529
2530 /*
2531  * Shift the EA entries in the inode to create space for the increased
2532  * i_extra_isize.
2533  */
2534 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2535                                      int value_offs_shift, void *to,
2536                                      void *from, size_t n)
2537 {
2538         struct ext4_xattr_entry *last = entry;
2539         int new_offs;
2540
2541         /* We always shift xattr headers further thus offsets get lower */
2542         BUG_ON(value_offs_shift > 0);
2543
2544         /* Adjust the value offsets of the entries */
2545         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2546                 if (!last->e_value_inum && last->e_value_size) {
2547                         new_offs = le16_to_cpu(last->e_value_offs) +
2548                                                         value_offs_shift;
2549                         last->e_value_offs = cpu_to_le16(new_offs);
2550                 }
2551         }
2552         /* Shift the entries by n bytes */
2553         memmove(to, from, n);
2554 }
2555
2556 /*
2557  * Move xattr pointed to by 'entry' from inode into external xattr block
2558  */
2559 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2560                                     struct ext4_inode *raw_inode,
2561                                     struct ext4_xattr_entry *entry)
2562 {
2563         struct ext4_xattr_ibody_find *is = NULL;
2564         struct ext4_xattr_block_find *bs = NULL;
2565         char *buffer = NULL, *b_entry_name = NULL;
2566         size_t value_size = le32_to_cpu(entry->e_value_size);
2567         struct ext4_xattr_info i = {
2568                 .value = NULL,
2569                 .value_len = 0,
2570                 .name_index = entry->e_name_index,
2571                 .in_inode = !!entry->e_value_inum,
2572         };
2573         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2574         int error;
2575
2576         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2577         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2578         buffer = kmalloc(value_size, GFP_NOFS);
2579         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2580         if (!is || !bs || !buffer || !b_entry_name) {
2581                 error = -ENOMEM;
2582                 goto out;
2583         }
2584
2585         is->s.not_found = -ENODATA;
2586         bs->s.not_found = -ENODATA;
2587         is->iloc.bh = NULL;
2588         bs->bh = NULL;
2589
2590         /* Save the entry name and the entry value */
2591         if (entry->e_value_inum) {
2592                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2593                 if (error)
2594                         goto out;
2595         } else {
2596                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2597                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2598         }
2599
2600         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2601         b_entry_name[entry->e_name_len] = '\0';
2602         i.name = b_entry_name;
2603
2604         error = ext4_get_inode_loc(inode, &is->iloc);
2605         if (error)
2606                 goto out;
2607
2608         error = ext4_xattr_ibody_find(inode, &i, is);
2609         if (error)
2610                 goto out;
2611
2612         /* Remove the chosen entry from the inode */
2613         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2614         if (error)
2615                 goto out;
2616
2617         i.value = buffer;
2618         i.value_len = value_size;
2619         error = ext4_xattr_block_find(inode, &i, bs);
2620         if (error)
2621                 goto out;
2622
2623         /* Add entry which was removed from the inode into the block */
2624         error = ext4_xattr_block_set(handle, inode, &i, bs);
2625         if (error)
2626                 goto out;
2627         error = 0;
2628 out:
2629         kfree(b_entry_name);
2630         kfree(buffer);
2631         if (is)
2632                 brelse(is->iloc.bh);
2633         if (bs)
2634                 brelse(bs->bh);
2635         kfree(is);
2636         kfree(bs);
2637
2638         return error;
2639 }
2640
2641 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2642                                        struct ext4_inode *raw_inode,
2643                                        int isize_diff, size_t ifree,
2644                                        size_t bfree, int *total_ino)
2645 {
2646         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2647         struct ext4_xattr_entry *small_entry;
2648         struct ext4_xattr_entry *entry;
2649         struct ext4_xattr_entry *last;
2650         unsigned int entry_size;        /* EA entry size */
2651         unsigned int total_size;        /* EA entry size + value size */
2652         unsigned int min_total_size;
2653         int error;
2654
2655         while (isize_diff > ifree) {
2656                 entry = NULL;
2657                 small_entry = NULL;
2658                 min_total_size = ~0U;
2659                 last = IFIRST(header);
2660                 /* Find the entry best suited to be pushed into EA block */
2661                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2662                         /* never move system.data out of the inode */
2663                         if ((last->e_name_len == 4) &&
2664                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2665                             !memcmp(last->e_name, "data", 4))
2666                                 continue;
2667                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2668                         if (!last->e_value_inum)
2669                                 total_size += EXT4_XATTR_SIZE(
2670                                                le32_to_cpu(last->e_value_size));
2671                         if (total_size <= bfree &&
2672                             total_size < min_total_size) {
2673                                 if (total_size + ifree < isize_diff) {
2674                                         small_entry = last;
2675                                 } else {
2676                                         entry = last;
2677                                         min_total_size = total_size;
2678                                 }
2679                         }
2680                 }
2681
2682                 if (entry == NULL) {
2683                         if (small_entry == NULL)
2684                                 return -ENOSPC;
2685                         entry = small_entry;
2686                 }
2687
2688                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2689                 total_size = entry_size;
2690                 if (!entry->e_value_inum)
2691                         total_size += EXT4_XATTR_SIZE(
2692                                               le32_to_cpu(entry->e_value_size));
2693                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2694                                                  entry);
2695                 if (error)
2696                         return error;
2697
2698                 *total_ino -= entry_size;
2699                 ifree += total_size;
2700                 bfree -= total_size;
2701         }
2702
2703         return 0;
2704 }
2705
2706 /*
2707  * Expand an inode by new_extra_isize bytes when EAs are present.
2708  * Returns 0 on success or negative error number on failure.
2709  */
2710 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2711                                struct ext4_inode *raw_inode, handle_t *handle)
2712 {
2713         struct ext4_xattr_ibody_header *header;
2714         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2715         static unsigned int mnt_count;
2716         size_t min_offs;
2717         size_t ifree, bfree;
2718         int total_ino;
2719         void *base, *end;
2720         int error = 0, tried_min_extra_isize = 0;
2721         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2722         int isize_diff; /* How much do we need to grow i_extra_isize */
2723
2724 retry:
2725         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2726         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2727                 return 0;
2728
2729         header = IHDR(inode, raw_inode);
2730
2731         /*
2732          * Check if enough free space is available in the inode to shift the
2733          * entries ahead by new_extra_isize.
2734          */
2735
2736         base = IFIRST(header);
2737         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2738         min_offs = end - base;
2739         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2740
2741         error = xattr_check_inode(inode, header, end);
2742         if (error)
2743                 goto cleanup;
2744
2745         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2746         if (ifree >= isize_diff)
2747                 goto shift;
2748
2749         /*
2750          * Enough free space isn't available in the inode, check if
2751          * EA block can hold new_extra_isize bytes.
2752          */
2753         if (EXT4_I(inode)->i_file_acl) {
2754                 struct buffer_head *bh;
2755
2756                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2757                 if (IS_ERR(bh)) {
2758                         error = PTR_ERR(bh);
2759                         goto cleanup;
2760                 }
2761                 error = ext4_xattr_check_block(inode, bh);
2762                 if (error) {
2763                         brelse(bh);
2764                         goto cleanup;
2765                 }
2766                 base = BHDR(bh);
2767                 end = bh->b_data + bh->b_size;
2768                 min_offs = end - base;
2769                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2770                                               NULL);
2771                 brelse(bh);
2772                 if (bfree + ifree < isize_diff) {
2773                         if (!tried_min_extra_isize && s_min_extra_isize) {
2774                                 tried_min_extra_isize++;
2775                                 new_extra_isize = s_min_extra_isize;
2776                                 goto retry;
2777                         }
2778                         error = -ENOSPC;
2779                         goto cleanup;
2780                 }
2781         } else {
2782                 bfree = inode->i_sb->s_blocksize;
2783         }
2784
2785         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2786                                             isize_diff, ifree, bfree,
2787                                             &total_ino);
2788         if (error) {
2789                 if (error == -ENOSPC && !tried_min_extra_isize &&
2790                     s_min_extra_isize) {
2791                         tried_min_extra_isize++;
2792                         new_extra_isize = s_min_extra_isize;
2793                         goto retry;
2794                 }
2795                 goto cleanup;
2796         }
2797 shift:
2798         /* Adjust the offsets and shift the remaining entries ahead */
2799         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2800                         - new_extra_isize, (void *)raw_inode +
2801                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2802                         (void *)header, total_ino);
2803         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2804
2805 cleanup:
2806         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2807                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2808                              inode->i_ino);
2809                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2810         }
2811         return error;
2812 }
2813
2814 #define EIA_INCR 16 /* must be 2^n */
2815 #define EIA_MASK (EIA_INCR - 1)
2816
2817 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2818  * If @ea_inode_array is new or full it will be grown and the old
2819  * contents copied over.
2820  */
2821 static int
2822 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2823                         struct inode *inode)
2824 {
2825         if (*ea_inode_array == NULL) {
2826                 /*
2827                  * Start with 15 inodes, so it fits into a power-of-two size.
2828                  * If *ea_inode_array is NULL, this is essentially offsetof()
2829                  */
2830                 (*ea_inode_array) =
2831                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2832                                          inodes[EIA_MASK]),
2833                                 GFP_NOFS);
2834                 if (*ea_inode_array == NULL)
2835                         return -ENOMEM;
2836                 (*ea_inode_array)->count = 0;
2837         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2838                 /* expand the array once all 15 + n * 16 slots are full */
2839                 struct ext4_xattr_inode_array *new_array = NULL;
2840                 int count = (*ea_inode_array)->count;
2841
2842                 /* if new_array is NULL, this is essentially offsetof() */
2843                 new_array = kmalloc(
2844                                 offsetof(struct ext4_xattr_inode_array,
2845                                          inodes[count + EIA_INCR]),
2846                                 GFP_NOFS);
2847                 if (new_array == NULL)
2848                         return -ENOMEM;
2849                 memcpy(new_array, *ea_inode_array,
2850                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2851                 kfree(*ea_inode_array);
2852                 *ea_inode_array = new_array;
2853         }
2854         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2855         return 0;
2856 }
2857
2858 /*
2859  * ext4_xattr_delete_inode()
2860  *
2861  * Free extended attribute resources associated with this inode. Traverse
2862  * all entries and decrement reference on any xattr inodes associated with this
2863  * inode. This is called immediately before an inode is freed. We have exclusive
2864  * access to the inode. If an orphan inode is deleted it will also release its
2865  * references on xattr block and xattr inodes.
2866  */
2867 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2868                             struct ext4_xattr_inode_array **ea_inode_array,
2869                             int extra_credits)
2870 {
2871         struct buffer_head *bh = NULL;
2872         struct ext4_xattr_ibody_header *header;
2873         struct ext4_iloc iloc = { .bh = NULL };
2874         struct ext4_xattr_entry *entry;
2875         struct inode *ea_inode;
2876         int error;
2877
2878         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2879                                           NULL /* bh */,
2880                                           false /* dirty */,
2881                                           false /* block_csum */);
2882         if (error) {
2883                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2884                 goto cleanup;
2885         }
2886
2887         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2888             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2889
2890                 error = ext4_get_inode_loc(inode, &iloc);
2891                 if (error) {
2892                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2893                         goto cleanup;
2894                 }
2895
2896                 error = ext4_journal_get_write_access(handle, iloc.bh);
2897                 if (error) {
2898                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2899                                          error);
2900                         goto cleanup;
2901                 }
2902
2903                 header = IHDR(inode, ext4_raw_inode(&iloc));
2904                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2905                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2906                                                      IFIRST(header),
2907                                                      false /* block_csum */,
2908                                                      ea_inode_array,
2909                                                      extra_credits,
2910                                                      false /* skip_quota */);
2911         }
2912
2913         if (EXT4_I(inode)->i_file_acl) {
2914                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2915                 if (IS_ERR(bh)) {
2916                         error = PTR_ERR(bh);
2917                         if (error == -EIO)
2918                                 EXT4_ERROR_INODE(inode, "block %llu read error",
2919                                                  EXT4_I(inode)->i_file_acl);
2920                         bh = NULL;
2921                         goto cleanup;
2922                 }
2923                 error = ext4_xattr_check_block(inode, bh);
2924                 if (error)
2925                         goto cleanup;
2926
2927                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2928                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2929                              entry = EXT4_XATTR_NEXT(entry)) {
2930                                 if (!entry->e_value_inum)
2931                                         continue;
2932                                 error = ext4_xattr_inode_iget(inode,
2933                                               le32_to_cpu(entry->e_value_inum),
2934                                               le32_to_cpu(entry->e_hash),
2935                                               &ea_inode);
2936                                 if (error)
2937                                         continue;
2938                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2939                                               le32_to_cpu(entry->e_value_size));
2940                                 iput(ea_inode);
2941                         }
2942
2943                 }
2944
2945                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2946                                          extra_credits);
2947                 /*
2948                  * Update i_file_acl value in the same transaction that releases
2949                  * block.
2950                  */
2951                 EXT4_I(inode)->i_file_acl = 0;
2952                 error = ext4_mark_inode_dirty(handle, inode);
2953                 if (error) {
2954                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2955                                          error);
2956                         goto cleanup;
2957                 }
2958         }
2959         error = 0;
2960 cleanup:
2961         brelse(iloc.bh);
2962         brelse(bh);
2963         return error;
2964 }
2965
2966 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2967 {
2968         int idx;
2969
2970         if (ea_inode_array == NULL)
2971                 return;
2972
2973         for (idx = 0; idx < ea_inode_array->count; ++idx)
2974                 iput(ea_inode_array->inodes[idx]);
2975         kfree(ea_inode_array);
2976 }
2977
2978 /*
2979  * ext4_xattr_block_cache_insert()
2980  *
2981  * Create a new entry in the extended attribute block cache, and insert
2982  * it unless such an entry is already in the cache.
2983  *
2984  * Returns 0, or a negative error number on failure.
2985  */
2986 static void
2987 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2988                               struct buffer_head *bh)
2989 {
2990         struct ext4_xattr_header *header = BHDR(bh);
2991         __u32 hash = le32_to_cpu(header->h_hash);
2992         int reusable = le32_to_cpu(header->h_refcount) <
2993                        EXT4_XATTR_REFCOUNT_MAX;
2994         int error;
2995
2996         if (!ea_block_cache)
2997                 return;
2998         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2999                                       bh->b_blocknr, reusable);
3000         if (error) {
3001                 if (error == -EBUSY)
3002                         ea_bdebug(bh, "already in cache");
3003         } else
3004                 ea_bdebug(bh, "inserting [%x]", (int)hash);
3005 }
3006
3007 /*
3008  * ext4_xattr_cmp()
3009  *
3010  * Compare two extended attribute blocks for equality.
3011  *
3012  * Returns 0 if the blocks are equal, 1 if they differ, and
3013  * a negative error number on errors.
3014  */
3015 static int
3016 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3017                struct ext4_xattr_header *header2)
3018 {
3019         struct ext4_xattr_entry *entry1, *entry2;
3020
3021         entry1 = ENTRY(header1+1);
3022         entry2 = ENTRY(header2+1);
3023         while (!IS_LAST_ENTRY(entry1)) {
3024                 if (IS_LAST_ENTRY(entry2))
3025                         return 1;
3026                 if (entry1->e_hash != entry2->e_hash ||
3027                     entry1->e_name_index != entry2->e_name_index ||
3028                     entry1->e_name_len != entry2->e_name_len ||
3029                     entry1->e_value_size != entry2->e_value_size ||
3030                     entry1->e_value_inum != entry2->e_value_inum ||
3031                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3032                         return 1;
3033                 if (!entry1->e_value_inum &&
3034                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3035                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3036                            le32_to_cpu(entry1->e_value_size)))
3037                         return 1;
3038
3039                 entry1 = EXT4_XATTR_NEXT(entry1);
3040                 entry2 = EXT4_XATTR_NEXT(entry2);
3041         }
3042         if (!IS_LAST_ENTRY(entry2))
3043                 return 1;
3044         return 0;
3045 }
3046
3047 /*
3048  * ext4_xattr_block_cache_find()
3049  *
3050  * Find an identical extended attribute block.
3051  *
3052  * Returns a pointer to the block found, or NULL if such a block was
3053  * not found or an error occurred.
3054  */
3055 static struct buffer_head *
3056 ext4_xattr_block_cache_find(struct inode *inode,
3057                             struct ext4_xattr_header *header,
3058                             struct mb_cache_entry **pce)
3059 {
3060         __u32 hash = le32_to_cpu(header->h_hash);
3061         struct mb_cache_entry *ce;
3062         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3063
3064         if (!ea_block_cache)
3065                 return NULL;
3066         if (!header->h_hash)
3067                 return NULL;  /* never share */
3068         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3069         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3070         while (ce) {
3071                 struct buffer_head *bh;
3072
3073                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3074                 if (IS_ERR(bh)) {
3075                         if (PTR_ERR(bh) == -ENOMEM)
3076                                 return NULL;
3077                         bh = NULL;
3078                         EXT4_ERROR_INODE(inode, "block %lu read error",
3079                                          (unsigned long)ce->e_value);
3080                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3081                         *pce = ce;
3082                         return bh;
3083                 }
3084                 brelse(bh);
3085                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3086         }
3087         return NULL;
3088 }
3089
3090 #define NAME_HASH_SHIFT 5
3091 #define VALUE_HASH_SHIFT 16
3092
3093 /*
3094  * ext4_xattr_hash_entry()
3095  *
3096  * Compute the hash of an extended attribute.
3097  */
3098 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3099                                     size_t value_count)
3100 {
3101         __u32 hash = 0;
3102
3103         while (name_len--) {
3104                 hash = (hash << NAME_HASH_SHIFT) ^
3105                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3106                        *name++;
3107         }
3108         while (value_count--) {
3109                 hash = (hash << VALUE_HASH_SHIFT) ^
3110                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3111                        le32_to_cpu(*value++);
3112         }
3113         return cpu_to_le32(hash);
3114 }
3115
3116 #undef NAME_HASH_SHIFT
3117 #undef VALUE_HASH_SHIFT
3118
3119 #define BLOCK_HASH_SHIFT 16
3120
3121 /*
3122  * ext4_xattr_rehash()
3123  *
3124  * Re-compute the extended attribute hash value after an entry has changed.
3125  */
3126 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3127 {
3128         struct ext4_xattr_entry *here;
3129         __u32 hash = 0;
3130
3131         here = ENTRY(header+1);
3132         while (!IS_LAST_ENTRY(here)) {
3133                 if (!here->e_hash) {
3134                         /* Block is not shared if an entry's hash value == 0 */
3135                         hash = 0;
3136                         break;
3137                 }
3138                 hash = (hash << BLOCK_HASH_SHIFT) ^
3139                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3140                        le32_to_cpu(here->e_hash);
3141                 here = EXT4_XATTR_NEXT(here);
3142         }
3143         header->h_hash = cpu_to_le32(hash);
3144 }
3145
3146 #undef BLOCK_HASH_SHIFT
3147
3148 #define HASH_BUCKET_BITS        10
3149
3150 struct mb_cache *
3151 ext4_xattr_create_cache(void)
3152 {
3153         return mb_cache_create(HASH_BUCKET_BITS);
3154 }
3155
3156 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3157 {
3158         if (cache)
3159                 mb_cache_destroy(cache);
3160 }
3161