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         [EXT4_XATTR_INDEX_HURD]              = &ext4_xattr_hurd_handler,
  97 };
  98
  99 const struct xattr_handler *ext4_xattr_handlers[] = {
 100         &ext4_xattr_user_handler,
 101         &ext4_xattr_trusted_handler,
 102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 103         &posix_acl_access_xattr_handler,
 104         &posix_acl_default_xattr_handler,
 105 #endif
 106 #ifdef CONFIG_EXT4_FS_SECURITY
 107         &ext4_xattr_security_handler,
 108 #endif
 109         &ext4_xattr_hurd_handler,
 110         NULL
 111 };
 112
 113 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 114                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
 115
 116 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 117                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
 118
 119 static int
 120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 121                         struct inode *inode);
 122
 123 #ifdef CONFIG_LOCKDEP
 124 void ext4_xattr_inode_set_class(struct inode *ea_inode)
 125 {
 126         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 127 }
 128 #endif
 129
 130 static __le32 ext4_xattr_block_csum(struct inode *inode,
 131                                     sector_t block_nr,
 132                                     struct ext4_xattr_header *hdr)
 133 {
 134         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 135         __u32 csum;
 136         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 137         __u32 dummy_csum = 0;
 138         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 139
 140         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 141                            sizeof(dsk_block_nr));
 142         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 143         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 144         offset += sizeof(dummy_csum);
 145         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 146                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 147
 148         return cpu_to_le32(csum);
 149 }
 150
 151 static int ext4_xattr_block_csum_verify(struct inode *inode,
 152                                         struct buffer_head *bh)
 153 {
 154         struct ext4_xattr_header *hdr = BHDR(bh);
 155         int ret = 1;
 156
 157         if (ext4_has_metadata_csum(inode->i_sb)) {
 158                 lock_buffer(bh);
 159                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 160                                                         bh->b_blocknr, hdr));
 161                 unlock_buffer(bh);
 162         }
 163         return ret;
 164 }
 165
 166 static void ext4_xattr_block_csum_set(struct inode *inode,
 167                                       struct buffer_head *bh)
 168 {
 169         if (ext4_has_metadata_csum(inode->i_sb))
 170                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 171                                                 bh->b_blocknr, BHDR(bh));
 172 }
 173
 174 static inline const struct xattr_handler *
 175 ext4_xattr_handler(int name_index)
 176 {
 177         const struct xattr_handler *handler = NULL;
 178
 179         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 180                 handler = ext4_xattr_handler_map[name_index];
 181         return handler;
 182 }
 183
 184 static int
 185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 186                          void *value_start)
 187 {
 188         struct ext4_xattr_entry *e = entry;
 189
 190         /* Find the end of the names list */
 191         while (!IS_LAST_ENTRY(e)) {
 192                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 193                 if ((void *)next >= end)
 194                         return -EFSCORRUPTED;
 195                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 196                         return -EFSCORRUPTED;
 197                 e = next;
 198         }
 199
 200         /* Check the values */
 201         while (!IS_LAST_ENTRY(entry)) {
 202                 u32 size = le32_to_cpu(entry->e_value_size);
 203
 204                 if (size > EXT4_XATTR_SIZE_MAX)
 205                         return -EFSCORRUPTED;
 206
 207                 if (size != 0 && entry->e_value_inum == 0) {
 208                         u16 offs = le16_to_cpu(entry->e_value_offs);
 209                         void *value;
 210
 211                         /*
 212                          * The value cannot overlap the names, and the value
 213                          * with padding cannot extend beyond 'end'.  Check both
 214                          * the padded and unpadded sizes, since the size may
 215                          * overflow to 0 when adding padding.
 216                          */
 217                         if (offs > end - value_start)
 218                                 return -EFSCORRUPTED;
 219                         value = value_start + offs;
 220                         if (value < (void *)e + sizeof(u32) ||
 221                             size > end - value ||
 222                             EXT4_XATTR_SIZE(size) > end - value)
 223                                 return -EFSCORRUPTED;
 224                 }
 225                 entry = EXT4_XATTR_NEXT(entry);
 226         }
 227
 228         return 0;
 229 }
 230
 231 static inline int
 232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 233                          const char *function, unsigned int line)
 234 {
 235         int error = -EFSCORRUPTED;
 236
 237         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 238             BHDR(bh)->h_blocks != cpu_to_le32(1))
 239                 goto errout;
 240         if (buffer_verified(bh))
 241                 return 0;
 242
 243         error = -EFSBADCRC;
 244         if (!ext4_xattr_block_csum_verify(inode, bh))
 245                 goto errout;
 246         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 247                                          bh->b_data);
 248 errout:
 249         if (error)
 250                 __ext4_error_inode(inode, function, line, 0, -error,
 251                                    "corrupted xattr block %llu",
 252                                    (unsigned long long) bh->b_blocknr);
 253         else
 254                 set_buffer_verified(bh);
 255         return error;
 256 }
 257
 258 #define ext4_xattr_check_block(inode, bh) \
 259         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 260
 261
 262 static int
 263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 264                          void *end, const char *function, unsigned int line)
 265 {
 266         int error = -EFSCORRUPTED;
 267
 268         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 269             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 270                 goto errout;
 271         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 272 errout:
 273         if (error)
 274                 __ext4_error_inode(inode, function, line, 0, -error,
 275                                    "corrupted in-inode xattr");
 276         return error;
 277 }
 278
 279 #define xattr_check_inode(inode, header, end) \
 280         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 281
 282 static int
 283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 284                  void *end, int name_index, const char *name, int sorted)
 285 {
 286         struct ext4_xattr_entry *entry, *next;
 287         size_t name_len;
 288         int cmp = 1;
 289
 290         if (name == NULL)
 291                 return -EINVAL;
 292         name_len = strlen(name);
 293         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 294                 next = EXT4_XATTR_NEXT(entry);
 295                 if ((void *) next >= end) {
 296                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 297                         return -EFSCORRUPTED;
 298                 }
 299                 cmp = name_index - entry->e_name_index;
 300                 if (!cmp)
 301                         cmp = name_len - entry->e_name_len;
 302                 if (!cmp)
 303                         cmp = memcmp(name, entry->e_name, name_len);
 304                 if (cmp <= 0 && (sorted || cmp == 0))
 305                         break;
 306         }
 307         *pentry = entry;
 308         return cmp ? -ENODATA : 0;
 309 }
 310
 311 static u32
 312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 313 {
 314         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 315 }
 316
 317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 318 {
 319         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 320                 (u32) inode_peek_iversion_raw(ea_inode);
 321 }
 322
 323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 324 {
 325         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 326         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 327 }
 328
 329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 330 {
 331         return (u32)ea_inode->i_atime.tv_sec;
 332 }
 333
 334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 335 {
 336         ea_inode->i_atime.tv_sec = hash;
 337 }
 338
 339 /*
 340  * Read the EA value from an inode.
 341  */
 342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 343 {
 344         int blocksize = 1 << ea_inode->i_blkbits;
 345         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 346         int tail_size = (size % blocksize) ?: blocksize;
 347         struct buffer_head *bhs_inline[8];
 348         struct buffer_head **bhs = bhs_inline;
 349         int i, ret;
 350
 351         if (bh_count > ARRAY_SIZE(bhs_inline)) {
 352                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 353                 if (!bhs)
 354                         return -ENOMEM;
 355         }
 356
 357         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 358                                true /* wait */, bhs);
 359         if (ret)
 360                 goto free_bhs;
 361
 362         for (i = 0; i < bh_count; i++) {
 363                 /* There shouldn't be any holes in ea_inode. */
 364                 if (!bhs[i]) {
 365                         ret = -EFSCORRUPTED;
 366                         goto put_bhs;
 367                 }
 368                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 369                        i < bh_count - 1 ? blocksize : tail_size);
 370         }
 371         ret = 0;
 372 put_bhs:
 373         for (i = 0; i < bh_count; i++)
 374                 brelse(bhs[i]);
 375 free_bhs:
 376         if (bhs != bhs_inline)
 377                 kfree(bhs);
 378         return ret;
 379 }
 380
 381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 382
 383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 384                                  u32 ea_inode_hash, struct inode **ea_inode)
 385 {
 386         struct inode *inode;
 387         int err;
 388
 389         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 390         if (IS_ERR(inode)) {
 391                 err = PTR_ERR(inode);
 392                 ext4_error(parent->i_sb,
 393                            "error while reading EA inode %lu err=%d", ea_ino,
 394                            err);
 395                 return err;
 396         }
 397
 398         if (is_bad_inode(inode)) {
 399                 ext4_error(parent->i_sb,
 400                            "error while reading EA inode %lu is_bad_inode",
 401                            ea_ino);
 402                 err = -EIO;
 403                 goto error;
 404         }
 405
 406         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 407                 ext4_error(parent->i_sb,
 408                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 409                             ea_ino);
 410                 err = -EINVAL;
 411                 goto error;
 412         }
 413
 414         ext4_xattr_inode_set_class(inode);
 415
 416         /*
 417          * Check whether this is an old Lustre-style xattr inode. Lustre
 418          * implementation does not have hash validation, rather it has a
 419          * backpointer from ea_inode to the parent inode.
 420          */
 421         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 422             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 423             inode->i_generation == parent->i_generation) {
 424                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 425                 ext4_xattr_inode_set_ref(inode, 1);
 426         } else {
 427                 inode_lock(inode);
 428                 inode->i_flags |= S_NOQUOTA;
 429                 inode_unlock(inode);
 430         }
 431
 432         *ea_inode = inode;
 433         return 0;
 434 error:
 435         iput(inode);
 436         return err;
 437 }
 438
 439 /* Remove entry from mbcache when EA inode is getting evicted */
 440 void ext4_evict_ea_inode(struct inode *inode)
 441 {
 442         struct mb_cache_entry *oe;
 443
 444         if (!EA_INODE_CACHE(inode))
 445                 return;
 446         /* Wait for entry to get unused so that we can remove it */
 447         while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
 448                         ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
 449                 mb_cache_entry_wait_unused(oe);
 450                 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
 451         }
 452 }
 453
 454 static int
 455 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 456                                struct ext4_xattr_entry *entry, void *buffer,
 457                                size_t size)
 458 {
 459         u32 hash;
 460
 461         /* Verify stored hash matches calculated hash. */
 462         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 463         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 464                 return -EFSCORRUPTED;
 465
 466         if (entry) {
 467                 __le32 e_hash, tmp_data;
 468
 469                 /* Verify entry hash. */
 470                 tmp_data = cpu_to_le32(hash);
 471                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 472                                                &tmp_data, 1);
 473                 if (e_hash != entry->e_hash)
 474                         return -EFSCORRUPTED;
 475         }
 476         return 0;
 477 }
 478
 479 /*
 480  * Read xattr value from the EA inode.
 481  */
 482 static int
 483 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 484                      void *buffer, size_t size)
 485 {
 486         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 487         struct inode *ea_inode;
 488         int err;
 489
 490         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 491                                     le32_to_cpu(entry->e_hash), &ea_inode);
 492         if (err) {
 493                 ea_inode = NULL;
 494                 goto out;
 495         }
 496
 497         if (i_size_read(ea_inode) != size) {
 498                 ext4_warning_inode(ea_inode,
 499                                    "ea_inode file size=%llu entry size=%zu",
 500                                    i_size_read(ea_inode), size);
 501                 err = -EFSCORRUPTED;
 502                 goto out;
 503         }
 504
 505         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 506         if (err)
 507                 goto out;
 508
 509         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 510                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 511                                                      size);
 512                 if (err) {
 513                         ext4_warning_inode(ea_inode,
 514                                            "EA inode hash validation failed");
 515                         goto out;
 516                 }
 517
 518                 if (ea_inode_cache)
 519                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 520                                         ext4_xattr_inode_get_hash(ea_inode),
 521                                         ea_inode->i_ino, true /* reusable */);
 522         }
 523 out:
 524         iput(ea_inode);
 525         return err;
 526 }
 527
 528 static int
 529 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 530                      void *buffer, size_t buffer_size)
 531 {
 532         struct buffer_head *bh = NULL;
 533         struct ext4_xattr_entry *entry;
 534         size_t size;
 535         void *end;
 536         int error;
 537         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 538
 539         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 540                   name_index, name, buffer, (long)buffer_size);
 541
 542         if (!EXT4_I(inode)->i_file_acl)
 543                 return -ENODATA;
 544         ea_idebug(inode, "reading block %llu",
 545                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 546         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 547         if (IS_ERR(bh))
 548                 return PTR_ERR(bh);
 549         ea_bdebug(bh, "b_count=%d, refcount=%d",
 550                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 551         error = ext4_xattr_check_block(inode, bh);
 552         if (error)
 553                 goto cleanup;
 554         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 555         entry = BFIRST(bh);
 556         end = bh->b_data + bh->b_size;
 557         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 558         if (error)
 559                 goto cleanup;
 560         size = le32_to_cpu(entry->e_value_size);
 561         error = -ERANGE;
 562         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 563                 goto cleanup;
 564         if (buffer) {
 565                 if (size > buffer_size)
 566                         goto cleanup;
 567                 if (entry->e_value_inum) {
 568                         error = ext4_xattr_inode_get(inode, entry, buffer,
 569                                                      size);
 570                         if (error)
 571                                 goto cleanup;
 572                 } else {
 573                         u16 offset = le16_to_cpu(entry->e_value_offs);
 574                         void *p = bh->b_data + offset;
 575
 576                         if (unlikely(p + size > end))
 577                                 goto cleanup;
 578                         memcpy(buffer, p, size);
 579                 }
 580         }
 581         error = size;
 582
 583 cleanup:
 584         brelse(bh);
 585         return error;
 586 }
 587
 588 int
 589 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 590                      void *buffer, size_t buffer_size)
 591 {
 592         struct ext4_xattr_ibody_header *header;
 593         struct ext4_xattr_entry *entry;
 594         struct ext4_inode *raw_inode;
 595         struct ext4_iloc iloc;
 596         size_t size;
 597         void *end;
 598         int error;
 599
 600         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 601                 return -ENODATA;
 602         error = ext4_get_inode_loc(inode, &iloc);
 603         if (error)
 604                 return error;
 605         raw_inode = ext4_raw_inode(&iloc);
 606         header = IHDR(inode, raw_inode);
 607         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 608         error = xattr_check_inode(inode, header, end);
 609         if (error)
 610                 goto cleanup;
 611         entry = IFIRST(header);
 612         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 613         if (error)
 614                 goto cleanup;
 615         size = le32_to_cpu(entry->e_value_size);
 616         error = -ERANGE;
 617         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 618                 goto cleanup;
 619         if (buffer) {
 620                 if (size > buffer_size)
 621                         goto cleanup;
 622                 if (entry->e_value_inum) {
 623                         error = ext4_xattr_inode_get(inode, entry, buffer,
 624                                                      size);
 625                         if (error)
 626                                 goto cleanup;
 627                 } else {
 628                         u16 offset = le16_to_cpu(entry->e_value_offs);
 629                         void *p = (void *)IFIRST(header) + offset;
 630
 631                         if (unlikely(p + size > end))
 632                                 goto cleanup;
 633                         memcpy(buffer, p, size);
 634                 }
 635         }
 636         error = size;
 637
 638 cleanup:
 639         brelse(iloc.bh);
 640         return error;
 641 }
 642
 643 /*
 644  * ext4_xattr_get()
 645  *
 646  * Copy an extended attribute into the buffer
 647  * provided, or compute the buffer size required.
 648  * Buffer is NULL to compute the size of the buffer required.
 649  *
 650  * Returns a negative error number on failure, or the number of bytes
 651  * used / required on success.
 652  */
 653 int
 654 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 655                void *buffer, size_t buffer_size)
 656 {
 657         int error;
 658
 659         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 660                 return -EIO;
 661
 662         if (strlen(name) > 255)
 663                 return -ERANGE;
 664
 665         down_read(&EXT4_I(inode)->xattr_sem);
 666         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 667                                      buffer_size);
 668         if (error == -ENODATA)
 669                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 670                                              buffer_size);
 671         up_read(&EXT4_I(inode)->xattr_sem);
 672         return error;
 673 }
 674
 675 static int
 676 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 677                         char *buffer, size_t buffer_size)
 678 {
 679         size_t rest = buffer_size;
 680
 681         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 682                 const struct xattr_handler *handler =
 683                         ext4_xattr_handler(entry->e_name_index);
 684
 685                 if (handler && (!handler->list || handler->list(dentry))) {
 686                         const char *prefix = handler->prefix ?: handler->name;
 687                         size_t prefix_len = strlen(prefix);
 688                         size_t size = prefix_len + entry->e_name_len + 1;
 689
 690                         if (buffer) {
 691                                 if (size > rest)
 692                                         return -ERANGE;
 693                                 memcpy(buffer, prefix, prefix_len);
 694                                 buffer += prefix_len;
 695                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 696                                 buffer += entry->e_name_len;
 697                                 *buffer++ = 0;
 698                         }
 699                         rest -= size;
 700                 }
 701         }
 702         return buffer_size - rest;  /* total size */
 703 }
 704
 705 static int
 706 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 707 {
 708         struct inode *inode = d_inode(dentry);
 709         struct buffer_head *bh = NULL;
 710         int error;
 711
 712         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 713                   buffer, (long)buffer_size);
 714
 715         if (!EXT4_I(inode)->i_file_acl)
 716                 return 0;
 717         ea_idebug(inode, "reading block %llu",
 718                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 719         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 720         if (IS_ERR(bh))
 721                 return PTR_ERR(bh);
 722         ea_bdebug(bh, "b_count=%d, refcount=%d",
 723                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 724         error = ext4_xattr_check_block(inode, bh);
 725         if (error)
 726                 goto cleanup;
 727         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 728         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 729                                         buffer_size);
 730 cleanup:
 731         brelse(bh);
 732         return error;
 733 }
 734
 735 static int
 736 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 737 {
 738         struct inode *inode = d_inode(dentry);
 739         struct ext4_xattr_ibody_header *header;
 740         struct ext4_inode *raw_inode;
 741         struct ext4_iloc iloc;
 742         void *end;
 743         int error;
 744
 745         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 746                 return 0;
 747         error = ext4_get_inode_loc(inode, &iloc);
 748         if (error)
 749                 return error;
 750         raw_inode = ext4_raw_inode(&iloc);
 751         header = IHDR(inode, raw_inode);
 752         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 753         error = xattr_check_inode(inode, header, end);
 754         if (error)
 755                 goto cleanup;
 756         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 757                                         buffer, buffer_size);
 758
 759 cleanup:
 760         brelse(iloc.bh);
 761         return error;
 762 }
 763
 764 /*
 765  * Inode operation listxattr()
 766  *
 767  * d_inode(dentry)->i_rwsem: don't care
 768  *
 769  * Copy a list of attribute names into the buffer
 770  * provided, or compute the buffer size required.
 771  * Buffer is NULL to compute the size of the buffer required.
 772  *
 773  * Returns a negative error number on failure, or the number of bytes
 774  * used / required on success.
 775  */
 776 ssize_t
 777 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 778 {
 779         int ret, ret2;
 780
 781         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 782         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 783         if (ret < 0)
 784                 goto errout;
 785         if (buffer) {
 786                 buffer += ret;
 787                 buffer_size -= ret;
 788         }
 789         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 790         if (ret < 0)
 791                 goto errout;
 792         ret += ret2;
 793 errout:
 794         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 795         return ret;
 796 }
 797
 798 /*
 799  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 800  * not set, set it.
 801  */
 802 static void ext4_xattr_update_super_block(handle_t *handle,
 803                                           struct super_block *sb)
 804 {
 805         if (ext4_has_feature_xattr(sb))
 806                 return;
 807
 808         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 809         if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 810                                           EXT4_JTR_NONE) == 0) {
 811                 lock_buffer(EXT4_SB(sb)->s_sbh);
 812                 ext4_set_feature_xattr(sb);
 813                 ext4_superblock_csum_set(sb);
 814                 unlock_buffer(EXT4_SB(sb)->s_sbh);
 815                 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 816         }
 817 }
 818
 819 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 820 {
 821         struct ext4_iloc iloc = { .bh = NULL };
 822         struct buffer_head *bh = NULL;
 823         struct ext4_inode *raw_inode;
 824         struct ext4_xattr_ibody_header *header;
 825         struct ext4_xattr_entry *entry;
 826         qsize_t ea_inode_refs = 0;
 827         void *end;
 828         int ret;
 829
 830         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 831
 832         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 833                 ret = ext4_get_inode_loc(inode, &iloc);
 834                 if (ret)
 835                         goto out;
 836                 raw_inode = ext4_raw_inode(&iloc);
 837                 header = IHDR(inode, raw_inode);
 838                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 839                 ret = xattr_check_inode(inode, header, end);
 840                 if (ret)
 841                         goto out;
 842
 843                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 844                      entry = EXT4_XATTR_NEXT(entry))
 845                         if (entry->e_value_inum)
 846                                 ea_inode_refs++;
 847         }
 848
 849         if (EXT4_I(inode)->i_file_acl) {
 850                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 851                 if (IS_ERR(bh)) {
 852                         ret = PTR_ERR(bh);
 853                         bh = NULL;
 854                         goto out;
 855                 }
 856
 857                 ret = ext4_xattr_check_block(inode, bh);
 858                 if (ret)
 859                         goto out;
 860
 861                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 862                      entry = EXT4_XATTR_NEXT(entry))
 863                         if (entry->e_value_inum)
 864                                 ea_inode_refs++;
 865         }
 866         *usage = ea_inode_refs + 1;
 867         ret = 0;
 868 out:
 869         brelse(iloc.bh);
 870         brelse(bh);
 871         return ret;
 872 }
 873
 874 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 875 {
 876         struct super_block *sb = inode->i_sb;
 877         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 878                                     inode->i_blkbits);
 879         size_t mask = ~(cluster_size - 1);
 880
 881         return (length + cluster_size - 1) & mask;
 882 }
 883
 884 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 885 {
 886         int err;
 887
 888         err = dquot_alloc_inode(inode);
 889         if (err)
 890                 return err;
 891         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 892         if (err)
 893                 dquot_free_inode(inode);
 894         return err;
 895 }
 896
 897 static void ext4_xattr_inode_free_quota(struct inode *parent,
 898                                         struct inode *ea_inode,
 899                                         size_t len)
 900 {
 901         if (ea_inode &&
 902             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 903                 return;
 904         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 905         dquot_free_inode(parent);
 906 }
 907
 908 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 909                              struct buffer_head *block_bh, size_t value_len,
 910                              bool is_create)
 911 {
 912         int credits;
 913         int blocks;
 914
 915         /*
 916          * 1) Owner inode update
 917          * 2) Ref count update on old xattr block
 918          * 3) new xattr block
 919          * 4) block bitmap update for new xattr block
 920          * 5) group descriptor for new xattr block
 921          * 6) block bitmap update for old xattr block
 922          * 7) group descriptor for old block
 923          *
 924          * 6 & 7 can happen if we have two racing threads T_a and T_b
 925          * which are each trying to set an xattr on inodes I_a and I_b
 926          * which were both initially sharing an xattr block.
 927          */
 928         credits = 7;
 929
 930         /* Quota updates. */
 931         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 932
 933         /*
 934          * In case of inline data, we may push out the data to a block,
 935          * so we need to reserve credits for this eventuality
 936          */
 937         if (inode && ext4_has_inline_data(inode))
 938                 credits += ext4_writepage_trans_blocks(inode) + 1;
 939
 940         /* We are done if ea_inode feature is not enabled. */
 941         if (!ext4_has_feature_ea_inode(sb))
 942                 return credits;
 943
 944         /* New ea_inode, inode map, block bitmap, group descriptor. */
 945         credits += 4;
 946
 947         /* Data blocks. */
 948         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 949
 950         /* Indirection block or one level of extent tree. */
 951         blocks += 1;
 952
 953         /* Block bitmap and group descriptor updates for each block. */
 954         credits += blocks * 2;
 955
 956         /* Blocks themselves. */
 957         credits += blocks;
 958
 959         if (!is_create) {
 960                 /* Dereference ea_inode holding old xattr value.
 961                  * Old ea_inode, inode map, block bitmap, group descriptor.
 962                  */
 963                 credits += 4;
 964
 965                 /* Data blocks for old ea_inode. */
 966                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 967
 968                 /* Indirection block or one level of extent tree for old
 969                  * ea_inode.
 970                  */
 971                 blocks += 1;
 972
 973                 /* Block bitmap and group descriptor updates for each block. */
 974                 credits += blocks * 2;
 975         }
 976
 977         /* We may need to clone the existing xattr block in which case we need
 978          * to increment ref counts for existing ea_inodes referenced by it.
 979          */
 980         if (block_bh) {
 981                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
 982
 983                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 984                         if (entry->e_value_inum)
 985                                 /* Ref count update on ea_inode. */
 986                                 credits += 1;
 987         }
 988         return credits;
 989 }
 990
 991 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
 992                                        int ref_change)
 993 {
 994         struct ext4_iloc iloc;
 995         s64 ref_count;
 996         int ret;
 997
 998         inode_lock(ea_inode);
 999
1000         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1001         if (ret)
1002                 goto out;
1003
1004         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1005         ref_count += ref_change;
1006         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1007
1008         if (ref_change > 0) {
1009                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1010                           ea_inode->i_ino, ref_count);
1011
1012                 if (ref_count == 1) {
1013                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1014                                   ea_inode->i_ino, ea_inode->i_nlink);
1015
1016                         set_nlink(ea_inode, 1);
1017                         ext4_orphan_del(handle, ea_inode);
1018                 }
1019         } else {
1020                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1021                           ea_inode->i_ino, ref_count);
1022
1023                 if (ref_count == 0) {
1024                         WARN_ONCE(ea_inode->i_nlink != 1,
1025                                   "EA inode %lu i_nlink=%u",
1026                                   ea_inode->i_ino, ea_inode->i_nlink);
1027
1028                         clear_nlink(ea_inode);
1029                         ext4_orphan_add(handle, ea_inode);
1030                 }
1031         }
1032
1033         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1034         if (ret)
1035                 ext4_warning_inode(ea_inode,
1036                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1037 out:
1038         inode_unlock(ea_inode);
1039         return ret;
1040 }
1041
1042 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1043 {
1044         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1045 }
1046
1047 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1048 {
1049         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1050 }
1051
1052 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1053                                         struct ext4_xattr_entry *first)
1054 {
1055         struct inode *ea_inode;
1056         struct ext4_xattr_entry *entry;
1057         struct ext4_xattr_entry *failed_entry;
1058         unsigned int ea_ino;
1059         int err, saved_err;
1060
1061         for (entry = first; !IS_LAST_ENTRY(entry);
1062              entry = EXT4_XATTR_NEXT(entry)) {
1063                 if (!entry->e_value_inum)
1064                         continue;
1065                 ea_ino = le32_to_cpu(entry->e_value_inum);
1066                 err = ext4_xattr_inode_iget(parent, ea_ino,
1067                                             le32_to_cpu(entry->e_hash),
1068                                             &ea_inode);
1069                 if (err)
1070                         goto cleanup;
1071                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1072                 if (err) {
1073                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1074                         iput(ea_inode);
1075                         goto cleanup;
1076                 }
1077                 iput(ea_inode);
1078         }
1079         return 0;
1080
1081 cleanup:
1082         saved_err = err;
1083         failed_entry = entry;
1084
1085         for (entry = first; entry != failed_entry;
1086              entry = EXT4_XATTR_NEXT(entry)) {
1087                 if (!entry->e_value_inum)
1088                         continue;
1089                 ea_ino = le32_to_cpu(entry->e_value_inum);
1090                 err = ext4_xattr_inode_iget(parent, ea_ino,
1091                                             le32_to_cpu(entry->e_hash),
1092                                             &ea_inode);
1093                 if (err) {
1094                         ext4_warning(parent->i_sb,
1095                                      "cleanup ea_ino %u iget error %d", ea_ino,
1096                                      err);
1097                         continue;
1098                 }
1099                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1100                 if (err)
1101                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1102                                            err);
1103                 iput(ea_inode);
1104         }
1105         return saved_err;
1106 }
1107
1108 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1109                         struct buffer_head *bh, bool block_csum, bool dirty)
1110 {
1111         int error;
1112
1113         if (bh && dirty) {
1114                 if (block_csum)
1115                         ext4_xattr_block_csum_set(inode, bh);
1116                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1117                 if (error) {
1118                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1119                                      error);
1120                         return error;
1121                 }
1122         }
1123         return 0;
1124 }
1125
1126 static void
1127 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1128                              struct buffer_head *bh,
1129                              struct ext4_xattr_entry *first, bool block_csum,
1130                              struct ext4_xattr_inode_array **ea_inode_array,
1131                              int extra_credits, bool skip_quota)
1132 {
1133         struct inode *ea_inode;
1134         struct ext4_xattr_entry *entry;
1135         bool dirty = false;
1136         unsigned int ea_ino;
1137         int err;
1138         int credits;
1139
1140         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1141         credits = 2 + extra_credits;
1142
1143         for (entry = first; !IS_LAST_ENTRY(entry);
1144              entry = EXT4_XATTR_NEXT(entry)) {
1145                 if (!entry->e_value_inum)
1146                         continue;
1147                 ea_ino = le32_to_cpu(entry->e_value_inum);
1148                 err = ext4_xattr_inode_iget(parent, ea_ino,
1149                                             le32_to_cpu(entry->e_hash),
1150                                             &ea_inode);
1151                 if (err)
1152                         continue;
1153
1154                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1155                 if (err) {
1156                         ext4_warning_inode(ea_inode,
1157                                            "Expand inode array err=%d", err);
1158                         iput(ea_inode);
1159                         continue;
1160                 }
1161
1162                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1163                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1164                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1165                                               dirty));
1166                 if (err < 0) {
1167                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1168                                            err);
1169                         continue;
1170                 }
1171                 if (err > 0) {
1172                         err = ext4_journal_get_write_access(handle,
1173                                         parent->i_sb, bh, EXT4_JTR_NONE);
1174                         if (err) {
1175                                 ext4_warning_inode(ea_inode,
1176                                                 "Re-get write access err=%d",
1177                                                 err);
1178                                 continue;
1179                         }
1180                 }
1181
1182                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1183                 if (err) {
1184                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1185                                            err);
1186                         continue;
1187                 }
1188
1189                 if (!skip_quota)
1190                         ext4_xattr_inode_free_quota(parent, ea_inode,
1191                                               le32_to_cpu(entry->e_value_size));
1192
1193                 /*
1194                  * Forget about ea_inode within the same transaction that
1195                  * decrements the ref count. This avoids duplicate decrements in
1196                  * case the rest of the work spills over to subsequent
1197                  * transactions.
1198                  */
1199                 entry->e_value_inum = 0;
1200                 entry->e_value_size = 0;
1201
1202                 dirty = true;
1203         }
1204
1205         if (dirty) {
1206                 /*
1207                  * Note that we are deliberately skipping csum calculation for
1208                  * the final update because we do not expect any journal
1209                  * restarts until xattr block is freed.
1210                  */
1211
1212                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1213                 if (err)
1214                         ext4_warning_inode(parent,
1215                                            "handle dirty metadata err=%d", err);
1216         }
1217 }
1218
1219 /*
1220  * Release the xattr block BH: If the reference count is > 1, decrement it;
1221  * otherwise free the block.
1222  */
1223 static void
1224 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1225                          struct buffer_head *bh,
1226                          struct ext4_xattr_inode_array **ea_inode_array,
1227                          int extra_credits)
1228 {
1229         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1230         u32 hash, ref;
1231         int error = 0;
1232
1233         BUFFER_TRACE(bh, "get_write_access");
1234         error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1235                                               EXT4_JTR_NONE);
1236         if (error)
1237                 goto out;
1238
1239 retry_ref:
1240         lock_buffer(bh);
1241         hash = le32_to_cpu(BHDR(bh)->h_hash);
1242         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1243         if (ref == 1) {
1244                 ea_bdebug(bh, "refcount now=0; freeing");
1245                 /*
1246                  * This must happen under buffer lock for
1247                  * ext4_xattr_block_set() to reliably detect freed block
1248                  */
1249                 if (ea_block_cache) {
1250                         struct mb_cache_entry *oe;
1251
1252                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1253                                                           bh->b_blocknr);
1254                         if (oe) {
1255                                 unlock_buffer(bh);
1256                                 mb_cache_entry_wait_unused(oe);
1257                                 mb_cache_entry_put(ea_block_cache, oe);
1258                                 goto retry_ref;
1259                         }
1260                 }
1261                 get_bh(bh);
1262                 unlock_buffer(bh);
1263
1264                 if (ext4_has_feature_ea_inode(inode->i_sb))
1265                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1266                                                      BFIRST(bh),
1267                                                      true /* block_csum */,
1268                                                      ea_inode_array,
1269                                                      extra_credits,
1270                                                      true /* skip_quota */);
1271                 ext4_free_blocks(handle, inode, bh, 0, 1,
1272                                  EXT4_FREE_BLOCKS_METADATA |
1273                                  EXT4_FREE_BLOCKS_FORGET);
1274         } else {
1275                 ref--;
1276                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1277                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1278                         struct mb_cache_entry *ce;
1279
1280                         if (ea_block_cache) {
1281                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1282                                                         bh->b_blocknr);
1283                                 if (ce) {
1284                                         set_bit(MBE_REUSABLE_B, &ce->e_flags);
1285                                         mb_cache_entry_put(ea_block_cache, ce);
1286                                 }
1287                         }
1288                 }
1289
1290                 ext4_xattr_block_csum_set(inode, bh);
1291                 /*
1292                  * Beware of this ugliness: Releasing of xattr block references
1293                  * from different inodes can race and so we have to protect
1294                  * from a race where someone else frees the block (and releases
1295                  * its journal_head) before we are done dirtying the buffer. In
1296                  * nojournal mode this race is harmless and we actually cannot
1297                  * call ext4_handle_dirty_metadata() with locked buffer as
1298                  * that function can call sync_dirty_buffer() so for that case
1299                  * we handle the dirtying after unlocking the buffer.
1300                  */
1301                 if (ext4_handle_valid(handle))
1302                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1303                 unlock_buffer(bh);
1304                 if (!ext4_handle_valid(handle))
1305                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1306                 if (IS_SYNC(inode))
1307                         ext4_handle_sync(handle);
1308                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1309                 ea_bdebug(bh, "refcount now=%d; releasing",
1310                           le32_to_cpu(BHDR(bh)->h_refcount));
1311         }
1312 out:
1313         ext4_std_error(inode->i_sb, error);
1314         return;
1315 }
1316
1317 /*
1318  * Find the available free space for EAs. This also returns the total number of
1319  * bytes used by EA entries.
1320  */
1321 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1322                                     size_t *min_offs, void *base, int *total)
1323 {
1324         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1325                 if (!last->e_value_inum && last->e_value_size) {
1326                         size_t offs = le16_to_cpu(last->e_value_offs);
1327                         if (offs < *min_offs)
1328                                 *min_offs = offs;
1329                 }
1330                 if (total)
1331                         *total += EXT4_XATTR_LEN(last->e_name_len);
1332         }
1333         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1334 }
1335
1336 /*
1337  * Write the value of the EA in an inode.
1338  */
1339 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1340                                   const void *buf, int bufsize)
1341 {
1342         struct buffer_head *bh = NULL;
1343         unsigned long block = 0;
1344         int blocksize = ea_inode->i_sb->s_blocksize;
1345         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1346         int csize, wsize = 0;
1347         int ret = 0, ret2 = 0;
1348         int retries = 0;
1349
1350 retry:
1351         while (ret >= 0 && ret < max_blocks) {
1352                 struct ext4_map_blocks map;
1353                 map.m_lblk = block += ret;
1354                 map.m_len = max_blocks -= ret;
1355
1356                 ret = ext4_map_blocks(handle, ea_inode, &map,
1357                                       EXT4_GET_BLOCKS_CREATE);
1358                 if (ret <= 0) {
1359                         ext4_mark_inode_dirty(handle, ea_inode);
1360                         if (ret == -ENOSPC &&
1361                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1362                                 ret = 0;
1363                                 goto retry;
1364                         }
1365                         break;
1366                 }
1367         }
1368
1369         if (ret < 0)
1370                 return ret;
1371
1372         block = 0;
1373         while (wsize < bufsize) {
1374                 brelse(bh);
1375                 csize = (bufsize - wsize) > blocksize ? blocksize :
1376                                                                 bufsize - wsize;
1377                 bh = ext4_getblk(handle, ea_inode, block, 0);
1378                 if (IS_ERR(bh))
1379                         return PTR_ERR(bh);
1380                 if (!bh) {
1381                         WARN_ON_ONCE(1);
1382                         EXT4_ERROR_INODE(ea_inode,
1383                                          "ext4_getblk() return bh = NULL");
1384                         return -EFSCORRUPTED;
1385                 }
1386                 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1387                                                    EXT4_JTR_NONE);
1388                 if (ret)
1389                         goto out;
1390
1391                 memcpy(bh->b_data, buf, csize);
1392                 set_buffer_uptodate(bh);
1393                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1394
1395                 buf += csize;
1396                 wsize += csize;
1397                 block += 1;
1398         }
1399
1400         inode_lock(ea_inode);
1401         i_size_write(ea_inode, wsize);
1402         ext4_update_i_disksize(ea_inode, wsize);
1403         inode_unlock(ea_inode);
1404
1405         ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1406         if (unlikely(ret2 && !ret))
1407                 ret = ret2;
1408
1409 out:
1410         brelse(bh);
1411
1412         return ret;
1413 }
1414
1415 /*
1416  * Create an inode to store the value of a large EA.
1417  */
1418 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1419                                              struct inode *inode, u32 hash)
1420 {
1421         struct inode *ea_inode = NULL;
1422         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1423         int err;
1424
1425         if (inode->i_sb->s_root == NULL) {
1426                 ext4_warning(inode->i_sb,
1427                              "refuse to create EA inode when umounting");
1428                 WARN_ON(1);
1429                 return ERR_PTR(-EINVAL);
1430         }
1431
1432         /*
1433          * Let the next inode be the goal, so we try and allocate the EA inode
1434          * in the same group, or nearby one.
1435          */
1436         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1437                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1438                                   EXT4_EA_INODE_FL);
1439         if (!IS_ERR(ea_inode)) {
1440                 ea_inode->i_op = &ext4_file_inode_operations;
1441                 ea_inode->i_fop = &ext4_file_operations;
1442                 ext4_set_aops(ea_inode);
1443                 ext4_xattr_inode_set_class(ea_inode);
1444                 unlock_new_inode(ea_inode);
1445                 ext4_xattr_inode_set_ref(ea_inode, 1);
1446                 ext4_xattr_inode_set_hash(ea_inode, hash);
1447                 err = ext4_mark_inode_dirty(handle, ea_inode);
1448                 if (!err)
1449                         err = ext4_inode_attach_jinode(ea_inode);
1450                 if (err) {
1451                         if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1452                                 ext4_warning_inode(ea_inode,
1453                                         "cleanup dec ref error %d", err);
1454                         iput(ea_inode);
1455                         return ERR_PTR(err);
1456                 }
1457
1458                 /*
1459                  * Xattr inodes are shared therefore quota charging is performed
1460                  * at a higher level.
1461                  */
1462                 dquot_free_inode(ea_inode);
1463                 dquot_drop(ea_inode);
1464                 inode_lock(ea_inode);
1465                 ea_inode->i_flags |= S_NOQUOTA;
1466                 inode_unlock(ea_inode);
1467         }
1468
1469         return ea_inode;
1470 }
1471
1472 static struct inode *
1473 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1474                             size_t value_len, u32 hash)
1475 {
1476         struct inode *ea_inode;
1477         struct mb_cache_entry *ce;
1478         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1479         void *ea_data;
1480
1481         if (!ea_inode_cache)
1482                 return NULL;
1483
1484         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1485         if (!ce)
1486                 return NULL;
1487
1488         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1489                      !(current->flags & PF_MEMALLOC_NOFS));
1490
1491         ea_data = kvmalloc(value_len, GFP_KERNEL);
1492         if (!ea_data) {
1493                 mb_cache_entry_put(ea_inode_cache, ce);
1494                 return NULL;
1495         }
1496
1497         while (ce) {
1498                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1499                                      EXT4_IGET_NORMAL);
1500                 if (!IS_ERR(ea_inode) &&
1501                     !is_bad_inode(ea_inode) &&
1502                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1503                     i_size_read(ea_inode) == value_len &&
1504                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1505                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1506                                                     value_len) &&
1507                     !memcmp(value, ea_data, value_len)) {
1508                         mb_cache_entry_touch(ea_inode_cache, ce);
1509                         mb_cache_entry_put(ea_inode_cache, ce);
1510                         kvfree(ea_data);
1511                         return ea_inode;
1512                 }
1513
1514                 if (!IS_ERR(ea_inode))
1515                         iput(ea_inode);
1516                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1517         }
1518         kvfree(ea_data);
1519         return NULL;
1520 }
1521
1522 /*
1523  * Add value of the EA in an inode.
1524  */
1525 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1526                                           const void *value, size_t value_len,
1527                                           struct inode **ret_inode)
1528 {
1529         struct inode *ea_inode;
1530         u32 hash;
1531         int err;
1532
1533         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1534         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1535         if (ea_inode) {
1536                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1537                 if (err) {
1538                         iput(ea_inode);
1539                         return err;
1540                 }
1541
1542                 *ret_inode = ea_inode;
1543                 return 0;
1544         }
1545
1546         /* Create an inode for the EA value */
1547         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1548         if (IS_ERR(ea_inode))
1549                 return PTR_ERR(ea_inode);
1550
1551         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1552         if (err) {
1553                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1554                 iput(ea_inode);
1555                 return err;
1556         }
1557
1558         if (EA_INODE_CACHE(inode))
1559                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1560                                       ea_inode->i_ino, true /* reusable */);
1561
1562         *ret_inode = ea_inode;
1563         return 0;
1564 }
1565
1566 /*
1567  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1568  * feature is enabled.
1569  */
1570 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1571
1572 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1573                                 struct ext4_xattr_search *s,
1574                                 handle_t *handle, struct inode *inode,
1575                                 bool is_block)
1576 {
1577         struct ext4_xattr_entry *last, *next;
1578         struct ext4_xattr_entry *here = s->here;
1579         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1580         int in_inode = i->in_inode;
1581         struct inode *old_ea_inode = NULL;
1582         struct inode *new_ea_inode = NULL;
1583         size_t old_size, new_size;
1584         int ret;
1585
1586         /* Space used by old and new values. */
1587         old_size = (!s->not_found && !here->e_value_inum) ?
1588                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1589         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1590
1591         /*
1592          * Optimization for the simple case when old and new values have the
1593          * same padded sizes. Not applicable if external inodes are involved.
1594          */
1595         if (new_size && new_size == old_size) {
1596                 size_t offs = le16_to_cpu(here->e_value_offs);
1597                 void *val = s->base + offs;
1598
1599                 here->e_value_size = cpu_to_le32(i->value_len);
1600                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1601                         memset(val, 0, new_size);
1602                 } else {
1603                         memcpy(val, i->value, i->value_len);
1604                         /* Clear padding bytes. */
1605                         memset(val + i->value_len, 0, new_size - i->value_len);
1606                 }
1607                 goto update_hash;
1608         }
1609
1610         /* Compute min_offs and last. */
1611         last = s->first;
1612         for (; !IS_LAST_ENTRY(last); last = next) {
1613                 next = EXT4_XATTR_NEXT(last);
1614                 if ((void *)next >= s->end) {
1615                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1616                         ret = -EFSCORRUPTED;
1617                         goto out;
1618                 }
1619                 if (!last->e_value_inum && last->e_value_size) {
1620                         size_t offs = le16_to_cpu(last->e_value_offs);
1621                         if (offs < min_offs)
1622                                 min_offs = offs;
1623                 }
1624         }
1625
1626         /* Check whether we have enough space. */
1627         if (i->value) {
1628                 size_t free;
1629
1630                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1631                 if (!s->not_found)
1632                         free += EXT4_XATTR_LEN(name_len) + old_size;
1633
1634                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1635                         ret = -ENOSPC;
1636                         goto out;
1637                 }
1638
1639                 /*
1640                  * If storing the value in an external inode is an option,
1641                  * reserve space for xattr entries/names in the external
1642                  * attribute block so that a long value does not occupy the
1643                  * whole space and prevent further entries being added.
1644                  */
1645                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1646                     new_size && is_block &&
1647                     (min_offs + old_size - new_size) <
1648                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1649                         ret = -ENOSPC;
1650                         goto out;
1651                 }
1652         }
1653
1654         /*
1655          * Getting access to old and new ea inodes is subject to failures.
1656          * Finish that work before doing any modifications to the xattr data.
1657          */
1658         if (!s->not_found && here->e_value_inum) {
1659                 ret = ext4_xattr_inode_iget(inode,
1660                                             le32_to_cpu(here->e_value_inum),
1661                                             le32_to_cpu(here->e_hash),
1662                                             &old_ea_inode);
1663                 if (ret) {
1664                         old_ea_inode = NULL;
1665                         goto out;
1666                 }
1667         }
1668         if (i->value && in_inode) {
1669                 WARN_ON_ONCE(!i->value_len);
1670
1671                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1672                 if (ret)
1673                         goto out;
1674
1675                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1676                                                      i->value_len,
1677                                                      &new_ea_inode);
1678                 if (ret) {
1679                         new_ea_inode = NULL;
1680                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1681                         goto out;
1682                 }
1683         }
1684
1685         if (old_ea_inode) {
1686                 /* We are ready to release ref count on the old_ea_inode. */
1687                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1688                 if (ret) {
1689                         /* Release newly required ref count on new_ea_inode. */
1690                         if (new_ea_inode) {
1691                                 int err;
1692
1693                                 err = ext4_xattr_inode_dec_ref(handle,
1694                                                                new_ea_inode);
1695                                 if (err)
1696                                         ext4_warning_inode(new_ea_inode,
1697                                                   "dec ref new_ea_inode err=%d",
1698                                                   err);
1699                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1700                                                             i->value_len);
1701                         }
1702                         goto out;
1703                 }
1704
1705                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1706                                             le32_to_cpu(here->e_value_size));
1707         }
1708
1709         /* No failures allowed past this point. */
1710
1711         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1712                 /* Remove the old value. */
1713                 void *first_val = s->base + min_offs;
1714                 size_t offs = le16_to_cpu(here->e_value_offs);
1715                 void *val = s->base + offs;
1716
1717                 memmove(first_val + old_size, first_val, val - first_val);
1718                 memset(first_val, 0, old_size);
1719                 min_offs += old_size;
1720
1721                 /* Adjust all value offsets. */
1722                 last = s->first;
1723                 while (!IS_LAST_ENTRY(last)) {
1724                         size_t o = le16_to_cpu(last->e_value_offs);
1725
1726                         if (!last->e_value_inum &&
1727                             last->e_value_size && o < offs)
1728                                 last->e_value_offs = cpu_to_le16(o + old_size);
1729                         last = EXT4_XATTR_NEXT(last);
1730                 }
1731         }
1732
1733         if (!i->value) {
1734                 /* Remove old name. */
1735                 size_t size = EXT4_XATTR_LEN(name_len);
1736
1737                 last = ENTRY((void *)last - size);
1738                 memmove(here, (void *)here + size,
1739                         (void *)last - (void *)here + sizeof(__u32));
1740                 memset(last, 0, size);
1741         } else if (s->not_found) {
1742                 /* Insert new name. */
1743                 size_t size = EXT4_XATTR_LEN(name_len);
1744                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1745
1746                 memmove((void *)here + size, here, rest);
1747                 memset(here, 0, size);
1748                 here->e_name_index = i->name_index;
1749                 here->e_name_len = name_len;
1750                 memcpy(here->e_name, i->name, name_len);
1751         } else {
1752                 /* This is an update, reset value info. */
1753                 here->e_value_inum = 0;
1754                 here->e_value_offs = 0;
1755                 here->e_value_size = 0;
1756         }
1757
1758         if (i->value) {
1759                 /* Insert new value. */
1760                 if (in_inode) {
1761                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1762                 } else if (i->value_len) {
1763                         void *val = s->base + min_offs - new_size;
1764
1765                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1766                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1767                                 memset(val, 0, new_size);
1768                         } else {
1769                                 memcpy(val, i->value, i->value_len);
1770                                 /* Clear padding bytes. */
1771                                 memset(val + i->value_len, 0,
1772                                        new_size - i->value_len);
1773                         }
1774                 }
1775                 here->e_value_size = cpu_to_le32(i->value_len);
1776         }
1777
1778 update_hash:
1779         if (i->value) {
1780                 __le32 hash = 0;
1781
1782                 /* Entry hash calculation. */
1783                 if (in_inode) {
1784                         __le32 crc32c_hash;
1785
1786                         /*
1787                          * Feed crc32c hash instead of the raw value for entry
1788                          * hash calculation. This is to avoid walking
1789                          * potentially long value buffer again.
1790                          */
1791                         crc32c_hash = cpu_to_le32(
1792                                        ext4_xattr_inode_get_hash(new_ea_inode));
1793                         hash = ext4_xattr_hash_entry(here->e_name,
1794                                                      here->e_name_len,
1795                                                      &crc32c_hash, 1);
1796                 } else if (is_block) {
1797                         __le32 *value = s->base + le16_to_cpu(
1798                                                         here->e_value_offs);
1799
1800                         hash = ext4_xattr_hash_entry(here->e_name,
1801                                                      here->e_name_len, value,
1802                                                      new_size >> 2);
1803                 }
1804                 here->e_hash = hash;
1805         }
1806
1807         if (is_block)
1808                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1809
1810         ret = 0;
1811 out:
1812         iput(old_ea_inode);
1813         iput(new_ea_inode);
1814         return ret;
1815 }
1816
1817 struct ext4_xattr_block_find {
1818         struct ext4_xattr_search s;
1819         struct buffer_head *bh;
1820 };
1821
1822 static int
1823 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1824                       struct ext4_xattr_block_find *bs)
1825 {
1826         struct super_block *sb = inode->i_sb;
1827         int error;
1828
1829         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1830                   i->name_index, i->name, i->value, (long)i->value_len);
1831
1832         if (EXT4_I(inode)->i_file_acl) {
1833                 /* The inode already has an extended attribute block. */
1834                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1835                 if (IS_ERR(bs->bh)) {
1836                         error = PTR_ERR(bs->bh);
1837                         bs->bh = NULL;
1838                         return error;
1839                 }
1840                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1841                         atomic_read(&(bs->bh->b_count)),
1842                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1843                 error = ext4_xattr_check_block(inode, bs->bh);
1844                 if (error)
1845                         return error;
1846                 /* Find the named attribute. */
1847                 bs->s.base = BHDR(bs->bh);
1848                 bs->s.first = BFIRST(bs->bh);
1849                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1850                 bs->s.here = bs->s.first;
1851                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1852                                          i->name_index, i->name, 1);
1853                 if (error && error != -ENODATA)
1854                         return error;
1855                 bs->s.not_found = error;
1856         }
1857         return 0;
1858 }
1859
1860 static int
1861 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1862                      struct ext4_xattr_info *i,
1863                      struct ext4_xattr_block_find *bs)
1864 {
1865         struct super_block *sb = inode->i_sb;
1866         struct buffer_head *new_bh = NULL;
1867         struct ext4_xattr_search s_copy = bs->s;
1868         struct ext4_xattr_search *s = &s_copy;
1869         struct mb_cache_entry *ce = NULL;
1870         int error = 0;
1871         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1872         struct inode *ea_inode = NULL, *tmp_inode;
1873         size_t old_ea_inode_quota = 0;
1874         unsigned int ea_ino;
1875
1876
1877 #define header(x) ((struct ext4_xattr_header *)(x))
1878
1879         if (s->base) {
1880                 int offset = (char *)s->here - bs->bh->b_data;
1881
1882                 BUFFER_TRACE(bs->bh, "get_write_access");
1883                 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1884                                                       EXT4_JTR_NONE);
1885                 if (error)
1886                         goto cleanup;
1887                 lock_buffer(bs->bh);
1888
1889                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1890                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1891
1892                         /*
1893                          * This must happen under buffer lock for
1894                          * ext4_xattr_block_set() to reliably detect modified
1895                          * block
1896                          */
1897                         if (ea_block_cache) {
1898                                 struct mb_cache_entry *oe;
1899
1900                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1901                                         hash, bs->bh->b_blocknr);
1902                                 if (oe) {
1903                                         /*
1904                                          * Xattr block is getting reused. Leave
1905                                          * it alone.
1906                                          */
1907                                         mb_cache_entry_put(ea_block_cache, oe);
1908                                         goto clone_block;
1909                                 }
1910                         }
1911                         ea_bdebug(bs->bh, "modifying in-place");
1912                         error = ext4_xattr_set_entry(i, s, handle, inode,
1913                                                      true /* is_block */);
1914                         ext4_xattr_block_csum_set(inode, bs->bh);
1915                         unlock_buffer(bs->bh);
1916                         if (error == -EFSCORRUPTED)
1917                                 goto bad_block;
1918                         if (!error)
1919                                 error = ext4_handle_dirty_metadata(handle,
1920                                                                    inode,
1921                                                                    bs->bh);
1922                         if (error)
1923                                 goto cleanup;
1924                         goto inserted;
1925                 }
1926 clone_block:
1927                 unlock_buffer(bs->bh);
1928                 ea_bdebug(bs->bh, "cloning");
1929                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1930                 error = -ENOMEM;
1931                 if (s->base == NULL)
1932                         goto cleanup;
1933                 s->first = ENTRY(header(s->base)+1);
1934                 header(s->base)->h_refcount = cpu_to_le32(1);
1935                 s->here = ENTRY(s->base + offset);
1936                 s->end = s->base + bs->bh->b_size;
1937
1938                 /*
1939                  * If existing entry points to an xattr inode, we need
1940                  * to prevent ext4_xattr_set_entry() from decrementing
1941                  * ref count on it because the reference belongs to the
1942                  * original block. In this case, make the entry look
1943                  * like it has an empty value.
1944                  */
1945                 if (!s->not_found && s->here->e_value_inum) {
1946                         ea_ino = le32_to_cpu(s->here->e_value_inum);
1947                         error = ext4_xattr_inode_iget(inode, ea_ino,
1948                                       le32_to_cpu(s->here->e_hash),
1949                                       &tmp_inode);
1950                         if (error)
1951                                 goto cleanup;
1952
1953                         if (!ext4_test_inode_state(tmp_inode,
1954                                         EXT4_STATE_LUSTRE_EA_INODE)) {
1955                                 /*
1956                                  * Defer quota free call for previous
1957                                  * inode until success is guaranteed.
1958                                  */
1959                                 old_ea_inode_quota = le32_to_cpu(
1960                                                 s->here->e_value_size);
1961                         }
1962                         iput(tmp_inode);
1963
1964                         s->here->e_value_inum = 0;
1965                         s->here->e_value_size = 0;
1966                 }
1967         } else {
1968                 /* Allocate a buffer where we construct the new block. */
1969                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1970                 error = -ENOMEM;
1971                 if (s->base == NULL)
1972                         goto cleanup;
1973                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1974                 header(s->base)->h_blocks = cpu_to_le32(1);
1975                 header(s->base)->h_refcount = cpu_to_le32(1);
1976                 s->first = ENTRY(header(s->base)+1);
1977                 s->here = ENTRY(header(s->base)+1);
1978                 s->end = s->base + sb->s_blocksize;
1979         }
1980
1981         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1982         if (error == -EFSCORRUPTED)
1983                 goto bad_block;
1984         if (error)
1985                 goto cleanup;
1986
1987         if (i->value && s->here->e_value_inum) {
1988                 /*
1989                  * A ref count on ea_inode has been taken as part of the call to
1990                  * ext4_xattr_set_entry() above. We would like to drop this
1991                  * extra ref but we have to wait until the xattr block is
1992                  * initialized and has its own ref count on the ea_inode.
1993                  */
1994                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1995                 error = ext4_xattr_inode_iget(inode, ea_ino,
1996                                               le32_to_cpu(s->here->e_hash),
1997                                               &ea_inode);
1998                 if (error) {
1999                         ea_inode = NULL;
2000                         goto cleanup;
2001                 }
2002         }
2003
2004 inserted:
2005         if (!IS_LAST_ENTRY(s->first)) {
2006                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2007                                                      &ce);
2008                 if (new_bh) {
2009                         /* We found an identical block in the cache. */
2010                         if (new_bh == bs->bh)
2011                                 ea_bdebug(new_bh, "keeping");
2012                         else {
2013                                 u32 ref;
2014
2015                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2016
2017                                 /* The old block is released after updating
2018                                    the inode. */
2019                                 error = dquot_alloc_block(inode,
2020                                                 EXT4_C2B(EXT4_SB(sb), 1));
2021                                 if (error)
2022                                         goto cleanup;
2023                                 BUFFER_TRACE(new_bh, "get_write_access");
2024                                 error = ext4_journal_get_write_access(
2025                                                 handle, sb, new_bh,
2026                                                 EXT4_JTR_NONE);
2027                                 if (error)
2028                                         goto cleanup_dquot;
2029                                 lock_buffer(new_bh);
2030                                 /*
2031                                  * We have to be careful about races with
2032                                  * adding references to xattr block. Once we
2033                                  * hold buffer lock xattr block's state is
2034                                  * stable so we can check the additional
2035                                  * reference fits.
2036                                  */
2037                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2038                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2039                                         /*
2040                                          * Undo everything and check mbcache
2041                                          * again.
2042                                          */
2043                                         unlock_buffer(new_bh);
2044                                         dquot_free_block(inode,
2045                                                          EXT4_C2B(EXT4_SB(sb),
2046                                                                   1));
2047                                         brelse(new_bh);
2048                                         mb_cache_entry_put(ea_block_cache, ce);
2049                                         ce = NULL;
2050                                         new_bh = NULL;
2051                                         goto inserted;
2052                                 }
2053                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2054                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2055                                         clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2056                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2057                                           ref);
2058                                 ext4_xattr_block_csum_set(inode, new_bh);
2059                                 unlock_buffer(new_bh);
2060                                 error = ext4_handle_dirty_metadata(handle,
2061                                                                    inode,
2062                                                                    new_bh);
2063                                 if (error)
2064                                         goto cleanup_dquot;
2065                         }
2066                         mb_cache_entry_touch(ea_block_cache, ce);
2067                         mb_cache_entry_put(ea_block_cache, ce);
2068                         ce = NULL;
2069                 } else if (bs->bh && s->base == bs->bh->b_data) {
2070                         /* We were modifying this block in-place. */
2071                         ea_bdebug(bs->bh, "keeping this block");
2072                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2073                         new_bh = bs->bh;
2074                         get_bh(new_bh);
2075                 } else {
2076                         /* We need to allocate a new block */
2077                         ext4_fsblk_t goal, block;
2078
2079                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2080
2081                         goal = ext4_group_first_block_no(sb,
2082                                                 EXT4_I(inode)->i_block_group);
2083                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2084                                                      NULL, &error);
2085                         if (error)
2086                                 goto cleanup;
2087
2088                         ea_idebug(inode, "creating block %llu",
2089                                   (unsigned long long)block);
2090
2091                         new_bh = sb_getblk(sb, block);
2092                         if (unlikely(!new_bh)) {
2093                                 error = -ENOMEM;
2094 getblk_failed:
2095                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2096                                                  EXT4_FREE_BLOCKS_METADATA);
2097                                 goto cleanup;
2098                         }
2099                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2100                                                       ENTRY(header(s->base)+1));
2101                         if (error)
2102                                 goto getblk_failed;
2103                         if (ea_inode) {
2104                                 /* Drop the extra ref on ea_inode. */
2105                                 error = ext4_xattr_inode_dec_ref(handle,
2106                                                                  ea_inode);
2107                                 if (error)
2108                                         ext4_warning_inode(ea_inode,
2109                                                            "dec ref error=%d",
2110                                                            error);
2111                                 iput(ea_inode);
2112                                 ea_inode = NULL;
2113                         }
2114
2115                         lock_buffer(new_bh);
2116                         error = ext4_journal_get_create_access(handle, sb,
2117                                                         new_bh, EXT4_JTR_NONE);
2118                         if (error) {
2119                                 unlock_buffer(new_bh);
2120                                 error = -EIO;
2121                                 goto getblk_failed;
2122                         }
2123                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2124                         ext4_xattr_block_csum_set(inode, new_bh);
2125                         set_buffer_uptodate(new_bh);
2126                         unlock_buffer(new_bh);
2127                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2128                         error = ext4_handle_dirty_metadata(handle, inode,
2129                                                            new_bh);
2130                         if (error)
2131                                 goto cleanup;
2132                 }
2133         }
2134
2135         if (old_ea_inode_quota)
2136                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2137
2138         /* Update the inode. */
2139         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2140
2141         /* Drop the previous xattr block. */
2142         if (bs->bh && bs->bh != new_bh) {
2143                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2144
2145                 ext4_xattr_release_block(handle, inode, bs->bh,
2146                                          &ea_inode_array,
2147                                          0 /* extra_credits */);
2148                 ext4_xattr_inode_array_free(ea_inode_array);
2149         }
2150         error = 0;
2151
2152 cleanup:
2153         if (ea_inode) {
2154                 int error2;
2155
2156                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2157                 if (error2)
2158                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2159                                            error2);
2160
2161                 /* If there was an error, revert the quota charge. */
2162                 if (error)
2163                         ext4_xattr_inode_free_quota(inode, ea_inode,
2164                                                     i_size_read(ea_inode));
2165                 iput(ea_inode);
2166         }
2167         if (ce)
2168                 mb_cache_entry_put(ea_block_cache, ce);
2169         brelse(new_bh);
2170         if (!(bs->bh && s->base == bs->bh->b_data))
2171                 kfree(s->base);
2172
2173         return error;
2174
2175 cleanup_dquot:
2176         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2177         goto cleanup;
2178
2179 bad_block:
2180         EXT4_ERROR_INODE(inode, "bad block %llu",
2181                          EXT4_I(inode)->i_file_acl);
2182         goto cleanup;
2183
2184 #undef header
2185 }
2186
2187 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2188                           struct ext4_xattr_ibody_find *is)
2189 {
2190         struct ext4_xattr_ibody_header *header;
2191         struct ext4_inode *raw_inode;
2192         int error;
2193
2194         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2195                 return 0;
2196
2197         raw_inode = ext4_raw_inode(&is->iloc);
2198         header = IHDR(inode, raw_inode);
2199         is->s.base = is->s.first = IFIRST(header);
2200         is->s.here = is->s.first;
2201         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2202         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2203                 error = xattr_check_inode(inode, header, is->s.end);
2204                 if (error)
2205                         return error;
2206                 /* Find the named attribute. */
2207                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2208                                          i->name_index, i->name, 0);
2209                 if (error && error != -ENODATA)
2210                         return error;
2211                 is->s.not_found = error;
2212         }
2213         return 0;
2214 }
2215
2216 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2217                                 struct ext4_xattr_info *i,
2218                                 struct ext4_xattr_ibody_find *is)
2219 {
2220         struct ext4_xattr_ibody_header *header;
2221         struct ext4_xattr_search *s = &is->s;
2222         int error;
2223
2224         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2225                 return -ENOSPC;
2226
2227         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2228         if (error)
2229                 return error;
2230         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2231         if (!IS_LAST_ENTRY(s->first)) {
2232                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2233                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2234         } else {
2235                 header->h_magic = cpu_to_le32(0);
2236                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2237         }
2238         return 0;
2239 }
2240
2241 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2242                                  struct ext4_xattr_info *i)
2243 {
2244         void *value;
2245
2246         /* When e_value_inum is set the value is stored externally. */
2247         if (s->here->e_value_inum)
2248                 return 0;
2249         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2250                 return 0;
2251         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2252         return !memcmp(value, i->value, i->value_len);
2253 }
2254
2255 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2256 {
2257         struct buffer_head *bh;
2258         int error;
2259
2260         if (!EXT4_I(inode)->i_file_acl)
2261                 return NULL;
2262         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2263         if (IS_ERR(bh))
2264                 return bh;
2265         error = ext4_xattr_check_block(inode, bh);
2266         if (error) {
2267                 brelse(bh);
2268                 return ERR_PTR(error);
2269         }
2270         return bh;
2271 }
2272
2273 /*
2274  * ext4_xattr_set_handle()
2275  *
2276  * Create, replace or remove an extended attribute for this inode.  Value
2277  * is NULL to remove an existing extended attribute, and non-NULL to
2278  * either replace an existing extended attribute, or create a new extended
2279  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2280  * specify that an extended attribute must exist and must not exist
2281  * previous to the call, respectively.
2282  *
2283  * Returns 0, or a negative error number on failure.
2284  */
2285 int
2286 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2287                       const char *name, const void *value, size_t value_len,
2288                       int flags)
2289 {
2290         struct ext4_xattr_info i = {
2291                 .name_index = name_index,
2292                 .name = name,
2293                 .value = value,
2294                 .value_len = value_len,
2295                 .in_inode = 0,
2296         };
2297         struct ext4_xattr_ibody_find is = {
2298                 .s = { .not_found = -ENODATA, },
2299         };
2300         struct ext4_xattr_block_find bs = {
2301                 .s = { .not_found = -ENODATA, },
2302         };
2303         int no_expand;
2304         int error;
2305
2306         if (!name)
2307                 return -EINVAL;
2308         if (strlen(name) > 255)
2309                 return -ERANGE;
2310
2311         ext4_write_lock_xattr(inode, &no_expand);
2312
2313         /* Check journal credits under write lock. */
2314         if (ext4_handle_valid(handle)) {
2315                 struct buffer_head *bh;
2316                 int credits;
2317
2318                 bh = ext4_xattr_get_block(inode);
2319                 if (IS_ERR(bh)) {
2320                         error = PTR_ERR(bh);
2321                         goto cleanup;
2322                 }
2323
2324                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2325                                                    value_len,
2326                                                    flags & XATTR_CREATE);
2327                 brelse(bh);
2328
2329                 if (jbd2_handle_buffer_credits(handle) < credits) {
2330                         error = -ENOSPC;
2331                         goto cleanup;
2332                 }
2333                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2334         }
2335
2336         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2337         if (error)
2338                 goto cleanup;
2339
2340         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2341                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2342                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2343                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2344         }
2345
2346         error = ext4_xattr_ibody_find(inode, &i, &is);
2347         if (error)
2348                 goto cleanup;
2349         if (is.s.not_found)
2350                 error = ext4_xattr_block_find(inode, &i, &bs);
2351         if (error)
2352                 goto cleanup;
2353         if (is.s.not_found && bs.s.not_found) {
2354                 error = -ENODATA;
2355                 if (flags & XATTR_REPLACE)
2356                         goto cleanup;
2357                 error = 0;
2358                 if (!value)
2359                         goto cleanup;
2360         } else {
2361                 error = -EEXIST;
2362                 if (flags & XATTR_CREATE)
2363                         goto cleanup;
2364         }
2365
2366         if (!value) {
2367                 if (!is.s.not_found)
2368                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2369                 else if (!bs.s.not_found)
2370                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2371         } else {
2372                 error = 0;
2373                 /* Xattr value did not change? Save us some work and bail out */
2374                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2375                         goto cleanup;
2376                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2377                         goto cleanup;
2378
2379                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2380                     (EXT4_XATTR_SIZE(i.value_len) >
2381                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2382                         i.in_inode = 1;
2383 retry_inode:
2384                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2385                 if (!error && !bs.s.not_found) {
2386                         i.value = NULL;
2387                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2388                 } else if (error == -ENOSPC) {
2389                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2390                                 brelse(bs.bh);
2391                                 bs.bh = NULL;
2392                                 error = ext4_xattr_block_find(inode, &i, &bs);
2393                                 if (error)
2394                                         goto cleanup;
2395                         }
2396                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2397                         if (!error && !is.s.not_found) {
2398                                 i.value = NULL;
2399                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2400                                                              &is);
2401                         } else if (error == -ENOSPC) {
2402                                 /*
2403                                  * Xattr does not fit in the block, store at
2404                                  * external inode if possible.
2405                                  */
2406                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2407                                     i.value_len && !i.in_inode) {
2408                                         i.in_inode = 1;
2409                                         goto retry_inode;
2410                                 }
2411                         }
2412                 }
2413         }
2414         if (!error) {
2415                 ext4_xattr_update_super_block(handle, inode->i_sb);
2416                 inode->i_ctime = current_time(inode);
2417                 inode_inc_iversion(inode);
2418                 if (!value)
2419                         no_expand = 0;
2420                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2421                 /*
2422                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2423                  * error != 0.
2424                  */
2425                 is.iloc.bh = NULL;
2426                 if (IS_SYNC(inode))
2427                         ext4_handle_sync(handle);
2428         }
2429         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2430
2431 cleanup:
2432         brelse(is.iloc.bh);
2433         brelse(bs.bh);
2434         ext4_write_unlock_xattr(inode, &no_expand);
2435         return error;
2436 }
2437
2438 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2439                            bool is_create, int *credits)
2440 {
2441         struct buffer_head *bh;
2442         int err;
2443
2444         *credits = 0;
2445
2446         if (!EXT4_SB(inode->i_sb)->s_journal)
2447                 return 0;
2448
2449         down_read(&EXT4_I(inode)->xattr_sem);
2450
2451         bh = ext4_xattr_get_block(inode);
2452         if (IS_ERR(bh)) {
2453                 err = PTR_ERR(bh);
2454         } else {
2455                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2456                                                     value_len, is_create);
2457                 brelse(bh);
2458                 err = 0;
2459         }
2460
2461         up_read(&EXT4_I(inode)->xattr_sem);
2462         return err;
2463 }
2464
2465 /*
2466  * ext4_xattr_set()
2467  *
2468  * Like ext4_xattr_set_handle, but start from an inode. This extended
2469  * attribute modification is a filesystem transaction by itself.
2470  *
2471  * Returns 0, or a negative error number on failure.
2472  */
2473 int
2474 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2475                const void *value, size_t value_len, int flags)
2476 {
2477         handle_t *handle;
2478         struct super_block *sb = inode->i_sb;
2479         int error, retries = 0;
2480         int credits;
2481
2482         error = dquot_initialize(inode);
2483         if (error)
2484                 return error;
2485
2486 retry:
2487         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2488                                        &credits);
2489         if (error)
2490                 return error;
2491
2492         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2493         if (IS_ERR(handle)) {
2494                 error = PTR_ERR(handle);
2495         } else {
2496                 int error2;
2497
2498                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2499                                               value, value_len, flags);
2500                 error2 = ext4_journal_stop(handle);
2501                 if (error == -ENOSPC &&
2502                     ext4_should_retry_alloc(sb, &retries))
2503                         goto retry;
2504                 if (error == 0)
2505                         error = error2;
2506         }
2507         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2508
2509         return error;
2510 }
2511
2512 /*
2513  * Shift the EA entries in the inode to create space for the increased
2514  * i_extra_isize.
2515  */
2516 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2517                                      int value_offs_shift, void *to,
2518                                      void *from, size_t n)
2519 {
2520         struct ext4_xattr_entry *last = entry;
2521         int new_offs;
2522
2523         /* We always shift xattr headers further thus offsets get lower */
2524         BUG_ON(value_offs_shift > 0);
2525
2526         /* Adjust the value offsets of the entries */
2527         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2528                 if (!last->e_value_inum && last->e_value_size) {
2529                         new_offs = le16_to_cpu(last->e_value_offs) +
2530                                                         value_offs_shift;
2531                         last->e_value_offs = cpu_to_le16(new_offs);
2532                 }
2533         }
2534         /* Shift the entries by n bytes */
2535         memmove(to, from, n);
2536 }
2537
2538 /*
2539  * Move xattr pointed to by 'entry' from inode into external xattr block
2540  */
2541 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2542                                     struct ext4_inode *raw_inode,
2543                                     struct ext4_xattr_entry *entry)
2544 {
2545         struct ext4_xattr_ibody_find *is = NULL;
2546         struct ext4_xattr_block_find *bs = NULL;
2547         char *buffer = NULL, *b_entry_name = NULL;
2548         size_t value_size = le32_to_cpu(entry->e_value_size);
2549         struct ext4_xattr_info i = {
2550                 .value = NULL,
2551                 .value_len = 0,
2552                 .name_index = entry->e_name_index,
2553                 .in_inode = !!entry->e_value_inum,
2554         };
2555         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2556         int error;
2557
2558         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2559         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2560         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2561         if (!is || !bs || !b_entry_name) {
2562                 error = -ENOMEM;
2563                 goto out;
2564         }
2565
2566         is->s.not_found = -ENODATA;
2567         bs->s.not_found = -ENODATA;
2568         is->iloc.bh = NULL;
2569         bs->bh = NULL;
2570
2571         /* Save the entry name and the entry value */
2572         if (entry->e_value_inum) {
2573                 buffer = kvmalloc(value_size, GFP_NOFS);
2574                 if (!buffer) {
2575                         error = -ENOMEM;
2576                         goto out;
2577                 }
2578
2579                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2580                 if (error)
2581                         goto out;
2582         } else {
2583                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2584                 buffer = (void *)IFIRST(header) + value_offs;
2585         }
2586
2587         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2588         b_entry_name[entry->e_name_len] = '\0';
2589         i.name = b_entry_name;
2590
2591         error = ext4_get_inode_loc(inode, &is->iloc);
2592         if (error)
2593                 goto out;
2594
2595         error = ext4_xattr_ibody_find(inode, &i, is);
2596         if (error)
2597                 goto out;
2598
2599         i.value = buffer;
2600         i.value_len = value_size;
2601         error = ext4_xattr_block_find(inode, &i, bs);
2602         if (error)
2603                 goto out;
2604
2605         /* Move ea entry from the inode into the block */
2606         error = ext4_xattr_block_set(handle, inode, &i, bs);
2607         if (error)
2608                 goto out;
2609
2610         /* Remove the chosen entry from the inode */
2611         i.value = NULL;
2612         i.value_len = 0;
2613         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2614
2615 out:
2616         kfree(b_entry_name);
2617         if (entry->e_value_inum && buffer)
2618                 kvfree(buffer);
2619         if (is)
2620                 brelse(is->iloc.bh);
2621         if (bs)
2622                 brelse(bs->bh);
2623         kfree(is);
2624         kfree(bs);
2625
2626         return error;
2627 }
2628
2629 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2630                                        struct ext4_inode *raw_inode,
2631                                        int isize_diff, size_t ifree,
2632                                        size_t bfree, int *total_ino)
2633 {
2634         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2635         struct ext4_xattr_entry *small_entry;
2636         struct ext4_xattr_entry *entry;
2637         struct ext4_xattr_entry *last;
2638         unsigned int entry_size;        /* EA entry size */
2639         unsigned int total_size;        /* EA entry size + value size */
2640         unsigned int min_total_size;
2641         int error;
2642
2643         while (isize_diff > ifree) {
2644                 entry = NULL;
2645                 small_entry = NULL;
2646                 min_total_size = ~0U;
2647                 last = IFIRST(header);
2648                 /* Find the entry best suited to be pushed into EA block */
2649                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2650                         /* never move system.data out of the inode */
2651                         if ((last->e_name_len == 4) &&
2652                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2653                             !memcmp(last->e_name, "data", 4))
2654                                 continue;
2655                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2656                         if (!last->e_value_inum)
2657                                 total_size += EXT4_XATTR_SIZE(
2658                                                le32_to_cpu(last->e_value_size));
2659                         if (total_size <= bfree &&
2660                             total_size < min_total_size) {
2661                                 if (total_size + ifree < isize_diff) {
2662                                         small_entry = last;
2663                                 } else {
2664                                         entry = last;
2665                                         min_total_size = total_size;
2666                                 }
2667                         }
2668                 }
2669
2670                 if (entry == NULL) {
2671                         if (small_entry == NULL)
2672                                 return -ENOSPC;
2673                         entry = small_entry;
2674                 }
2675
2676                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2677                 total_size = entry_size;
2678                 if (!entry->e_value_inum)
2679                         total_size += EXT4_XATTR_SIZE(
2680                                               le32_to_cpu(entry->e_value_size));
2681                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2682                                                  entry);
2683                 if (error)
2684                         return error;
2685
2686                 *total_ino -= entry_size;
2687                 ifree += total_size;
2688                 bfree -= total_size;
2689         }
2690
2691         return 0;
2692 }
2693
2694 /*
2695  * Expand an inode by new_extra_isize bytes when EAs are present.
2696  * Returns 0 on success or negative error number on failure.
2697  */
2698 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2699                                struct ext4_inode *raw_inode, handle_t *handle)
2700 {
2701         struct ext4_xattr_ibody_header *header;
2702         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2703         static unsigned int mnt_count;
2704         size_t min_offs;
2705         size_t ifree, bfree;
2706         int total_ino;
2707         void *base, *end;
2708         int error = 0, tried_min_extra_isize = 0;
2709         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2710         int isize_diff; /* How much do we need to grow i_extra_isize */
2711
2712 retry:
2713         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2714         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2715                 return 0;
2716
2717         header = IHDR(inode, raw_inode);
2718
2719         /*
2720          * Check if enough free space is available in the inode to shift the
2721          * entries ahead by new_extra_isize.
2722          */
2723
2724         base = IFIRST(header);
2725         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2726         min_offs = end - base;
2727         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2728
2729         error = xattr_check_inode(inode, header, end);
2730         if (error)
2731                 goto cleanup;
2732
2733         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2734         if (ifree >= isize_diff)
2735                 goto shift;
2736
2737         /*
2738          * Enough free space isn't available in the inode, check if
2739          * EA block can hold new_extra_isize bytes.
2740          */
2741         if (EXT4_I(inode)->i_file_acl) {
2742                 struct buffer_head *bh;
2743
2744                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2745                 if (IS_ERR(bh)) {
2746                         error = PTR_ERR(bh);
2747                         goto cleanup;
2748                 }
2749                 error = ext4_xattr_check_block(inode, bh);
2750                 if (error) {
2751                         brelse(bh);
2752                         goto cleanup;
2753                 }
2754                 base = BHDR(bh);
2755                 end = bh->b_data + bh->b_size;
2756                 min_offs = end - base;
2757                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2758                                               NULL);
2759                 brelse(bh);
2760                 if (bfree + ifree < isize_diff) {
2761                         if (!tried_min_extra_isize && s_min_extra_isize) {
2762                                 tried_min_extra_isize++;
2763                                 new_extra_isize = s_min_extra_isize;
2764                                 goto retry;
2765                         }
2766                         error = -ENOSPC;
2767                         goto cleanup;
2768                 }
2769         } else {
2770                 bfree = inode->i_sb->s_blocksize;
2771         }
2772
2773         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2774                                             isize_diff, ifree, bfree,
2775                                             &total_ino);
2776         if (error) {
2777                 if (error == -ENOSPC && !tried_min_extra_isize &&
2778                     s_min_extra_isize) {
2779                         tried_min_extra_isize++;
2780                         new_extra_isize = s_min_extra_isize;
2781                         goto retry;
2782                 }
2783                 goto cleanup;
2784         }
2785 shift:
2786         /* Adjust the offsets and shift the remaining entries ahead */
2787         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2788                         - new_extra_isize, (void *)raw_inode +
2789                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2790                         (void *)header, total_ino);
2791         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2792
2793 cleanup:
2794         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2795                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2796                              inode->i_ino);
2797                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2798         }
2799         return error;
2800 }
2801
2802 #define EIA_INCR 16 /* must be 2^n */
2803 #define EIA_MASK (EIA_INCR - 1)
2804
2805 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2806  * If @ea_inode_array is new or full it will be grown and the old
2807  * contents copied over.
2808  */
2809 static int
2810 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2811                         struct inode *inode)
2812 {
2813         if (*ea_inode_array == NULL) {
2814                 /*
2815                  * Start with 15 inodes, so it fits into a power-of-two size.
2816                  * If *ea_inode_array is NULL, this is essentially offsetof()
2817                  */
2818                 (*ea_inode_array) =
2819                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2820                                          inodes[EIA_MASK]),
2821                                 GFP_NOFS);
2822                 if (*ea_inode_array == NULL)
2823                         return -ENOMEM;
2824                 (*ea_inode_array)->count = 0;
2825         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2826                 /* expand the array once all 15 + n * 16 slots are full */
2827                 struct ext4_xattr_inode_array *new_array = NULL;
2828                 int count = (*ea_inode_array)->count;
2829
2830                 /* if new_array is NULL, this is essentially offsetof() */
2831                 new_array = kmalloc(
2832                                 offsetof(struct ext4_xattr_inode_array,
2833                                          inodes[count + EIA_INCR]),
2834                                 GFP_NOFS);
2835                 if (new_array == NULL)
2836                         return -ENOMEM;
2837                 memcpy(new_array, *ea_inode_array,
2838                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2839                 kfree(*ea_inode_array);
2840                 *ea_inode_array = new_array;
2841         }
2842         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2843         return 0;
2844 }
2845
2846 /*
2847  * ext4_xattr_delete_inode()
2848  *
2849  * Free extended attribute resources associated with this inode. Traverse
2850  * all entries and decrement reference on any xattr inodes associated with this
2851  * inode. This is called immediately before an inode is freed. We have exclusive
2852  * access to the inode. If an orphan inode is deleted it will also release its
2853  * references on xattr block and xattr inodes.
2854  */
2855 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2856                             struct ext4_xattr_inode_array **ea_inode_array,
2857                             int extra_credits)
2858 {
2859         struct buffer_head *bh = NULL;
2860         struct ext4_xattr_ibody_header *header;
2861         struct ext4_iloc iloc = { .bh = NULL };
2862         struct ext4_xattr_entry *entry;
2863         struct inode *ea_inode;
2864         int error;
2865
2866         error = ext4_journal_ensure_credits(handle, extra_credits,
2867                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2868         if (error < 0) {
2869                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2870                 goto cleanup;
2871         }
2872
2873         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2874             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2875
2876                 error = ext4_get_inode_loc(inode, &iloc);
2877                 if (error) {
2878                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2879                         goto cleanup;
2880                 }
2881
2882                 error = ext4_journal_get_write_access(handle, inode->i_sb,
2883                                                 iloc.bh, EXT4_JTR_NONE);
2884                 if (error) {
2885                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2886                                          error);
2887                         goto cleanup;
2888                 }
2889
2890                 header = IHDR(inode, ext4_raw_inode(&iloc));
2891                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2892                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2893                                                      IFIRST(header),
2894                                                      false /* block_csum */,
2895                                                      ea_inode_array,
2896                                                      extra_credits,
2897                                                      false /* skip_quota */);
2898         }
2899
2900         if (EXT4_I(inode)->i_file_acl) {
2901                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2902                 if (IS_ERR(bh)) {
2903                         error = PTR_ERR(bh);
2904                         if (error == -EIO) {
2905                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2906                                                      "block %llu read error",
2907                                                      EXT4_I(inode)->i_file_acl);
2908                         }
2909                         bh = NULL;
2910                         goto cleanup;
2911                 }
2912                 error = ext4_xattr_check_block(inode, bh);
2913                 if (error)
2914                         goto cleanup;
2915
2916                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2917                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2918                              entry = EXT4_XATTR_NEXT(entry)) {
2919                                 if (!entry->e_value_inum)
2920                                         continue;
2921                                 error = ext4_xattr_inode_iget(inode,
2922                                               le32_to_cpu(entry->e_value_inum),
2923                                               le32_to_cpu(entry->e_hash),
2924                                               &ea_inode);
2925                                 if (error)
2926                                         continue;
2927                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2928                                               le32_to_cpu(entry->e_value_size));
2929                                 iput(ea_inode);
2930                         }
2931
2932                 }
2933
2934                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2935                                          extra_credits);
2936                 /*
2937                  * Update i_file_acl value in the same transaction that releases
2938                  * block.
2939                  */
2940                 EXT4_I(inode)->i_file_acl = 0;
2941                 error = ext4_mark_inode_dirty(handle, inode);
2942                 if (error) {
2943                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2944                                          error);
2945                         goto cleanup;
2946                 }
2947                 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2948         }
2949         error = 0;
2950 cleanup:
2951         brelse(iloc.bh);
2952         brelse(bh);
2953         return error;
2954 }
2955
2956 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2957 {
2958         int idx;
2959
2960         if (ea_inode_array == NULL)
2961                 return;
2962
2963         for (idx = 0; idx < ea_inode_array->count; ++idx)
2964                 iput(ea_inode_array->inodes[idx]);
2965         kfree(ea_inode_array);
2966 }
2967
2968 /*
2969  * ext4_xattr_block_cache_insert()
2970  *
2971  * Create a new entry in the extended attribute block cache, and insert
2972  * it unless such an entry is already in the cache.
2973  *
2974  * Returns 0, or a negative error number on failure.
2975  */
2976 static void
2977 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2978                               struct buffer_head *bh)
2979 {
2980         struct ext4_xattr_header *header = BHDR(bh);
2981         __u32 hash = le32_to_cpu(header->h_hash);
2982         int reusable = le32_to_cpu(header->h_refcount) <
2983                        EXT4_XATTR_REFCOUNT_MAX;
2984         int error;
2985
2986         if (!ea_block_cache)
2987                 return;
2988         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2989                                       bh->b_blocknr, reusable);
2990         if (error) {
2991                 if (error == -EBUSY)
2992                         ea_bdebug(bh, "already in cache");
2993         } else
2994                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2995 }
2996
2997 /*
2998  * ext4_xattr_cmp()
2999  *
3000  * Compare two extended attribute blocks for equality.
3001  *
3002  * Returns 0 if the blocks are equal, 1 if they differ, and
3003  * a negative error number on errors.
3004  */
3005 static int
3006 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3007                struct ext4_xattr_header *header2)
3008 {
3009         struct ext4_xattr_entry *entry1, *entry2;
3010
3011         entry1 = ENTRY(header1+1);
3012         entry2 = ENTRY(header2+1);
3013         while (!IS_LAST_ENTRY(entry1)) {
3014                 if (IS_LAST_ENTRY(entry2))
3015                         return 1;
3016                 if (entry1->e_hash != entry2->e_hash ||
3017                     entry1->e_name_index != entry2->e_name_index ||
3018                     entry1->e_name_len != entry2->e_name_len ||
3019                     entry1->e_value_size != entry2->e_value_size ||
3020                     entry1->e_value_inum != entry2->e_value_inum ||
3021                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3022                         return 1;
3023                 if (!entry1->e_value_inum &&
3024                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3025                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3026                            le32_to_cpu(entry1->e_value_size)))
3027                         return 1;
3028
3029                 entry1 = EXT4_XATTR_NEXT(entry1);
3030                 entry2 = EXT4_XATTR_NEXT(entry2);
3031         }
3032         if (!IS_LAST_ENTRY(entry2))
3033                 return 1;
3034         return 0;
3035 }
3036
3037 /*
3038  * ext4_xattr_block_cache_find()
3039  *
3040  * Find an identical extended attribute block.
3041  *
3042  * Returns a pointer to the block found, or NULL if such a block was
3043  * not found or an error occurred.
3044  */
3045 static struct buffer_head *
3046 ext4_xattr_block_cache_find(struct inode *inode,
3047                             struct ext4_xattr_header *header,
3048                             struct mb_cache_entry **pce)
3049 {
3050         __u32 hash = le32_to_cpu(header->h_hash);
3051         struct mb_cache_entry *ce;
3052         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3053
3054         if (!ea_block_cache)
3055                 return NULL;
3056         if (!header->h_hash)
3057                 return NULL;  /* never share */
3058         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3059         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3060         while (ce) {
3061                 struct buffer_head *bh;
3062
3063                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3064                 if (IS_ERR(bh)) {
3065                         if (PTR_ERR(bh) == -ENOMEM)
3066                                 return NULL;
3067                         bh = NULL;
3068                         EXT4_ERROR_INODE(inode, "block %lu read error",
3069                                          (unsigned long)ce->e_value);
3070                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3071                         *pce = ce;
3072                         return bh;
3073                 }
3074                 brelse(bh);
3075                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3076         }
3077         return NULL;
3078 }
3079
3080 #define NAME_HASH_SHIFT 5
3081 #define VALUE_HASH_SHIFT 16
3082
3083 /*
3084  * ext4_xattr_hash_entry()
3085  *
3086  * Compute the hash of an extended attribute.
3087  */
3088 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3089                                     size_t value_count)
3090 {
3091         __u32 hash = 0;
3092
3093         while (name_len--) {
3094                 hash = (hash << NAME_HASH_SHIFT) ^
3095                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3096                        *name++;
3097         }
3098         while (value_count--) {
3099                 hash = (hash << VALUE_HASH_SHIFT) ^
3100                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3101                        le32_to_cpu(*value++);
3102         }
3103         return cpu_to_le32(hash);
3104 }
3105
3106 #undef NAME_HASH_SHIFT
3107 #undef VALUE_HASH_SHIFT
3108
3109 #define BLOCK_HASH_SHIFT 16
3110
3111 /*
3112  * ext4_xattr_rehash()
3113  *
3114  * Re-compute the extended attribute hash value after an entry has changed.
3115  */
3116 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3117 {
3118         struct ext4_xattr_entry *here;
3119         __u32 hash = 0;
3120
3121         here = ENTRY(header+1);
3122         while (!IS_LAST_ENTRY(here)) {
3123                 if (!here->e_hash) {
3124                         /* Block is not shared if an entry's hash value == 0 */
3125                         hash = 0;
3126                         break;
3127                 }
3128                 hash = (hash << BLOCK_HASH_SHIFT) ^
3129                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3130                        le32_to_cpu(here->e_hash);
3131                 here = EXT4_XATTR_NEXT(here);
3132         }
3133         header->h_hash = cpu_to_le32(hash);
3134 }
3135
3136 #undef BLOCK_HASH_SHIFT
3137
3138 #define HASH_BUCKET_BITS        10
3139
3140 struct mb_cache *
3141 ext4_xattr_create_cache(void)
3142 {
3143         return mb_cache_create(HASH_BUCKET_BITS);
3144 }
3145
3146 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3147 {
3148         if (cache)
3149                 mb_cache_destroy(cache);
3150 }
3151