fs/ext4/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/fs/ext4/xattr.c
   4  *
   5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6  *
   7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9  * Extended attributes for symlinks and special files added per
  10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12  *  Red Hat Inc.
  13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14  *  and Andreas Gruenbacher <agruen@suse.de>.
  15  */
  16
  17 /*
  18  * Extended attributes are stored directly in inodes (on file systems with
  19  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20  * field contains the block number if an inode uses an additional block. All
  21  * attributes must fit in the inode and one additional block. Blocks that
  22  * contain the identical set of attributes may be shared among several inodes.
  23  * Identical blocks are detected by keeping a cache of blocks that have
  24  * recently been accessed.
  25  *
  26  * The attributes in inodes and on blocks have a different header; the entries
  27  * are stored in the same format:
  28  *
  29  *   +------------------+
  30  *   | header           |
  31  *   | entry 1          | |
  32  *   | entry 2          | | growing downwards
  33  *   | entry 3          | v
  34  *   | four null bytes  |
  35  *   | . . .            |
  36  *   | value 1          | ^
  37  *   | value 3          | | growing upwards
  38  *   | value 2          | |
  39  *   +------------------+
  40  *
  41  * The header is followed by multiple entry descriptors. In disk blocks, the
  42  * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43  * attribute values are aligned to the end of the block in no specific order.
  44  *
  45  * Locking strategy
  46  * ----------------
  47  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48  * EA blocks are only changed if they are exclusive to an inode, so
  49  * holding xattr_sem also means that nothing but the EA block's reference
  50  * count can change. Multiple writers to the same block are synchronized
  51  * by the buffer lock.
  52  */
  53
  54 #include <linux/init.h>
  55 #include <linux/fs.h>
  56 #include <linux/slab.h>
  57 #include <linux/mbcache.h>
  58 #include <linux/quotaops.h>
  59 #include <linux/iversion.h>
  60 #include "ext4_jbd2.h"
  61 #include "ext4.h"
  62 #include "xattr.h"
  63 #include "acl.h"
  64
  65 #ifdef EXT4_XATTR_DEBUG
  66 # define ea_idebug(inode, fmt, ...)                                     \
  67         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  68                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69 # define ea_bdebug(bh, fmt, ...)                                        \
  70         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  71                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72 #else
  73 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  74 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  75 #endif
  76
  77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78                                           struct buffer_head *);
  79 static struct buffer_head *
  80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81                             struct mb_cache_entry **);
  82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83                                     size_t value_count);
  84 static void ext4_xattr_rehash(struct ext4_xattr_header *);
  85
  86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  87         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  89         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  90         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  91 #endif
  92         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  93 #ifdef CONFIG_EXT4_FS_SECURITY
  94         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  95 #endif
  96 };
  97
  98 const struct xattr_handler *ext4_xattr_handlers[] = {
  99         &ext4_xattr_user_handler,
 100         &ext4_xattr_trusted_handler,
 101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 102         &posix_acl_access_xattr_handler,
 103         &posix_acl_default_xattr_handler,
 104 #endif
 105 #ifdef CONFIG_EXT4_FS_SECURITY
 106         &ext4_xattr_security_handler,
 107 #endif
 108         NULL
 109 };
 110
 111 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 112                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 115                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117 static int
 118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119                         struct inode *inode);
 120
 121 #ifdef CONFIG_LOCKDEP
 122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123 {
 124         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 125 }
 126 #endif
 127
 128 static __le32 ext4_xattr_block_csum(struct inode *inode,
 129                                     sector_t block_nr,
 130                                     struct ext4_xattr_header *hdr)
 131 {
 132         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133         __u32 csum;
 134         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 135         __u32 dummy_csum = 0;
 136         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 137
 138         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 139                            sizeof(dsk_block_nr));
 140         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 141         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 142         offset += sizeof(dummy_csum);
 143         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 144                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 145
 146         return cpu_to_le32(csum);
 147 }
 148
 149 static int ext4_xattr_block_csum_verify(struct inode *inode,
 150                                         struct buffer_head *bh)
 151 {
 152         struct ext4_xattr_header *hdr = BHDR(bh);
 153         int ret = 1;
 154
 155         if (ext4_has_metadata_csum(inode->i_sb)) {
 156                 lock_buffer(bh);
 157                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 158                                                         bh->b_blocknr, hdr));
 159                 unlock_buffer(bh);
 160         }
 161         return ret;
 162 }
 163
 164 static void ext4_xattr_block_csum_set(struct inode *inode,
 165                                       struct buffer_head *bh)
 166 {
 167         if (ext4_has_metadata_csum(inode->i_sb))
 168                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 169                                                 bh->b_blocknr, BHDR(bh));
 170 }
 171
 172 static inline const struct xattr_handler *
 173 ext4_xattr_handler(int name_index)
 174 {
 175         const struct xattr_handler *handler = NULL;
 176
 177         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 178                 handler = ext4_xattr_handler_map[name_index];
 179         return handler;
 180 }
 181
 182 static int
 183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 184                          void *value_start)
 185 {
 186         struct ext4_xattr_entry *e = entry;
 187
 188         /* Find the end of the names list */
 189         while (!IS_LAST_ENTRY(e)) {
 190                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 191                 if ((void *)next >= end)
 192                         return -EFSCORRUPTED;
 193                 e = next;
 194         }
 195
 196         /* Check the values */
 197         while (!IS_LAST_ENTRY(entry)) {
 198                 u32 size = le32_to_cpu(entry->e_value_size);
 199
 200                 if (size > EXT4_XATTR_SIZE_MAX)
 201                         return -EFSCORRUPTED;
 202
 203                 if (size != 0 && entry->e_value_inum == 0) {
 204                         u16 offs = le16_to_cpu(entry->e_value_offs);
 205                         void *value;
 206
 207                         /*
 208                          * The value cannot overlap the names, and the value
 209                          * with padding cannot extend beyond 'end'.  Check both
 210                          * the padded and unpadded sizes, since the size may
 211                          * overflow to 0 when adding padding.
 212                          */
 213                         if (offs > end - value_start)
 214                                 return -EFSCORRUPTED;
 215                         value = value_start + offs;
 216                         if (value < (void *)e + sizeof(u32) ||
 217                             size > end - value ||
 218                             EXT4_XATTR_SIZE(size) > end - value)
 219                                 return -EFSCORRUPTED;
 220                 }
 221                 entry = EXT4_XATTR_NEXT(entry);
 222         }
 223
 224         return 0;
 225 }
 226
 227 static inline int
 228 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 229                          const char *function, unsigned int line)
 230 {
 231         int error = -EFSCORRUPTED;
 232
 233         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 234             BHDR(bh)->h_blocks != cpu_to_le32(1))
 235                 goto errout;
 236         if (buffer_verified(bh))
 237                 return 0;
 238
 239         error = -EFSBADCRC;
 240         if (!ext4_xattr_block_csum_verify(inode, bh))
 241                 goto errout;
 242         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 243                                          bh->b_data);
 244 errout:
 245         if (error)
 246                 __ext4_error_inode(inode, function, line, 0,
 247                                    "corrupted xattr block %llu",
 248                                    (unsigned long long) bh->b_blocknr);
 249         else
 250                 set_buffer_verified(bh);
 251         return error;
 252 }
 253
 254 #define ext4_xattr_check_block(inode, bh) \
 255         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 256
 257
 258 static int
 259 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 260                          void *end, const char *function, unsigned int line)
 261 {
 262         int error = -EFSCORRUPTED;
 263
 264         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 265             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 266                 goto errout;
 267         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 268 errout:
 269         if (error)
 270                 __ext4_error_inode(inode, function, line, 0,
 271                                    "corrupted in-inode xattr");
 272         return error;
 273 }
 274
 275 #define xattr_check_inode(inode, header, end) \
 276         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 277
 278 static int
 279 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 280                  void *end, int name_index, const char *name, int sorted)
 281 {
 282         struct ext4_xattr_entry *entry, *next;
 283         size_t name_len;
 284         int cmp = 1;
 285
 286         if (name == NULL)
 287                 return -EINVAL;
 288         name_len = strlen(name);
 289         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 290                 next = EXT4_XATTR_NEXT(entry);
 291                 if ((void *) next >= end) {
 292                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 293                         return -EFSCORRUPTED;
 294                 }
 295                 cmp = name_index - entry->e_name_index;
 296                 if (!cmp)
 297                         cmp = name_len - entry->e_name_len;
 298                 if (!cmp)
 299                         cmp = memcmp(name, entry->e_name, name_len);
 300                 if (cmp <= 0 && (sorted || cmp == 0))
 301                         break;
 302         }
 303         *pentry = entry;
 304         return cmp ? -ENODATA : 0;
 305 }
 306
 307 static u32
 308 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 309 {
 310         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 311 }
 312
 313 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 314 {
 315         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 316                 (u32) inode_peek_iversion_raw(ea_inode);
 317 }
 318
 319 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 320 {
 321         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 322         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 323 }
 324
 325 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 326 {
 327         return (u32)ea_inode->i_atime.tv_sec;
 328 }
 329
 330 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 331 {
 332         ea_inode->i_atime.tv_sec = hash;
 333 }
 334
 335 /*
 336  * Read the EA value from an inode.
 337  */
 338 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 339 {
 340         int blocksize = 1 << ea_inode->i_blkbits;
 341         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 342         int tail_size = (size % blocksize) ?: blocksize;
 343         struct buffer_head *bhs_inline[8];
 344         struct buffer_head **bhs = bhs_inline;
 345         int i, ret;
 346
 347         if (bh_count > ARRAY_SIZE(bhs_inline)) {
 348                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 349                 if (!bhs)
 350                         return -ENOMEM;
 351         }
 352
 353         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 354                                true /* wait */, bhs);
 355         if (ret)
 356                 goto free_bhs;
 357
 358         for (i = 0; i < bh_count; i++) {
 359                 /* There shouldn't be any holes in ea_inode. */
 360                 if (!bhs[i]) {
 361                         ret = -EFSCORRUPTED;
 362                         goto put_bhs;
 363                 }
 364                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 365                        i < bh_count - 1 ? blocksize : tail_size);
 366         }
 367         ret = 0;
 368 put_bhs:
 369         for (i = 0; i < bh_count; i++)
 370                 brelse(bhs[i]);
 371 free_bhs:
 372         if (bhs != bhs_inline)
 373                 kfree(bhs);
 374         return ret;
 375 }
 376
 377 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 378
 379 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 380                                  u32 ea_inode_hash, struct inode **ea_inode)
 381 {
 382         struct inode *inode;
 383         int err;
 384
 385         inode = ext4_iget(parent->i_sb, ea_ino);
 386         if (IS_ERR(inode)) {
 387                 err = PTR_ERR(inode);
 388                 ext4_error(parent->i_sb,
 389                            "error while reading EA inode %lu err=%d", ea_ino,
 390                            err);
 391                 return err;
 392         }
 393
 394         if (is_bad_inode(inode)) {
 395                 ext4_error(parent->i_sb,
 396                            "error while reading EA inode %lu is_bad_inode",
 397                            ea_ino);
 398                 err = -EIO;
 399                 goto error;
 400         }
 401
 402         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 403                 ext4_error(parent->i_sb,
 404                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 405                             ea_ino);
 406                 err = -EINVAL;
 407                 goto error;
 408         }
 409
 410         ext4_xattr_inode_set_class(inode);
 411
 412         /*
 413          * Check whether this is an old Lustre-style xattr inode. Lustre
 414          * implementation does not have hash validation, rather it has a
 415          * backpointer from ea_inode to the parent inode.
 416          */
 417         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 418             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 419             inode->i_generation == parent->i_generation) {
 420                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 421                 ext4_xattr_inode_set_ref(inode, 1);
 422         } else {
 423                 inode_lock(inode);
 424                 inode->i_flags |= S_NOQUOTA;
 425                 inode_unlock(inode);
 426         }
 427
 428         *ea_inode = inode;
 429         return 0;
 430 error:
 431         iput(inode);
 432         return err;
 433 }
 434
 435 static int
 436 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 437                                struct ext4_xattr_entry *entry, void *buffer,
 438                                size_t size)
 439 {
 440         u32 hash;
 441
 442         /* Verify stored hash matches calculated hash. */
 443         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 444         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 445                 return -EFSCORRUPTED;
 446
 447         if (entry) {
 448                 __le32 e_hash, tmp_data;
 449
 450                 /* Verify entry hash. */
 451                 tmp_data = cpu_to_le32(hash);
 452                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 453                                                &tmp_data, 1);
 454                 if (e_hash != entry->e_hash)
 455                         return -EFSCORRUPTED;
 456         }
 457         return 0;
 458 }
 459
 460 /*
 461  * Read xattr value from the EA inode.
 462  */
 463 static int
 464 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 465                      void *buffer, size_t size)
 466 {
 467         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 468         struct inode *ea_inode;
 469         int err;
 470
 471         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 472                                     le32_to_cpu(entry->e_hash), &ea_inode);
 473         if (err) {
 474                 ea_inode = NULL;
 475                 goto out;
 476         }
 477
 478         if (i_size_read(ea_inode) != size) {
 479                 ext4_warning_inode(ea_inode,
 480                                    "ea_inode file size=%llu entry size=%zu",
 481                                    i_size_read(ea_inode), size);
 482                 err = -EFSCORRUPTED;
 483                 goto out;
 484         }
 485
 486         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 487         if (err)
 488                 goto out;
 489
 490         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 491                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 492                                                      size);
 493                 if (err) {
 494                         ext4_warning_inode(ea_inode,
 495                                            "EA inode hash validation failed");
 496                         goto out;
 497                 }
 498
 499                 if (ea_inode_cache)
 500                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 501                                         ext4_xattr_inode_get_hash(ea_inode),
 502                                         ea_inode->i_ino, true /* reusable */);
 503         }
 504 out:
 505         iput(ea_inode);
 506         return err;
 507 }
 508
 509 static int
 510 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 511                      void *buffer, size_t buffer_size)
 512 {
 513         struct buffer_head *bh = NULL;
 514         struct ext4_xattr_entry *entry;
 515         size_t size;
 516         void *end;
 517         int error;
 518         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 519
 520         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 521                   name_index, name, buffer, (long)buffer_size);
 522
 523         error = -ENODATA;
 524         if (!EXT4_I(inode)->i_file_acl)
 525                 goto cleanup;
 526         ea_idebug(inode, "reading block %llu",
 527                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 528         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 529         if (!bh)
 530                 goto cleanup;
 531         ea_bdebug(bh, "b_count=%d, refcount=%d",
 532                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 533         error = ext4_xattr_check_block(inode, bh);
 534         if (error)
 535                 goto cleanup;
 536         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 537         entry = BFIRST(bh);
 538         end = bh->b_data + bh->b_size;
 539         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 540         if (error)
 541                 goto cleanup;
 542         size = le32_to_cpu(entry->e_value_size);
 543         error = -ERANGE;
 544         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 545                 goto cleanup;
 546         if (buffer) {
 547                 if (size > buffer_size)
 548                         goto cleanup;
 549                 if (entry->e_value_inum) {
 550                         error = ext4_xattr_inode_get(inode, entry, buffer,
 551                                                      size);
 552                         if (error)
 553                                 goto cleanup;
 554                 } else {
 555                         u16 offset = le16_to_cpu(entry->e_value_offs);
 556                         void *p = bh->b_data + offset;
 557
 558                         if (unlikely(p + size > end))
 559                                 goto cleanup;
 560                         memcpy(buffer, p, size);
 561                 }
 562         }
 563         error = size;
 564
 565 cleanup:
 566         brelse(bh);
 567         return error;
 568 }
 569
 570 int
 571 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 572                      void *buffer, size_t buffer_size)
 573 {
 574         struct ext4_xattr_ibody_header *header;
 575         struct ext4_xattr_entry *entry;
 576         struct ext4_inode *raw_inode;
 577         struct ext4_iloc iloc;
 578         size_t size;
 579         void *end;
 580         int error;
 581
 582         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 583                 return -ENODATA;
 584         error = ext4_get_inode_loc(inode, &iloc);
 585         if (error)
 586                 return error;
 587         raw_inode = ext4_raw_inode(&iloc);
 588         header = IHDR(inode, raw_inode);
 589         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 590         error = xattr_check_inode(inode, header, end);
 591         if (error)
 592                 goto cleanup;
 593         entry = IFIRST(header);
 594         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 595         if (error)
 596                 goto cleanup;
 597         size = le32_to_cpu(entry->e_value_size);
 598         error = -ERANGE;
 599         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 600                 goto cleanup;
 601         if (buffer) {
 602                 if (size > buffer_size)
 603                         goto cleanup;
 604                 if (entry->e_value_inum) {
 605                         error = ext4_xattr_inode_get(inode, entry, buffer,
 606                                                      size);
 607                         if (error)
 608                                 goto cleanup;
 609                 } else {
 610                         u16 offset = le16_to_cpu(entry->e_value_offs);
 611                         void *p = (void *)IFIRST(header) + offset;
 612
 613                         if (unlikely(p + size > end))
 614                                 goto cleanup;
 615                         memcpy(buffer, p, size);
 616                 }
 617         }
 618         error = size;
 619
 620 cleanup:
 621         brelse(iloc.bh);
 622         return error;
 623 }
 624
 625 /*
 626  * ext4_xattr_get()
 627  *
 628  * Copy an extended attribute into the buffer
 629  * provided, or compute the buffer size required.
 630  * Buffer is NULL to compute the size of the buffer required.
 631  *
 632  * Returns a negative error number on failure, or the number of bytes
 633  * used / required on success.
 634  */
 635 int
 636 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 637                void *buffer, size_t buffer_size)
 638 {
 639         int error;
 640
 641         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 642                 return -EIO;
 643
 644         if (strlen(name) > 255)
 645                 return -ERANGE;
 646
 647         down_read(&EXT4_I(inode)->xattr_sem);
 648         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 649                                      buffer_size);
 650         if (error == -ENODATA)
 651                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 652                                              buffer_size);
 653         up_read(&EXT4_I(inode)->xattr_sem);
 654         return error;
 655 }
 656
 657 static int
 658 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 659                         char *buffer, size_t buffer_size)
 660 {
 661         size_t rest = buffer_size;
 662
 663         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 664                 const struct xattr_handler *handler =
 665                         ext4_xattr_handler(entry->e_name_index);
 666
 667                 if (handler && (!handler->list || handler->list(dentry))) {
 668                         const char *prefix = handler->prefix ?: handler->name;
 669                         size_t prefix_len = strlen(prefix);
 670                         size_t size = prefix_len + entry->e_name_len + 1;
 671
 672                         if (buffer) {
 673                                 if (size > rest)
 674                                         return -ERANGE;
 675                                 memcpy(buffer, prefix, prefix_len);
 676                                 buffer += prefix_len;
 677                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 678                                 buffer += entry->e_name_len;
 679                                 *buffer++ = 0;
 680                         }
 681                         rest -= size;
 682                 }
 683         }
 684         return buffer_size - rest;  /* total size */
 685 }
 686
 687 static int
 688 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 689 {
 690         struct inode *inode = d_inode(dentry);
 691         struct buffer_head *bh = NULL;
 692         int error;
 693
 694         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 695                   buffer, (long)buffer_size);
 696
 697         error = 0;
 698         if (!EXT4_I(inode)->i_file_acl)
 699                 goto cleanup;
 700         ea_idebug(inode, "reading block %llu",
 701                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 702         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 703         error = -EIO;
 704         if (!bh)
 705                 goto cleanup;
 706         ea_bdebug(bh, "b_count=%d, refcount=%d",
 707                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 708         error = ext4_xattr_check_block(inode, bh);
 709         if (error)
 710                 goto cleanup;
 711         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 712         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 713
 714 cleanup:
 715         brelse(bh);
 716
 717         return error;
 718 }
 719
 720 static int
 721 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 722 {
 723         struct inode *inode = d_inode(dentry);
 724         struct ext4_xattr_ibody_header *header;
 725         struct ext4_inode *raw_inode;
 726         struct ext4_iloc iloc;
 727         void *end;
 728         int error;
 729
 730         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 731                 return 0;
 732         error = ext4_get_inode_loc(inode, &iloc);
 733         if (error)
 734                 return error;
 735         raw_inode = ext4_raw_inode(&iloc);
 736         header = IHDR(inode, raw_inode);
 737         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 738         error = xattr_check_inode(inode, header, end);
 739         if (error)
 740                 goto cleanup;
 741         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 742                                         buffer, buffer_size);
 743
 744 cleanup:
 745         brelse(iloc.bh);
 746         return error;
 747 }
 748
 749 /*
 750  * Inode operation listxattr()
 751  *
 752  * d_inode(dentry)->i_rwsem: don't care
 753  *
 754  * Copy a list of attribute names into the buffer
 755  * provided, or compute the buffer size required.
 756  * Buffer is NULL to compute the size of the buffer required.
 757  *
 758  * Returns a negative error number on failure, or the number of bytes
 759  * used / required on success.
 760  */
 761 ssize_t
 762 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 763 {
 764         int ret, ret2;
 765
 766         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 767         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 768         if (ret < 0)
 769                 goto errout;
 770         if (buffer) {
 771                 buffer += ret;
 772                 buffer_size -= ret;
 773         }
 774         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 775         if (ret < 0)
 776                 goto errout;
 777         ret += ret2;
 778 errout:
 779         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 780         return ret;
 781 }
 782
 783 /*
 784  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 785  * not set, set it.
 786  */
 787 static void ext4_xattr_update_super_block(handle_t *handle,
 788                                           struct super_block *sb)
 789 {
 790         if (ext4_has_feature_xattr(sb))
 791                 return;
 792
 793         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 794         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 795                 ext4_set_feature_xattr(sb);
 796                 ext4_handle_dirty_super(handle, sb);
 797         }
 798 }
 799
 800 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 801 {
 802         struct ext4_iloc iloc = { .bh = NULL };
 803         struct buffer_head *bh = NULL;
 804         struct ext4_inode *raw_inode;
 805         struct ext4_xattr_ibody_header *header;
 806         struct ext4_xattr_entry *entry;
 807         qsize_t ea_inode_refs = 0;
 808         void *end;
 809         int ret;
 810
 811         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 812
 813         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 814                 ret = ext4_get_inode_loc(inode, &iloc);
 815                 if (ret)
 816                         goto out;
 817                 raw_inode = ext4_raw_inode(&iloc);
 818                 header = IHDR(inode, raw_inode);
 819                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 820                 ret = xattr_check_inode(inode, header, end);
 821                 if (ret)
 822                         goto out;
 823
 824                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 825                      entry = EXT4_XATTR_NEXT(entry))
 826                         if (entry->e_value_inum)
 827                                 ea_inode_refs++;
 828         }
 829
 830         if (EXT4_I(inode)->i_file_acl) {
 831                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 832                 if (!bh) {
 833                         ret = -EIO;
 834                         goto out;
 835                 }
 836
 837                 ret = ext4_xattr_check_block(inode, bh);
 838                 if (ret)
 839                         goto out;
 840
 841                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 842                      entry = EXT4_XATTR_NEXT(entry))
 843                         if (entry->e_value_inum)
 844                                 ea_inode_refs++;
 845         }
 846         *usage = ea_inode_refs + 1;
 847         ret = 0;
 848 out:
 849         brelse(iloc.bh);
 850         brelse(bh);
 851         return ret;
 852 }
 853
 854 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 855 {
 856         struct super_block *sb = inode->i_sb;
 857         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 858                                     inode->i_blkbits);
 859         size_t mask = ~(cluster_size - 1);
 860
 861         return (length + cluster_size - 1) & mask;
 862 }
 863
 864 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 865 {
 866         int err;
 867
 868         err = dquot_alloc_inode(inode);
 869         if (err)
 870                 return err;
 871         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 872         if (err)
 873                 dquot_free_inode(inode);
 874         return err;
 875 }
 876
 877 static void ext4_xattr_inode_free_quota(struct inode *parent,
 878                                         struct inode *ea_inode,
 879                                         size_t len)
 880 {
 881         if (ea_inode &&
 882             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 883                 return;
 884         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 885         dquot_free_inode(parent);
 886 }
 887
 888 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 889                              struct buffer_head *block_bh, size_t value_len,
 890                              bool is_create)
 891 {
 892         int credits;
 893         int blocks;
 894
 895         /*
 896          * 1) Owner inode update
 897          * 2) Ref count update on old xattr block
 898          * 3) new xattr block
 899          * 4) block bitmap update for new xattr block
 900          * 5) group descriptor for new xattr block
 901          * 6) block bitmap update for old xattr block
 902          * 7) group descriptor for old block
 903          *
 904          * 6 & 7 can happen if we have two racing threads T_a and T_b
 905          * which are each trying to set an xattr on inodes I_a and I_b
 906          * which were both initially sharing an xattr block.
 907          */
 908         credits = 7;
 909
 910         /* Quota updates. */
 911         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 912
 913         /*
 914          * In case of inline data, we may push out the data to a block,
 915          * so we need to reserve credits for this eventuality
 916          */
 917         if (inode && ext4_has_inline_data(inode))
 918                 credits += ext4_writepage_trans_blocks(inode) + 1;
 919
 920         /* We are done if ea_inode feature is not enabled. */
 921         if (!ext4_has_feature_ea_inode(sb))
 922                 return credits;
 923
 924         /* New ea_inode, inode map, block bitmap, group descriptor. */
 925         credits += 4;
 926
 927         /* Data blocks. */
 928         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 929
 930         /* Indirection block or one level of extent tree. */
 931         blocks += 1;
 932
 933         /* Block bitmap and group descriptor updates for each block. */
 934         credits += blocks * 2;
 935
 936         /* Blocks themselves. */
 937         credits += blocks;
 938
 939         if (!is_create) {
 940                 /* Dereference ea_inode holding old xattr value.
 941                  * Old ea_inode, inode map, block bitmap, group descriptor.
 942                  */
 943                 credits += 4;
 944
 945                 /* Data blocks for old ea_inode. */
 946                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 947
 948                 /* Indirection block or one level of extent tree for old
 949                  * ea_inode.
 950                  */
 951                 blocks += 1;
 952
 953                 /* Block bitmap and group descriptor updates for each block. */
 954                 credits += blocks * 2;
 955         }
 956
 957         /* We may need to clone the existing xattr block in which case we need
 958          * to increment ref counts for existing ea_inodes referenced by it.
 959          */
 960         if (block_bh) {
 961                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
 962
 963                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 964                         if (entry->e_value_inum)
 965                                 /* Ref count update on ea_inode. */
 966                                 credits += 1;
 967         }
 968         return credits;
 969 }
 970
 971 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 972                                      int credits, struct buffer_head *bh,
 973                                      bool dirty, bool block_csum)
 974 {
 975         int error;
 976
 977         if (!ext4_handle_valid(handle))
 978                 return 0;
 979
 980         if (handle->h_buffer_credits >= credits)
 981                 return 0;
 982
 983         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 984         if (!error)
 985                 return 0;
 986         if (error < 0) {
 987                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 988                 return error;
 989         }
 990
 991         if (bh && dirty) {
 992                 if (block_csum)
 993                         ext4_xattr_block_csum_set(inode, bh);
 994                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
 995                 if (error) {
 996                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 997                                      error);
 998                         return error;
 999                 }
1000         }
1001
1002         error = ext4_journal_restart(handle, credits);
1003         if (error) {
1004                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1005                 return error;
1006         }
1007
1008         if (bh) {
1009                 error = ext4_journal_get_write_access(handle, bh);
1010                 if (error) {
1011                         ext4_warning(inode->i_sb,
1012                                      "Get write access failed (error %d)",
1013                                      error);
1014                         return error;
1015                 }
1016         }
1017         return 0;
1018 }
1019
1020 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1021                                        int ref_change)
1022 {
1023         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1024         struct ext4_iloc iloc;
1025         s64 ref_count;
1026         u32 hash;
1027         int ret;
1028
1029         inode_lock(ea_inode);
1030
1031         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1032         if (ret) {
1033                 iloc.bh = NULL;
1034                 goto out;
1035         }
1036
1037         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1038         ref_count += ref_change;
1039         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1040
1041         if (ref_change > 0) {
1042                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1043                           ea_inode->i_ino, ref_count);
1044
1045                 if (ref_count == 1) {
1046                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1047                                   ea_inode->i_ino, ea_inode->i_nlink);
1048
1049                         set_nlink(ea_inode, 1);
1050                         ext4_orphan_del(handle, ea_inode);
1051
1052                         if (ea_inode_cache) {
1053                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1054                                 mb_cache_entry_create(ea_inode_cache,
1055                                                       GFP_NOFS, hash,
1056                                                       ea_inode->i_ino,
1057                                                       true /* reusable */);
1058                         }
1059                 }
1060         } else {
1061                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1062                           ea_inode->i_ino, ref_count);
1063
1064                 if (ref_count == 0) {
1065                         WARN_ONCE(ea_inode->i_nlink != 1,
1066                                   "EA inode %lu i_nlink=%u",
1067                                   ea_inode->i_ino, ea_inode->i_nlink);
1068
1069                         clear_nlink(ea_inode);
1070                         ext4_orphan_add(handle, ea_inode);
1071
1072                         if (ea_inode_cache) {
1073                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1074                                 mb_cache_entry_delete(ea_inode_cache, hash,
1075                                                       ea_inode->i_ino);
1076                         }
1077                 }
1078         }
1079
1080         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1081         iloc.bh = NULL;
1082         if (ret)
1083                 ext4_warning_inode(ea_inode,
1084                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1085 out:
1086         brelse(iloc.bh);
1087         inode_unlock(ea_inode);
1088         return ret;
1089 }
1090
1091 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1092 {
1093         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1094 }
1095
1096 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1097 {
1098         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1099 }
1100
1101 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1102                                         struct ext4_xattr_entry *first)
1103 {
1104         struct inode *ea_inode;
1105         struct ext4_xattr_entry *entry;
1106         struct ext4_xattr_entry *failed_entry;
1107         unsigned int ea_ino;
1108         int err, saved_err;
1109
1110         for (entry = first; !IS_LAST_ENTRY(entry);
1111              entry = EXT4_XATTR_NEXT(entry)) {
1112                 if (!entry->e_value_inum)
1113                         continue;
1114                 ea_ino = le32_to_cpu(entry->e_value_inum);
1115                 err = ext4_xattr_inode_iget(parent, ea_ino,
1116                                             le32_to_cpu(entry->e_hash),
1117                                             &ea_inode);
1118                 if (err)
1119                         goto cleanup;
1120                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1121                 if (err) {
1122                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1123                         iput(ea_inode);
1124                         goto cleanup;
1125                 }
1126                 iput(ea_inode);
1127         }
1128         return 0;
1129
1130 cleanup:
1131         saved_err = err;
1132         failed_entry = entry;
1133
1134         for (entry = first; entry != failed_entry;
1135              entry = EXT4_XATTR_NEXT(entry)) {
1136                 if (!entry->e_value_inum)
1137                         continue;
1138                 ea_ino = le32_to_cpu(entry->e_value_inum);
1139                 err = ext4_xattr_inode_iget(parent, ea_ino,
1140                                             le32_to_cpu(entry->e_hash),
1141                                             &ea_inode);
1142                 if (err) {
1143                         ext4_warning(parent->i_sb,
1144                                      "cleanup ea_ino %u iget error %d", ea_ino,
1145                                      err);
1146                         continue;
1147                 }
1148                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1149                 if (err)
1150                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1151                                            err);
1152                 iput(ea_inode);
1153         }
1154         return saved_err;
1155 }
1156
1157 static void
1158 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1159                              struct buffer_head *bh,
1160                              struct ext4_xattr_entry *first, bool block_csum,
1161                              struct ext4_xattr_inode_array **ea_inode_array,
1162                              int extra_credits, bool skip_quota)
1163 {
1164         struct inode *ea_inode;
1165         struct ext4_xattr_entry *entry;
1166         bool dirty = false;
1167         unsigned int ea_ino;
1168         int err;
1169         int credits;
1170
1171         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1172         credits = 2 + extra_credits;
1173
1174         for (entry = first; !IS_LAST_ENTRY(entry);
1175              entry = EXT4_XATTR_NEXT(entry)) {
1176                 if (!entry->e_value_inum)
1177                         continue;
1178                 ea_ino = le32_to_cpu(entry->e_value_inum);
1179                 err = ext4_xattr_inode_iget(parent, ea_ino,
1180                                             le32_to_cpu(entry->e_hash),
1181                                             &ea_inode);
1182                 if (err)
1183                         continue;
1184
1185                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1186                 if (err) {
1187                         ext4_warning_inode(ea_inode,
1188                                            "Expand inode array err=%d", err);
1189                         iput(ea_inode);
1190                         continue;
1191                 }
1192
1193                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1194                                                 dirty, block_csum);
1195                 if (err) {
1196                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1197                                            err);
1198                         continue;
1199                 }
1200
1201                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1202                 if (err) {
1203                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1204                                            err);
1205                         continue;
1206                 }
1207
1208                 if (!skip_quota)
1209                         ext4_xattr_inode_free_quota(parent, ea_inode,
1210                                               le32_to_cpu(entry->e_value_size));
1211
1212                 /*
1213                  * Forget about ea_inode within the same transaction that
1214                  * decrements the ref count. This avoids duplicate decrements in
1215                  * case the rest of the work spills over to subsequent
1216                  * transactions.
1217                  */
1218                 entry->e_value_inum = 0;
1219                 entry->e_value_size = 0;
1220
1221                 dirty = true;
1222         }
1223
1224         if (dirty) {
1225                 /*
1226                  * Note that we are deliberately skipping csum calculation for
1227                  * the final update because we do not expect any journal
1228                  * restarts until xattr block is freed.
1229                  */
1230
1231                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1232                 if (err)
1233                         ext4_warning_inode(parent,
1234                                            "handle dirty metadata err=%d", err);
1235         }
1236 }
1237
1238 /*
1239  * Release the xattr block BH: If the reference count is > 1, decrement it;
1240  * otherwise free the block.
1241  */
1242 static void
1243 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1244                          struct buffer_head *bh,
1245                          struct ext4_xattr_inode_array **ea_inode_array,
1246                          int extra_credits)
1247 {
1248         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1249         u32 hash, ref;
1250         int error = 0;
1251
1252         BUFFER_TRACE(bh, "get_write_access");
1253         error = ext4_journal_get_write_access(handle, bh);
1254         if (error)
1255                 goto out;
1256
1257         lock_buffer(bh);
1258         hash = le32_to_cpu(BHDR(bh)->h_hash);
1259         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1260         if (ref == 1) {
1261                 ea_bdebug(bh, "refcount now=0; freeing");
1262                 /*
1263                  * This must happen under buffer lock for
1264                  * ext4_xattr_block_set() to reliably detect freed block
1265                  */
1266                 if (ea_block_cache)
1267                         mb_cache_entry_delete(ea_block_cache, hash,
1268                                               bh->b_blocknr);
1269                 get_bh(bh);
1270                 unlock_buffer(bh);
1271
1272                 if (ext4_has_feature_ea_inode(inode->i_sb))
1273                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1274                                                      BFIRST(bh),
1275                                                      true /* block_csum */,
1276                                                      ea_inode_array,
1277                                                      extra_credits,
1278                                                      true /* skip_quota */);
1279                 ext4_free_blocks(handle, inode, bh, 0, 1,
1280                                  EXT4_FREE_BLOCKS_METADATA |
1281                                  EXT4_FREE_BLOCKS_FORGET);
1282         } else {
1283                 ref--;
1284                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1285                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1286                         struct mb_cache_entry *ce;
1287
1288                         if (ea_block_cache) {
1289                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1290                                                         bh->b_blocknr);
1291                                 if (ce) {
1292                                         ce->e_reusable = 1;
1293                                         mb_cache_entry_put(ea_block_cache, ce);
1294                                 }
1295                         }
1296                 }
1297
1298                 ext4_xattr_block_csum_set(inode, bh);
1299                 /*
1300                  * Beware of this ugliness: Releasing of xattr block references
1301                  * from different inodes can race and so we have to protect
1302                  * from a race where someone else frees the block (and releases
1303                  * its journal_head) before we are done dirtying the buffer. In
1304                  * nojournal mode this race is harmless and we actually cannot
1305                  * call ext4_handle_dirty_metadata() with locked buffer as
1306                  * that function can call sync_dirty_buffer() so for that case
1307                  * we handle the dirtying after unlocking the buffer.
1308                  */
1309                 if (ext4_handle_valid(handle))
1310                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1311                 unlock_buffer(bh);
1312                 if (!ext4_handle_valid(handle))
1313                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1314                 if (IS_SYNC(inode))
1315                         ext4_handle_sync(handle);
1316                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1317                 ea_bdebug(bh, "refcount now=%d; releasing",
1318                           le32_to_cpu(BHDR(bh)->h_refcount));
1319         }
1320 out:
1321         ext4_std_error(inode->i_sb, error);
1322         return;
1323 }
1324
1325 /*
1326  * Find the available free space for EAs. This also returns the total number of
1327  * bytes used by EA entries.
1328  */
1329 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1330                                     size_t *min_offs, void *base, int *total)
1331 {
1332         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1333                 if (!last->e_value_inum && last->e_value_size) {
1334                         size_t offs = le16_to_cpu(last->e_value_offs);
1335                         if (offs < *min_offs)
1336                                 *min_offs = offs;
1337                 }
1338                 if (total)
1339                         *total += EXT4_XATTR_LEN(last->e_name_len);
1340         }
1341         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1342 }
1343
1344 /*
1345  * Write the value of the EA in an inode.
1346  */
1347 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1348                                   const void *buf, int bufsize)
1349 {
1350         struct buffer_head *bh = NULL;
1351         unsigned long block = 0;
1352         int blocksize = ea_inode->i_sb->s_blocksize;
1353         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1354         int csize, wsize = 0;
1355         int ret = 0;
1356         int retries = 0;
1357
1358 retry:
1359         while (ret >= 0 && ret < max_blocks) {
1360                 struct ext4_map_blocks map;
1361                 map.m_lblk = block += ret;
1362                 map.m_len = max_blocks -= ret;
1363
1364                 ret = ext4_map_blocks(handle, ea_inode, &map,
1365                                       EXT4_GET_BLOCKS_CREATE);
1366                 if (ret <= 0) {
1367                         ext4_mark_inode_dirty(handle, ea_inode);
1368                         if (ret == -ENOSPC &&
1369                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1370                                 ret = 0;
1371                                 goto retry;
1372                         }
1373                         break;
1374                 }
1375         }
1376
1377         if (ret < 0)
1378                 return ret;
1379
1380         block = 0;
1381         while (wsize < bufsize) {
1382                 if (bh != NULL)
1383                         brelse(bh);
1384                 csize = (bufsize - wsize) > blocksize ? blocksize :
1385                                                                 bufsize - wsize;
1386                 bh = ext4_getblk(handle, ea_inode, block, 0);
1387                 if (IS_ERR(bh))
1388                         return PTR_ERR(bh);
1389                 ret = ext4_journal_get_write_access(handle, bh);
1390                 if (ret)
1391                         goto out;
1392
1393                 memcpy(bh->b_data, buf, csize);
1394                 set_buffer_uptodate(bh);
1395                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1396
1397                 buf += csize;
1398                 wsize += csize;
1399                 block += 1;
1400         }
1401
1402         inode_lock(ea_inode);
1403         i_size_write(ea_inode, wsize);
1404         ext4_update_i_disksize(ea_inode, wsize);
1405         inode_unlock(ea_inode);
1406
1407         ext4_mark_inode_dirty(handle, ea_inode);
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         /*
1426          * Let the next inode be the goal, so we try and allocate the EA inode
1427          * in the same group, or nearby one.
1428          */
1429         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1430                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1431                                   EXT4_EA_INODE_FL);
1432         if (!IS_ERR(ea_inode)) {
1433                 ea_inode->i_op = &ext4_file_inode_operations;
1434                 ea_inode->i_fop = &ext4_file_operations;
1435                 ext4_set_aops(ea_inode);
1436                 ext4_xattr_inode_set_class(ea_inode);
1437                 unlock_new_inode(ea_inode);
1438                 ext4_xattr_inode_set_ref(ea_inode, 1);
1439                 ext4_xattr_inode_set_hash(ea_inode, hash);
1440                 err = ext4_mark_inode_dirty(handle, ea_inode);
1441                 if (!err)
1442                         err = ext4_inode_attach_jinode(ea_inode);
1443                 if (err) {
1444                         iput(ea_inode);
1445                         return ERR_PTR(err);
1446                 }
1447
1448                 /*
1449                  * Xattr inodes are shared therefore quota charging is performed
1450                  * at a higher level.
1451                  */
1452                 dquot_free_inode(ea_inode);
1453                 dquot_drop(ea_inode);
1454                 inode_lock(ea_inode);
1455                 ea_inode->i_flags |= S_NOQUOTA;
1456                 inode_unlock(ea_inode);
1457         }
1458
1459         return ea_inode;
1460 }
1461
1462 static struct inode *
1463 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1464                             size_t value_len, u32 hash)
1465 {
1466         struct inode *ea_inode;
1467         struct mb_cache_entry *ce;
1468         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1469         void *ea_data;
1470
1471         if (!ea_inode_cache)
1472                 return NULL;
1473
1474         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1475         if (!ce)
1476                 return NULL;
1477
1478         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1479         if (!ea_data) {
1480                 mb_cache_entry_put(ea_inode_cache, ce);
1481                 return NULL;
1482         }
1483
1484         while (ce) {
1485                 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1486                 if (!IS_ERR(ea_inode) &&
1487                     !is_bad_inode(ea_inode) &&
1488                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1489                     i_size_read(ea_inode) == value_len &&
1490                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1491                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1492                                                     value_len) &&
1493                     !memcmp(value, ea_data, value_len)) {
1494                         mb_cache_entry_touch(ea_inode_cache, ce);
1495                         mb_cache_entry_put(ea_inode_cache, ce);
1496                         kvfree(ea_data);
1497                         return ea_inode;
1498                 }
1499
1500                 if (!IS_ERR(ea_inode))
1501                         iput(ea_inode);
1502                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1503         }
1504         kvfree(ea_data);
1505         return NULL;
1506 }
1507
1508 /*
1509  * Add value of the EA in an inode.
1510  */
1511 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1512                                           const void *value, size_t value_len,
1513                                           struct inode **ret_inode)
1514 {
1515         struct inode *ea_inode;
1516         u32 hash;
1517         int err;
1518
1519         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1520         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1521         if (ea_inode) {
1522                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1523                 if (err) {
1524                         iput(ea_inode);
1525                         return err;
1526                 }
1527
1528                 *ret_inode = ea_inode;
1529                 return 0;
1530         }
1531
1532         /* Create an inode for the EA value */
1533         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1534         if (IS_ERR(ea_inode))
1535                 return PTR_ERR(ea_inode);
1536
1537         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1538         if (err) {
1539                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1540                 iput(ea_inode);
1541                 return err;
1542         }
1543
1544         if (EA_INODE_CACHE(inode))
1545                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1546                                       ea_inode->i_ino, true /* reusable */);
1547
1548         *ret_inode = ea_inode;
1549         return 0;
1550 }
1551
1552 /*
1553  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1554  * feature is enabled.
1555  */
1556 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1557
1558 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1559                                 struct ext4_xattr_search *s,
1560                                 handle_t *handle, struct inode *inode,
1561                                 bool is_block)
1562 {
1563         struct ext4_xattr_entry *last, *next;
1564         struct ext4_xattr_entry *here = s->here;
1565         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1566         int in_inode = i->in_inode;
1567         struct inode *old_ea_inode = NULL;
1568         struct inode *new_ea_inode = NULL;
1569         size_t old_size, new_size;
1570         int ret;
1571
1572         /* Space used by old and new values. */
1573         old_size = (!s->not_found && !here->e_value_inum) ?
1574                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1575         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1576
1577         /*
1578          * Optimization for the simple case when old and new values have the
1579          * same padded sizes. Not applicable if external inodes are involved.
1580          */
1581         if (new_size && new_size == old_size) {
1582                 size_t offs = le16_to_cpu(here->e_value_offs);
1583                 void *val = s->base + offs;
1584
1585                 here->e_value_size = cpu_to_le32(i->value_len);
1586                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1587                         memset(val, 0, new_size);
1588                 } else {
1589                         memcpy(val, i->value, i->value_len);
1590                         /* Clear padding bytes. */
1591                         memset(val + i->value_len, 0, new_size - i->value_len);
1592                 }
1593                 goto update_hash;
1594         }
1595
1596         /* Compute min_offs and last. */
1597         last = s->first;
1598         for (; !IS_LAST_ENTRY(last); last = next) {
1599                 next = EXT4_XATTR_NEXT(last);
1600                 if ((void *)next >= s->end) {
1601                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1602                         ret = -EFSCORRUPTED;
1603                         goto out;
1604                 }
1605                 if (!last->e_value_inum && last->e_value_size) {
1606                         size_t offs = le16_to_cpu(last->e_value_offs);
1607                         if (offs < min_offs)
1608                                 min_offs = offs;
1609                 }
1610         }
1611
1612         /* Check whether we have enough space. */
1613         if (i->value) {
1614                 size_t free;
1615
1616                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1617                 if (!s->not_found)
1618                         free += EXT4_XATTR_LEN(name_len) + old_size;
1619
1620                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1621                         ret = -ENOSPC;
1622                         goto out;
1623                 }
1624
1625                 /*
1626                  * If storing the value in an external inode is an option,
1627                  * reserve space for xattr entries/names in the external
1628                  * attribute block so that a long value does not occupy the
1629                  * whole space and prevent futher entries being added.
1630                  */
1631                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1632                     new_size && is_block &&
1633                     (min_offs + old_size - new_size) <
1634                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1635                         ret = -ENOSPC;
1636                         goto out;
1637                 }
1638         }
1639
1640         /*
1641          * Getting access to old and new ea inodes is subject to failures.
1642          * Finish that work before doing any modifications to the xattr data.
1643          */
1644         if (!s->not_found && here->e_value_inum) {
1645                 ret = ext4_xattr_inode_iget(inode,
1646                                             le32_to_cpu(here->e_value_inum),
1647                                             le32_to_cpu(here->e_hash),
1648                                             &old_ea_inode);
1649                 if (ret) {
1650                         old_ea_inode = NULL;
1651                         goto out;
1652                 }
1653         }
1654         if (i->value && in_inode) {
1655                 WARN_ON_ONCE(!i->value_len);
1656
1657                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1658                 if (ret)
1659                         goto out;
1660
1661                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1662                                                      i->value_len,
1663                                                      &new_ea_inode);
1664                 if (ret) {
1665                         new_ea_inode = NULL;
1666                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1667                         goto out;
1668                 }
1669         }
1670
1671         if (old_ea_inode) {
1672                 /* We are ready to release ref count on the old_ea_inode. */
1673                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1674                 if (ret) {
1675                         /* Release newly required ref count on new_ea_inode. */
1676                         if (new_ea_inode) {
1677                                 int err;
1678
1679                                 err = ext4_xattr_inode_dec_ref(handle,
1680                                                                new_ea_inode);
1681                                 if (err)
1682                                         ext4_warning_inode(new_ea_inode,
1683                                                   "dec ref new_ea_inode err=%d",
1684                                                   err);
1685                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1686                                                             i->value_len);
1687                         }
1688                         goto out;
1689                 }
1690
1691                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1692                                             le32_to_cpu(here->e_value_size));
1693         }
1694
1695         /* No failures allowed past this point. */
1696
1697         if (!s->not_found && here->e_value_size && here->e_value_offs) {
1698                 /* Remove the old value. */
1699                 void *first_val = s->base + min_offs;
1700                 size_t offs = le16_to_cpu(here->e_value_offs);
1701                 void *val = s->base + offs;
1702
1703                 memmove(first_val + old_size, first_val, val - first_val);
1704                 memset(first_val, 0, old_size);
1705                 min_offs += old_size;
1706
1707                 /* Adjust all value offsets. */
1708                 last = s->first;
1709                 while (!IS_LAST_ENTRY(last)) {
1710                         size_t o = le16_to_cpu(last->e_value_offs);
1711
1712                         if (!last->e_value_inum &&
1713                             last->e_value_size && o < offs)
1714                                 last->e_value_offs = cpu_to_le16(o + old_size);
1715                         last = EXT4_XATTR_NEXT(last);
1716                 }
1717         }
1718
1719         if (!i->value) {
1720                 /* Remove old name. */
1721                 size_t size = EXT4_XATTR_LEN(name_len);
1722
1723                 last = ENTRY((void *)last - size);
1724                 memmove(here, (void *)here + size,
1725                         (void *)last - (void *)here + sizeof(__u32));
1726                 memset(last, 0, size);
1727         } else if (s->not_found) {
1728                 /* Insert new name. */
1729                 size_t size = EXT4_XATTR_LEN(name_len);
1730                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1731
1732                 memmove((void *)here + size, here, rest);
1733                 memset(here, 0, size);
1734                 here->e_name_index = i->name_index;
1735                 here->e_name_len = name_len;
1736                 memcpy(here->e_name, i->name, name_len);
1737         } else {
1738                 /* This is an update, reset value info. */
1739                 here->e_value_inum = 0;
1740                 here->e_value_offs = 0;
1741                 here->e_value_size = 0;
1742         }
1743
1744         if (i->value) {
1745                 /* Insert new value. */
1746                 if (in_inode) {
1747                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1748                 } else if (i->value_len) {
1749                         void *val = s->base + min_offs - new_size;
1750
1751                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1752                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1753                                 memset(val, 0, new_size);
1754                         } else {
1755                                 memcpy(val, i->value, i->value_len);
1756                                 /* Clear padding bytes. */
1757                                 memset(val + i->value_len, 0,
1758                                        new_size - i->value_len);
1759                         }
1760                 }
1761                 here->e_value_size = cpu_to_le32(i->value_len);
1762         }
1763
1764 update_hash:
1765         if (i->value) {
1766                 __le32 hash = 0;
1767
1768                 /* Entry hash calculation. */
1769                 if (in_inode) {
1770                         __le32 crc32c_hash;
1771
1772                         /*
1773                          * Feed crc32c hash instead of the raw value for entry
1774                          * hash calculation. This is to avoid walking
1775                          * potentially long value buffer again.
1776                          */
1777                         crc32c_hash = cpu_to_le32(
1778                                        ext4_xattr_inode_get_hash(new_ea_inode));
1779                         hash = ext4_xattr_hash_entry(here->e_name,
1780                                                      here->e_name_len,
1781                                                      &crc32c_hash, 1);
1782                 } else if (is_block) {
1783                         __le32 *value = s->base + le16_to_cpu(
1784                                                         here->e_value_offs);
1785
1786                         hash = ext4_xattr_hash_entry(here->e_name,
1787                                                      here->e_name_len, value,
1788                                                      new_size >> 2);
1789                 }
1790                 here->e_hash = hash;
1791         }
1792
1793         if (is_block)
1794                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1795
1796         ret = 0;
1797 out:
1798         iput(old_ea_inode);
1799         iput(new_ea_inode);
1800         return ret;
1801 }
1802
1803 struct ext4_xattr_block_find {
1804         struct ext4_xattr_search s;
1805         struct buffer_head *bh;
1806 };
1807
1808 static int
1809 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1810                       struct ext4_xattr_block_find *bs)
1811 {
1812         struct super_block *sb = inode->i_sb;
1813         int error;
1814
1815         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1816                   i->name_index, i->name, i->value, (long)i->value_len);
1817
1818         if (EXT4_I(inode)->i_file_acl) {
1819                 /* The inode already has an extended attribute block. */
1820                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1821                 error = -EIO;
1822                 if (!bs->bh)
1823                         goto cleanup;
1824                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1825                         atomic_read(&(bs->bh->b_count)),
1826                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1827                 error = ext4_xattr_check_block(inode, bs->bh);
1828                 if (error)
1829                         goto cleanup;
1830                 /* Find the named attribute. */
1831                 bs->s.base = BHDR(bs->bh);
1832                 bs->s.first = BFIRST(bs->bh);
1833                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1834                 bs->s.here = bs->s.first;
1835                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1836                                          i->name_index, i->name, 1);
1837                 if (error && error != -ENODATA)
1838                         goto cleanup;
1839                 bs->s.not_found = error;
1840         }
1841         error = 0;
1842
1843 cleanup:
1844         return error;
1845 }
1846
1847 static int
1848 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1849                      struct ext4_xattr_info *i,
1850                      struct ext4_xattr_block_find *bs)
1851 {
1852         struct super_block *sb = inode->i_sb;
1853         struct buffer_head *new_bh = NULL;
1854         struct ext4_xattr_search s_copy = bs->s;
1855         struct ext4_xattr_search *s = &s_copy;
1856         struct mb_cache_entry *ce = NULL;
1857         int error = 0;
1858         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1859         struct inode *ea_inode = NULL, *tmp_inode;
1860         size_t old_ea_inode_quota = 0;
1861         unsigned int ea_ino;
1862
1863
1864 #define header(x) ((struct ext4_xattr_header *)(x))
1865
1866         if (s->base) {
1867                 BUFFER_TRACE(bs->bh, "get_write_access");
1868                 error = ext4_journal_get_write_access(handle, bs->bh);
1869                 if (error)
1870                         goto cleanup;
1871                 lock_buffer(bs->bh);
1872
1873                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1874                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1875
1876                         /*
1877                          * This must happen under buffer lock for
1878                          * ext4_xattr_block_set() to reliably detect modified
1879                          * block
1880                          */
1881                         if (ea_block_cache)
1882                                 mb_cache_entry_delete(ea_block_cache, hash,
1883                                                       bs->bh->b_blocknr);
1884                         ea_bdebug(bs->bh, "modifying in-place");
1885                         error = ext4_xattr_set_entry(i, s, handle, inode,
1886                                                      true /* is_block */);
1887                         ext4_xattr_block_csum_set(inode, bs->bh);
1888                         unlock_buffer(bs->bh);
1889                         if (error == -EFSCORRUPTED)
1890                                 goto bad_block;
1891                         if (!error)
1892                                 error = ext4_handle_dirty_metadata(handle,
1893                                                                    inode,
1894                                                                    bs->bh);
1895                         if (error)
1896                                 goto cleanup;
1897                         goto inserted;
1898                 } else {
1899                         int offset = (char *)s->here - bs->bh->b_data;
1900
1901                         unlock_buffer(bs->bh);
1902                         ea_bdebug(bs->bh, "cloning");
1903                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1904                         error = -ENOMEM;
1905                         if (s->base == NULL)
1906                                 goto cleanup;
1907                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1908                         s->first = ENTRY(header(s->base)+1);
1909                         header(s->base)->h_refcount = cpu_to_le32(1);
1910                         s->here = ENTRY(s->base + offset);
1911                         s->end = s->base + bs->bh->b_size;
1912
1913                         /*
1914                          * If existing entry points to an xattr inode, we need
1915                          * to prevent ext4_xattr_set_entry() from decrementing
1916                          * ref count on it because the reference belongs to the
1917                          * original block. In this case, make the entry look
1918                          * like it has an empty value.
1919                          */
1920                         if (!s->not_found && s->here->e_value_inum) {
1921                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1922                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1923                                               le32_to_cpu(s->here->e_hash),
1924                                               &tmp_inode);
1925                                 if (error)
1926                                         goto cleanup;
1927
1928                                 if (!ext4_test_inode_state(tmp_inode,
1929                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1930                                         /*
1931                                          * Defer quota free call for previous
1932                                          * inode until success is guaranteed.
1933                                          */
1934                                         old_ea_inode_quota = le32_to_cpu(
1935                                                         s->here->e_value_size);
1936                                 }
1937                                 iput(tmp_inode);
1938
1939                                 s->here->e_value_inum = 0;
1940                                 s->here->e_value_size = 0;
1941                         }
1942                 }
1943         } else {
1944                 /* Allocate a buffer where we construct the new block. */
1945                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1946                 /* assert(header == s->base) */
1947                 error = -ENOMEM;
1948                 if (s->base == NULL)
1949                         goto cleanup;
1950                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1951                 header(s->base)->h_blocks = cpu_to_le32(1);
1952                 header(s->base)->h_refcount = cpu_to_le32(1);
1953                 s->first = ENTRY(header(s->base)+1);
1954                 s->here = ENTRY(header(s->base)+1);
1955                 s->end = s->base + sb->s_blocksize;
1956         }
1957
1958         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1959         if (error == -EFSCORRUPTED)
1960                 goto bad_block;
1961         if (error)
1962                 goto cleanup;
1963
1964         if (i->value && s->here->e_value_inum) {
1965                 /*
1966                  * A ref count on ea_inode has been taken as part of the call to
1967                  * ext4_xattr_set_entry() above. We would like to drop this
1968                  * extra ref but we have to wait until the xattr block is
1969                  * initialized and has its own ref count on the ea_inode.
1970                  */
1971                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1972                 error = ext4_xattr_inode_iget(inode, ea_ino,
1973                                               le32_to_cpu(s->here->e_hash),
1974                                               &ea_inode);
1975                 if (error) {
1976                         ea_inode = NULL;
1977                         goto cleanup;
1978                 }
1979         }
1980
1981 inserted:
1982         if (!IS_LAST_ENTRY(s->first)) {
1983                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1984                                                      &ce);
1985                 if (new_bh) {
1986                         /* We found an identical block in the cache. */
1987                         if (new_bh == bs->bh)
1988                                 ea_bdebug(new_bh, "keeping");
1989                         else {
1990                                 u32 ref;
1991
1992                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1993
1994                                 /* The old block is released after updating
1995                                    the inode. */
1996                                 error = dquot_alloc_block(inode,
1997                                                 EXT4_C2B(EXT4_SB(sb), 1));
1998                                 if (error)
1999                                         goto cleanup;
2000                                 BUFFER_TRACE(new_bh, "get_write_access");
2001                                 error = ext4_journal_get_write_access(handle,
2002                                                                       new_bh);
2003                                 if (error)
2004                                         goto cleanup_dquot;
2005                                 lock_buffer(new_bh);
2006                                 /*
2007                                  * We have to be careful about races with
2008                                  * freeing, rehashing or adding references to
2009                                  * xattr block. Once we hold buffer lock xattr
2010                                  * block's state is stable so we can check
2011                                  * whether the block got freed / rehashed or
2012                                  * not.  Since we unhash mbcache entry under
2013                                  * buffer lock when freeing / rehashing xattr
2014                                  * block, checking whether entry is still
2015                                  * hashed is reliable. Same rules hold for
2016                                  * e_reusable handling.
2017                                  */
2018                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2019                                     !ce->e_reusable) {
2020                                         /*
2021                                          * Undo everything and check mbcache
2022                                          * again.
2023                                          */
2024                                         unlock_buffer(new_bh);
2025                                         dquot_free_block(inode,
2026                                                          EXT4_C2B(EXT4_SB(sb),
2027                                                                   1));
2028                                         brelse(new_bh);
2029                                         mb_cache_entry_put(ea_block_cache, ce);
2030                                         ce = NULL;
2031                                         new_bh = NULL;
2032                                         goto inserted;
2033                                 }
2034                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2035                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2036                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2037                                         ce->e_reusable = 0;
2038                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2039                                           ref);
2040                                 ext4_xattr_block_csum_set(inode, new_bh);
2041                                 unlock_buffer(new_bh);
2042                                 error = ext4_handle_dirty_metadata(handle,
2043                                                                    inode,
2044                                                                    new_bh);
2045                                 if (error)
2046                                         goto cleanup_dquot;
2047                         }
2048                         mb_cache_entry_touch(ea_block_cache, ce);
2049                         mb_cache_entry_put(ea_block_cache, ce);
2050                         ce = NULL;
2051                 } else if (bs->bh && s->base == bs->bh->b_data) {
2052                         /* We were modifying this block in-place. */
2053                         ea_bdebug(bs->bh, "keeping this block");
2054                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2055                         new_bh = bs->bh;
2056                         get_bh(new_bh);
2057                 } else {
2058                         /* We need to allocate a new block */
2059                         ext4_fsblk_t goal, block;
2060
2061                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2062
2063                         goal = ext4_group_first_block_no(sb,
2064                                                 EXT4_I(inode)->i_block_group);
2065
2066                         /* non-extent files can't have physical blocks past 2^32 */
2067                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2068                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2069
2070                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2071                                                      NULL, &error);
2072                         if (error)
2073                                 goto cleanup;
2074
2075                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2076                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2077
2078                         ea_idebug(inode, "creating block %llu",
2079                                   (unsigned long long)block);
2080
2081                         new_bh = sb_getblk(sb, block);
2082                         if (unlikely(!new_bh)) {
2083                                 error = -ENOMEM;
2084 getblk_failed:
2085                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2086                                                  EXT4_FREE_BLOCKS_METADATA);
2087                                 goto cleanup;
2088                         }
2089                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2090                                                       ENTRY(header(s->base)+1));
2091                         if (error)
2092                                 goto getblk_failed;
2093                         if (ea_inode) {
2094                                 /* Drop the extra ref on ea_inode. */
2095                                 error = ext4_xattr_inode_dec_ref(handle,
2096                                                                  ea_inode);
2097                                 if (error)
2098                                         ext4_warning_inode(ea_inode,
2099                                                            "dec ref error=%d",
2100                                                            error);
2101                                 iput(ea_inode);
2102                                 ea_inode = NULL;
2103                         }
2104
2105                         lock_buffer(new_bh);
2106                         error = ext4_journal_get_create_access(handle, new_bh);
2107                         if (error) {
2108                                 unlock_buffer(new_bh);
2109                                 error = -EIO;
2110                                 goto getblk_failed;
2111                         }
2112                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2113                         ext4_xattr_block_csum_set(inode, new_bh);
2114                         set_buffer_uptodate(new_bh);
2115                         unlock_buffer(new_bh);
2116                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2117                         error = ext4_handle_dirty_metadata(handle, inode,
2118                                                            new_bh);
2119                         if (error)
2120                                 goto cleanup;
2121                 }
2122         }
2123
2124         if (old_ea_inode_quota)
2125                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2126
2127         /* Update the inode. */
2128         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2129
2130         /* Drop the previous xattr block. */
2131         if (bs->bh && bs->bh != new_bh) {
2132                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2133
2134                 ext4_xattr_release_block(handle, inode, bs->bh,
2135                                          &ea_inode_array,
2136                                          0 /* extra_credits */);
2137                 ext4_xattr_inode_array_free(ea_inode_array);
2138         }
2139         error = 0;
2140
2141 cleanup:
2142         if (ea_inode) {
2143                 int error2;
2144
2145                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2146                 if (error2)
2147                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2148                                            error2);
2149
2150                 /* If there was an error, revert the quota charge. */
2151                 if (error)
2152                         ext4_xattr_inode_free_quota(inode, ea_inode,
2153                                                     i_size_read(ea_inode));
2154                 iput(ea_inode);
2155         }
2156         if (ce)
2157                 mb_cache_entry_put(ea_block_cache, ce);
2158         brelse(new_bh);
2159         if (!(bs->bh && s->base == bs->bh->b_data))
2160                 kfree(s->base);
2161
2162         return error;
2163
2164 cleanup_dquot:
2165         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2166         goto cleanup;
2167
2168 bad_block:
2169         EXT4_ERROR_INODE(inode, "bad block %llu",
2170                          EXT4_I(inode)->i_file_acl);
2171         goto cleanup;
2172
2173 #undef header
2174 }
2175
2176 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2177                           struct ext4_xattr_ibody_find *is)
2178 {
2179         struct ext4_xattr_ibody_header *header;
2180         struct ext4_inode *raw_inode;
2181         int error;
2182
2183         if (EXT4_I(inode)->i_extra_isize == 0)
2184                 return 0;
2185         raw_inode = ext4_raw_inode(&is->iloc);
2186         header = IHDR(inode, raw_inode);
2187         is->s.base = is->s.first = IFIRST(header);
2188         is->s.here = is->s.first;
2189         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2190         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2191                 error = xattr_check_inode(inode, header, is->s.end);
2192                 if (error)
2193                         return error;
2194                 /* Find the named attribute. */
2195                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2196                                          i->name_index, i->name, 0);
2197                 if (error && error != -ENODATA)
2198                         return error;
2199                 is->s.not_found = error;
2200         }
2201         return 0;
2202 }
2203
2204 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2205                                 struct ext4_xattr_info *i,
2206                                 struct ext4_xattr_ibody_find *is)
2207 {
2208         struct ext4_xattr_ibody_header *header;
2209         struct ext4_xattr_search *s = &is->s;
2210         int error;
2211
2212         if (EXT4_I(inode)->i_extra_isize == 0)
2213                 return -ENOSPC;
2214         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2215         if (error)
2216                 return error;
2217         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2218         if (!IS_LAST_ENTRY(s->first)) {
2219                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2220                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2221         } else {
2222                 header->h_magic = cpu_to_le32(0);
2223                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2224         }
2225         return 0;
2226 }
2227
2228 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2229                                 struct ext4_xattr_info *i,
2230                                 struct ext4_xattr_ibody_find *is)
2231 {
2232         struct ext4_xattr_ibody_header *header;
2233         struct ext4_xattr_search *s = &is->s;
2234         int error;
2235
2236         if (EXT4_I(inode)->i_extra_isize == 0)
2237                 return -ENOSPC;
2238         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2239         if (error)
2240                 return error;
2241         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2242         if (!IS_LAST_ENTRY(s->first)) {
2243                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2244                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2245         } else {
2246                 header->h_magic = cpu_to_le32(0);
2247                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2248         }
2249         return 0;
2250 }
2251
2252 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2253                                  struct ext4_xattr_info *i)
2254 {
2255         void *value;
2256
2257         /* When e_value_inum is set the value is stored externally. */
2258         if (s->here->e_value_inum)
2259                 return 0;
2260         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2261                 return 0;
2262         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2263         return !memcmp(value, i->value, i->value_len);
2264 }
2265
2266 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2267 {
2268         struct buffer_head *bh;
2269         int error;
2270
2271         if (!EXT4_I(inode)->i_file_acl)
2272                 return NULL;
2273         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2274         if (!bh)
2275                 return ERR_PTR(-EIO);
2276         error = ext4_xattr_check_block(inode, bh);
2277         if (error)
2278                 return ERR_PTR(error);
2279         return bh;
2280 }
2281
2282 /*
2283  * ext4_xattr_set_handle()
2284  *
2285  * Create, replace or remove an extended attribute for this inode.  Value
2286  * is NULL to remove an existing extended attribute, and non-NULL to
2287  * either replace an existing extended attribute, or create a new extended
2288  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2289  * specify that an extended attribute must exist and must not exist
2290  * previous to the call, respectively.
2291  *
2292  * Returns 0, or a negative error number on failure.
2293  */
2294 int
2295 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2296                       const char *name, const void *value, size_t value_len,
2297                       int flags)
2298 {
2299         struct ext4_xattr_info i = {
2300                 .name_index = name_index,
2301                 .name = name,
2302                 .value = value,
2303                 .value_len = value_len,
2304                 .in_inode = 0,
2305         };
2306         struct ext4_xattr_ibody_find is = {
2307                 .s = { .not_found = -ENODATA, },
2308         };
2309         struct ext4_xattr_block_find bs = {
2310                 .s = { .not_found = -ENODATA, },
2311         };
2312         int no_expand;
2313         int error;
2314
2315         if (!name)
2316                 return -EINVAL;
2317         if (strlen(name) > 255)
2318                 return -ERANGE;
2319
2320         ext4_write_lock_xattr(inode, &no_expand);
2321
2322         /* Check journal credits under write lock. */
2323         if (ext4_handle_valid(handle)) {
2324                 struct buffer_head *bh;
2325                 int credits;
2326
2327                 bh = ext4_xattr_get_block(inode);
2328                 if (IS_ERR(bh)) {
2329                         error = PTR_ERR(bh);
2330                         goto cleanup;
2331                 }
2332
2333                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2334                                                    value_len,
2335                                                    flags & XATTR_CREATE);
2336                 brelse(bh);
2337
2338                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2339                         error = -ENOSPC;
2340                         goto cleanup;
2341                 }
2342         }
2343
2344         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2345         if (error)
2346                 goto cleanup;
2347
2348         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2349                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2350                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2351                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2352         }
2353
2354         error = ext4_xattr_ibody_find(inode, &i, &is);
2355         if (error)
2356                 goto cleanup;
2357         if (is.s.not_found)
2358                 error = ext4_xattr_block_find(inode, &i, &bs);
2359         if (error)
2360                 goto cleanup;
2361         if (is.s.not_found && bs.s.not_found) {
2362                 error = -ENODATA;
2363                 if (flags & XATTR_REPLACE)
2364                         goto cleanup;
2365                 error = 0;
2366                 if (!value)
2367                         goto cleanup;
2368         } else {
2369                 error = -EEXIST;
2370                 if (flags & XATTR_CREATE)
2371                         goto cleanup;
2372         }
2373
2374         if (!value) {
2375                 if (!is.s.not_found)
2376                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2377                 else if (!bs.s.not_found)
2378                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2379         } else {
2380                 error = 0;
2381                 /* Xattr value did not change? Save us some work and bail out */
2382                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2383                         goto cleanup;
2384                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2385                         goto cleanup;
2386
2387                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2388                     (EXT4_XATTR_SIZE(i.value_len) >
2389                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2390                         i.in_inode = 1;
2391 retry_inode:
2392                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2393                 if (!error && !bs.s.not_found) {
2394                         i.value = NULL;
2395                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2396                 } else if (error == -ENOSPC) {
2397                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2398                                 error = ext4_xattr_block_find(inode, &i, &bs);
2399                                 if (error)
2400                                         goto cleanup;
2401                         }
2402                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2403                         if (!error && !is.s.not_found) {
2404                                 i.value = NULL;
2405                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2406                                                              &is);
2407                         } else if (error == -ENOSPC) {
2408                                 /*
2409                                  * Xattr does not fit in the block, store at
2410                                  * external inode if possible.
2411                                  */
2412                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2413                                     !i.in_inode) {
2414                                         i.in_inode = 1;
2415                                         goto retry_inode;
2416                                 }
2417                         }
2418                 }
2419         }
2420         if (!error) {
2421                 ext4_xattr_update_super_block(handle, inode->i_sb);
2422                 inode->i_ctime = current_time(inode);
2423                 if (!value)
2424                         no_expand = 0;
2425                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2426                 /*
2427                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2428                  * error != 0.
2429                  */
2430                 is.iloc.bh = NULL;
2431                 if (IS_SYNC(inode))
2432                         ext4_handle_sync(handle);
2433         }
2434
2435 cleanup:
2436         brelse(is.iloc.bh);
2437         brelse(bs.bh);
2438         ext4_write_unlock_xattr(inode, &no_expand);
2439         return error;
2440 }
2441
2442 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2443                            bool is_create, int *credits)
2444 {
2445         struct buffer_head *bh;
2446         int err;
2447
2448         *credits = 0;
2449
2450         if (!EXT4_SB(inode->i_sb)->s_journal)
2451                 return 0;
2452
2453         down_read(&EXT4_I(inode)->xattr_sem);
2454
2455         bh = ext4_xattr_get_block(inode);
2456         if (IS_ERR(bh)) {
2457                 err = PTR_ERR(bh);
2458         } else {
2459                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2460                                                     value_len, is_create);
2461                 brelse(bh);
2462                 err = 0;
2463         }
2464
2465         up_read(&EXT4_I(inode)->xattr_sem);
2466         return err;
2467 }
2468
2469 /*
2470  * ext4_xattr_set()
2471  *
2472  * Like ext4_xattr_set_handle, but start from an inode. This extended
2473  * attribute modification is a filesystem transaction by itself.
2474  *
2475  * Returns 0, or a negative error number on failure.
2476  */
2477 int
2478 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2479                const void *value, size_t value_len, int flags)
2480 {
2481         handle_t *handle;
2482         struct super_block *sb = inode->i_sb;
2483         int error, retries = 0;
2484         int credits;
2485
2486         error = dquot_initialize(inode);
2487         if (error)
2488                 return error;
2489
2490 retry:
2491         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2492                                        &credits);
2493         if (error)
2494                 return error;
2495
2496         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2497         if (IS_ERR(handle)) {
2498                 error = PTR_ERR(handle);
2499         } else {
2500                 int error2;
2501
2502                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2503                                               value, value_len, flags);
2504                 error2 = ext4_journal_stop(handle);
2505                 if (error == -ENOSPC &&
2506                     ext4_should_retry_alloc(sb, &retries))
2507                         goto retry;
2508                 if (error == 0)
2509                         error = error2;
2510         }
2511
2512         return error;
2513 }
2514
2515 /*
2516  * Shift the EA entries in the inode to create space for the increased
2517  * i_extra_isize.
2518  */
2519 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2520                                      int value_offs_shift, void *to,
2521                                      void *from, size_t n)
2522 {
2523         struct ext4_xattr_entry *last = entry;
2524         int new_offs;
2525
2526         /* We always shift xattr headers further thus offsets get lower */
2527         BUG_ON(value_offs_shift > 0);
2528
2529         /* Adjust the value offsets of the entries */
2530         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2531                 if (!last->e_value_inum && last->e_value_size) {
2532                         new_offs = le16_to_cpu(last->e_value_offs) +
2533                                                         value_offs_shift;
2534                         last->e_value_offs = cpu_to_le16(new_offs);
2535                 }
2536         }
2537         /* Shift the entries by n bytes */
2538         memmove(to, from, n);
2539 }
2540
2541 /*
2542  * Move xattr pointed to by 'entry' from inode into external xattr block
2543  */
2544 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2545                                     struct ext4_inode *raw_inode,
2546                                     struct ext4_xattr_entry *entry)
2547 {
2548         struct ext4_xattr_ibody_find *is = NULL;
2549         struct ext4_xattr_block_find *bs = NULL;
2550         char *buffer = NULL, *b_entry_name = NULL;
2551         size_t value_size = le32_to_cpu(entry->e_value_size);
2552         struct ext4_xattr_info i = {
2553                 .value = NULL,
2554                 .value_len = 0,
2555                 .name_index = entry->e_name_index,
2556                 .in_inode = !!entry->e_value_inum,
2557         };
2558         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2559         int error;
2560
2561         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2562         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2563         buffer = kmalloc(value_size, GFP_NOFS);
2564         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2565         if (!is || !bs || !buffer || !b_entry_name) {
2566                 error = -ENOMEM;
2567                 goto out;
2568         }
2569
2570         is->s.not_found = -ENODATA;
2571         bs->s.not_found = -ENODATA;
2572         is->iloc.bh = NULL;
2573         bs->bh = NULL;
2574
2575         /* Save the entry name and the entry value */
2576         if (entry->e_value_inum) {
2577                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2578                 if (error)
2579                         goto out;
2580         } else {
2581                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2582                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2583         }
2584
2585         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2586         b_entry_name[entry->e_name_len] = '\0';
2587         i.name = b_entry_name;
2588
2589         error = ext4_get_inode_loc(inode, &is->iloc);
2590         if (error)
2591                 goto out;
2592
2593         error = ext4_xattr_ibody_find(inode, &i, is);
2594         if (error)
2595                 goto out;
2596
2597         /* Remove the chosen entry from the inode */
2598         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2599         if (error)
2600                 goto out;
2601
2602         i.value = buffer;
2603         i.value_len = value_size;
2604         error = ext4_xattr_block_find(inode, &i, bs);
2605         if (error)
2606                 goto out;
2607
2608         /* Add entry which was removed from the inode into the block */
2609         error = ext4_xattr_block_set(handle, inode, &i, bs);
2610         if (error)
2611                 goto out;
2612         error = 0;
2613 out:
2614         kfree(b_entry_name);
2615         kfree(buffer);
2616         if (is)
2617                 brelse(is->iloc.bh);
2618         kfree(is);
2619         kfree(bs);
2620
2621         return error;
2622 }
2623
2624 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2625                                        struct ext4_inode *raw_inode,
2626                                        int isize_diff, size_t ifree,
2627                                        size_t bfree, int *total_ino)
2628 {
2629         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2630         struct ext4_xattr_entry *small_entry;
2631         struct ext4_xattr_entry *entry;
2632         struct ext4_xattr_entry *last;
2633         unsigned int entry_size;        /* EA entry size */
2634         unsigned int total_size;        /* EA entry size + value size */
2635         unsigned int min_total_size;
2636         int error;
2637
2638         while (isize_diff > ifree) {
2639                 entry = NULL;
2640                 small_entry = NULL;
2641                 min_total_size = ~0U;
2642                 last = IFIRST(header);
2643                 /* Find the entry best suited to be pushed into EA block */
2644                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2645                         /* never move system.data out of the inode */
2646                         if ((last->e_name_len == 4) &&
2647                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2648                             !memcmp(last->e_name, "data", 4))
2649                                 continue;
2650                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2651                         if (!last->e_value_inum)
2652                                 total_size += EXT4_XATTR_SIZE(
2653                                                le32_to_cpu(last->e_value_size));
2654                         if (total_size <= bfree &&
2655                             total_size < min_total_size) {
2656                                 if (total_size + ifree < isize_diff) {
2657                                         small_entry = last;
2658                                 } else {
2659                                         entry = last;
2660                                         min_total_size = total_size;
2661                                 }
2662                         }
2663                 }
2664
2665                 if (entry == NULL) {
2666                         if (small_entry == NULL)
2667                                 return -ENOSPC;
2668                         entry = small_entry;
2669                 }
2670
2671                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2672                 total_size = entry_size;
2673                 if (!entry->e_value_inum)
2674                         total_size += EXT4_XATTR_SIZE(
2675                                               le32_to_cpu(entry->e_value_size));
2676                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2677                                                  entry);
2678                 if (error)
2679                         return error;
2680
2681                 *total_ino -= entry_size;
2682                 ifree += total_size;
2683                 bfree -= total_size;
2684         }
2685
2686         return 0;
2687 }
2688
2689 /*
2690  * Expand an inode by new_extra_isize bytes when EAs are present.
2691  * Returns 0 on success or negative error number on failure.
2692  */
2693 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2694                                struct ext4_inode *raw_inode, handle_t *handle)
2695 {
2696         struct ext4_xattr_ibody_header *header;
2697         struct buffer_head *bh;
2698         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2699         static unsigned int mnt_count;
2700         size_t min_offs;
2701         size_t ifree, bfree;
2702         int total_ino;
2703         void *base, *end;
2704         int error = 0, tried_min_extra_isize = 0;
2705         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2706         int isize_diff; /* How much do we need to grow i_extra_isize */
2707
2708 retry:
2709         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2710         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2711                 return 0;
2712
2713         header = IHDR(inode, raw_inode);
2714
2715         /*
2716          * Check if enough free space is available in the inode to shift the
2717          * entries ahead by new_extra_isize.
2718          */
2719
2720         base = IFIRST(header);
2721         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2722         min_offs = end - base;
2723         total_ino = sizeof(struct ext4_xattr_ibody_header);
2724
2725         error = xattr_check_inode(inode, header, end);
2726         if (error)
2727                 goto cleanup;
2728
2729         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2730         if (ifree >= isize_diff)
2731                 goto shift;
2732
2733         /*
2734          * Enough free space isn't available in the inode, check if
2735          * EA block can hold new_extra_isize bytes.
2736          */
2737         if (EXT4_I(inode)->i_file_acl) {
2738                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2739                 error = -EIO;
2740                 if (!bh)
2741                         goto cleanup;
2742                 error = ext4_xattr_check_block(inode, bh);
2743                 if (error)
2744                         goto cleanup;
2745                 base = BHDR(bh);
2746                 end = bh->b_data + bh->b_size;
2747                 min_offs = end - base;
2748                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2749                                               NULL);
2750                 brelse(bh);
2751                 if (bfree + ifree < isize_diff) {
2752                         if (!tried_min_extra_isize && s_min_extra_isize) {
2753                                 tried_min_extra_isize++;
2754                                 new_extra_isize = s_min_extra_isize;
2755                                 goto retry;
2756                         }
2757                         error = -ENOSPC;
2758                         goto cleanup;
2759                 }
2760         } else {
2761                 bfree = inode->i_sb->s_blocksize;
2762         }
2763
2764         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2765                                             isize_diff, ifree, bfree,
2766                                             &total_ino);
2767         if (error) {
2768                 if (error == -ENOSPC && !tried_min_extra_isize &&
2769                     s_min_extra_isize) {
2770                         tried_min_extra_isize++;
2771                         new_extra_isize = s_min_extra_isize;
2772                         goto retry;
2773                 }
2774                 goto cleanup;
2775         }
2776 shift:
2777         /* Adjust the offsets and shift the remaining entries ahead */
2778         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2779                         - new_extra_isize, (void *)raw_inode +
2780                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2781                         (void *)header, total_ino);
2782         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2783
2784 cleanup:
2785         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2786                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2787                              inode->i_ino);
2788                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2789         }
2790         return error;
2791 }
2792
2793 #define EIA_INCR 16 /* must be 2^n */
2794 #define EIA_MASK (EIA_INCR - 1)
2795
2796 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2797  * If @ea_inode_array is new or full it will be grown and the old
2798  * contents copied over.
2799  */
2800 static int
2801 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2802                         struct inode *inode)
2803 {
2804         if (*ea_inode_array == NULL) {
2805                 /*
2806                  * Start with 15 inodes, so it fits into a power-of-two size.
2807                  * If *ea_inode_array is NULL, this is essentially offsetof()
2808                  */
2809                 (*ea_inode_array) =
2810                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2811                                          inodes[EIA_MASK]),
2812                                 GFP_NOFS);
2813                 if (*ea_inode_array == NULL)
2814                         return -ENOMEM;
2815                 (*ea_inode_array)->count = 0;
2816         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2817                 /* expand the array once all 15 + n * 16 slots are full */
2818                 struct ext4_xattr_inode_array *new_array = NULL;
2819                 int count = (*ea_inode_array)->count;
2820
2821                 /* if new_array is NULL, this is essentially offsetof() */
2822                 new_array = kmalloc(
2823                                 offsetof(struct ext4_xattr_inode_array,
2824                                          inodes[count + EIA_INCR]),
2825                                 GFP_NOFS);
2826                 if (new_array == NULL)
2827                         return -ENOMEM;
2828                 memcpy(new_array, *ea_inode_array,
2829                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2830                 kfree(*ea_inode_array);
2831                 *ea_inode_array = new_array;
2832         }
2833         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2834         return 0;
2835 }
2836
2837 /*
2838  * ext4_xattr_delete_inode()
2839  *
2840  * Free extended attribute resources associated with this inode. Traverse
2841  * all entries and decrement reference on any xattr inodes associated with this
2842  * inode. This is called immediately before an inode is freed. We have exclusive
2843  * access to the inode. If an orphan inode is deleted it will also release its
2844  * references on xattr block and xattr inodes.
2845  */
2846 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2847                             struct ext4_xattr_inode_array **ea_inode_array,
2848                             int extra_credits)
2849 {
2850         struct buffer_head *bh = NULL;
2851         struct ext4_xattr_ibody_header *header;
2852         struct ext4_iloc iloc = { .bh = NULL };
2853         struct ext4_xattr_entry *entry;
2854         struct inode *ea_inode;
2855         int error;
2856
2857         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2858                                           NULL /* bh */,
2859                                           false /* dirty */,
2860                                           false /* block_csum */);
2861         if (error) {
2862                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2863                 goto cleanup;
2864         }
2865
2866         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2867             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2868
2869                 error = ext4_get_inode_loc(inode, &iloc);
2870                 if (error) {
2871                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2872                         goto cleanup;
2873                 }
2874
2875                 error = ext4_journal_get_write_access(handle, iloc.bh);
2876                 if (error) {
2877                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2878                                          error);
2879                         goto cleanup;
2880                 }
2881
2882                 header = IHDR(inode, ext4_raw_inode(&iloc));
2883                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2884                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2885                                                      IFIRST(header),
2886                                                      false /* block_csum */,
2887                                                      ea_inode_array,
2888                                                      extra_credits,
2889                                                      false /* skip_quota */);
2890         }
2891
2892         if (EXT4_I(inode)->i_file_acl) {
2893                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2894                 if (!bh) {
2895                         EXT4_ERROR_INODE(inode, "block %llu read error",
2896                                          EXT4_I(inode)->i_file_acl);
2897                         error = -EIO;
2898                         goto cleanup;
2899                 }
2900                 error = ext4_xattr_check_block(inode, bh);
2901                 if (error)
2902                         goto cleanup;
2903
2904                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2905                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2906                              entry = EXT4_XATTR_NEXT(entry)) {
2907                                 if (!entry->e_value_inum)
2908                                         continue;
2909                                 error = ext4_xattr_inode_iget(inode,
2910                                               le32_to_cpu(entry->e_value_inum),
2911                                               le32_to_cpu(entry->e_hash),
2912                                               &ea_inode);
2913                                 if (error)
2914                                         continue;
2915                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2916                                               le32_to_cpu(entry->e_value_size));
2917                                 iput(ea_inode);
2918                         }
2919
2920                 }
2921
2922                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2923                                          extra_credits);
2924                 /*
2925                  * Update i_file_acl value in the same transaction that releases
2926                  * block.
2927                  */
2928                 EXT4_I(inode)->i_file_acl = 0;
2929                 error = ext4_mark_inode_dirty(handle, inode);
2930                 if (error) {
2931                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2932                                          error);
2933                         goto cleanup;
2934                 }
2935         }
2936         error = 0;
2937 cleanup:
2938         brelse(iloc.bh);
2939         brelse(bh);
2940         return error;
2941 }
2942
2943 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2944 {
2945         int idx;
2946
2947         if (ea_inode_array == NULL)
2948                 return;
2949
2950         for (idx = 0; idx < ea_inode_array->count; ++idx)
2951                 iput(ea_inode_array->inodes[idx]);
2952         kfree(ea_inode_array);
2953 }
2954
2955 /*
2956  * ext4_xattr_block_cache_insert()
2957  *
2958  * Create a new entry in the extended attribute block cache, and insert
2959  * it unless such an entry is already in the cache.
2960  *
2961  * Returns 0, or a negative error number on failure.
2962  */
2963 static void
2964 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2965                               struct buffer_head *bh)
2966 {
2967         struct ext4_xattr_header *header = BHDR(bh);
2968         __u32 hash = le32_to_cpu(header->h_hash);
2969         int reusable = le32_to_cpu(header->h_refcount) <
2970                        EXT4_XATTR_REFCOUNT_MAX;
2971         int error;
2972
2973         if (!ea_block_cache)
2974                 return;
2975         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2976                                       bh->b_blocknr, reusable);
2977         if (error) {
2978                 if (error == -EBUSY)
2979                         ea_bdebug(bh, "already in cache");
2980         } else
2981                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2982 }
2983
2984 /*
2985  * ext4_xattr_cmp()
2986  *
2987  * Compare two extended attribute blocks for equality.
2988  *
2989  * Returns 0 if the blocks are equal, 1 if they differ, and
2990  * a negative error number on errors.
2991  */
2992 static int
2993 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2994                struct ext4_xattr_header *header2)
2995 {
2996         struct ext4_xattr_entry *entry1, *entry2;
2997
2998         entry1 = ENTRY(header1+1);
2999         entry2 = ENTRY(header2+1);
3000         while (!IS_LAST_ENTRY(entry1)) {
3001                 if (IS_LAST_ENTRY(entry2))
3002                         return 1;
3003                 if (entry1->e_hash != entry2->e_hash ||
3004                     entry1->e_name_index != entry2->e_name_index ||
3005                     entry1->e_name_len != entry2->e_name_len ||
3006                     entry1->e_value_size != entry2->e_value_size ||
3007                     entry1->e_value_inum != entry2->e_value_inum ||
3008                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3009                         return 1;
3010                 if (!entry1->e_value_inum &&
3011                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3012                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3013                            le32_to_cpu(entry1->e_value_size)))
3014                         return 1;
3015
3016                 entry1 = EXT4_XATTR_NEXT(entry1);
3017                 entry2 = EXT4_XATTR_NEXT(entry2);
3018         }
3019         if (!IS_LAST_ENTRY(entry2))
3020                 return 1;
3021         return 0;
3022 }
3023
3024 /*
3025  * ext4_xattr_block_cache_find()
3026  *
3027  * Find an identical extended attribute block.
3028  *
3029  * Returns a pointer to the block found, or NULL if such a block was
3030  * not found or an error occurred.
3031  */
3032 static struct buffer_head *
3033 ext4_xattr_block_cache_find(struct inode *inode,
3034                             struct ext4_xattr_header *header,
3035                             struct mb_cache_entry **pce)
3036 {
3037         __u32 hash = le32_to_cpu(header->h_hash);
3038         struct mb_cache_entry *ce;
3039         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3040
3041         if (!ea_block_cache)
3042                 return NULL;
3043         if (!header->h_hash)
3044                 return NULL;  /* never share */
3045         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3046         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3047         while (ce) {
3048                 struct buffer_head *bh;
3049
3050                 bh = sb_bread(inode->i_sb, ce->e_value);
3051                 if (!bh) {
3052                         EXT4_ERROR_INODE(inode, "block %lu read error",
3053                                          (unsigned long)ce->e_value);
3054                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3055                         *pce = ce;
3056                         return bh;
3057                 }
3058                 brelse(bh);
3059                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3060         }
3061         return NULL;
3062 }
3063
3064 #define NAME_HASH_SHIFT 5
3065 #define VALUE_HASH_SHIFT 16
3066
3067 /*
3068  * ext4_xattr_hash_entry()
3069  *
3070  * Compute the hash of an extended attribute.
3071  */
3072 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3073                                     size_t value_count)
3074 {
3075         __u32 hash = 0;
3076
3077         while (name_len--) {
3078                 hash = (hash << NAME_HASH_SHIFT) ^
3079                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3080                        *name++;
3081         }
3082         while (value_count--) {
3083                 hash = (hash << VALUE_HASH_SHIFT) ^
3084                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3085                        le32_to_cpu(*value++);
3086         }
3087         return cpu_to_le32(hash);
3088 }
3089
3090 #undef NAME_HASH_SHIFT
3091 #undef VALUE_HASH_SHIFT
3092
3093 #define BLOCK_HASH_SHIFT 16
3094
3095 /*
3096  * ext4_xattr_rehash()
3097  *
3098  * Re-compute the extended attribute hash value after an entry has changed.
3099  */
3100 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3101 {
3102         struct ext4_xattr_entry *here;
3103         __u32 hash = 0;
3104
3105         here = ENTRY(header+1);
3106         while (!IS_LAST_ENTRY(here)) {
3107                 if (!here->e_hash) {
3108                         /* Block is not shared if an entry's hash value == 0 */
3109                         hash = 0;
3110                         break;
3111                 }
3112                 hash = (hash << BLOCK_HASH_SHIFT) ^
3113                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3114                        le32_to_cpu(here->e_hash);
3115                 here = EXT4_XATTR_NEXT(here);
3116         }
3117         header->h_hash = cpu_to_le32(hash);
3118 }
3119
3120 #undef BLOCK_HASH_SHIFT
3121
3122 #define HASH_BUCKET_BITS        10
3123
3124 struct mb_cache *
3125 ext4_xattr_create_cache(void)
3126 {
3127         return mb_cache_create(HASH_BUCKET_BITS);
3128 }
3129
3130 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3131 {
3132         if (cache)
3133                 mb_cache_destroy(cache);
3134 }
3135