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 __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
  85                                     size_t value_count);
  86 static void ext4_xattr_rehash(struct ext4_xattr_header *);
  87
  88 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  89         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  90 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  91         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &nop_posix_acl_access,
  92         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
  93 #endif
  94         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  95 #ifdef CONFIG_EXT4_FS_SECURITY
  96         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  97 #endif
  98         [EXT4_XATTR_INDEX_HURD]              = &ext4_xattr_hurd_handler,
  99 };
 100
 101 const struct xattr_handler * const ext4_xattr_handlers[] = {
 102         &ext4_xattr_user_handler,
 103         &ext4_xattr_trusted_handler,
 104 #ifdef CONFIG_EXT4_FS_SECURITY
 105         &ext4_xattr_security_handler,
 106 #endif
 107         &ext4_xattr_hurd_handler,
 108         NULL
 109 };
 110
 111 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 112                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 115                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117 static int
 118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119                         struct inode *inode);
 120
 121 #ifdef CONFIG_LOCKDEP
 122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123 {
 124         struct ext4_inode_info *ei = EXT4_I(ea_inode);
 125
 126         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 127         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
 128         lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
 129 }
 130 #endif
 131
 132 static __le32 ext4_xattr_block_csum(struct inode *inode,
 133                                     sector_t block_nr,
 134                                     struct ext4_xattr_header *hdr)
 135 {
 136         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 137         __u32 csum;
 138         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 139         __u32 dummy_csum = 0;
 140         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 141
 142         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 143                            sizeof(dsk_block_nr));
 144         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 145         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 146         offset += sizeof(dummy_csum);
 147         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 148                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 149
 150         return cpu_to_le32(csum);
 151 }
 152
 153 static int ext4_xattr_block_csum_verify(struct inode *inode,
 154                                         struct buffer_head *bh)
 155 {
 156         struct ext4_xattr_header *hdr = BHDR(bh);
 157         int ret = 1;
 158
 159         if (ext4_has_metadata_csum(inode->i_sb)) {
 160                 lock_buffer(bh);
 161                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 162                                                         bh->b_blocknr, hdr));
 163                 unlock_buffer(bh);
 164         }
 165         return ret;
 166 }
 167
 168 static void ext4_xattr_block_csum_set(struct inode *inode,
 169                                       struct buffer_head *bh)
 170 {
 171         if (ext4_has_metadata_csum(inode->i_sb))
 172                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 173                                                 bh->b_blocknr, BHDR(bh));
 174 }
 175
 176 static inline const char *ext4_xattr_prefix(int name_index,
 177                                             struct dentry *dentry)
 178 {
 179         const struct xattr_handler *handler = NULL;
 180
 181         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 182                 handler = ext4_xattr_handler_map[name_index];
 183
 184         if (!xattr_handler_can_list(handler, dentry))
 185                 return NULL;
 186
 187         return xattr_prefix(handler);
 188 }
 189
 190 static int
 191 check_xattrs(struct inode *inode, struct buffer_head *bh,
 192              struct ext4_xattr_entry *entry, void *end, void *value_start,
 193              const char *function, unsigned int line)
 194 {
 195         struct ext4_xattr_entry *e = entry;
 196         int err = -EFSCORRUPTED;
 197         char *err_str;
 198
 199         if (bh) {
 200                 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 201                     BHDR(bh)->h_blocks != cpu_to_le32(1)) {
 202                         err_str = "invalid header";
 203                         goto errout;
 204                 }
 205                 if (buffer_verified(bh))
 206                         return 0;
 207                 if (!ext4_xattr_block_csum_verify(inode, bh)) {
 208                         err = -EFSBADCRC;
 209                         err_str = "invalid checksum";
 210                         goto errout;
 211                 }
 212         } else {
 213                 struct ext4_xattr_ibody_header *header = value_start;
 214
 215                 header -= 1;
 216                 if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
 217                         err_str = "in-inode xattr block too small";
 218                         goto errout;
 219                 }
 220                 if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
 221                         err_str = "bad magic number in in-inode xattr";
 222                         goto errout;
 223                 }
 224         }
 225
 226         /* Find the end of the names list */
 227         while (!IS_LAST_ENTRY(e)) {
 228                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 229                 if ((void *)next >= end) {
 230                         err_str = "e_name out of bounds";
 231                         goto errout;
 232                 }
 233                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
 234                         err_str = "bad e_name length";
 235                         goto errout;
 236                 }
 237                 e = next;
 238         }
 239
 240         /* Check the values */
 241         while (!IS_LAST_ENTRY(entry)) {
 242                 u32 size = le32_to_cpu(entry->e_value_size);
 243                 unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
 244
 245                 if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
 246                         err_str = "ea_inode specified without ea_inode feature enabled";
 247                         goto errout;
 248                 }
 249                 if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
 250                                !ext4_valid_inum(inode->i_sb, ea_ino))) {
 251                         err_str = "invalid ea_ino";
 252                         goto errout;
 253                 }
 254                 if (size > EXT4_XATTR_SIZE_MAX) {
 255                         err_str = "e_value size too large";
 256                         goto errout;
 257                 }
 258
 259                 if (size != 0 && entry->e_value_inum == 0) {
 260                         u16 offs = le16_to_cpu(entry->e_value_offs);
 261                         void *value;
 262
 263                         /*
 264                          * The value cannot overlap the names, and the value
 265                          * with padding cannot extend beyond 'end'.  Check both
 266                          * the padded and unpadded sizes, since the size may
 267                          * overflow to 0 when adding padding.
 268                          */
 269                         if (offs > end - value_start) {
 270                                 err_str = "e_value out of bounds";
 271                                 goto errout;
 272                         }
 273                         value = value_start + offs;
 274                         if (value < (void *)e + sizeof(u32) ||
 275                             size > end - value ||
 276                             EXT4_XATTR_SIZE(size) > end - value) {
 277                                 err_str = "overlapping e_value ";
 278                                 goto errout;
 279                         }
 280                 }
 281                 entry = EXT4_XATTR_NEXT(entry);
 282         }
 283         if (bh)
 284                 set_buffer_verified(bh);
 285         return 0;
 286
 287 errout:
 288         if (bh)
 289                 __ext4_error_inode(inode, function, line, 0, -err,
 290                                    "corrupted xattr block %llu: %s",
 291                                    (unsigned long long) bh->b_blocknr,
 292                                    err_str);
 293         else
 294                 __ext4_error_inode(inode, function, line, 0, -err,
 295                                    "corrupted in-inode xattr: %s", err_str);
 296         return err;
 297 }
 298
 299 static inline int
 300 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 301                          const char *function, unsigned int line)
 302 {
 303         return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
 304                             bh->b_data, function, line);
 305 }
 306
 307 #define ext4_xattr_check_block(inode, bh) \
 308         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 309
 310
 311 static inline int
 312 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 313                          void *end, const char *function, unsigned int line)
 314 {
 315         return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
 316                             function, line);
 317 }
 318
 319 #define xattr_check_inode(inode, header, end) \
 320         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 321
 322 static int
 323 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 324                  void *end, int name_index, const char *name, int sorted)
 325 {
 326         struct ext4_xattr_entry *entry, *next;
 327         size_t name_len;
 328         int cmp = 1;
 329
 330         if (name == NULL)
 331                 return -EINVAL;
 332         name_len = strlen(name);
 333         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 334                 next = EXT4_XATTR_NEXT(entry);
 335                 if ((void *) next >= end) {
 336                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 337                         return -EFSCORRUPTED;
 338                 }
 339                 cmp = name_index - entry->e_name_index;
 340                 if (!cmp)
 341                         cmp = name_len - entry->e_name_len;
 342                 if (!cmp)
 343                         cmp = memcmp(name, entry->e_name, name_len);
 344                 if (cmp <= 0 && (sorted || cmp == 0))
 345                         break;
 346         }
 347         *pentry = entry;
 348         return cmp ? -ENODATA : 0;
 349 }
 350
 351 static u32
 352 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 353 {
 354         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 355 }
 356
 357 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 358 {
 359         return ((u64) inode_get_ctime_sec(ea_inode) << 32) |
 360                 (u32) inode_peek_iversion_raw(ea_inode);
 361 }
 362
 363 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 364 {
 365         inode_set_ctime(ea_inode, (u32)(ref_count >> 32), 0);
 366         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 367 }
 368
 369 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 370 {
 371         return (u32) inode_get_atime_sec(ea_inode);
 372 }
 373
 374 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 375 {
 376         inode_set_atime(ea_inode, hash, 0);
 377 }
 378
 379 /*
 380  * Read the EA value from an inode.
 381  */
 382 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 383 {
 384         int blocksize = 1 << ea_inode->i_blkbits;
 385         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 386         int tail_size = (size % blocksize) ?: blocksize;
 387         struct buffer_head *bhs_inline[8];
 388         struct buffer_head **bhs = bhs_inline;
 389         int i, ret;
 390
 391         if (bh_count > ARRAY_SIZE(bhs_inline)) {
 392                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 393                 if (!bhs)
 394                         return -ENOMEM;
 395         }
 396
 397         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 398                                true /* wait */, bhs);
 399         if (ret)
 400                 goto free_bhs;
 401
 402         for (i = 0; i < bh_count; i++) {
 403                 /* There shouldn't be any holes in ea_inode. */
 404                 if (!bhs[i]) {
 405                         ret = -EFSCORRUPTED;
 406                         goto put_bhs;
 407                 }
 408                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 409                        i < bh_count - 1 ? blocksize : tail_size);
 410         }
 411         ret = 0;
 412 put_bhs:
 413         for (i = 0; i < bh_count; i++)
 414                 brelse(bhs[i]);
 415 free_bhs:
 416         if (bhs != bhs_inline)
 417                 kfree(bhs);
 418         return ret;
 419 }
 420
 421 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode_get_mtime_sec(inode)))
 422
 423 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 424                                  u32 ea_inode_hash, struct inode **ea_inode)
 425 {
 426         struct inode *inode;
 427         int err;
 428
 429         /*
 430          * We have to check for this corruption early as otherwise
 431          * iget_locked() could wait indefinitely for the state of our
 432          * parent inode.
 433          */
 434         if (parent->i_ino == ea_ino) {
 435                 ext4_error(parent->i_sb,
 436                            "Parent and EA inode have the same ino %lu", ea_ino);
 437                 return -EFSCORRUPTED;
 438         }
 439
 440         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
 441         if (IS_ERR(inode)) {
 442                 err = PTR_ERR(inode);
 443                 ext4_error(parent->i_sb,
 444                            "error while reading EA inode %lu err=%d", ea_ino,
 445                            err);
 446                 return err;
 447         }
 448         ext4_xattr_inode_set_class(inode);
 449
 450         /*
 451          * Check whether this is an old Lustre-style xattr inode. Lustre
 452          * implementation does not have hash validation, rather it has a
 453          * backpointer from ea_inode to the parent inode.
 454          */
 455         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 456             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 457             inode->i_generation == parent->i_generation) {
 458                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 459                 ext4_xattr_inode_set_ref(inode, 1);
 460         } else {
 461                 inode_lock(inode);
 462                 inode->i_flags |= S_NOQUOTA;
 463                 inode_unlock(inode);
 464         }
 465
 466         *ea_inode = inode;
 467         return 0;
 468 }
 469
 470 /* Remove entry from mbcache when EA inode is getting evicted */
 471 void ext4_evict_ea_inode(struct inode *inode)
 472 {
 473         struct mb_cache_entry *oe;
 474
 475         if (!EA_INODE_CACHE(inode))
 476                 return;
 477         /* Wait for entry to get unused so that we can remove it */
 478         while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
 479                         ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
 480                 mb_cache_entry_wait_unused(oe);
 481                 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
 482         }
 483 }
 484
 485 static int
 486 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 487                                struct ext4_xattr_entry *entry, void *buffer,
 488                                size_t size)
 489 {
 490         u32 hash;
 491
 492         /* Verify stored hash matches calculated hash. */
 493         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 494         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 495                 return -EFSCORRUPTED;
 496
 497         if (entry) {
 498                 __le32 e_hash, tmp_data;
 499
 500                 /* Verify entry hash. */
 501                 tmp_data = cpu_to_le32(hash);
 502                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 503                                                &tmp_data, 1);
 504                 /* All good? */
 505                 if (e_hash == entry->e_hash)
 506                         return 0;
 507
 508                 /*
 509                  * Not good. Maybe the entry hash was calculated
 510                  * using the buggy signed char version?
 511                  */
 512                 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
 513                                                         &tmp_data, 1);
 514                 /* Still no match - bad */
 515                 if (e_hash != entry->e_hash)
 516                         return -EFSCORRUPTED;
 517
 518                 /* Let people know about old hash */
 519                 pr_warn_once("ext4: filesystem with signed xattr name hash");
 520         }
 521         return 0;
 522 }
 523
 524 /*
 525  * Read xattr value from the EA inode.
 526  */
 527 static int
 528 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 529                      void *buffer, size_t size)
 530 {
 531         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 532         struct inode *ea_inode;
 533         int err;
 534
 535         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 536                                     le32_to_cpu(entry->e_hash), &ea_inode);
 537         if (err) {
 538                 ea_inode = NULL;
 539                 goto out;
 540         }
 541
 542         if (i_size_read(ea_inode) != size) {
 543                 ext4_warning_inode(ea_inode,
 544                                    "ea_inode file size=%llu entry size=%zu",
 545                                    i_size_read(ea_inode), size);
 546                 err = -EFSCORRUPTED;
 547                 goto out;
 548         }
 549
 550         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 551         if (err)
 552                 goto out;
 553
 554         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 555                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 556                                                      size);
 557                 if (err) {
 558                         ext4_warning_inode(ea_inode,
 559                                            "EA inode hash validation failed");
 560                         goto out;
 561                 }
 562
 563                 if (ea_inode_cache)
 564                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 565                                         ext4_xattr_inode_get_hash(ea_inode),
 566                                         ea_inode->i_ino, true /* reusable */);
 567         }
 568 out:
 569         iput(ea_inode);
 570         return err;
 571 }
 572
 573 static int
 574 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 575                      void *buffer, size_t buffer_size)
 576 {
 577         struct buffer_head *bh = NULL;
 578         struct ext4_xattr_entry *entry;
 579         size_t size;
 580         void *end;
 581         int error;
 582         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 583
 584         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 585                   name_index, name, buffer, (long)buffer_size);
 586
 587         if (!EXT4_I(inode)->i_file_acl)
 588                 return -ENODATA;
 589         ea_idebug(inode, "reading block %llu",
 590                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 591         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 592         if (IS_ERR(bh))
 593                 return PTR_ERR(bh);
 594         ea_bdebug(bh, "b_count=%d, refcount=%d",
 595                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 596         error = ext4_xattr_check_block(inode, bh);
 597         if (error)
 598                 goto cleanup;
 599         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 600         entry = BFIRST(bh);
 601         end = bh->b_data + bh->b_size;
 602         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 603         if (error)
 604                 goto cleanup;
 605         size = le32_to_cpu(entry->e_value_size);
 606         error = -ERANGE;
 607         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 608                 goto cleanup;
 609         if (buffer) {
 610                 if (size > buffer_size)
 611                         goto cleanup;
 612                 if (entry->e_value_inum) {
 613                         error = ext4_xattr_inode_get(inode, entry, buffer,
 614                                                      size);
 615                         if (error)
 616                                 goto cleanup;
 617                 } else {
 618                         u16 offset = le16_to_cpu(entry->e_value_offs);
 619                         void *p = bh->b_data + offset;
 620
 621                         if (unlikely(p + size > end))
 622                                 goto cleanup;
 623                         memcpy(buffer, p, size);
 624                 }
 625         }
 626         error = size;
 627
 628 cleanup:
 629         brelse(bh);
 630         return error;
 631 }
 632
 633 int
 634 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 635                      void *buffer, size_t buffer_size)
 636 {
 637         struct ext4_xattr_ibody_header *header;
 638         struct ext4_xattr_entry *entry;
 639         struct ext4_inode *raw_inode;
 640         struct ext4_iloc iloc;
 641         size_t size;
 642         void *end;
 643         int error;
 644
 645         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 646                 return -ENODATA;
 647         error = ext4_get_inode_loc(inode, &iloc);
 648         if (error)
 649                 return error;
 650         raw_inode = ext4_raw_inode(&iloc);
 651         header = IHDR(inode, raw_inode);
 652         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 653         error = xattr_check_inode(inode, header, end);
 654         if (error)
 655                 goto cleanup;
 656         entry = IFIRST(header);
 657         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 658         if (error)
 659                 goto cleanup;
 660         size = le32_to_cpu(entry->e_value_size);
 661         error = -ERANGE;
 662         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 663                 goto cleanup;
 664         if (buffer) {
 665                 if (size > buffer_size)
 666                         goto cleanup;
 667                 if (entry->e_value_inum) {
 668                         error = ext4_xattr_inode_get(inode, entry, buffer,
 669                                                      size);
 670                         if (error)
 671                                 goto cleanup;
 672                 } else {
 673                         u16 offset = le16_to_cpu(entry->e_value_offs);
 674                         void *p = (void *)IFIRST(header) + offset;
 675
 676                         if (unlikely(p + size > end))
 677                                 goto cleanup;
 678                         memcpy(buffer, p, size);
 679                 }
 680         }
 681         error = size;
 682
 683 cleanup:
 684         brelse(iloc.bh);
 685         return error;
 686 }
 687
 688 /*
 689  * ext4_xattr_get()
 690  *
 691  * Copy an extended attribute into the buffer
 692  * provided, or compute the buffer size required.
 693  * Buffer is NULL to compute the size of the buffer required.
 694  *
 695  * Returns a negative error number on failure, or the number of bytes
 696  * used / required on success.
 697  */
 698 int
 699 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 700                void *buffer, size_t buffer_size)
 701 {
 702         int error;
 703
 704         if (unlikely(ext4_forced_shutdown(inode->i_sb)))
 705                 return -EIO;
 706
 707         if (strlen(name) > 255)
 708                 return -ERANGE;
 709
 710         down_read(&EXT4_I(inode)->xattr_sem);
 711         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 712                                      buffer_size);
 713         if (error == -ENODATA)
 714                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 715                                              buffer_size);
 716         up_read(&EXT4_I(inode)->xattr_sem);
 717         return error;
 718 }
 719
 720 static int
 721 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 722                         char *buffer, size_t buffer_size)
 723 {
 724         size_t rest = buffer_size;
 725
 726         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 727                 const char *prefix;
 728
 729                 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
 730                 if (prefix) {
 731                         size_t prefix_len = strlen(prefix);
 732                         size_t size = prefix_len + entry->e_name_len + 1;
 733
 734                         if (buffer) {
 735                                 if (size > rest)
 736                                         return -ERANGE;
 737                                 memcpy(buffer, prefix, prefix_len);
 738                                 buffer += prefix_len;
 739                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 740                                 buffer += entry->e_name_len;
 741                                 *buffer++ = 0;
 742                         }
 743                         rest -= size;
 744                 }
 745         }
 746         return buffer_size - rest;  /* total size */
 747 }
 748
 749 static int
 750 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 751 {
 752         struct inode *inode = d_inode(dentry);
 753         struct buffer_head *bh = NULL;
 754         int error;
 755
 756         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 757                   buffer, (long)buffer_size);
 758
 759         if (!EXT4_I(inode)->i_file_acl)
 760                 return 0;
 761         ea_idebug(inode, "reading block %llu",
 762                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 763         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 764         if (IS_ERR(bh))
 765                 return PTR_ERR(bh);
 766         ea_bdebug(bh, "b_count=%d, refcount=%d",
 767                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 768         error = ext4_xattr_check_block(inode, bh);
 769         if (error)
 770                 goto cleanup;
 771         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 772         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 773                                         buffer_size);
 774 cleanup:
 775         brelse(bh);
 776         return error;
 777 }
 778
 779 static int
 780 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 781 {
 782         struct inode *inode = d_inode(dentry);
 783         struct ext4_xattr_ibody_header *header;
 784         struct ext4_inode *raw_inode;
 785         struct ext4_iloc iloc;
 786         void *end;
 787         int error;
 788
 789         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 790                 return 0;
 791         error = ext4_get_inode_loc(inode, &iloc);
 792         if (error)
 793                 return error;
 794         raw_inode = ext4_raw_inode(&iloc);
 795         header = IHDR(inode, raw_inode);
 796         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 797         error = xattr_check_inode(inode, header, end);
 798         if (error)
 799                 goto cleanup;
 800         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 801                                         buffer, buffer_size);
 802
 803 cleanup:
 804         brelse(iloc.bh);
 805         return error;
 806 }
 807
 808 /*
 809  * Inode operation listxattr()
 810  *
 811  * d_inode(dentry)->i_rwsem: don't care
 812  *
 813  * Copy a list of attribute names into the buffer
 814  * provided, or compute the buffer size required.
 815  * Buffer is NULL to compute the size of the buffer required.
 816  *
 817  * Returns a negative error number on failure, or the number of bytes
 818  * used / required on success.
 819  */
 820 ssize_t
 821 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 822 {
 823         int ret, ret2;
 824
 825         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 826         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 827         if (ret < 0)
 828                 goto errout;
 829         if (buffer) {
 830                 buffer += ret;
 831                 buffer_size -= ret;
 832         }
 833         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 834         if (ret < 0)
 835                 goto errout;
 836         ret += ret2;
 837 errout:
 838         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 839         return ret;
 840 }
 841
 842 /*
 843  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 844  * not set, set it.
 845  */
 846 static void ext4_xattr_update_super_block(handle_t *handle,
 847                                           struct super_block *sb)
 848 {
 849         if (ext4_has_feature_xattr(sb))
 850                 return;
 851
 852         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 853         if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 854                                           EXT4_JTR_NONE) == 0) {
 855                 lock_buffer(EXT4_SB(sb)->s_sbh);
 856                 ext4_set_feature_xattr(sb);
 857                 ext4_superblock_csum_set(sb);
 858                 unlock_buffer(EXT4_SB(sb)->s_sbh);
 859                 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 860         }
 861 }
 862
 863 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 864 {
 865         struct ext4_iloc iloc = { .bh = NULL };
 866         struct buffer_head *bh = NULL;
 867         struct ext4_inode *raw_inode;
 868         struct ext4_xattr_ibody_header *header;
 869         struct ext4_xattr_entry *entry;
 870         qsize_t ea_inode_refs = 0;
 871         void *end;
 872         int ret;
 873
 874         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 875
 876         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 877                 ret = ext4_get_inode_loc(inode, &iloc);
 878                 if (ret)
 879                         goto out;
 880                 raw_inode = ext4_raw_inode(&iloc);
 881                 header = IHDR(inode, raw_inode);
 882                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 883                 ret = xattr_check_inode(inode, header, end);
 884                 if (ret)
 885                         goto out;
 886
 887                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 888                      entry = EXT4_XATTR_NEXT(entry))
 889                         if (entry->e_value_inum)
 890                                 ea_inode_refs++;
 891         }
 892
 893         if (EXT4_I(inode)->i_file_acl) {
 894                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 895                 if (IS_ERR(bh)) {
 896                         ret = PTR_ERR(bh);
 897                         bh = NULL;
 898                         goto out;
 899                 }
 900
 901                 ret = ext4_xattr_check_block(inode, bh);
 902                 if (ret)
 903                         goto out;
 904
 905                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 906                      entry = EXT4_XATTR_NEXT(entry))
 907                         if (entry->e_value_inum)
 908                                 ea_inode_refs++;
 909         }
 910         *usage = ea_inode_refs + 1;
 911         ret = 0;
 912 out:
 913         brelse(iloc.bh);
 914         brelse(bh);
 915         return ret;
 916 }
 917
 918 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 919 {
 920         struct super_block *sb = inode->i_sb;
 921         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 922                                     inode->i_blkbits);
 923         size_t mask = ~(cluster_size - 1);
 924
 925         return (length + cluster_size - 1) & mask;
 926 }
 927
 928 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 929 {
 930         int err;
 931
 932         err = dquot_alloc_inode(inode);
 933         if (err)
 934                 return err;
 935         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 936         if (err)
 937                 dquot_free_inode(inode);
 938         return err;
 939 }
 940
 941 static void ext4_xattr_inode_free_quota(struct inode *parent,
 942                                         struct inode *ea_inode,
 943                                         size_t len)
 944 {
 945         if (ea_inode &&
 946             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 947                 return;
 948         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 949         dquot_free_inode(parent);
 950 }
 951
 952 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 953                              struct buffer_head *block_bh, size_t value_len,
 954                              bool is_create)
 955 {
 956         int credits;
 957         int blocks;
 958
 959         /*
 960          * 1) Owner inode update
 961          * 2) Ref count update on old xattr block
 962          * 3) new xattr block
 963          * 4) block bitmap update for new xattr block
 964          * 5) group descriptor for new xattr block
 965          * 6) block bitmap update for old xattr block
 966          * 7) group descriptor for old block
 967          *
 968          * 6 & 7 can happen if we have two racing threads T_a and T_b
 969          * which are each trying to set an xattr on inodes I_a and I_b
 970          * which were both initially sharing an xattr block.
 971          */
 972         credits = 7;
 973
 974         /* Quota updates. */
 975         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 976
 977         /*
 978          * In case of inline data, we may push out the data to a block,
 979          * so we need to reserve credits for this eventuality
 980          */
 981         if (inode && ext4_has_inline_data(inode))
 982                 credits += ext4_writepage_trans_blocks(inode) + 1;
 983
 984         /* We are done if ea_inode feature is not enabled. */
 985         if (!ext4_has_feature_ea_inode(sb))
 986                 return credits;
 987
 988         /* New ea_inode, inode map, block bitmap, group descriptor. */
 989         credits += 4;
 990
 991         /* Data blocks. */
 992         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 993
 994         /* Indirection block or one level of extent tree. */
 995         blocks += 1;
 996
 997         /* Block bitmap and group descriptor updates for each block. */
 998         credits += blocks * 2;
 999
1000         /* Blocks themselves. */
1001         credits += blocks;
1002
1003         if (!is_create) {
1004                 /* Dereference ea_inode holding old xattr value.
1005                  * Old ea_inode, inode map, block bitmap, group descriptor.
1006                  */
1007                 credits += 4;
1008
1009                 /* Data blocks for old ea_inode. */
1010                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1011
1012                 /* Indirection block or one level of extent tree for old
1013                  * ea_inode.
1014                  */
1015                 blocks += 1;
1016
1017                 /* Block bitmap and group descriptor updates for each block. */
1018                 credits += blocks * 2;
1019         }
1020
1021         /* We may need to clone the existing xattr block in which case we need
1022          * to increment ref counts for existing ea_inodes referenced by it.
1023          */
1024         if (block_bh) {
1025                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1026
1027                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1028                         if (entry->e_value_inum)
1029                                 /* Ref count update on ea_inode. */
1030                                 credits += 1;
1031         }
1032         return credits;
1033 }
1034
1035 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1036                                        int ref_change)
1037 {
1038         struct ext4_iloc iloc;
1039         s64 ref_count;
1040         int ret;
1041
1042         inode_lock(ea_inode);
1043
1044         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1045         if (ret)
1046                 goto out;
1047
1048         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1049         ref_count += ref_change;
1050         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1051
1052         if (ref_change > 0) {
1053                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1054                           ea_inode->i_ino, ref_count);
1055
1056                 if (ref_count == 1) {
1057                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1058                                   ea_inode->i_ino, ea_inode->i_nlink);
1059
1060                         set_nlink(ea_inode, 1);
1061                         ext4_orphan_del(handle, ea_inode);
1062                 }
1063         } else {
1064                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1065                           ea_inode->i_ino, ref_count);
1066
1067                 if (ref_count == 0) {
1068                         WARN_ONCE(ea_inode->i_nlink != 1,
1069                                   "EA inode %lu i_nlink=%u",
1070                                   ea_inode->i_ino, ea_inode->i_nlink);
1071
1072                         clear_nlink(ea_inode);
1073                         ext4_orphan_add(handle, ea_inode);
1074                 }
1075         }
1076
1077         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1078         if (ret)
1079                 ext4_warning_inode(ea_inode,
1080                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1081 out:
1082         inode_unlock(ea_inode);
1083         return ret;
1084 }
1085
1086 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1087 {
1088         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1089 }
1090
1091 static int ext4_xattr_inode_dec_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_inc_ref_all(handle_t *handle, struct inode *parent,
1097                                         struct ext4_xattr_entry *first)
1098 {
1099         struct inode *ea_inode;
1100         struct ext4_xattr_entry *entry;
1101         struct ext4_xattr_entry *failed_entry;
1102         unsigned int ea_ino;
1103         int err, saved_err;
1104
1105         for (entry = first; !IS_LAST_ENTRY(entry);
1106              entry = EXT4_XATTR_NEXT(entry)) {
1107                 if (!entry->e_value_inum)
1108                         continue;
1109                 ea_ino = le32_to_cpu(entry->e_value_inum);
1110                 err = ext4_xattr_inode_iget(parent, ea_ino,
1111                                             le32_to_cpu(entry->e_hash),
1112                                             &ea_inode);
1113                 if (err)
1114                         goto cleanup;
1115                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1116                 if (err) {
1117                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1118                         iput(ea_inode);
1119                         goto cleanup;
1120                 }
1121                 iput(ea_inode);
1122         }
1123         return 0;
1124
1125 cleanup:
1126         saved_err = err;
1127         failed_entry = entry;
1128
1129         for (entry = first; entry != failed_entry;
1130              entry = EXT4_XATTR_NEXT(entry)) {
1131                 if (!entry->e_value_inum)
1132                         continue;
1133                 ea_ino = le32_to_cpu(entry->e_value_inum);
1134                 err = ext4_xattr_inode_iget(parent, ea_ino,
1135                                             le32_to_cpu(entry->e_hash),
1136                                             &ea_inode);
1137                 if (err) {
1138                         ext4_warning(parent->i_sb,
1139                                      "cleanup ea_ino %u iget error %d", ea_ino,
1140                                      err);
1141                         continue;
1142                 }
1143                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1144                 if (err)
1145                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1146                                            err);
1147                 iput(ea_inode);
1148         }
1149         return saved_err;
1150 }
1151
1152 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1153                         struct buffer_head *bh, bool block_csum, bool dirty)
1154 {
1155         int error;
1156
1157         if (bh && dirty) {
1158                 if (block_csum)
1159                         ext4_xattr_block_csum_set(inode, bh);
1160                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1161                 if (error) {
1162                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1163                                      error);
1164                         return error;
1165                 }
1166         }
1167         return 0;
1168 }
1169
1170 static void
1171 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1172                              struct buffer_head *bh,
1173                              struct ext4_xattr_entry *first, bool block_csum,
1174                              struct ext4_xattr_inode_array **ea_inode_array,
1175                              int extra_credits, bool skip_quota)
1176 {
1177         struct inode *ea_inode;
1178         struct ext4_xattr_entry *entry;
1179         bool dirty = false;
1180         unsigned int ea_ino;
1181         int err;
1182         int credits;
1183
1184         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1185         credits = 2 + extra_credits;
1186
1187         for (entry = first; !IS_LAST_ENTRY(entry);
1188              entry = EXT4_XATTR_NEXT(entry)) {
1189                 if (!entry->e_value_inum)
1190                         continue;
1191                 ea_ino = le32_to_cpu(entry->e_value_inum);
1192                 err = ext4_xattr_inode_iget(parent, ea_ino,
1193                                             le32_to_cpu(entry->e_hash),
1194                                             &ea_inode);
1195                 if (err)
1196                         continue;
1197
1198                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1199                 if (err) {
1200                         ext4_warning_inode(ea_inode,
1201                                            "Expand inode array err=%d", err);
1202                         iput(ea_inode);
1203                         continue;
1204                 }
1205
1206                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1207                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1208                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1209                                               dirty));
1210                 if (err < 0) {
1211                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1212                                            err);
1213                         continue;
1214                 }
1215                 if (err > 0) {
1216                         err = ext4_journal_get_write_access(handle,
1217                                         parent->i_sb, bh, EXT4_JTR_NONE);
1218                         if (err) {
1219                                 ext4_warning_inode(ea_inode,
1220                                                 "Re-get write access err=%d",
1221                                                 err);
1222                                 continue;
1223                         }
1224                 }
1225
1226                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1227                 if (err) {
1228                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1229                                            err);
1230                         continue;
1231                 }
1232
1233                 if (!skip_quota)
1234                         ext4_xattr_inode_free_quota(parent, ea_inode,
1235                                               le32_to_cpu(entry->e_value_size));
1236
1237                 /*
1238                  * Forget about ea_inode within the same transaction that
1239                  * decrements the ref count. This avoids duplicate decrements in
1240                  * case the rest of the work spills over to subsequent
1241                  * transactions.
1242                  */
1243                 entry->e_value_inum = 0;
1244                 entry->e_value_size = 0;
1245
1246                 dirty = true;
1247         }
1248
1249         if (dirty) {
1250                 /*
1251                  * Note that we are deliberately skipping csum calculation for
1252                  * the final update because we do not expect any journal
1253                  * restarts until xattr block is freed.
1254                  */
1255
1256                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1257                 if (err)
1258                         ext4_warning_inode(parent,
1259                                            "handle dirty metadata err=%d", err);
1260         }
1261 }
1262
1263 /*
1264  * Release the xattr block BH: If the reference count is > 1, decrement it;
1265  * otherwise free the block.
1266  */
1267 static void
1268 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1269                          struct buffer_head *bh,
1270                          struct ext4_xattr_inode_array **ea_inode_array,
1271                          int extra_credits)
1272 {
1273         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1274         u32 hash, ref;
1275         int error = 0;
1276
1277         BUFFER_TRACE(bh, "get_write_access");
1278         error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1279                                               EXT4_JTR_NONE);
1280         if (error)
1281                 goto out;
1282
1283 retry_ref:
1284         lock_buffer(bh);
1285         hash = le32_to_cpu(BHDR(bh)->h_hash);
1286         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1287         if (ref == 1) {
1288                 ea_bdebug(bh, "refcount now=0; freeing");
1289                 /*
1290                  * This must happen under buffer lock for
1291                  * ext4_xattr_block_set() to reliably detect freed block
1292                  */
1293                 if (ea_block_cache) {
1294                         struct mb_cache_entry *oe;
1295
1296                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1297                                                           bh->b_blocknr);
1298                         if (oe) {
1299                                 unlock_buffer(bh);
1300                                 mb_cache_entry_wait_unused(oe);
1301                                 mb_cache_entry_put(ea_block_cache, oe);
1302                                 goto retry_ref;
1303                         }
1304                 }
1305                 get_bh(bh);
1306                 unlock_buffer(bh);
1307
1308                 if (ext4_has_feature_ea_inode(inode->i_sb))
1309                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1310                                                      BFIRST(bh),
1311                                                      true /* block_csum */,
1312                                                      ea_inode_array,
1313                                                      extra_credits,
1314                                                      true /* skip_quota */);
1315                 ext4_free_blocks(handle, inode, bh, 0, 1,
1316                                  EXT4_FREE_BLOCKS_METADATA |
1317                                  EXT4_FREE_BLOCKS_FORGET);
1318         } else {
1319                 ref--;
1320                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1321                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1322                         struct mb_cache_entry *ce;
1323
1324                         if (ea_block_cache) {
1325                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1326                                                         bh->b_blocknr);
1327                                 if (ce) {
1328                                         set_bit(MBE_REUSABLE_B, &ce->e_flags);
1329                                         mb_cache_entry_put(ea_block_cache, ce);
1330                                 }
1331                         }
1332                 }
1333
1334                 ext4_xattr_block_csum_set(inode, bh);
1335                 /*
1336                  * Beware of this ugliness: Releasing of xattr block references
1337                  * from different inodes can race and so we have to protect
1338                  * from a race where someone else frees the block (and releases
1339                  * its journal_head) before we are done dirtying the buffer. In
1340                  * nojournal mode this race is harmless and we actually cannot
1341                  * call ext4_handle_dirty_metadata() with locked buffer as
1342                  * that function can call sync_dirty_buffer() so for that case
1343                  * we handle the dirtying after unlocking the buffer.
1344                  */
1345                 if (ext4_handle_valid(handle))
1346                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1347                 unlock_buffer(bh);
1348                 if (!ext4_handle_valid(handle))
1349                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1350                 if (IS_SYNC(inode))
1351                         ext4_handle_sync(handle);
1352                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1353                 ea_bdebug(bh, "refcount now=%d; releasing",
1354                           le32_to_cpu(BHDR(bh)->h_refcount));
1355         }
1356 out:
1357         ext4_std_error(inode->i_sb, error);
1358         return;
1359 }
1360
1361 /*
1362  * Find the available free space for EAs. This also returns the total number of
1363  * bytes used by EA entries.
1364  */
1365 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1366                                     size_t *min_offs, void *base, int *total)
1367 {
1368         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1369                 if (!last->e_value_inum && last->e_value_size) {
1370                         size_t offs = le16_to_cpu(last->e_value_offs);
1371                         if (offs < *min_offs)
1372                                 *min_offs = offs;
1373                 }
1374                 if (total)
1375                         *total += EXT4_XATTR_LEN(last->e_name_len);
1376         }
1377         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1378 }
1379
1380 /*
1381  * Write the value of the EA in an inode.
1382  */
1383 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1384                                   const void *buf, int bufsize)
1385 {
1386         struct buffer_head *bh = NULL;
1387         unsigned long block = 0;
1388         int blocksize = ea_inode->i_sb->s_blocksize;
1389         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1390         int csize, wsize = 0;
1391         int ret = 0, ret2 = 0;
1392         int retries = 0;
1393
1394 retry:
1395         while (ret >= 0 && ret < max_blocks) {
1396                 struct ext4_map_blocks map;
1397                 map.m_lblk = block += ret;
1398                 map.m_len = max_blocks -= ret;
1399
1400                 ret = ext4_map_blocks(handle, ea_inode, &map,
1401                                       EXT4_GET_BLOCKS_CREATE);
1402                 if (ret <= 0) {
1403                         ext4_mark_inode_dirty(handle, ea_inode);
1404                         if (ret == -ENOSPC &&
1405                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1406                                 ret = 0;
1407                                 goto retry;
1408                         }
1409                         break;
1410                 }
1411         }
1412
1413         if (ret < 0)
1414                 return ret;
1415
1416         block = 0;
1417         while (wsize < bufsize) {
1418                 brelse(bh);
1419                 csize = (bufsize - wsize) > blocksize ? blocksize :
1420                                                                 bufsize - wsize;
1421                 bh = ext4_getblk(handle, ea_inode, block, 0);
1422                 if (IS_ERR(bh))
1423                         return PTR_ERR(bh);
1424                 if (!bh) {
1425                         WARN_ON_ONCE(1);
1426                         EXT4_ERROR_INODE(ea_inode,
1427                                          "ext4_getblk() return bh = NULL");
1428                         return -EFSCORRUPTED;
1429                 }
1430                 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1431                                                    EXT4_JTR_NONE);
1432                 if (ret)
1433                         goto out;
1434
1435                 memcpy(bh->b_data, buf, csize);
1436                 set_buffer_uptodate(bh);
1437                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1438
1439                 buf += csize;
1440                 wsize += csize;
1441                 block += 1;
1442         }
1443
1444         inode_lock(ea_inode);
1445         i_size_write(ea_inode, wsize);
1446         ext4_update_i_disksize(ea_inode, wsize);
1447         inode_unlock(ea_inode);
1448
1449         ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1450         if (unlikely(ret2 && !ret))
1451                 ret = ret2;
1452
1453 out:
1454         brelse(bh);
1455
1456         return ret;
1457 }
1458
1459 /*
1460  * Create an inode to store the value of a large EA.
1461  */
1462 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1463                                              struct inode *inode, u32 hash)
1464 {
1465         struct inode *ea_inode = NULL;
1466         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1467         int err;
1468
1469         if (inode->i_sb->s_root == NULL) {
1470                 ext4_warning(inode->i_sb,
1471                              "refuse to create EA inode when umounting");
1472                 WARN_ON(1);
1473                 return ERR_PTR(-EINVAL);
1474         }
1475
1476         /*
1477          * Let the next inode be the goal, so we try and allocate the EA inode
1478          * in the same group, or nearby one.
1479          */
1480         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1481                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1482                                   EXT4_EA_INODE_FL);
1483         if (!IS_ERR(ea_inode)) {
1484                 ea_inode->i_op = &ext4_file_inode_operations;
1485                 ea_inode->i_fop = &ext4_file_operations;
1486                 ext4_set_aops(ea_inode);
1487                 ext4_xattr_inode_set_class(ea_inode);
1488                 unlock_new_inode(ea_inode);
1489                 ext4_xattr_inode_set_ref(ea_inode, 1);
1490                 ext4_xattr_inode_set_hash(ea_inode, hash);
1491                 err = ext4_mark_inode_dirty(handle, ea_inode);
1492                 if (!err)
1493                         err = ext4_inode_attach_jinode(ea_inode);
1494                 if (err) {
1495                         if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1496                                 ext4_warning_inode(ea_inode,
1497                                         "cleanup dec ref error %d", err);
1498                         iput(ea_inode);
1499                         return ERR_PTR(err);
1500                 }
1501
1502                 /*
1503                  * Xattr inodes are shared therefore quota charging is performed
1504                  * at a higher level.
1505                  */
1506                 dquot_free_inode(ea_inode);
1507                 dquot_drop(ea_inode);
1508                 inode_lock(ea_inode);
1509                 ea_inode->i_flags |= S_NOQUOTA;
1510                 inode_unlock(ea_inode);
1511         }
1512
1513         return ea_inode;
1514 }
1515
1516 static struct inode *
1517 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1518                             size_t value_len, u32 hash)
1519 {
1520         struct inode *ea_inode;
1521         struct mb_cache_entry *ce;
1522         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1523         void *ea_data;
1524
1525         if (!ea_inode_cache)
1526                 return NULL;
1527
1528         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1529         if (!ce)
1530                 return NULL;
1531
1532         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1533                      !(current->flags & PF_MEMALLOC_NOFS));
1534
1535         ea_data = kvmalloc(value_len, GFP_KERNEL);
1536         if (!ea_data) {
1537                 mb_cache_entry_put(ea_inode_cache, ce);
1538                 return NULL;
1539         }
1540
1541         while (ce) {
1542                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1543                                      EXT4_IGET_EA_INODE);
1544                 if (IS_ERR(ea_inode))
1545                         goto next_entry;
1546                 ext4_xattr_inode_set_class(ea_inode);
1547                 if (i_size_read(ea_inode) == value_len &&
1548                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1549                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1550                                                     value_len) &&
1551                     !memcmp(value, ea_data, value_len)) {
1552                         mb_cache_entry_touch(ea_inode_cache, ce);
1553                         mb_cache_entry_put(ea_inode_cache, ce);
1554                         kvfree(ea_data);
1555                         return ea_inode;
1556                 }
1557                 iput(ea_inode);
1558         next_entry:
1559                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1560         }
1561         kvfree(ea_data);
1562         return NULL;
1563 }
1564
1565 /*
1566  * Add value of the EA in an inode.
1567  */
1568 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1569                                           const void *value, size_t value_len,
1570                                           struct inode **ret_inode)
1571 {
1572         struct inode *ea_inode;
1573         u32 hash;
1574         int err;
1575
1576         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1577         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1578         if (ea_inode) {
1579                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1580                 if (err) {
1581                         iput(ea_inode);
1582                         return err;
1583                 }
1584
1585                 *ret_inode = ea_inode;
1586                 return 0;
1587         }
1588
1589         /* Create an inode for the EA value */
1590         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1591         if (IS_ERR(ea_inode))
1592                 return PTR_ERR(ea_inode);
1593
1594         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1595         if (err) {
1596                 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1597                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1598                 iput(ea_inode);
1599                 return err;
1600         }
1601
1602         if (EA_INODE_CACHE(inode))
1603                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1604                                       ea_inode->i_ino, true /* reusable */);
1605
1606         *ret_inode = ea_inode;
1607         return 0;
1608 }
1609
1610 /*
1611  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1612  * feature is enabled.
1613  */
1614 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1615
1616 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1617                                 struct ext4_xattr_search *s,
1618                                 handle_t *handle, struct inode *inode,
1619                                 bool is_block)
1620 {
1621         struct ext4_xattr_entry *last, *next;
1622         struct ext4_xattr_entry *here = s->here;
1623         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1624         int in_inode = i->in_inode;
1625         struct inode *old_ea_inode = NULL;
1626         struct inode *new_ea_inode = NULL;
1627         size_t old_size, new_size;
1628         int ret;
1629
1630         /* Space used by old and new values. */
1631         old_size = (!s->not_found && !here->e_value_inum) ?
1632                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1633         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1634
1635         /*
1636          * Optimization for the simple case when old and new values have the
1637          * same padded sizes. Not applicable if external inodes are involved.
1638          */
1639         if (new_size && new_size == old_size) {
1640                 size_t offs = le16_to_cpu(here->e_value_offs);
1641                 void *val = s->base + offs;
1642
1643                 here->e_value_size = cpu_to_le32(i->value_len);
1644                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1645                         memset(val, 0, new_size);
1646                 } else {
1647                         memcpy(val, i->value, i->value_len);
1648                         /* Clear padding bytes. */
1649                         memset(val + i->value_len, 0, new_size - i->value_len);
1650                 }
1651                 goto update_hash;
1652         }
1653
1654         /* Compute min_offs and last. */
1655         last = s->first;
1656         for (; !IS_LAST_ENTRY(last); last = next) {
1657                 next = EXT4_XATTR_NEXT(last);
1658                 if ((void *)next >= s->end) {
1659                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1660                         ret = -EFSCORRUPTED;
1661                         goto out;
1662                 }
1663                 if (!last->e_value_inum && last->e_value_size) {
1664                         size_t offs = le16_to_cpu(last->e_value_offs);
1665                         if (offs < min_offs)
1666                                 min_offs = offs;
1667                 }
1668         }
1669
1670         /* Check whether we have enough space. */
1671         if (i->value) {
1672                 size_t free;
1673
1674                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1675                 if (!s->not_found)
1676                         free += EXT4_XATTR_LEN(name_len) + old_size;
1677
1678                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1679                         ret = -ENOSPC;
1680                         goto out;
1681                 }
1682
1683                 /*
1684                  * If storing the value in an external inode is an option,
1685                  * reserve space for xattr entries/names in the external
1686                  * attribute block so that a long value does not occupy the
1687                  * whole space and prevent further entries being added.
1688                  */
1689                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1690                     new_size && is_block &&
1691                     (min_offs + old_size - new_size) <
1692                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1693                         ret = -ENOSPC;
1694                         goto out;
1695                 }
1696         }
1697
1698         /*
1699          * Getting access to old and new ea inodes is subject to failures.
1700          * Finish that work before doing any modifications to the xattr data.
1701          */
1702         if (!s->not_found && here->e_value_inum) {
1703                 ret = ext4_xattr_inode_iget(inode,
1704                                             le32_to_cpu(here->e_value_inum),
1705                                             le32_to_cpu(here->e_hash),
1706                                             &old_ea_inode);
1707                 if (ret) {
1708                         old_ea_inode = NULL;
1709                         goto out;
1710                 }
1711         }
1712         if (i->value && in_inode) {
1713                 WARN_ON_ONCE(!i->value_len);
1714
1715                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1716                 if (ret)
1717                         goto out;
1718
1719                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1720                                                      i->value_len,
1721                                                      &new_ea_inode);
1722                 if (ret) {
1723                         new_ea_inode = NULL;
1724                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1725                         goto out;
1726                 }
1727         }
1728
1729         if (old_ea_inode) {
1730                 /* We are ready to release ref count on the old_ea_inode. */
1731                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1732                 if (ret) {
1733                         /* Release newly required ref count on new_ea_inode. */
1734                         if (new_ea_inode) {
1735                                 int err;
1736
1737                                 err = ext4_xattr_inode_dec_ref(handle,
1738                                                                new_ea_inode);
1739                                 if (err)
1740                                         ext4_warning_inode(new_ea_inode,
1741                                                   "dec ref new_ea_inode err=%d",
1742                                                   err);
1743                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1744                                                             i->value_len);
1745                         }
1746                         goto out;
1747                 }
1748
1749                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1750                                             le32_to_cpu(here->e_value_size));
1751         }
1752
1753         /* No failures allowed past this point. */
1754
1755         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1756                 /* Remove the old value. */
1757                 void *first_val = s->base + min_offs;
1758                 size_t offs = le16_to_cpu(here->e_value_offs);
1759                 void *val = s->base + offs;
1760
1761                 memmove(first_val + old_size, first_val, val - first_val);
1762                 memset(first_val, 0, old_size);
1763                 min_offs += old_size;
1764
1765                 /* Adjust all value offsets. */
1766                 last = s->first;
1767                 while (!IS_LAST_ENTRY(last)) {
1768                         size_t o = le16_to_cpu(last->e_value_offs);
1769
1770                         if (!last->e_value_inum &&
1771                             last->e_value_size && o < offs)
1772                                 last->e_value_offs = cpu_to_le16(o + old_size);
1773                         last = EXT4_XATTR_NEXT(last);
1774                 }
1775         }
1776
1777         if (!i->value) {
1778                 /* Remove old name. */
1779                 size_t size = EXT4_XATTR_LEN(name_len);
1780
1781                 last = ENTRY((void *)last - size);
1782                 memmove(here, (void *)here + size,
1783                         (void *)last - (void *)here + sizeof(__u32));
1784                 memset(last, 0, size);
1785
1786                 /*
1787                  * Update i_inline_off - moved ibody region might contain
1788                  * system.data attribute.  Handling a failure here won't
1789                  * cause other complications for setting an xattr.
1790                  */
1791                 if (!is_block && ext4_has_inline_data(inode)) {
1792                         ret = ext4_find_inline_data_nolock(inode);
1793                         if (ret) {
1794                                 ext4_warning_inode(inode,
1795                                         "unable to update i_inline_off");
1796                                 goto out;
1797                         }
1798                 }
1799         } else if (s->not_found) {
1800                 /* Insert new name. */
1801                 size_t size = EXT4_XATTR_LEN(name_len);
1802                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1803
1804                 memmove((void *)here + size, here, rest);
1805                 memset(here, 0, size);
1806                 here->e_name_index = i->name_index;
1807                 here->e_name_len = name_len;
1808                 memcpy(here->e_name, i->name, name_len);
1809         } else {
1810                 /* This is an update, reset value info. */
1811                 here->e_value_inum = 0;
1812                 here->e_value_offs = 0;
1813                 here->e_value_size = 0;
1814         }
1815
1816         if (i->value) {
1817                 /* Insert new value. */
1818                 if (in_inode) {
1819                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1820                 } else if (i->value_len) {
1821                         void *val = s->base + min_offs - new_size;
1822
1823                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1824                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1825                                 memset(val, 0, new_size);
1826                         } else {
1827                                 memcpy(val, i->value, i->value_len);
1828                                 /* Clear padding bytes. */
1829                                 memset(val + i->value_len, 0,
1830                                        new_size - i->value_len);
1831                         }
1832                 }
1833                 here->e_value_size = cpu_to_le32(i->value_len);
1834         }
1835
1836 update_hash:
1837         if (i->value) {
1838                 __le32 hash = 0;
1839
1840                 /* Entry hash calculation. */
1841                 if (in_inode) {
1842                         __le32 crc32c_hash;
1843
1844                         /*
1845                          * Feed crc32c hash instead of the raw value for entry
1846                          * hash calculation. This is to avoid walking
1847                          * potentially long value buffer again.
1848                          */
1849                         crc32c_hash = cpu_to_le32(
1850                                        ext4_xattr_inode_get_hash(new_ea_inode));
1851                         hash = ext4_xattr_hash_entry(here->e_name,
1852                                                      here->e_name_len,
1853                                                      &crc32c_hash, 1);
1854                 } else if (is_block) {
1855                         __le32 *value = s->base + le16_to_cpu(
1856                                                         here->e_value_offs);
1857
1858                         hash = ext4_xattr_hash_entry(here->e_name,
1859                                                      here->e_name_len, value,
1860                                                      new_size >> 2);
1861                 }
1862                 here->e_hash = hash;
1863         }
1864
1865         if (is_block)
1866                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1867
1868         ret = 0;
1869 out:
1870         iput(old_ea_inode);
1871         iput(new_ea_inode);
1872         return ret;
1873 }
1874
1875 struct ext4_xattr_block_find {
1876         struct ext4_xattr_search s;
1877         struct buffer_head *bh;
1878 };
1879
1880 static int
1881 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1882                       struct ext4_xattr_block_find *bs)
1883 {
1884         struct super_block *sb = inode->i_sb;
1885         int error;
1886
1887         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1888                   i->name_index, i->name, i->value, (long)i->value_len);
1889
1890         if (EXT4_I(inode)->i_file_acl) {
1891                 /* The inode already has an extended attribute block. */
1892                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1893                 if (IS_ERR(bs->bh)) {
1894                         error = PTR_ERR(bs->bh);
1895                         bs->bh = NULL;
1896                         return error;
1897                 }
1898                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1899                         atomic_read(&(bs->bh->b_count)),
1900                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1901                 error = ext4_xattr_check_block(inode, bs->bh);
1902                 if (error)
1903                         return error;
1904                 /* Find the named attribute. */
1905                 bs->s.base = BHDR(bs->bh);
1906                 bs->s.first = BFIRST(bs->bh);
1907                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1908                 bs->s.here = bs->s.first;
1909                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1910                                          i->name_index, i->name, 1);
1911                 if (error && error != -ENODATA)
1912                         return error;
1913                 bs->s.not_found = error;
1914         }
1915         return 0;
1916 }
1917
1918 static int
1919 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1920                      struct ext4_xattr_info *i,
1921                      struct ext4_xattr_block_find *bs)
1922 {
1923         struct super_block *sb = inode->i_sb;
1924         struct buffer_head *new_bh = NULL;
1925         struct ext4_xattr_search s_copy = bs->s;
1926         struct ext4_xattr_search *s = &s_copy;
1927         struct mb_cache_entry *ce = NULL;
1928         int error = 0;
1929         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1930         struct inode *ea_inode = NULL, *tmp_inode;
1931         size_t old_ea_inode_quota = 0;
1932         unsigned int ea_ino;
1933
1934
1935 #define header(x) ((struct ext4_xattr_header *)(x))
1936
1937         if (s->base) {
1938                 int offset = (char *)s->here - bs->bh->b_data;
1939
1940                 BUFFER_TRACE(bs->bh, "get_write_access");
1941                 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1942                                                       EXT4_JTR_NONE);
1943                 if (error)
1944                         goto cleanup;
1945                 lock_buffer(bs->bh);
1946
1947                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1948                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1949
1950                         /*
1951                          * This must happen under buffer lock for
1952                          * ext4_xattr_block_set() to reliably detect modified
1953                          * block
1954                          */
1955                         if (ea_block_cache) {
1956                                 struct mb_cache_entry *oe;
1957
1958                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1959                                         hash, bs->bh->b_blocknr);
1960                                 if (oe) {
1961                                         /*
1962                                          * Xattr block is getting reused. Leave
1963                                          * it alone.
1964                                          */
1965                                         mb_cache_entry_put(ea_block_cache, oe);
1966                                         goto clone_block;
1967                                 }
1968                         }
1969                         ea_bdebug(bs->bh, "modifying in-place");
1970                         error = ext4_xattr_set_entry(i, s, handle, inode,
1971                                                      true /* is_block */);
1972                         ext4_xattr_block_csum_set(inode, bs->bh);
1973                         unlock_buffer(bs->bh);
1974                         if (error == -EFSCORRUPTED)
1975                                 goto bad_block;
1976                         if (!error)
1977                                 error = ext4_handle_dirty_metadata(handle,
1978                                                                    inode,
1979                                                                    bs->bh);
1980                         if (error)
1981                                 goto cleanup;
1982                         goto inserted;
1983                 }
1984 clone_block:
1985                 unlock_buffer(bs->bh);
1986                 ea_bdebug(bs->bh, "cloning");
1987                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1988                 error = -ENOMEM;
1989                 if (s->base == NULL)
1990                         goto cleanup;
1991                 s->first = ENTRY(header(s->base)+1);
1992                 header(s->base)->h_refcount = cpu_to_le32(1);
1993                 s->here = ENTRY(s->base + offset);
1994                 s->end = s->base + bs->bh->b_size;
1995
1996                 /*
1997                  * If existing entry points to an xattr inode, we need
1998                  * to prevent ext4_xattr_set_entry() from decrementing
1999                  * ref count on it because the reference belongs to the
2000                  * original block. In this case, make the entry look
2001                  * like it has an empty value.
2002                  */
2003                 if (!s->not_found && s->here->e_value_inum) {
2004                         ea_ino = le32_to_cpu(s->here->e_value_inum);
2005                         error = ext4_xattr_inode_iget(inode, ea_ino,
2006                                       le32_to_cpu(s->here->e_hash),
2007                                       &tmp_inode);
2008                         if (error)
2009                                 goto cleanup;
2010
2011                         if (!ext4_test_inode_state(tmp_inode,
2012                                         EXT4_STATE_LUSTRE_EA_INODE)) {
2013                                 /*
2014                                  * Defer quota free call for previous
2015                                  * inode until success is guaranteed.
2016                                  */
2017                                 old_ea_inode_quota = le32_to_cpu(
2018                                                 s->here->e_value_size);
2019                         }
2020                         iput(tmp_inode);
2021
2022                         s->here->e_value_inum = 0;
2023                         s->here->e_value_size = 0;
2024                 }
2025         } else {
2026                 /* Allocate a buffer where we construct the new block. */
2027                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2028                 error = -ENOMEM;
2029                 if (s->base == NULL)
2030                         goto cleanup;
2031                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2032                 header(s->base)->h_blocks = cpu_to_le32(1);
2033                 header(s->base)->h_refcount = cpu_to_le32(1);
2034                 s->first = ENTRY(header(s->base)+1);
2035                 s->here = ENTRY(header(s->base)+1);
2036                 s->end = s->base + sb->s_blocksize;
2037         }
2038
2039         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2040         if (error == -EFSCORRUPTED)
2041                 goto bad_block;
2042         if (error)
2043                 goto cleanup;
2044
2045         if (i->value && s->here->e_value_inum) {
2046                 /*
2047                  * A ref count on ea_inode has been taken as part of the call to
2048                  * ext4_xattr_set_entry() above. We would like to drop this
2049                  * extra ref but we have to wait until the xattr block is
2050                  * initialized and has its own ref count on the ea_inode.
2051                  */
2052                 ea_ino = le32_to_cpu(s->here->e_value_inum);
2053                 error = ext4_xattr_inode_iget(inode, ea_ino,
2054                                               le32_to_cpu(s->here->e_hash),
2055                                               &ea_inode);
2056                 if (error) {
2057                         ea_inode = NULL;
2058                         goto cleanup;
2059                 }
2060         }
2061
2062 inserted:
2063         if (!IS_LAST_ENTRY(s->first)) {
2064                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2065                                                      &ce);
2066                 if (new_bh) {
2067                         /* We found an identical block in the cache. */
2068                         if (new_bh == bs->bh)
2069                                 ea_bdebug(new_bh, "keeping");
2070                         else {
2071                                 u32 ref;
2072
2073 #ifdef EXT4_XATTR_DEBUG
2074                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2075 #endif
2076                                 /* The old block is released after updating
2077                                    the inode. */
2078                                 error = dquot_alloc_block(inode,
2079                                                 EXT4_C2B(EXT4_SB(sb), 1));
2080                                 if (error)
2081                                         goto cleanup;
2082                                 BUFFER_TRACE(new_bh, "get_write_access");
2083                                 error = ext4_journal_get_write_access(
2084                                                 handle, sb, new_bh,
2085                                                 EXT4_JTR_NONE);
2086                                 if (error)
2087                                         goto cleanup_dquot;
2088                                 lock_buffer(new_bh);
2089                                 /*
2090                                  * We have to be careful about races with
2091                                  * adding references to xattr block. Once we
2092                                  * hold buffer lock xattr block's state is
2093                                  * stable so we can check the additional
2094                                  * reference fits.
2095                                  */
2096                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2097                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2098                                         /*
2099                                          * Undo everything and check mbcache
2100                                          * again.
2101                                          */
2102                                         unlock_buffer(new_bh);
2103                                         dquot_free_block(inode,
2104                                                          EXT4_C2B(EXT4_SB(sb),
2105                                                                   1));
2106                                         brelse(new_bh);
2107                                         mb_cache_entry_put(ea_block_cache, ce);
2108                                         ce = NULL;
2109                                         new_bh = NULL;
2110                                         goto inserted;
2111                                 }
2112                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2113                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2114                                         clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2115                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2116                                           ref);
2117                                 ext4_xattr_block_csum_set(inode, new_bh);
2118                                 unlock_buffer(new_bh);
2119                                 error = ext4_handle_dirty_metadata(handle,
2120                                                                    inode,
2121                                                                    new_bh);
2122                                 if (error)
2123                                         goto cleanup_dquot;
2124                         }
2125                         mb_cache_entry_touch(ea_block_cache, ce);
2126                         mb_cache_entry_put(ea_block_cache, ce);
2127                         ce = NULL;
2128                 } else if (bs->bh && s->base == bs->bh->b_data) {
2129                         /* We were modifying this block in-place. */
2130                         ea_bdebug(bs->bh, "keeping this block");
2131                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2132                         new_bh = bs->bh;
2133                         get_bh(new_bh);
2134                 } else {
2135                         /* We need to allocate a new block */
2136                         ext4_fsblk_t goal, block;
2137
2138 #ifdef EXT4_XATTR_DEBUG
2139                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2140 #endif
2141                         goal = ext4_group_first_block_no(sb,
2142                                                 EXT4_I(inode)->i_block_group);
2143                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2144                                                      NULL, &error);
2145                         if (error)
2146                                 goto cleanup;
2147
2148                         ea_idebug(inode, "creating block %llu",
2149                                   (unsigned long long)block);
2150
2151                         new_bh = sb_getblk(sb, block);
2152                         if (unlikely(!new_bh)) {
2153                                 error = -ENOMEM;
2154 getblk_failed:
2155                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2156                                                  EXT4_FREE_BLOCKS_METADATA);
2157                                 goto cleanup;
2158                         }
2159                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2160                                                       ENTRY(header(s->base)+1));
2161                         if (error)
2162                                 goto getblk_failed;
2163                         if (ea_inode) {
2164                                 /* Drop the extra ref on ea_inode. */
2165                                 error = ext4_xattr_inode_dec_ref(handle,
2166                                                                  ea_inode);
2167                                 if (error)
2168                                         ext4_warning_inode(ea_inode,
2169                                                            "dec ref error=%d",
2170                                                            error);
2171                                 iput(ea_inode);
2172                                 ea_inode = NULL;
2173                         }
2174
2175                         lock_buffer(new_bh);
2176                         error = ext4_journal_get_create_access(handle, sb,
2177                                                         new_bh, EXT4_JTR_NONE);
2178                         if (error) {
2179                                 unlock_buffer(new_bh);
2180                                 error = -EIO;
2181                                 goto getblk_failed;
2182                         }
2183                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2184                         ext4_xattr_block_csum_set(inode, new_bh);
2185                         set_buffer_uptodate(new_bh);
2186                         unlock_buffer(new_bh);
2187                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2188                         error = ext4_handle_dirty_metadata(handle, inode,
2189                                                            new_bh);
2190                         if (error)
2191                                 goto cleanup;
2192                 }
2193         }
2194
2195         if (old_ea_inode_quota)
2196                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2197
2198         /* Update the inode. */
2199         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2200
2201         /* Drop the previous xattr block. */
2202         if (bs->bh && bs->bh != new_bh) {
2203                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2204
2205                 ext4_xattr_release_block(handle, inode, bs->bh,
2206                                          &ea_inode_array,
2207                                          0 /* extra_credits */);
2208                 ext4_xattr_inode_array_free(ea_inode_array);
2209         }
2210         error = 0;
2211
2212 cleanup:
2213         if (ea_inode) {
2214                 int error2;
2215
2216                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2217                 if (error2)
2218                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2219                                            error2);
2220
2221                 /* If there was an error, revert the quota charge. */
2222                 if (error)
2223                         ext4_xattr_inode_free_quota(inode, ea_inode,
2224                                                     i_size_read(ea_inode));
2225                 iput(ea_inode);
2226         }
2227         if (ce)
2228                 mb_cache_entry_put(ea_block_cache, ce);
2229         brelse(new_bh);
2230         if (!(bs->bh && s->base == bs->bh->b_data))
2231                 kfree(s->base);
2232
2233         return error;
2234
2235 cleanup_dquot:
2236         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2237         goto cleanup;
2238
2239 bad_block:
2240         EXT4_ERROR_INODE(inode, "bad block %llu",
2241                          EXT4_I(inode)->i_file_acl);
2242         goto cleanup;
2243
2244 #undef header
2245 }
2246
2247 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2248                           struct ext4_xattr_ibody_find *is)
2249 {
2250         struct ext4_xattr_ibody_header *header;
2251         struct ext4_inode *raw_inode;
2252         int error;
2253
2254         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2255                 return 0;
2256
2257         raw_inode = ext4_raw_inode(&is->iloc);
2258         header = IHDR(inode, raw_inode);
2259         is->s.base = is->s.first = IFIRST(header);
2260         is->s.here = is->s.first;
2261         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2262         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2263                 error = xattr_check_inode(inode, header, is->s.end);
2264                 if (error)
2265                         return error;
2266                 /* Find the named attribute. */
2267                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2268                                          i->name_index, i->name, 0);
2269                 if (error && error != -ENODATA)
2270                         return error;
2271                 is->s.not_found = error;
2272         }
2273         return 0;
2274 }
2275
2276 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2277                                 struct ext4_xattr_info *i,
2278                                 struct ext4_xattr_ibody_find *is)
2279 {
2280         struct ext4_xattr_ibody_header *header;
2281         struct ext4_xattr_search *s = &is->s;
2282         int error;
2283
2284         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2285                 return -ENOSPC;
2286
2287         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2288         if (error)
2289                 return error;
2290         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2291         if (!IS_LAST_ENTRY(s->first)) {
2292                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2293                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2294         } else {
2295                 header->h_magic = cpu_to_le32(0);
2296                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2297         }
2298         return 0;
2299 }
2300
2301 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2302                                  struct ext4_xattr_info *i)
2303 {
2304         void *value;
2305
2306         /* When e_value_inum is set the value is stored externally. */
2307         if (s->here->e_value_inum)
2308                 return 0;
2309         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2310                 return 0;
2311         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2312         return !memcmp(value, i->value, i->value_len);
2313 }
2314
2315 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2316 {
2317         struct buffer_head *bh;
2318         int error;
2319
2320         if (!EXT4_I(inode)->i_file_acl)
2321                 return NULL;
2322         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2323         if (IS_ERR(bh))
2324                 return bh;
2325         error = ext4_xattr_check_block(inode, bh);
2326         if (error) {
2327                 brelse(bh);
2328                 return ERR_PTR(error);
2329         }
2330         return bh;
2331 }
2332
2333 /*
2334  * ext4_xattr_set_handle()
2335  *
2336  * Create, replace or remove an extended attribute for this inode.  Value
2337  * is NULL to remove an existing extended attribute, and non-NULL to
2338  * either replace an existing extended attribute, or create a new extended
2339  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2340  * specify that an extended attribute must exist and must not exist
2341  * previous to the call, respectively.
2342  *
2343  * Returns 0, or a negative error number on failure.
2344  */
2345 int
2346 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2347                       const char *name, const void *value, size_t value_len,
2348                       int flags)
2349 {
2350         struct ext4_xattr_info i = {
2351                 .name_index = name_index,
2352                 .name = name,
2353                 .value = value,
2354                 .value_len = value_len,
2355                 .in_inode = 0,
2356         };
2357         struct ext4_xattr_ibody_find is = {
2358                 .s = { .not_found = -ENODATA, },
2359         };
2360         struct ext4_xattr_block_find bs = {
2361                 .s = { .not_found = -ENODATA, },
2362         };
2363         int no_expand;
2364         int error;
2365
2366         if (!name)
2367                 return -EINVAL;
2368         if (strlen(name) > 255)
2369                 return -ERANGE;
2370
2371         ext4_write_lock_xattr(inode, &no_expand);
2372
2373         /* Check journal credits under write lock. */
2374         if (ext4_handle_valid(handle)) {
2375                 struct buffer_head *bh;
2376                 int credits;
2377
2378                 bh = ext4_xattr_get_block(inode);
2379                 if (IS_ERR(bh)) {
2380                         error = PTR_ERR(bh);
2381                         goto cleanup;
2382                 }
2383
2384                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2385                                                    value_len,
2386                                                    flags & XATTR_CREATE);
2387                 brelse(bh);
2388
2389                 if (jbd2_handle_buffer_credits(handle) < credits) {
2390                         error = -ENOSPC;
2391                         goto cleanup;
2392                 }
2393                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2394         }
2395
2396         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2397         if (error)
2398                 goto cleanup;
2399
2400         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2401                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2402                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2403                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2404         }
2405
2406         error = ext4_xattr_ibody_find(inode, &i, &is);
2407         if (error)
2408                 goto cleanup;
2409         if (is.s.not_found)
2410                 error = ext4_xattr_block_find(inode, &i, &bs);
2411         if (error)
2412                 goto cleanup;
2413         if (is.s.not_found && bs.s.not_found) {
2414                 error = -ENODATA;
2415                 if (flags & XATTR_REPLACE)
2416                         goto cleanup;
2417                 error = 0;
2418                 if (!value)
2419                         goto cleanup;
2420         } else {
2421                 error = -EEXIST;
2422                 if (flags & XATTR_CREATE)
2423                         goto cleanup;
2424         }
2425
2426         if (!value) {
2427                 if (!is.s.not_found)
2428                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2429                 else if (!bs.s.not_found)
2430                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2431         } else {
2432                 error = 0;
2433                 /* Xattr value did not change? Save us some work and bail out */
2434                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2435                         goto cleanup;
2436                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2437                         goto cleanup;
2438
2439                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2440                     (EXT4_XATTR_SIZE(i.value_len) >
2441                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2442                         i.in_inode = 1;
2443 retry_inode:
2444                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2445                 if (!error && !bs.s.not_found) {
2446                         i.value = NULL;
2447                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2448                 } else if (error == -ENOSPC) {
2449                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2450                                 brelse(bs.bh);
2451                                 bs.bh = NULL;
2452                                 error = ext4_xattr_block_find(inode, &i, &bs);
2453                                 if (error)
2454                                         goto cleanup;
2455                         }
2456                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2457                         if (!error && !is.s.not_found) {
2458                                 i.value = NULL;
2459                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2460                                                              &is);
2461                         } else if (error == -ENOSPC) {
2462                                 /*
2463                                  * Xattr does not fit in the block, store at
2464                                  * external inode if possible.
2465                                  */
2466                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2467                                     i.value_len && !i.in_inode) {
2468                                         i.in_inode = 1;
2469                                         goto retry_inode;
2470                                 }
2471                         }
2472                 }
2473         }
2474         if (!error) {
2475                 ext4_xattr_update_super_block(handle, inode->i_sb);
2476                 inode_set_ctime_current(inode);
2477                 inode_inc_iversion(inode);
2478                 if (!value)
2479                         no_expand = 0;
2480                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2481                 /*
2482                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2483                  * error != 0.
2484                  */
2485                 is.iloc.bh = NULL;
2486                 if (IS_SYNC(inode))
2487                         ext4_handle_sync(handle);
2488         }
2489         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2490
2491 cleanup:
2492         brelse(is.iloc.bh);
2493         brelse(bs.bh);
2494         ext4_write_unlock_xattr(inode, &no_expand);
2495         return error;
2496 }
2497
2498 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2499                            bool is_create, int *credits)
2500 {
2501         struct buffer_head *bh;
2502         int err;
2503
2504         *credits = 0;
2505
2506         if (!EXT4_SB(inode->i_sb)->s_journal)
2507                 return 0;
2508
2509         down_read(&EXT4_I(inode)->xattr_sem);
2510
2511         bh = ext4_xattr_get_block(inode);
2512         if (IS_ERR(bh)) {
2513                 err = PTR_ERR(bh);
2514         } else {
2515                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2516                                                     value_len, is_create);
2517                 brelse(bh);
2518                 err = 0;
2519         }
2520
2521         up_read(&EXT4_I(inode)->xattr_sem);
2522         return err;
2523 }
2524
2525 /*
2526  * ext4_xattr_set()
2527  *
2528  * Like ext4_xattr_set_handle, but start from an inode. This extended
2529  * attribute modification is a filesystem transaction by itself.
2530  *
2531  * Returns 0, or a negative error number on failure.
2532  */
2533 int
2534 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2535                const void *value, size_t value_len, int flags)
2536 {
2537         handle_t *handle;
2538         struct super_block *sb = inode->i_sb;
2539         int error, retries = 0;
2540         int credits;
2541
2542         error = dquot_initialize(inode);
2543         if (error)
2544                 return error;
2545
2546 retry:
2547         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2548                                        &credits);
2549         if (error)
2550                 return error;
2551
2552         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2553         if (IS_ERR(handle)) {
2554                 error = PTR_ERR(handle);
2555         } else {
2556                 int error2;
2557
2558                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2559                                               value, value_len, flags);
2560                 error2 = ext4_journal_stop(handle);
2561                 if (error == -ENOSPC &&
2562                     ext4_should_retry_alloc(sb, &retries))
2563                         goto retry;
2564                 if (error == 0)
2565                         error = error2;
2566         }
2567         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2568
2569         return error;
2570 }
2571
2572 /*
2573  * Shift the EA entries in the inode to create space for the increased
2574  * i_extra_isize.
2575  */
2576 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2577                                      int value_offs_shift, void *to,
2578                                      void *from, size_t n)
2579 {
2580         struct ext4_xattr_entry *last = entry;
2581         int new_offs;
2582
2583         /* We always shift xattr headers further thus offsets get lower */
2584         BUG_ON(value_offs_shift > 0);
2585
2586         /* Adjust the value offsets of the entries */
2587         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2588                 if (!last->e_value_inum && last->e_value_size) {
2589                         new_offs = le16_to_cpu(last->e_value_offs) +
2590                                                         value_offs_shift;
2591                         last->e_value_offs = cpu_to_le16(new_offs);
2592                 }
2593         }
2594         /* Shift the entries by n bytes */
2595         memmove(to, from, n);
2596 }
2597
2598 /*
2599  * Move xattr pointed to by 'entry' from inode into external xattr block
2600  */
2601 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2602                                     struct ext4_inode *raw_inode,
2603                                     struct ext4_xattr_entry *entry)
2604 {
2605         struct ext4_xattr_ibody_find *is = NULL;
2606         struct ext4_xattr_block_find *bs = NULL;
2607         char *buffer = NULL, *b_entry_name = NULL;
2608         size_t value_size = le32_to_cpu(entry->e_value_size);
2609         struct ext4_xattr_info i = {
2610                 .value = NULL,
2611                 .value_len = 0,
2612                 .name_index = entry->e_name_index,
2613                 .in_inode = !!entry->e_value_inum,
2614         };
2615         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2616         int needs_kvfree = 0;
2617         int error;
2618
2619         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2620         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2621         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2622         if (!is || !bs || !b_entry_name) {
2623                 error = -ENOMEM;
2624                 goto out;
2625         }
2626
2627         is->s.not_found = -ENODATA;
2628         bs->s.not_found = -ENODATA;
2629         is->iloc.bh = NULL;
2630         bs->bh = NULL;
2631
2632         /* Save the entry name and the entry value */
2633         if (entry->e_value_inum) {
2634                 buffer = kvmalloc(value_size, GFP_NOFS);
2635                 if (!buffer) {
2636                         error = -ENOMEM;
2637                         goto out;
2638                 }
2639                 needs_kvfree = 1;
2640                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2641                 if (error)
2642                         goto out;
2643         } else {
2644                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2645                 buffer = (void *)IFIRST(header) + value_offs;
2646         }
2647
2648         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2649         b_entry_name[entry->e_name_len] = '\0';
2650         i.name = b_entry_name;
2651
2652         error = ext4_get_inode_loc(inode, &is->iloc);
2653         if (error)
2654                 goto out;
2655
2656         error = ext4_xattr_ibody_find(inode, &i, is);
2657         if (error)
2658                 goto out;
2659
2660         i.value = buffer;
2661         i.value_len = value_size;
2662         error = ext4_xattr_block_find(inode, &i, bs);
2663         if (error)
2664                 goto out;
2665
2666         /* Move ea entry from the inode into the block */
2667         error = ext4_xattr_block_set(handle, inode, &i, bs);
2668         if (error)
2669                 goto out;
2670
2671         /* Remove the chosen entry from the inode */
2672         i.value = NULL;
2673         i.value_len = 0;
2674         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2675
2676 out:
2677         kfree(b_entry_name);
2678         if (needs_kvfree && buffer)
2679                 kvfree(buffer);
2680         if (is)
2681                 brelse(is->iloc.bh);
2682         if (bs)
2683                 brelse(bs->bh);
2684         kfree(is);
2685         kfree(bs);
2686
2687         return error;
2688 }
2689
2690 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2691                                        struct ext4_inode *raw_inode,
2692                                        int isize_diff, size_t ifree,
2693                                        size_t bfree, int *total_ino)
2694 {
2695         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2696         struct ext4_xattr_entry *small_entry;
2697         struct ext4_xattr_entry *entry;
2698         struct ext4_xattr_entry *last;
2699         unsigned int entry_size;        /* EA entry size */
2700         unsigned int total_size;        /* EA entry size + value size */
2701         unsigned int min_total_size;
2702         int error;
2703
2704         while (isize_diff > ifree) {
2705                 entry = NULL;
2706                 small_entry = NULL;
2707                 min_total_size = ~0U;
2708                 last = IFIRST(header);
2709                 /* Find the entry best suited to be pushed into EA block */
2710                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2711                         /* never move system.data out of the inode */
2712                         if ((last->e_name_len == 4) &&
2713                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2714                             !memcmp(last->e_name, "data", 4))
2715                                 continue;
2716                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2717                         if (!last->e_value_inum)
2718                                 total_size += EXT4_XATTR_SIZE(
2719                                                le32_to_cpu(last->e_value_size));
2720                         if (total_size <= bfree &&
2721                             total_size < min_total_size) {
2722                                 if (total_size + ifree < isize_diff) {
2723                                         small_entry = last;
2724                                 } else {
2725                                         entry = last;
2726                                         min_total_size = total_size;
2727                                 }
2728                         }
2729                 }
2730
2731                 if (entry == NULL) {
2732                         if (small_entry == NULL)
2733                                 return -ENOSPC;
2734                         entry = small_entry;
2735                 }
2736
2737                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2738                 total_size = entry_size;
2739                 if (!entry->e_value_inum)
2740                         total_size += EXT4_XATTR_SIZE(
2741                                               le32_to_cpu(entry->e_value_size));
2742                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2743                                                  entry);
2744                 if (error)
2745                         return error;
2746
2747                 *total_ino -= entry_size;
2748                 ifree += total_size;
2749                 bfree -= total_size;
2750         }
2751
2752         return 0;
2753 }
2754
2755 /*
2756  * Expand an inode by new_extra_isize bytes when EAs are present.
2757  * Returns 0 on success or negative error number on failure.
2758  */
2759 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2760                                struct ext4_inode *raw_inode, handle_t *handle)
2761 {
2762         struct ext4_xattr_ibody_header *header;
2763         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2764         static unsigned int mnt_count;
2765         size_t min_offs;
2766         size_t ifree, bfree;
2767         int total_ino;
2768         void *base, *end;
2769         int error = 0, tried_min_extra_isize = 0;
2770         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2771         int isize_diff; /* How much do we need to grow i_extra_isize */
2772
2773 retry:
2774         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2775         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2776                 return 0;
2777
2778         header = IHDR(inode, raw_inode);
2779
2780         /*
2781          * Check if enough free space is available in the inode to shift the
2782          * entries ahead by new_extra_isize.
2783          */
2784
2785         base = IFIRST(header);
2786         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2787         min_offs = end - base;
2788         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2789
2790         error = xattr_check_inode(inode, header, end);
2791         if (error)
2792                 goto cleanup;
2793
2794         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2795         if (ifree >= isize_diff)
2796                 goto shift;
2797
2798         /*
2799          * Enough free space isn't available in the inode, check if
2800          * EA block can hold new_extra_isize bytes.
2801          */
2802         if (EXT4_I(inode)->i_file_acl) {
2803                 struct buffer_head *bh;
2804
2805                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2806                 if (IS_ERR(bh)) {
2807                         error = PTR_ERR(bh);
2808                         goto cleanup;
2809                 }
2810                 error = ext4_xattr_check_block(inode, bh);
2811                 if (error) {
2812                         brelse(bh);
2813                         goto cleanup;
2814                 }
2815                 base = BHDR(bh);
2816                 end = bh->b_data + bh->b_size;
2817                 min_offs = end - base;
2818                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2819                                               NULL);
2820                 brelse(bh);
2821                 if (bfree + ifree < isize_diff) {
2822                         if (!tried_min_extra_isize && s_min_extra_isize) {
2823                                 tried_min_extra_isize++;
2824                                 new_extra_isize = s_min_extra_isize;
2825                                 goto retry;
2826                         }
2827                         error = -ENOSPC;
2828                         goto cleanup;
2829                 }
2830         } else {
2831                 bfree = inode->i_sb->s_blocksize;
2832         }
2833
2834         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2835                                             isize_diff, ifree, bfree,
2836                                             &total_ino);
2837         if (error) {
2838                 if (error == -ENOSPC && !tried_min_extra_isize &&
2839                     s_min_extra_isize) {
2840                         tried_min_extra_isize++;
2841                         new_extra_isize = s_min_extra_isize;
2842                         goto retry;
2843                 }
2844                 goto cleanup;
2845         }
2846 shift:
2847         /* Adjust the offsets and shift the remaining entries ahead */
2848         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2849                         - new_extra_isize, (void *)raw_inode +
2850                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2851                         (void *)header, total_ino);
2852         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2853
2854         if (ext4_has_inline_data(inode))
2855                 error = ext4_find_inline_data_nolock(inode);
2856
2857 cleanup:
2858         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2859                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2860                              inode->i_ino);
2861                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2862         }
2863         return error;
2864 }
2865
2866 #define EIA_INCR 16 /* must be 2^n */
2867 #define EIA_MASK (EIA_INCR - 1)
2868
2869 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2870  * If @ea_inode_array is new or full it will be grown and the old
2871  * contents copied over.
2872  */
2873 static int
2874 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2875                         struct inode *inode)
2876 {
2877         if (*ea_inode_array == NULL) {
2878                 /*
2879                  * Start with 15 inodes, so it fits into a power-of-two size.
2880                  * If *ea_inode_array is NULL, this is essentially offsetof()
2881                  */
2882                 (*ea_inode_array) =
2883                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2884                                          inodes[EIA_MASK]),
2885                                 GFP_NOFS);
2886                 if (*ea_inode_array == NULL)
2887                         return -ENOMEM;
2888                 (*ea_inode_array)->count = 0;
2889         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2890                 /* expand the array once all 15 + n * 16 slots are full */
2891                 struct ext4_xattr_inode_array *new_array = NULL;
2892                 int count = (*ea_inode_array)->count;
2893
2894                 /* if new_array is NULL, this is essentially offsetof() */
2895                 new_array = kmalloc(
2896                                 offsetof(struct ext4_xattr_inode_array,
2897                                          inodes[count + EIA_INCR]),
2898                                 GFP_NOFS);
2899                 if (new_array == NULL)
2900                         return -ENOMEM;
2901                 memcpy(new_array, *ea_inode_array,
2902                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2903                 kfree(*ea_inode_array);
2904                 *ea_inode_array = new_array;
2905         }
2906         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2907         return 0;
2908 }
2909
2910 /*
2911  * ext4_xattr_delete_inode()
2912  *
2913  * Free extended attribute resources associated with this inode. Traverse
2914  * all entries and decrement reference on any xattr inodes associated with this
2915  * inode. This is called immediately before an inode is freed. We have exclusive
2916  * access to the inode. If an orphan inode is deleted it will also release its
2917  * references on xattr block and xattr inodes.
2918  */
2919 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2920                             struct ext4_xattr_inode_array **ea_inode_array,
2921                             int extra_credits)
2922 {
2923         struct buffer_head *bh = NULL;
2924         struct ext4_xattr_ibody_header *header;
2925         struct ext4_iloc iloc = { .bh = NULL };
2926         struct ext4_xattr_entry *entry;
2927         struct inode *ea_inode;
2928         int error;
2929
2930         error = ext4_journal_ensure_credits(handle, extra_credits,
2931                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2932         if (error < 0) {
2933                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2934                 goto cleanup;
2935         }
2936
2937         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2938             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2939
2940                 error = ext4_get_inode_loc(inode, &iloc);
2941                 if (error) {
2942                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2943                         goto cleanup;
2944                 }
2945
2946                 error = ext4_journal_get_write_access(handle, inode->i_sb,
2947                                                 iloc.bh, EXT4_JTR_NONE);
2948                 if (error) {
2949                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2950                                          error);
2951                         goto cleanup;
2952                 }
2953
2954                 header = IHDR(inode, ext4_raw_inode(&iloc));
2955                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2956                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2957                                                      IFIRST(header),
2958                                                      false /* block_csum */,
2959                                                      ea_inode_array,
2960                                                      extra_credits,
2961                                                      false /* skip_quota */);
2962         }
2963
2964         if (EXT4_I(inode)->i_file_acl) {
2965                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2966                 if (IS_ERR(bh)) {
2967                         error = PTR_ERR(bh);
2968                         if (error == -EIO) {
2969                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2970                                                      "block %llu read error",
2971                                                      EXT4_I(inode)->i_file_acl);
2972                         }
2973                         bh = NULL;
2974                         goto cleanup;
2975                 }
2976                 error = ext4_xattr_check_block(inode, bh);
2977                 if (error)
2978                         goto cleanup;
2979
2980                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2981                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2982                              entry = EXT4_XATTR_NEXT(entry)) {
2983                                 if (!entry->e_value_inum)
2984                                         continue;
2985                                 error = ext4_xattr_inode_iget(inode,
2986                                               le32_to_cpu(entry->e_value_inum),
2987                                               le32_to_cpu(entry->e_hash),
2988                                               &ea_inode);
2989                                 if (error)
2990                                         continue;
2991                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2992                                               le32_to_cpu(entry->e_value_size));
2993                                 iput(ea_inode);
2994                         }
2995
2996                 }
2997
2998                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2999                                          extra_credits);
3000                 /*
3001                  * Update i_file_acl value in the same transaction that releases
3002                  * block.
3003                  */
3004                 EXT4_I(inode)->i_file_acl = 0;
3005                 error = ext4_mark_inode_dirty(handle, inode);
3006                 if (error) {
3007                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3008                                          error);
3009                         goto cleanup;
3010                 }
3011                 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3012         }
3013         error = 0;
3014 cleanup:
3015         brelse(iloc.bh);
3016         brelse(bh);
3017         return error;
3018 }
3019
3020 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3021 {
3022         int idx;
3023
3024         if (ea_inode_array == NULL)
3025                 return;
3026
3027         for (idx = 0; idx < ea_inode_array->count; ++idx)
3028                 iput(ea_inode_array->inodes[idx]);
3029         kfree(ea_inode_array);
3030 }
3031
3032 /*
3033  * ext4_xattr_block_cache_insert()
3034  *
3035  * Create a new entry in the extended attribute block cache, and insert
3036  * it unless such an entry is already in the cache.
3037  *
3038  * Returns 0, or a negative error number on failure.
3039  */
3040 static void
3041 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3042                               struct buffer_head *bh)
3043 {
3044         struct ext4_xattr_header *header = BHDR(bh);
3045         __u32 hash = le32_to_cpu(header->h_hash);
3046         int reusable = le32_to_cpu(header->h_refcount) <
3047                        EXT4_XATTR_REFCOUNT_MAX;
3048         int error;
3049
3050         if (!ea_block_cache)
3051                 return;
3052         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3053                                       bh->b_blocknr, reusable);
3054         if (error) {
3055                 if (error == -EBUSY)
3056                         ea_bdebug(bh, "already in cache");
3057         } else
3058                 ea_bdebug(bh, "inserting [%x]", (int)hash);
3059 }
3060
3061 /*
3062  * ext4_xattr_cmp()
3063  *
3064  * Compare two extended attribute blocks for equality.
3065  *
3066  * Returns 0 if the blocks are equal, 1 if they differ, and
3067  * a negative error number on errors.
3068  */
3069 static int
3070 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3071                struct ext4_xattr_header *header2)
3072 {
3073         struct ext4_xattr_entry *entry1, *entry2;
3074
3075         entry1 = ENTRY(header1+1);
3076         entry2 = ENTRY(header2+1);
3077         while (!IS_LAST_ENTRY(entry1)) {
3078                 if (IS_LAST_ENTRY(entry2))
3079                         return 1;
3080                 if (entry1->e_hash != entry2->e_hash ||
3081                     entry1->e_name_index != entry2->e_name_index ||
3082                     entry1->e_name_len != entry2->e_name_len ||
3083                     entry1->e_value_size != entry2->e_value_size ||
3084                     entry1->e_value_inum != entry2->e_value_inum ||
3085                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3086                         return 1;
3087                 if (!entry1->e_value_inum &&
3088                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3089                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3090                            le32_to_cpu(entry1->e_value_size)))
3091                         return 1;
3092
3093                 entry1 = EXT4_XATTR_NEXT(entry1);
3094                 entry2 = EXT4_XATTR_NEXT(entry2);
3095         }
3096         if (!IS_LAST_ENTRY(entry2))
3097                 return 1;
3098         return 0;
3099 }
3100
3101 /*
3102  * ext4_xattr_block_cache_find()
3103  *
3104  * Find an identical extended attribute block.
3105  *
3106  * Returns a pointer to the block found, or NULL if such a block was
3107  * not found or an error occurred.
3108  */
3109 static struct buffer_head *
3110 ext4_xattr_block_cache_find(struct inode *inode,
3111                             struct ext4_xattr_header *header,
3112                             struct mb_cache_entry **pce)
3113 {
3114         __u32 hash = le32_to_cpu(header->h_hash);
3115         struct mb_cache_entry *ce;
3116         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3117
3118         if (!ea_block_cache)
3119                 return NULL;
3120         if (!header->h_hash)
3121                 return NULL;  /* never share */
3122         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3123         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3124         while (ce) {
3125                 struct buffer_head *bh;
3126
3127                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3128                 if (IS_ERR(bh)) {
3129                         if (PTR_ERR(bh) == -ENOMEM)
3130                                 return NULL;
3131                         bh = NULL;
3132                         EXT4_ERROR_INODE(inode, "block %lu read error",
3133                                          (unsigned long)ce->e_value);
3134                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3135                         *pce = ce;
3136                         return bh;
3137                 }
3138                 brelse(bh);
3139                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3140         }
3141         return NULL;
3142 }
3143
3144 #define NAME_HASH_SHIFT 5
3145 #define VALUE_HASH_SHIFT 16
3146
3147 /*
3148  * ext4_xattr_hash_entry()
3149  *
3150  * Compute the hash of an extended attribute.
3151  */
3152 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3153                                     size_t value_count)
3154 {
3155         __u32 hash = 0;
3156
3157         while (name_len--) {
3158                 hash = (hash << NAME_HASH_SHIFT) ^
3159                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3160                        (unsigned char)*name++;
3161         }
3162         while (value_count--) {
3163                 hash = (hash << VALUE_HASH_SHIFT) ^
3164                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3165                        le32_to_cpu(*value++);
3166         }
3167         return cpu_to_le32(hash);
3168 }
3169
3170 /*
3171  * ext4_xattr_hash_entry_signed()
3172  *
3173  * Compute the hash of an extended attribute incorrectly.
3174  */
3175 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3176 {
3177         __u32 hash = 0;
3178
3179         while (name_len--) {
3180                 hash = (hash << NAME_HASH_SHIFT) ^
3181                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3182                        (signed char)*name++;
3183         }
3184         while (value_count--) {
3185                 hash = (hash << VALUE_HASH_SHIFT) ^
3186                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3187                        le32_to_cpu(*value++);
3188         }
3189         return cpu_to_le32(hash);
3190 }
3191
3192 #undef NAME_HASH_SHIFT
3193 #undef VALUE_HASH_SHIFT
3194
3195 #define BLOCK_HASH_SHIFT 16
3196
3197 /*
3198  * ext4_xattr_rehash()
3199  *
3200  * Re-compute the extended attribute hash value after an entry has changed.
3201  */
3202 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3203 {
3204         struct ext4_xattr_entry *here;
3205         __u32 hash = 0;
3206
3207         here = ENTRY(header+1);
3208         while (!IS_LAST_ENTRY(here)) {
3209                 if (!here->e_hash) {
3210                         /* Block is not shared if an entry's hash value == 0 */
3211                         hash = 0;
3212                         break;
3213                 }
3214                 hash = (hash << BLOCK_HASH_SHIFT) ^
3215                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3216                        le32_to_cpu(here->e_hash);
3217                 here = EXT4_XATTR_NEXT(here);
3218         }
3219         header->h_hash = cpu_to_le32(hash);
3220 }
3221
3222 #undef BLOCK_HASH_SHIFT
3223
3224 #define HASH_BUCKET_BITS        10
3225
3226 struct mb_cache *
3227 ext4_xattr_create_cache(void)
3228 {
3229         return mb_cache_create(HASH_BUCKET_BITS);
3230 }
3231
3232 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3233 {
3234         if (cache)
3235                 mb_cache_destroy(cache);
3236 }
3237