fs/ext4/xattr.c

   1 /*
   2  * linux/fs/ext4/xattr.c
   3  *
   4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   5  *
   6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   7  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   8  * Extended attributes for symlinks and special files added per
   9  *  suggestion of Luka Renko <luka.renko@hermes.si>.
  10  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  11  *  Red Hat Inc.
  12  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  13  *  and Andreas Gruenbacher <agruen@suse.de>.
  14  */
  15
  16 /*
  17  * Extended attributes are stored directly in inodes (on file systems with
  18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  19  * field contains the block number if an inode uses an additional block. All
  20  * attributes must fit in the inode and one additional block. Blocks that
  21  * contain the identical set of attributes may be shared among several inodes.
  22  * Identical blocks are detected by keeping a cache of blocks that have
  23  * recently been accessed.
  24  *
  25  * The attributes in inodes and on blocks have a different header; the entries
  26  * are stored in the same format:
  27  *
  28  *   +------------------+
  29  *   | header           |
  30  *   | entry 1          | |
  31  *   | entry 2          | | growing downwards
  32  *   | entry 3          | v
  33  *   | four null bytes  |
  34  *   | . . .            |
  35  *   | value 1          | ^
  36  *   | value 3          | | growing upwards
  37  *   | value 2          | |
  38  *   +------------------+
  39  *
  40  * The header is followed by multiple entry descriptors. In disk blocks, the
  41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
  42  * attribute values are aligned to the end of the block in no specific order.
  43  *
  44  * Locking strategy
  45  * ----------------
  46  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  47  * EA blocks are only changed if they are exclusive to an inode, so
  48  * holding xattr_sem also means that nothing but the EA block's reference
  49  * count can change. Multiple writers to the same block are synchronized
  50  * by the buffer lock.
  51  */
  52
  53 #include <linux/init.h>
  54 #include <linux/fs.h>
  55 #include <linux/slab.h>
  56 #include <linux/mbcache.h>
  57 #include <linux/quotaops.h>
  58 #include "ext4_jbd2.h"
  59 #include "ext4.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62
  63 #ifdef EXT4_XATTR_DEBUG
  64 # define ea_idebug(inode, fmt, ...)                                     \
  65         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  66                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  67 # define ea_bdebug(bh, fmt, ...)                                        \
  68         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  69                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  70 #else
  71 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  72 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  73 #endif
  74
  75 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  76 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
  77                                                  struct ext4_xattr_header *,
  78                                                  struct mb_cache_entry **);
  79 static void ext4_xattr_rehash(struct ext4_xattr_header *,
  80                               struct ext4_xattr_entry *);
  81
  82 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  83         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  84 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  85         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  86         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  87 #endif
  88         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  89 #ifdef CONFIG_EXT4_FS_SECURITY
  90         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  91 #endif
  92 };
  93
  94 const struct xattr_handler *ext4_xattr_handlers[] = {
  95         &ext4_xattr_user_handler,
  96         &ext4_xattr_trusted_handler,
  97 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  98         &posix_acl_access_xattr_handler,
  99         &posix_acl_default_xattr_handler,
 100 #endif
 101 #ifdef CONFIG_EXT4_FS_SECURITY
 102         &ext4_xattr_security_handler,
 103 #endif
 104         NULL
 105 };
 106
 107 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
 108                                 inode->i_sb->s_fs_info)->s_mb_cache)
 109
 110 static __le32 ext4_xattr_block_csum(struct inode *inode,
 111                                     sector_t block_nr,
 112                                     struct ext4_xattr_header *hdr)
 113 {
 114         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 115         __u32 csum;
 116         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 117         __u32 dummy_csum = 0;
 118         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 119
 120         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 121                            sizeof(dsk_block_nr));
 122         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 123         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 124         offset += sizeof(dummy_csum);
 125         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 126                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_xattr_block_csum_verify(struct inode *inode,
 132                                         struct buffer_head *bh)
 133 {
 134         struct ext4_xattr_header *hdr = BHDR(bh);
 135         int ret = 1;
 136
 137         if (ext4_has_metadata_csum(inode->i_sb)) {
 138                 lock_buffer(bh);
 139                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 140                                                         bh->b_blocknr, hdr));
 141                 unlock_buffer(bh);
 142         }
 143         return ret;
 144 }
 145
 146 static void ext4_xattr_block_csum_set(struct inode *inode,
 147                                       struct buffer_head *bh)
 148 {
 149         if (ext4_has_metadata_csum(inode->i_sb))
 150                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 151                                                 bh->b_blocknr, BHDR(bh));
 152 }
 153
 154 static inline const struct xattr_handler *
 155 ext4_xattr_handler(int name_index)
 156 {
 157         const struct xattr_handler *handler = NULL;
 158
 159         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 160                 handler = ext4_xattr_handler_map[name_index];
 161         return handler;
 162 }
 163
 164 static int
 165 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 166                          void *value_start)
 167 {
 168         struct ext4_xattr_entry *e = entry;
 169
 170         /* Find the end of the names list */
 171         while (!IS_LAST_ENTRY(e)) {
 172                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 173                 if ((void *)next >= end)
 174                         return -EFSCORRUPTED;
 175                 e = next;
 176         }
 177
 178         /* Check the values */
 179         while (!IS_LAST_ENTRY(entry)) {
 180                 if (entry->e_value_block != 0)
 181                         return -EFSCORRUPTED;
 182                 if (entry->e_value_size != 0) {
 183                         u16 offs = le16_to_cpu(entry->e_value_offs);
 184                         u32 size = le32_to_cpu(entry->e_value_size);
 185                         void *value;
 186
 187                         /*
 188                          * The value cannot overlap the names, and the value
 189                          * with padding cannot extend beyond 'end'.  Check both
 190                          * the padded and unpadded sizes, since the size may
 191                          * overflow to 0 when adding padding.
 192                          */
 193                         if (offs > end - value_start)
 194                                 return -EFSCORRUPTED;
 195                         value = value_start + offs;
 196                         if (value < (void *)e + sizeof(u32) ||
 197                             size > end - value ||
 198                             EXT4_XATTR_SIZE(size) > end - value)
 199                                 return -EFSCORRUPTED;
 200                 }
 201                 entry = EXT4_XATTR_NEXT(entry);
 202         }
 203
 204         return 0;
 205 }
 206
 207 static inline int
 208 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 209 {
 210         int error;
 211
 212         if (buffer_verified(bh))
 213                 return 0;
 214
 215         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 216             BHDR(bh)->h_blocks != cpu_to_le32(1))
 217                 return -EFSCORRUPTED;
 218         if (!ext4_xattr_block_csum_verify(inode, bh))
 219                 return -EFSBADCRC;
 220         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 221                                          bh->b_data);
 222         if (!error)
 223                 set_buffer_verified(bh);
 224         return error;
 225 }
 226
 227 static int
 228 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 229                          void *end, const char *function, unsigned int line)
 230 {
 231         int error = -EFSCORRUPTED;
 232
 233         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 234             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 235                 goto errout;
 236         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 237 errout:
 238         if (error)
 239                 __ext4_error_inode(inode, function, line, 0,
 240                                    "corrupted in-inode xattr");
 241         return error;
 242 }
 243
 244 #define xattr_check_inode(inode, header, end) \
 245         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 246
 247 static int
 248 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 249                       const char *name, int sorted)
 250 {
 251         struct ext4_xattr_entry *entry;
 252         size_t name_len;
 253         int cmp = 1;
 254
 255         if (name == NULL)
 256                 return -EINVAL;
 257         name_len = strlen(name);
 258         entry = *pentry;
 259         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 260                 cmp = name_index - entry->e_name_index;
 261                 if (!cmp)
 262                         cmp = name_len - entry->e_name_len;
 263                 if (!cmp)
 264                         cmp = memcmp(name, entry->e_name, name_len);
 265                 if (cmp <= 0 && (sorted || cmp == 0))
 266                         break;
 267         }
 268         *pentry = entry;
 269         return cmp ? -ENODATA : 0;
 270 }
 271
 272 static int
 273 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 274                      void *buffer, size_t buffer_size)
 275 {
 276         struct buffer_head *bh = NULL;
 277         struct ext4_xattr_entry *entry;
 278         size_t size;
 279         int error;
 280         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 281
 282         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 283                   name_index, name, buffer, (long)buffer_size);
 284
 285         error = -ENODATA;
 286         if (!EXT4_I(inode)->i_file_acl)
 287                 goto cleanup;
 288         ea_idebug(inode, "reading block %llu",
 289                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 290         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 291         if (!bh)
 292                 goto cleanup;
 293         ea_bdebug(bh, "b_count=%d, refcount=%d",
 294                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 295         if (ext4_xattr_check_block(inode, bh)) {
 296                 EXT4_ERROR_INODE(inode, "bad block %llu",
 297                                  EXT4_I(inode)->i_file_acl);
 298                 error = -EFSCORRUPTED;
 299                 goto cleanup;
 300         }
 301         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 302         entry = BFIRST(bh);
 303         error = ext4_xattr_find_entry(&entry, name_index, name, 1);
 304         if (error)
 305                 goto cleanup;
 306         size = le32_to_cpu(entry->e_value_size);
 307         if (buffer) {
 308                 error = -ERANGE;
 309                 if (size > buffer_size)
 310                         goto cleanup;
 311                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 312                        size);
 313         }
 314         error = size;
 315
 316 cleanup:
 317         brelse(bh);
 318         return error;
 319 }
 320
 321 int
 322 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 323                      void *buffer, size_t buffer_size)
 324 {
 325         struct ext4_xattr_ibody_header *header;
 326         struct ext4_xattr_entry *entry;
 327         struct ext4_inode *raw_inode;
 328         struct ext4_iloc iloc;
 329         size_t size;
 330         void *end;
 331         int error;
 332
 333         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 334                 return -ENODATA;
 335         error = ext4_get_inode_loc(inode, &iloc);
 336         if (error)
 337                 return error;
 338         raw_inode = ext4_raw_inode(&iloc);
 339         header = IHDR(inode, raw_inode);
 340         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 341         error = xattr_check_inode(inode, header, end);
 342         if (error)
 343                 goto cleanup;
 344         entry = IFIRST(header);
 345         error = ext4_xattr_find_entry(&entry, name_index, name, 0);
 346         if (error)
 347                 goto cleanup;
 348         size = le32_to_cpu(entry->e_value_size);
 349         if (buffer) {
 350                 error = -ERANGE;
 351                 if (size > buffer_size)
 352                         goto cleanup;
 353                 memcpy(buffer, (void *)IFIRST(header) +
 354                        le16_to_cpu(entry->e_value_offs), size);
 355         }
 356         error = size;
 357
 358 cleanup:
 359         brelse(iloc.bh);
 360         return error;
 361 }
 362
 363 /*
 364  * ext4_xattr_get()
 365  *
 366  * Copy an extended attribute into the buffer
 367  * provided, or compute the buffer size required.
 368  * Buffer is NULL to compute the size of the buffer required.
 369  *
 370  * Returns a negative error number on failure, or the number of bytes
 371  * used / required on success.
 372  */
 373 int
 374 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 375                void *buffer, size_t buffer_size)
 376 {
 377         int error;
 378
 379         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 380                 return -EIO;
 381
 382         if (strlen(name) > 255)
 383                 return -ERANGE;
 384
 385         down_read(&EXT4_I(inode)->xattr_sem);
 386         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 387                                      buffer_size);
 388         if (error == -ENODATA)
 389                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 390                                              buffer_size);
 391         up_read(&EXT4_I(inode)->xattr_sem);
 392         return error;
 393 }
 394
 395 static int
 396 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 397                         char *buffer, size_t buffer_size)
 398 {
 399         size_t rest = buffer_size;
 400
 401         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 402                 const struct xattr_handler *handler =
 403                         ext4_xattr_handler(entry->e_name_index);
 404
 405                 if (handler && (!handler->list || handler->list(dentry))) {
 406                         const char *prefix = handler->prefix ?: handler->name;
 407                         size_t prefix_len = strlen(prefix);
 408                         size_t size = prefix_len + entry->e_name_len + 1;
 409
 410                         if (buffer) {
 411                                 if (size > rest)
 412                                         return -ERANGE;
 413                                 memcpy(buffer, prefix, prefix_len);
 414                                 buffer += prefix_len;
 415                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 416                                 buffer += entry->e_name_len;
 417                                 *buffer++ = 0;
 418                         }
 419                         rest -= size;
 420                 }
 421         }
 422         return buffer_size - rest;  /* total size */
 423 }
 424
 425 static int
 426 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 427 {
 428         struct inode *inode = d_inode(dentry);
 429         struct buffer_head *bh = NULL;
 430         int error;
 431         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 432
 433         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 434                   buffer, (long)buffer_size);
 435
 436         error = 0;
 437         if (!EXT4_I(inode)->i_file_acl)
 438                 goto cleanup;
 439         ea_idebug(inode, "reading block %llu",
 440                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 441         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 442         error = -EIO;
 443         if (!bh)
 444                 goto cleanup;
 445         ea_bdebug(bh, "b_count=%d, refcount=%d",
 446                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 447         if (ext4_xattr_check_block(inode, bh)) {
 448                 EXT4_ERROR_INODE(inode, "bad block %llu",
 449                                  EXT4_I(inode)->i_file_acl);
 450                 error = -EFSCORRUPTED;
 451                 goto cleanup;
 452         }
 453         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 454         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 455
 456 cleanup:
 457         brelse(bh);
 458
 459         return error;
 460 }
 461
 462 static int
 463 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 464 {
 465         struct inode *inode = d_inode(dentry);
 466         struct ext4_xattr_ibody_header *header;
 467         struct ext4_inode *raw_inode;
 468         struct ext4_iloc iloc;
 469         void *end;
 470         int error;
 471
 472         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 473                 return 0;
 474         error = ext4_get_inode_loc(inode, &iloc);
 475         if (error)
 476                 return error;
 477         raw_inode = ext4_raw_inode(&iloc);
 478         header = IHDR(inode, raw_inode);
 479         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 480         error = xattr_check_inode(inode, header, end);
 481         if (error)
 482                 goto cleanup;
 483         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 484                                         buffer, buffer_size);
 485
 486 cleanup:
 487         brelse(iloc.bh);
 488         return error;
 489 }
 490
 491 /*
 492  * Inode operation listxattr()
 493  *
 494  * d_inode(dentry)->i_rwsem: don't care
 495  *
 496  * Copy a list of attribute names into the buffer
 497  * provided, or compute the buffer size required.
 498  * Buffer is NULL to compute the size of the buffer required.
 499  *
 500  * Returns a negative error number on failure, or the number of bytes
 501  * used / required on success.
 502  */
 503 ssize_t
 504 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 505 {
 506         int ret, ret2;
 507
 508         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 509         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 510         if (ret < 0)
 511                 goto errout;
 512         if (buffer) {
 513                 buffer += ret;
 514                 buffer_size -= ret;
 515         }
 516         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 517         if (ret < 0)
 518                 goto errout;
 519         ret += ret2;
 520 errout:
 521         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 522         return ret;
 523 }
 524
 525 /*
 526  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 527  * not set, set it.
 528  */
 529 static void ext4_xattr_update_super_block(handle_t *handle,
 530                                           struct super_block *sb)
 531 {
 532         if (ext4_has_feature_xattr(sb))
 533                 return;
 534
 535         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 536         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 537                 ext4_set_feature_xattr(sb);
 538                 ext4_handle_dirty_super(handle, sb);
 539         }
 540 }
 541
 542 /*
 543  * Release the xattr block BH: If the reference count is > 1, decrement it;
 544  * otherwise free the block.
 545  */
 546 static void
 547 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 548                          struct buffer_head *bh)
 549 {
 550         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 551         u32 hash, ref;
 552         int error = 0;
 553
 554         BUFFER_TRACE(bh, "get_write_access");
 555         error = ext4_journal_get_write_access(handle, bh);
 556         if (error)
 557                 goto out;
 558
 559         lock_buffer(bh);
 560         hash = le32_to_cpu(BHDR(bh)->h_hash);
 561         ref = le32_to_cpu(BHDR(bh)->h_refcount);
 562         if (ref == 1) {
 563                 ea_bdebug(bh, "refcount now=0; freeing");
 564                 /*
 565                  * This must happen under buffer lock for
 566                  * ext4_xattr_block_set() to reliably detect freed block
 567                  */
 568                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
 569                 get_bh(bh);
 570                 unlock_buffer(bh);
 571                 ext4_free_blocks(handle, inode, bh, 0, 1,
 572                                  EXT4_FREE_BLOCKS_METADATA |
 573                                  EXT4_FREE_BLOCKS_FORGET);
 574         } else {
 575                 ref--;
 576                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
 577                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
 578                         struct mb_cache_entry *ce;
 579
 580                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
 581                                                 bh->b_blocknr);
 582                         if (ce) {
 583                                 ce->e_reusable = 1;
 584                                 mb_cache_entry_put(ext4_mb_cache, ce);
 585                         }
 586                 }
 587
 588                 ext4_xattr_block_csum_set(inode, bh);
 589                 /*
 590                  * Beware of this ugliness: Releasing of xattr block references
 591                  * from different inodes can race and so we have to protect
 592                  * from a race where someone else frees the block (and releases
 593                  * its journal_head) before we are done dirtying the buffer. In
 594                  * nojournal mode this race is harmless and we actually cannot
 595                  * call ext4_handle_dirty_metadata() with locked buffer as
 596                  * that function can call sync_dirty_buffer() so for that case
 597                  * we handle the dirtying after unlocking the buffer.
 598                  */
 599                 if (ext4_handle_valid(handle))
 600                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 601                 unlock_buffer(bh);
 602                 if (!ext4_handle_valid(handle))
 603                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 604                 if (IS_SYNC(inode))
 605                         ext4_handle_sync(handle);
 606                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
 607                 ea_bdebug(bh, "refcount now=%d; releasing",
 608                           le32_to_cpu(BHDR(bh)->h_refcount));
 609         }
 610 out:
 611         ext4_std_error(inode->i_sb, error);
 612         return;
 613 }
 614
 615 /*
 616  * Find the available free space for EAs. This also returns the total number of
 617  * bytes used by EA entries.
 618  */
 619 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
 620                                     size_t *min_offs, void *base, int *total)
 621 {
 622         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 623                 if (last->e_value_size) {
 624                         size_t offs = le16_to_cpu(last->e_value_offs);
 625                         if (offs < *min_offs)
 626                                 *min_offs = offs;
 627                 }
 628                 if (total)
 629                         *total += EXT4_XATTR_LEN(last->e_name_len);
 630         }
 631         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 632 }
 633
 634 static int
 635 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 636 {
 637         struct ext4_xattr_entry *last;
 638         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 639
 640         /* Compute min_offs and last. */
 641         last = s->first;
 642         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 643                 if (last->e_value_size) {
 644                         size_t offs = le16_to_cpu(last->e_value_offs);
 645                         if (offs < min_offs)
 646                                 min_offs = offs;
 647                 }
 648         }
 649         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 650         if (!s->not_found) {
 651                 if (s->here->e_value_size) {
 652                         size_t size = le32_to_cpu(s->here->e_value_size);
 653                         free += EXT4_XATTR_SIZE(size);
 654                 }
 655                 free += EXT4_XATTR_LEN(name_len);
 656         }
 657         if (i->value) {
 658                 if (free < EXT4_XATTR_LEN(name_len) +
 659                            EXT4_XATTR_SIZE(i->value_len))
 660                         return -ENOSPC;
 661         }
 662
 663         if (i->value && s->not_found) {
 664                 /* Insert the new name. */
 665                 size_t size = EXT4_XATTR_LEN(name_len);
 666                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
 667                 memmove((void *)s->here + size, s->here, rest);
 668                 memset(s->here, 0, size);
 669                 s->here->e_name_index = i->name_index;
 670                 s->here->e_name_len = name_len;
 671                 memcpy(s->here->e_name, i->name, name_len);
 672         } else {
 673                 if (s->here->e_value_size) {
 674                         void *first_val = s->base + min_offs;
 675                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 676                         void *val = s->base + offs;
 677                         size_t size = EXT4_XATTR_SIZE(
 678                                 le32_to_cpu(s->here->e_value_size));
 679
 680                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
 681                                 /* The old and the new value have the same
 682                                    size. Just replace. */
 683                                 s->here->e_value_size =
 684                                         cpu_to_le32(i->value_len);
 685                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
 686                                         memset(val, 0, size);
 687                                 } else {
 688                                         /* Clear pad bytes first. */
 689                                         memset(val + size - EXT4_XATTR_PAD, 0,
 690                                                EXT4_XATTR_PAD);
 691                                         memcpy(val, i->value, i->value_len);
 692                                 }
 693                                 return 0;
 694                         }
 695
 696                         /* Remove the old value. */
 697                         memmove(first_val + size, first_val, val - first_val);
 698                         memset(first_val, 0, size);
 699                         s->here->e_value_size = 0;
 700                         s->here->e_value_offs = 0;
 701                         min_offs += size;
 702
 703                         /* Adjust all value offsets. */
 704                         last = s->first;
 705                         while (!IS_LAST_ENTRY(last)) {
 706                                 size_t o = le16_to_cpu(last->e_value_offs);
 707                                 if (last->e_value_size && o < offs)
 708                                         last->e_value_offs =
 709                                                 cpu_to_le16(o + size);
 710                                 last = EXT4_XATTR_NEXT(last);
 711                         }
 712                 }
 713                 if (!i->value) {
 714                         /* Remove the old name. */
 715                         size_t size = EXT4_XATTR_LEN(name_len);
 716                         last = ENTRY((void *)last - size);
 717                         memmove(s->here, (void *)s->here + size,
 718                                 (void *)last - (void *)s->here + sizeof(__u32));
 719                         memset(last, 0, size);
 720                 }
 721         }
 722
 723         if (i->value) {
 724                 /* Insert the new value. */
 725                 s->here->e_value_size = cpu_to_le32(i->value_len);
 726                 if (i->value_len) {
 727                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 728                         void *val = s->base + min_offs - size;
 729                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 730                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 731                                 memset(val, 0, size);
 732                         } else {
 733                                 /* Clear the pad bytes first. */
 734                                 memset(val + size - EXT4_XATTR_PAD, 0,
 735                                        EXT4_XATTR_PAD);
 736                                 memcpy(val, i->value, i->value_len);
 737                         }
 738                 }
 739         }
 740         return 0;
 741 }
 742
 743 struct ext4_xattr_block_find {
 744         struct ext4_xattr_search s;
 745         struct buffer_head *bh;
 746 };
 747
 748 static int
 749 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
 750                       struct ext4_xattr_block_find *bs)
 751 {
 752         struct super_block *sb = inode->i_sb;
 753         int error;
 754
 755         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 756                   i->name_index, i->name, i->value, (long)i->value_len);
 757
 758         if (EXT4_I(inode)->i_file_acl) {
 759                 /* The inode already has an extended attribute block. */
 760                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
 761                 error = -EIO;
 762                 if (!bs->bh)
 763                         goto cleanup;
 764                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
 765                         atomic_read(&(bs->bh->b_count)),
 766                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
 767                 if (ext4_xattr_check_block(inode, bs->bh)) {
 768                         EXT4_ERROR_INODE(inode, "bad block %llu",
 769                                          EXT4_I(inode)->i_file_acl);
 770                         error = -EFSCORRUPTED;
 771                         goto cleanup;
 772                 }
 773                 /* Find the named attribute. */
 774                 bs->s.base = BHDR(bs->bh);
 775                 bs->s.first = BFIRST(bs->bh);
 776                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 777                 bs->s.here = bs->s.first;
 778                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 779                                               i->name, 1);
 780                 if (error && error != -ENODATA)
 781                         goto cleanup;
 782                 bs->s.not_found = error;
 783         }
 784         error = 0;
 785
 786 cleanup:
 787         return error;
 788 }
 789
 790 static int
 791 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 792                      struct ext4_xattr_info *i,
 793                      struct ext4_xattr_block_find *bs)
 794 {
 795         struct super_block *sb = inode->i_sb;
 796         struct buffer_head *new_bh = NULL;
 797         struct ext4_xattr_search *s = &bs->s;
 798         struct mb_cache_entry *ce = NULL;
 799         int error = 0;
 800         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 801
 802 #define header(x) ((struct ext4_xattr_header *)(x))
 803
 804         if (i->value && i->value_len > sb->s_blocksize)
 805                 return -ENOSPC;
 806         if (s->base) {
 807                 BUFFER_TRACE(bs->bh, "get_write_access");
 808                 error = ext4_journal_get_write_access(handle, bs->bh);
 809                 if (error)
 810                         goto cleanup;
 811                 lock_buffer(bs->bh);
 812
 813                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
 814                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
 815
 816                         /*
 817                          * This must happen under buffer lock for
 818                          * ext4_xattr_block_set() to reliably detect modified
 819                          * block
 820                          */
 821                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
 822                                                     bs->bh->b_blocknr);
 823                         ea_bdebug(bs->bh, "modifying in-place");
 824                         error = ext4_xattr_set_entry(i, s);
 825                         if (!error) {
 826                                 if (!IS_LAST_ENTRY(s->first))
 827                                         ext4_xattr_rehash(header(s->base),
 828                                                           s->here);
 829                                 ext4_xattr_cache_insert(ext4_mb_cache,
 830                                         bs->bh);
 831                         }
 832                         ext4_xattr_block_csum_set(inode, bs->bh);
 833                         unlock_buffer(bs->bh);
 834                         if (error == -EFSCORRUPTED)
 835                                 goto bad_block;
 836                         if (!error)
 837                                 error = ext4_handle_dirty_metadata(handle,
 838                                                                    inode,
 839                                                                    bs->bh);
 840                         if (error)
 841                                 goto cleanup;
 842                         goto inserted;
 843                 } else {
 844                         int offset = (char *)s->here - bs->bh->b_data;
 845
 846                         unlock_buffer(bs->bh);
 847                         ea_bdebug(bs->bh, "cloning");
 848                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
 849                         error = -ENOMEM;
 850                         if (s->base == NULL)
 851                                 goto cleanup;
 852                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
 853                         s->first = ENTRY(header(s->base)+1);
 854                         header(s->base)->h_refcount = cpu_to_le32(1);
 855                         s->here = ENTRY(s->base + offset);
 856                         s->end = s->base + bs->bh->b_size;
 857                 }
 858         } else {
 859                 /* Allocate a buffer where we construct the new block. */
 860                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 861                 /* assert(header == s->base) */
 862                 error = -ENOMEM;
 863                 if (s->base == NULL)
 864                         goto cleanup;
 865                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
 866                 header(s->base)->h_blocks = cpu_to_le32(1);
 867                 header(s->base)->h_refcount = cpu_to_le32(1);
 868                 s->first = ENTRY(header(s->base)+1);
 869                 s->here = ENTRY(header(s->base)+1);
 870                 s->end = s->base + sb->s_blocksize;
 871         }
 872
 873         error = ext4_xattr_set_entry(i, s);
 874         if (error == -EFSCORRUPTED)
 875                 goto bad_block;
 876         if (error)
 877                 goto cleanup;
 878         if (!IS_LAST_ENTRY(s->first))
 879                 ext4_xattr_rehash(header(s->base), s->here);
 880
 881 inserted:
 882         if (!IS_LAST_ENTRY(s->first)) {
 883                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
 884                 if (new_bh) {
 885                         /* We found an identical block in the cache. */
 886                         if (new_bh == bs->bh)
 887                                 ea_bdebug(new_bh, "keeping");
 888                         else {
 889                                 u32 ref;
 890
 891                                 /* The old block is released after updating
 892                                    the inode. */
 893                                 error = dquot_alloc_block(inode,
 894                                                 EXT4_C2B(EXT4_SB(sb), 1));
 895                                 if (error)
 896                                         goto cleanup;
 897                                 BUFFER_TRACE(new_bh, "get_write_access");
 898                                 error = ext4_journal_get_write_access(handle,
 899                                                                       new_bh);
 900                                 if (error)
 901                                         goto cleanup_dquot;
 902                                 lock_buffer(new_bh);
 903                                 /*
 904                                  * We have to be careful about races with
 905                                  * freeing, rehashing or adding references to
 906                                  * xattr block. Once we hold buffer lock xattr
 907                                  * block's state is stable so we can check
 908                                  * whether the block got freed / rehashed or
 909                                  * not.  Since we unhash mbcache entry under
 910                                  * buffer lock when freeing / rehashing xattr
 911                                  * block, checking whether entry is still
 912                                  * hashed is reliable. Same rules hold for
 913                                  * e_reusable handling.
 914                                  */
 915                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
 916                                     !ce->e_reusable) {
 917                                         /*
 918                                          * Undo everything and check mbcache
 919                                          * again.
 920                                          */
 921                                         unlock_buffer(new_bh);
 922                                         dquot_free_block(inode,
 923                                                          EXT4_C2B(EXT4_SB(sb),
 924                                                                   1));
 925                                         brelse(new_bh);
 926                                         mb_cache_entry_put(ext4_mb_cache, ce);
 927                                         ce = NULL;
 928                                         new_bh = NULL;
 929                                         goto inserted;
 930                                 }
 931                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
 932                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
 933                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
 934                                         ce->e_reusable = 0;
 935                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
 936                                           ref);
 937                                 ext4_xattr_block_csum_set(inode, new_bh);
 938                                 unlock_buffer(new_bh);
 939                                 error = ext4_handle_dirty_metadata(handle,
 940                                                                    inode,
 941                                                                    new_bh);
 942                                 if (error)
 943                                         goto cleanup_dquot;
 944                         }
 945                         mb_cache_entry_touch(ext4_mb_cache, ce);
 946                         mb_cache_entry_put(ext4_mb_cache, ce);
 947                         ce = NULL;
 948                 } else if (bs->bh && s->base == bs->bh->b_data) {
 949                         /* We were modifying this block in-place. */
 950                         ea_bdebug(bs->bh, "keeping this block");
 951                         new_bh = bs->bh;
 952                         get_bh(new_bh);
 953                 } else {
 954                         /* We need to allocate a new block */
 955                         ext4_fsblk_t goal, block;
 956
 957                         goal = ext4_group_first_block_no(sb,
 958                                                 EXT4_I(inode)->i_block_group);
 959
 960                         /* non-extent files can't have physical blocks past 2^32 */
 961                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 962                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 963
 964                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
 965                                                      NULL, &error);
 966                         if (error)
 967                                 goto cleanup;
 968
 969                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 970                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
 971
 972                         ea_idebug(inode, "creating block %llu",
 973                                   (unsigned long long)block);
 974
 975                         new_bh = sb_getblk(sb, block);
 976                         if (unlikely(!new_bh)) {
 977                                 error = -ENOMEM;
 978 getblk_failed:
 979                                 ext4_free_blocks(handle, inode, NULL, block, 1,
 980                                                  EXT4_FREE_BLOCKS_METADATA);
 981                                 goto cleanup;
 982                         }
 983                         lock_buffer(new_bh);
 984                         error = ext4_journal_get_create_access(handle, new_bh);
 985                         if (error) {
 986                                 unlock_buffer(new_bh);
 987                                 error = -EIO;
 988                                 goto getblk_failed;
 989                         }
 990                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
 991                         ext4_xattr_block_csum_set(inode, new_bh);
 992                         set_buffer_uptodate(new_bh);
 993                         unlock_buffer(new_bh);
 994                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
 995                         error = ext4_handle_dirty_metadata(handle, inode,
 996                                                            new_bh);
 997                         if (error)
 998                                 goto cleanup;
 999                 }
1000         }
1001
1002         /* Update the inode. */
1003         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1004
1005         /* Drop the previous xattr block. */
1006         if (bs->bh && bs->bh != new_bh)
1007                 ext4_xattr_release_block(handle, inode, bs->bh);
1008         error = 0;
1009
1010 cleanup:
1011         if (ce)
1012                 mb_cache_entry_put(ext4_mb_cache, ce);
1013         brelse(new_bh);
1014         if (!(bs->bh && s->base == bs->bh->b_data))
1015                 kfree(s->base);
1016
1017         return error;
1018
1019 cleanup_dquot:
1020         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1021         goto cleanup;
1022
1023 bad_block:
1024         EXT4_ERROR_INODE(inode, "bad block %llu",
1025                          EXT4_I(inode)->i_file_acl);
1026         goto cleanup;
1027
1028 #undef header
1029 }
1030
1031 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1032                           struct ext4_xattr_ibody_find *is)
1033 {
1034         struct ext4_xattr_ibody_header *header;
1035         struct ext4_inode *raw_inode;
1036         int error;
1037
1038         if (EXT4_I(inode)->i_extra_isize == 0)
1039                 return 0;
1040         raw_inode = ext4_raw_inode(&is->iloc);
1041         header = IHDR(inode, raw_inode);
1042         is->s.base = is->s.first = IFIRST(header);
1043         is->s.here = is->s.first;
1044         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1045         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1046                 error = xattr_check_inode(inode, header, is->s.end);
1047                 if (error)
1048                         return error;
1049                 /* Find the named attribute. */
1050                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1051                                               i->name, 0);
1052                 if (error && error != -ENODATA)
1053                         return error;
1054                 is->s.not_found = error;
1055         }
1056         return 0;
1057 }
1058
1059 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1060                                 struct ext4_xattr_info *i,
1061                                 struct ext4_xattr_ibody_find *is)
1062 {
1063         struct ext4_xattr_ibody_header *header;
1064         struct ext4_xattr_search *s = &is->s;
1065         int error;
1066
1067         if (EXT4_I(inode)->i_extra_isize == 0)
1068                 return -ENOSPC;
1069         error = ext4_xattr_set_entry(i, s);
1070         if (error) {
1071                 if (error == -ENOSPC &&
1072                     ext4_has_inline_data(inode)) {
1073                         error = ext4_try_to_evict_inline_data(handle, inode,
1074                                         EXT4_XATTR_LEN(strlen(i->name) +
1075                                         EXT4_XATTR_SIZE(i->value_len)));
1076                         if (error)
1077                                 return error;
1078                         error = ext4_xattr_ibody_find(inode, i, is);
1079                         if (error)
1080                                 return error;
1081                         error = ext4_xattr_set_entry(i, s);
1082                 }
1083                 if (error)
1084                         return error;
1085         }
1086         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1087         if (!IS_LAST_ENTRY(s->first)) {
1088                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1089                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1090         } else {
1091                 header->h_magic = cpu_to_le32(0);
1092                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1093         }
1094         return 0;
1095 }
1096
1097 static int ext4_xattr_ibody_set(struct inode *inode,
1098                                 struct ext4_xattr_info *i,
1099                                 struct ext4_xattr_ibody_find *is)
1100 {
1101         struct ext4_xattr_ibody_header *header;
1102         struct ext4_xattr_search *s = &is->s;
1103         int error;
1104
1105         if (EXT4_I(inode)->i_extra_isize == 0)
1106                 return -ENOSPC;
1107         error = ext4_xattr_set_entry(i, s);
1108         if (error)
1109                 return error;
1110         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1111         if (!IS_LAST_ENTRY(s->first)) {
1112                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1113                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1114         } else {
1115                 header->h_magic = cpu_to_le32(0);
1116                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1117         }
1118         return 0;
1119 }
1120
1121 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1122                                  struct ext4_xattr_info *i)
1123 {
1124         void *value;
1125
1126         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1127                 return 0;
1128         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1129         return !memcmp(value, i->value, i->value_len);
1130 }
1131
1132 /*
1133  * ext4_xattr_set_handle()
1134  *
1135  * Create, replace or remove an extended attribute for this inode.  Value
1136  * is NULL to remove an existing extended attribute, and non-NULL to
1137  * either replace an existing extended attribute, or create a new extended
1138  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1139  * specify that an extended attribute must exist and must not exist
1140  * previous to the call, respectively.
1141  *
1142  * Returns 0, or a negative error number on failure.
1143  */
1144 int
1145 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1146                       const char *name, const void *value, size_t value_len,
1147                       int flags)
1148 {
1149         struct ext4_xattr_info i = {
1150                 .name_index = name_index,
1151                 .name = name,
1152                 .value = value,
1153                 .value_len = value_len,
1154
1155         };
1156         struct ext4_xattr_ibody_find is = {
1157                 .s = { .not_found = -ENODATA, },
1158         };
1159         struct ext4_xattr_block_find bs = {
1160                 .s = { .not_found = -ENODATA, },
1161         };
1162         int no_expand;
1163         int error;
1164
1165         if (!name)
1166                 return -EINVAL;
1167         if (strlen(name) > 255)
1168                 return -ERANGE;
1169         ext4_write_lock_xattr(inode, &no_expand);
1170
1171         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1172         if (error)
1173                 goto cleanup;
1174
1175         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1176                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1177                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1178                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1179         }
1180
1181         error = ext4_xattr_ibody_find(inode, &i, &is);
1182         if (error)
1183                 goto cleanup;
1184         if (is.s.not_found)
1185                 error = ext4_xattr_block_find(inode, &i, &bs);
1186         if (error)
1187                 goto cleanup;
1188         if (is.s.not_found && bs.s.not_found) {
1189                 error = -ENODATA;
1190                 if (flags & XATTR_REPLACE)
1191                         goto cleanup;
1192                 error = 0;
1193                 if (!value)
1194                         goto cleanup;
1195         } else {
1196                 error = -EEXIST;
1197                 if (flags & XATTR_CREATE)
1198                         goto cleanup;
1199         }
1200         if (!value) {
1201                 if (!is.s.not_found)
1202                         error = ext4_xattr_ibody_set(inode, &i, &is);
1203                 else if (!bs.s.not_found)
1204                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1205         } else {
1206                 error = 0;
1207                 /* Xattr value did not change? Save us some work and bail out */
1208                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1209                         goto cleanup;
1210                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1211                         goto cleanup;
1212
1213                 error = ext4_xattr_ibody_set(inode, &i, &is);
1214                 if (!error && !bs.s.not_found) {
1215                         i.value = NULL;
1216                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1217                 } else if (error == -ENOSPC) {
1218                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1219                                 error = ext4_xattr_block_find(inode, &i, &bs);
1220                                 if (error)
1221                                         goto cleanup;
1222                         }
1223                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1224                         if (error)
1225                                 goto cleanup;
1226                         if (!is.s.not_found) {
1227                                 i.value = NULL;
1228                                 error = ext4_xattr_ibody_set(inode, &i, &is);
1229                         }
1230                 }
1231         }
1232         if (!error) {
1233                 ext4_xattr_update_super_block(handle, inode->i_sb);
1234                 inode->i_ctime = current_time(inode);
1235                 if (!value)
1236                         no_expand = 0;
1237                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1238                 /*
1239                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1240                  * error != 0.
1241                  */
1242                 is.iloc.bh = NULL;
1243                 if (IS_SYNC(inode))
1244                         ext4_handle_sync(handle);
1245         }
1246
1247 cleanup:
1248         brelse(is.iloc.bh);
1249         brelse(bs.bh);
1250         ext4_write_unlock_xattr(inode, &no_expand);
1251         return error;
1252 }
1253
1254 /*
1255  * ext4_xattr_set()
1256  *
1257  * Like ext4_xattr_set_handle, but start from an inode. This extended
1258  * attribute modification is a filesystem transaction by itself.
1259  *
1260  * Returns 0, or a negative error number on failure.
1261  */
1262 int
1263 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1264                const void *value, size_t value_len, int flags)
1265 {
1266         handle_t *handle;
1267         int error, retries = 0;
1268         int credits = ext4_jbd2_credits_xattr(inode);
1269
1270 retry:
1271         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1272         if (IS_ERR(handle)) {
1273                 error = PTR_ERR(handle);
1274         } else {
1275                 int error2;
1276
1277                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1278                                               value, value_len, flags);
1279                 error2 = ext4_journal_stop(handle);
1280                 if (error == -ENOSPC &&
1281                     ext4_should_retry_alloc(inode->i_sb, &retries))
1282                         goto retry;
1283                 if (error == 0)
1284                         error = error2;
1285         }
1286
1287         return error;
1288 }
1289
1290 /*
1291  * Shift the EA entries in the inode to create space for the increased
1292  * i_extra_isize.
1293  */
1294 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1295                                      int value_offs_shift, void *to,
1296                                      void *from, size_t n)
1297 {
1298         struct ext4_xattr_entry *last = entry;
1299         int new_offs;
1300
1301         /* We always shift xattr headers further thus offsets get lower */
1302         BUG_ON(value_offs_shift > 0);
1303
1304         /* Adjust the value offsets of the entries */
1305         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1306                 if (last->e_value_size) {
1307                         new_offs = le16_to_cpu(last->e_value_offs) +
1308                                                         value_offs_shift;
1309                         last->e_value_offs = cpu_to_le16(new_offs);
1310                 }
1311         }
1312         /* Shift the entries by n bytes */
1313         memmove(to, from, n);
1314 }
1315
1316 /*
1317  * Move xattr pointed to by 'entry' from inode into external xattr block
1318  */
1319 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1320                                     struct ext4_inode *raw_inode,
1321                                     struct ext4_xattr_entry *entry)
1322 {
1323         struct ext4_xattr_ibody_find *is = NULL;
1324         struct ext4_xattr_block_find *bs = NULL;
1325         char *buffer = NULL, *b_entry_name = NULL;
1326         size_t value_offs, value_size;
1327         struct ext4_xattr_info i = {
1328                 .value = NULL,
1329                 .value_len = 0,
1330                 .name_index = entry->e_name_index,
1331         };
1332         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1333         int error;
1334
1335         value_offs = le16_to_cpu(entry->e_value_offs);
1336         value_size = le32_to_cpu(entry->e_value_size);
1337
1338         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1339         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1340         buffer = kmalloc(value_size, GFP_NOFS);
1341         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1342         if (!is || !bs || !buffer || !b_entry_name) {
1343                 error = -ENOMEM;
1344                 goto out;
1345         }
1346
1347         is->s.not_found = -ENODATA;
1348         bs->s.not_found = -ENODATA;
1349         is->iloc.bh = NULL;
1350         bs->bh = NULL;
1351
1352         /* Save the entry name and the entry value */
1353         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1354         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1355         b_entry_name[entry->e_name_len] = '\0';
1356         i.name = b_entry_name;
1357
1358         error = ext4_get_inode_loc(inode, &is->iloc);
1359         if (error)
1360                 goto out;
1361
1362         error = ext4_xattr_ibody_find(inode, &i, is);
1363         if (error)
1364                 goto out;
1365
1366         /* Remove the chosen entry from the inode */
1367         error = ext4_xattr_ibody_set(inode, &i, is);
1368         if (error)
1369                 goto out;
1370
1371         i.name = b_entry_name;
1372         i.value = buffer;
1373         i.value_len = value_size;
1374         error = ext4_xattr_block_find(inode, &i, bs);
1375         if (error)
1376                 goto out;
1377
1378         /* Add entry which was removed from the inode into the block */
1379         error = ext4_xattr_block_set(handle, inode, &i, bs);
1380         if (error)
1381                 goto out;
1382         error = 0;
1383 out:
1384         kfree(b_entry_name);
1385         kfree(buffer);
1386         if (is)
1387                 brelse(is->iloc.bh);
1388         kfree(is);
1389         kfree(bs);
1390
1391         return error;
1392 }
1393
1394 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1395                                        struct ext4_inode *raw_inode,
1396                                        int isize_diff, size_t ifree,
1397                                        size_t bfree, int *total_ino)
1398 {
1399         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1400         struct ext4_xattr_entry *small_entry;
1401         struct ext4_xattr_entry *entry;
1402         struct ext4_xattr_entry *last;
1403         unsigned int entry_size;        /* EA entry size */
1404         unsigned int total_size;        /* EA entry size + value size */
1405         unsigned int min_total_size;
1406         int error;
1407
1408         while (isize_diff > ifree) {
1409                 entry = NULL;
1410                 small_entry = NULL;
1411                 min_total_size = ~0U;
1412                 last = IFIRST(header);
1413                 /* Find the entry best suited to be pushed into EA block */
1414                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1415                         total_size =
1416                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1417                                         EXT4_XATTR_LEN(last->e_name_len);
1418                         if (total_size <= bfree &&
1419                             total_size < min_total_size) {
1420                                 if (total_size + ifree < isize_diff) {
1421                                         small_entry = last;
1422                                 } else {
1423                                         entry = last;
1424                                         min_total_size = total_size;
1425                                 }
1426                         }
1427                 }
1428
1429                 if (entry == NULL) {
1430                         if (small_entry == NULL)
1431                                 return -ENOSPC;
1432                         entry = small_entry;
1433                 }
1434
1435                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1436                 total_size = entry_size +
1437                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1438                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1439                                                  entry);
1440                 if (error)
1441                         return error;
1442
1443                 *total_ino -= entry_size;
1444                 ifree += total_size;
1445                 bfree -= total_size;
1446         }
1447
1448         return 0;
1449 }
1450
1451 /*
1452  * Expand an inode by new_extra_isize bytes when EAs are present.
1453  * Returns 0 on success or negative error number on failure.
1454  */
1455 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1456                                struct ext4_inode *raw_inode, handle_t *handle)
1457 {
1458         struct ext4_xattr_ibody_header *header;
1459         struct buffer_head *bh = NULL;
1460         size_t min_offs;
1461         size_t ifree, bfree;
1462         int total_ino;
1463         void *base, *end;
1464         int error = 0, tried_min_extra_isize = 0;
1465         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1466         int isize_diff; /* How much do we need to grow i_extra_isize */
1467         int no_expand;
1468
1469         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
1470                 return 0;
1471
1472 retry:
1473         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1474         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1475                 goto out;
1476
1477         header = IHDR(inode, raw_inode);
1478
1479         /*
1480          * Check if enough free space is available in the inode to shift the
1481          * entries ahead by new_extra_isize.
1482          */
1483
1484         base = IFIRST(header);
1485         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1486         min_offs = end - base;
1487         total_ino = sizeof(struct ext4_xattr_ibody_header);
1488
1489         error = xattr_check_inode(inode, header, end);
1490         if (error)
1491                 goto cleanup;
1492
1493         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1494         if (ifree >= isize_diff)
1495                 goto shift;
1496
1497         /*
1498          * Enough free space isn't available in the inode, check if
1499          * EA block can hold new_extra_isize bytes.
1500          */
1501         if (EXT4_I(inode)->i_file_acl) {
1502                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1503                 error = -EIO;
1504                 if (!bh)
1505                         goto cleanup;
1506                 if (ext4_xattr_check_block(inode, bh)) {
1507                         EXT4_ERROR_INODE(inode, "bad block %llu",
1508                                          EXT4_I(inode)->i_file_acl);
1509                         error = -EFSCORRUPTED;
1510                         goto cleanup;
1511                 }
1512                 base = BHDR(bh);
1513                 end = bh->b_data + bh->b_size;
1514                 min_offs = end - base;
1515                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1516                                               NULL);
1517                 if (bfree + ifree < isize_diff) {
1518                         if (!tried_min_extra_isize && s_min_extra_isize) {
1519                                 tried_min_extra_isize++;
1520                                 new_extra_isize = s_min_extra_isize;
1521                                 brelse(bh);
1522                                 goto retry;
1523                         }
1524                         error = -ENOSPC;
1525                         goto cleanup;
1526                 }
1527         } else {
1528                 bfree = inode->i_sb->s_blocksize;
1529         }
1530
1531         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1532                                             isize_diff, ifree, bfree,
1533                                             &total_ino);
1534         if (error) {
1535                 if (error == -ENOSPC && !tried_min_extra_isize &&
1536                     s_min_extra_isize) {
1537                         tried_min_extra_isize++;
1538                         new_extra_isize = s_min_extra_isize;
1539                         brelse(bh);
1540                         goto retry;
1541                 }
1542                 goto cleanup;
1543         }
1544 shift:
1545         /* Adjust the offsets and shift the remaining entries ahead */
1546         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1547                         - new_extra_isize, (void *)raw_inode +
1548                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1549                         (void *)header, total_ino);
1550         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1551         brelse(bh);
1552 out:
1553         ext4_write_unlock_xattr(inode, &no_expand);
1554         return 0;
1555
1556 cleanup:
1557         brelse(bh);
1558         /*
1559          * Inode size expansion failed; don't try again
1560          */
1561         no_expand = 1;
1562         ext4_write_unlock_xattr(inode, &no_expand);
1563         return error;
1564 }
1565
1566
1567
1568 /*
1569  * ext4_xattr_delete_inode()
1570  *
1571  * Free extended attribute resources associated with this inode. This
1572  * is called immediately before an inode is freed. We have exclusive
1573  * access to the inode.
1574  */
1575 void
1576 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1577 {
1578         struct buffer_head *bh = NULL;
1579
1580         if (!EXT4_I(inode)->i_file_acl)
1581                 goto cleanup;
1582         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1583         if (!bh) {
1584                 EXT4_ERROR_INODE(inode, "block %llu read error",
1585                                  EXT4_I(inode)->i_file_acl);
1586                 goto cleanup;
1587         }
1588         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1589             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1590                 EXT4_ERROR_INODE(inode, "bad block %llu",
1591                                  EXT4_I(inode)->i_file_acl);
1592                 goto cleanup;
1593         }
1594         ext4_xattr_release_block(handle, inode, bh);
1595         EXT4_I(inode)->i_file_acl = 0;
1596
1597 cleanup:
1598         brelse(bh);
1599 }
1600
1601 /*
1602  * ext4_xattr_cache_insert()
1603  *
1604  * Create a new entry in the extended attribute cache, and insert
1605  * it unless such an entry is already in the cache.
1606  *
1607  * Returns 0, or a negative error number on failure.
1608  */
1609 static void
1610 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1611 {
1612         struct ext4_xattr_header *header = BHDR(bh);
1613         __u32 hash = le32_to_cpu(header->h_hash);
1614         int reusable = le32_to_cpu(header->h_refcount) <
1615                        EXT4_XATTR_REFCOUNT_MAX;
1616         int error;
1617
1618         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1619                                       bh->b_blocknr, reusable);
1620         if (error) {
1621                 if (error == -EBUSY)
1622                         ea_bdebug(bh, "already in cache");
1623         } else
1624                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1625 }
1626
1627 /*
1628  * ext4_xattr_cmp()
1629  *
1630  * Compare two extended attribute blocks for equality.
1631  *
1632  * Returns 0 if the blocks are equal, 1 if they differ, and
1633  * a negative error number on errors.
1634  */
1635 static int
1636 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1637                struct ext4_xattr_header *header2)
1638 {
1639         struct ext4_xattr_entry *entry1, *entry2;
1640
1641         entry1 = ENTRY(header1+1);
1642         entry2 = ENTRY(header2+1);
1643         while (!IS_LAST_ENTRY(entry1)) {
1644                 if (IS_LAST_ENTRY(entry2))
1645                         return 1;
1646                 if (entry1->e_hash != entry2->e_hash ||
1647                     entry1->e_name_index != entry2->e_name_index ||
1648                     entry1->e_name_len != entry2->e_name_len ||
1649                     entry1->e_value_size != entry2->e_value_size ||
1650                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1651                         return 1;
1652                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1653                         return -EFSCORRUPTED;
1654                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1655                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1656                            le32_to_cpu(entry1->e_value_size)))
1657                         return 1;
1658
1659                 entry1 = EXT4_XATTR_NEXT(entry1);
1660                 entry2 = EXT4_XATTR_NEXT(entry2);
1661         }
1662         if (!IS_LAST_ENTRY(entry2))
1663                 return 1;
1664         return 0;
1665 }
1666
1667 /*
1668  * ext4_xattr_cache_find()
1669  *
1670  * Find an identical extended attribute block.
1671  *
1672  * Returns a pointer to the block found, or NULL if such a block was
1673  * not found or an error occurred.
1674  */
1675 static struct buffer_head *
1676 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1677                       struct mb_cache_entry **pce)
1678 {
1679         __u32 hash = le32_to_cpu(header->h_hash);
1680         struct mb_cache_entry *ce;
1681         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1682
1683         if (!header->h_hash)
1684                 return NULL;  /* never share */
1685         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1686         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1687         while (ce) {
1688                 struct buffer_head *bh;
1689
1690                 bh = sb_bread(inode->i_sb, ce->e_block);
1691                 if (!bh) {
1692                         EXT4_ERROR_INODE(inode, "block %lu read error",
1693                                          (unsigned long) ce->e_block);
1694                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1695                         *pce = ce;
1696                         return bh;
1697                 }
1698                 brelse(bh);
1699                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1700         }
1701         return NULL;
1702 }
1703
1704 #define NAME_HASH_SHIFT 5
1705 #define VALUE_HASH_SHIFT 16
1706
1707 /*
1708  * ext4_xattr_hash_entry()
1709  *
1710  * Compute the hash of an extended attribute.
1711  */
1712 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1713                                          struct ext4_xattr_entry *entry)
1714 {
1715         __u32 hash = 0;
1716         char *name = entry->e_name;
1717         int n;
1718
1719         for (n = 0; n < entry->e_name_len; n++) {
1720                 hash = (hash << NAME_HASH_SHIFT) ^
1721                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1722                        *name++;
1723         }
1724
1725         if (entry->e_value_size != 0) {
1726                 __le32 *value = (__le32 *)((char *)header +
1727                         le16_to_cpu(entry->e_value_offs));
1728                 for (n = (le32_to_cpu(entry->e_value_size) +
1729                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1730                         hash = (hash << VALUE_HASH_SHIFT) ^
1731                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1732                                le32_to_cpu(*value++);
1733                 }
1734         }
1735         entry->e_hash = cpu_to_le32(hash);
1736 }
1737
1738 #undef NAME_HASH_SHIFT
1739 #undef VALUE_HASH_SHIFT
1740
1741 #define BLOCK_HASH_SHIFT 16
1742
1743 /*
1744  * ext4_xattr_rehash()
1745  *
1746  * Re-compute the extended attribute hash value after an entry has changed.
1747  */
1748 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1749                               struct ext4_xattr_entry *entry)
1750 {
1751         struct ext4_xattr_entry *here;
1752         __u32 hash = 0;
1753
1754         ext4_xattr_hash_entry(header, entry);
1755         here = ENTRY(header+1);
1756         while (!IS_LAST_ENTRY(here)) {
1757                 if (!here->e_hash) {
1758                         /* Block is not shared if an entry's hash value == 0 */
1759                         hash = 0;
1760                         break;
1761                 }
1762                 hash = (hash << BLOCK_HASH_SHIFT) ^
1763                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1764                        le32_to_cpu(here->e_hash);
1765                 here = EXT4_XATTR_NEXT(here);
1766         }
1767         header->h_hash = cpu_to_le32(hash);
1768 }
1769
1770 #undef BLOCK_HASH_SHIFT
1771
1772 #define HASH_BUCKET_BITS        10
1773
1774 struct mb_cache *
1775 ext4_xattr_create_cache(void)
1776 {
1777         return mb_cache_create(HASH_BUCKET_BITS);
1778 }
1779
1780 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1781 {
1782         if (cache)
1783                 mb_cache_destroy(cache);
1784 }
1785