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, f...) do { \
  65                 printk(KERN_DEBUG "inode %s:%lu: ", \
  66                         inode->i_sb->s_id, inode->i_ino); \
  67                 printk(f); \
  68                 printk("\n"); \
  69         } while (0)
  70 # define ea_bdebug(bh, f...) do { \
  71                 printk(KERN_DEBUG "block %pg:%lu: ",               \
  72                        bh->b_bdev, (unsigned long) bh->b_blocknr); \
  73                 printk(f); \
  74                 printk("\n"); \
  75         } while (0)
  76 #else
  77 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  78 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  79 #endif
  80
  81 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  82 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
  83                                                  struct ext4_xattr_header *,
  84                                                  struct mb_cache_entry **);
  85 static void ext4_xattr_rehash(struct ext4_xattr_header *,
  86                               struct ext4_xattr_entry *);
  87 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
  88                            size_t buffer_size);
  89
  90 static const struct xattr_handler *ext4_xattr_handler_map[] = {
  91         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  92 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  93         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  94         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  95 #endif
  96         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  97 #ifdef CONFIG_EXT4_FS_SECURITY
  98         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  99 #endif
 100 };
 101
 102 const struct xattr_handler *ext4_xattr_handlers[] = {
 103         &ext4_xattr_user_handler,
 104         &ext4_xattr_trusted_handler,
 105 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 106         &posix_acl_access_xattr_handler,
 107         &posix_acl_default_xattr_handler,
 108 #endif
 109 #ifdef CONFIG_EXT4_FS_SECURITY
 110         &ext4_xattr_security_handler,
 111 #endif
 112         NULL
 113 };
 114
 115 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
 116                                 inode->i_sb->s_fs_info)->s_mb_cache)
 117
 118 static __le32 ext4_xattr_block_csum(struct inode *inode,
 119                                     sector_t block_nr,
 120                                     struct ext4_xattr_header *hdr)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 123         __u32 csum;
 124         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 125         __u32 dummy_csum = 0;
 126         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 127
 128         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 129                            sizeof(dsk_block_nr));
 130         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 131         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 132         offset += sizeof(dummy_csum);
 133         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 134                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 135
 136         return cpu_to_le32(csum);
 137 }
 138
 139 static int ext4_xattr_block_csum_verify(struct inode *inode,
 140                                         sector_t block_nr,
 141                                         struct ext4_xattr_header *hdr)
 142 {
 143         if (ext4_has_metadata_csum(inode->i_sb) &&
 144             (hdr->h_checksum != ext4_xattr_block_csum(inode, block_nr, hdr)))
 145                 return 0;
 146         return 1;
 147 }
 148
 149 static void ext4_xattr_block_csum_set(struct inode *inode,
 150                                       sector_t block_nr,
 151                                       struct ext4_xattr_header *hdr)
 152 {
 153         if (!ext4_has_metadata_csum(inode->i_sb))
 154                 return;
 155
 156         hdr->h_checksum = ext4_xattr_block_csum(inode, block_nr, hdr);
 157 }
 158
 159 static inline int ext4_handle_dirty_xattr_block(handle_t *handle,
 160                                                 struct inode *inode,
 161                                                 struct buffer_head *bh)
 162 {
 163         ext4_xattr_block_csum_set(inode, bh->b_blocknr, BHDR(bh));
 164         return ext4_handle_dirty_metadata(handle, inode, bh);
 165 }
 166
 167 static inline const struct xattr_handler *
 168 ext4_xattr_handler(int name_index)
 169 {
 170         const struct xattr_handler *handler = NULL;
 171
 172         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 173                 handler = ext4_xattr_handler_map[name_index];
 174         return handler;
 175 }
 176
 177 /*
 178  * Inode operation listxattr()
 179  *
 180  * d_inode(dentry)->i_mutex: don't care
 181  */
 182 ssize_t
 183 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
 184 {
 185         return ext4_xattr_list(dentry, buffer, size);
 186 }
 187
 188 static int
 189 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
 190                        void *value_start)
 191 {
 192         struct ext4_xattr_entry *e = entry;
 193
 194         while (!IS_LAST_ENTRY(e)) {
 195                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 196                 if ((void *)next >= end)
 197                         return -EFSCORRUPTED;
 198                 e = next;
 199         }
 200
 201         while (!IS_LAST_ENTRY(entry)) {
 202                 if (entry->e_value_size != 0 &&
 203                     (value_start + le16_to_cpu(entry->e_value_offs) <
 204                      (void *)e + sizeof(__u32) ||
 205                      value_start + le16_to_cpu(entry->e_value_offs) +
 206                     le32_to_cpu(entry->e_value_size) > end))
 207                         return -EFSCORRUPTED;
 208                 entry = EXT4_XATTR_NEXT(entry);
 209         }
 210
 211         return 0;
 212 }
 213
 214 static inline int
 215 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 216 {
 217         int error;
 218
 219         if (buffer_verified(bh))
 220                 return 0;
 221
 222         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 223             BHDR(bh)->h_blocks != cpu_to_le32(1))
 224                 return -EFSCORRUPTED;
 225         if (!ext4_xattr_block_csum_verify(inode, bh->b_blocknr, BHDR(bh)))
 226                 return -EFSBADCRC;
 227         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
 228                                        bh->b_data);
 229         if (!error)
 230                 set_buffer_verified(bh);
 231         return error;
 232 }
 233
 234 static int
 235 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 236                          void *end, const char *function, unsigned int line)
 237 {
 238         struct ext4_xattr_entry *entry = IFIRST(header);
 239         int error = -EFSCORRUPTED;
 240
 241         if (((void *) header >= end) ||
 242             (header->h_magic != le32_to_cpu(EXT4_XATTR_MAGIC)))
 243                 goto errout;
 244         error = ext4_xattr_check_names(entry, end, entry);
 245 errout:
 246         if (error)
 247                 __ext4_error_inode(inode, function, line, 0,
 248                                    "corrupted in-inode xattr");
 249         return error;
 250 }
 251
 252 #define xattr_check_inode(inode, header, end) \
 253         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 254
 255 static inline int
 256 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 257 {
 258         size_t value_size = le32_to_cpu(entry->e_value_size);
 259
 260         if (entry->e_value_block != 0 || value_size > size ||
 261             le16_to_cpu(entry->e_value_offs) + value_size > size)
 262                 return -EFSCORRUPTED;
 263         return 0;
 264 }
 265
 266 static int
 267 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 268                       const char *name, size_t size, int sorted)
 269 {
 270         struct ext4_xattr_entry *entry;
 271         size_t name_len;
 272         int cmp = 1;
 273
 274         if (name == NULL)
 275                 return -EINVAL;
 276         name_len = strlen(name);
 277         entry = *pentry;
 278         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 279                 cmp = name_index - entry->e_name_index;
 280                 if (!cmp)
 281                         cmp = name_len - entry->e_name_len;
 282                 if (!cmp)
 283                         cmp = memcmp(name, entry->e_name, name_len);
 284                 if (cmp <= 0 && (sorted || cmp == 0))
 285                         break;
 286         }
 287         *pentry = entry;
 288         if (!cmp && ext4_xattr_check_entry(entry, size))
 289                 return -EFSCORRUPTED;
 290         return cmp ? -ENODATA : 0;
 291 }
 292
 293 static int
 294 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 295                      void *buffer, size_t buffer_size)
 296 {
 297         struct buffer_head *bh = NULL;
 298         struct ext4_xattr_entry *entry;
 299         size_t size;
 300         int error;
 301         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 302
 303         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 304                   name_index, name, buffer, (long)buffer_size);
 305
 306         error = -ENODATA;
 307         if (!EXT4_I(inode)->i_file_acl)
 308                 goto cleanup;
 309         ea_idebug(inode, "reading block %llu",
 310                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 311         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 312         if (!bh)
 313                 goto cleanup;
 314         ea_bdebug(bh, "b_count=%d, refcount=%d",
 315                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 316         if (ext4_xattr_check_block(inode, bh)) {
 317 bad_block:
 318                 EXT4_ERROR_INODE(inode, "bad block %llu",
 319                                  EXT4_I(inode)->i_file_acl);
 320                 error = -EFSCORRUPTED;
 321                 goto cleanup;
 322         }
 323         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 324         entry = BFIRST(bh);
 325         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 326         if (error == -EFSCORRUPTED)
 327                 goto bad_block;
 328         if (error)
 329                 goto cleanup;
 330         size = le32_to_cpu(entry->e_value_size);
 331         if (buffer) {
 332                 error = -ERANGE;
 333                 if (size > buffer_size)
 334                         goto cleanup;
 335                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 336                        size);
 337         }
 338         error = size;
 339
 340 cleanup:
 341         brelse(bh);
 342         return error;
 343 }
 344
 345 int
 346 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 347                      void *buffer, size_t buffer_size)
 348 {
 349         struct ext4_xattr_ibody_header *header;
 350         struct ext4_xattr_entry *entry;
 351         struct ext4_inode *raw_inode;
 352         struct ext4_iloc iloc;
 353         size_t size;
 354         void *end;
 355         int error;
 356
 357         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 358                 return -ENODATA;
 359         error = ext4_get_inode_loc(inode, &iloc);
 360         if (error)
 361                 return error;
 362         raw_inode = ext4_raw_inode(&iloc);
 363         header = IHDR(inode, raw_inode);
 364         entry = IFIRST(header);
 365         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 366         error = xattr_check_inode(inode, header, end);
 367         if (error)
 368                 goto cleanup;
 369         error = ext4_xattr_find_entry(&entry, name_index, name,
 370                                       end - (void *)entry, 0);
 371         if (error)
 372                 goto cleanup;
 373         size = le32_to_cpu(entry->e_value_size);
 374         if (buffer) {
 375                 error = -ERANGE;
 376                 if (size > buffer_size)
 377                         goto cleanup;
 378                 memcpy(buffer, (void *)IFIRST(header) +
 379                        le16_to_cpu(entry->e_value_offs), size);
 380         }
 381         error = size;
 382
 383 cleanup:
 384         brelse(iloc.bh);
 385         return error;
 386 }
 387
 388 /*
 389  * ext4_xattr_get()
 390  *
 391  * Copy an extended attribute into the buffer
 392  * provided, or compute the buffer size required.
 393  * Buffer is NULL to compute the size of the buffer required.
 394  *
 395  * Returns a negative error number on failure, or the number of bytes
 396  * used / required on success.
 397  */
 398 int
 399 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 400                void *buffer, size_t buffer_size)
 401 {
 402         int error;
 403
 404         if (strlen(name) > 255)
 405                 return -ERANGE;
 406
 407         down_read(&EXT4_I(inode)->xattr_sem);
 408         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 409                                      buffer_size);
 410         if (error == -ENODATA)
 411                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 412                                              buffer_size);
 413         up_read(&EXT4_I(inode)->xattr_sem);
 414         return error;
 415 }
 416
 417 static int
 418 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 419                         char *buffer, size_t buffer_size)
 420 {
 421         size_t rest = buffer_size;
 422
 423         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 424                 const struct xattr_handler *handler =
 425                         ext4_xattr_handler(entry->e_name_index);
 426
 427                 if (handler && (!handler->list || handler->list(dentry))) {
 428                         const char *prefix = handler->prefix ?: handler->name;
 429                         size_t prefix_len = strlen(prefix);
 430                         size_t size = prefix_len + entry->e_name_len + 1;
 431
 432                         if (buffer) {
 433                                 if (size > rest)
 434                                         return -ERANGE;
 435                                 memcpy(buffer, prefix, prefix_len);
 436                                 buffer += prefix_len;
 437                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 438                                 buffer += entry->e_name_len;
 439                                 *buffer++ = 0;
 440                         }
 441                         rest -= size;
 442                 }
 443         }
 444         return buffer_size - rest;  /* total size */
 445 }
 446
 447 static int
 448 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 449 {
 450         struct inode *inode = d_inode(dentry);
 451         struct buffer_head *bh = NULL;
 452         int error;
 453         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 454
 455         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 456                   buffer, (long)buffer_size);
 457
 458         error = 0;
 459         if (!EXT4_I(inode)->i_file_acl)
 460                 goto cleanup;
 461         ea_idebug(inode, "reading block %llu",
 462                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 463         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 464         error = -EIO;
 465         if (!bh)
 466                 goto cleanup;
 467         ea_bdebug(bh, "b_count=%d, refcount=%d",
 468                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 469         if (ext4_xattr_check_block(inode, bh)) {
 470                 EXT4_ERROR_INODE(inode, "bad block %llu",
 471                                  EXT4_I(inode)->i_file_acl);
 472                 error = -EFSCORRUPTED;
 473                 goto cleanup;
 474         }
 475         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 476         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 477
 478 cleanup:
 479         brelse(bh);
 480
 481         return error;
 482 }
 483
 484 static int
 485 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 486 {
 487         struct inode *inode = d_inode(dentry);
 488         struct ext4_xattr_ibody_header *header;
 489         struct ext4_inode *raw_inode;
 490         struct ext4_iloc iloc;
 491         void *end;
 492         int error;
 493
 494         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 495                 return 0;
 496         error = ext4_get_inode_loc(inode, &iloc);
 497         if (error)
 498                 return error;
 499         raw_inode = ext4_raw_inode(&iloc);
 500         header = IHDR(inode, raw_inode);
 501         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 502         error = xattr_check_inode(inode, header, end);
 503         if (error)
 504                 goto cleanup;
 505         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 506                                         buffer, buffer_size);
 507
 508 cleanup:
 509         brelse(iloc.bh);
 510         return error;
 511 }
 512
 513 /*
 514  * ext4_xattr_list()
 515  *
 516  * Copy a list of attribute names into the buffer
 517  * provided, or compute the buffer size required.
 518  * Buffer is NULL to compute the size of the buffer required.
 519  *
 520  * Returns a negative error number on failure, or the number of bytes
 521  * used / required on success.
 522  */
 523 static int
 524 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 525 {
 526         int ret, ret2;
 527
 528         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 529         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 530         if (ret < 0)
 531                 goto errout;
 532         if (buffer) {
 533                 buffer += ret;
 534                 buffer_size -= ret;
 535         }
 536         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 537         if (ret < 0)
 538                 goto errout;
 539         ret += ret2;
 540 errout:
 541         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 542         return ret;
 543 }
 544
 545 /*
 546  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 547  * not set, set it.
 548  */
 549 static void ext4_xattr_update_super_block(handle_t *handle,
 550                                           struct super_block *sb)
 551 {
 552         if (ext4_has_feature_xattr(sb))
 553                 return;
 554
 555         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 556         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 557                 ext4_set_feature_xattr(sb);
 558                 ext4_handle_dirty_super(handle, sb);
 559         }
 560 }
 561
 562 /*
 563  * Release the xattr block BH: If the reference count is > 1, decrement it;
 564  * otherwise free the block.
 565  */
 566 static void
 567 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 568                          struct buffer_head *bh)
 569 {
 570         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 571         u32 hash, ref;
 572         int error = 0;
 573
 574         BUFFER_TRACE(bh, "get_write_access");
 575         error = ext4_journal_get_write_access(handle, bh);
 576         if (error)
 577                 goto out;
 578
 579         lock_buffer(bh);
 580         hash = le32_to_cpu(BHDR(bh)->h_hash);
 581         ref = le32_to_cpu(BHDR(bh)->h_refcount);
 582         if (ref == 1) {
 583                 ea_bdebug(bh, "refcount now=0; freeing");
 584                 /*
 585                  * This must happen under buffer lock for
 586                  * ext4_xattr_block_set() to reliably detect freed block
 587                  */
 588                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
 589                 get_bh(bh);
 590                 unlock_buffer(bh);
 591                 ext4_free_blocks(handle, inode, bh, 0, 1,
 592                                  EXT4_FREE_BLOCKS_METADATA |
 593                                  EXT4_FREE_BLOCKS_FORGET);
 594         } else {
 595                 ref--;
 596                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
 597                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
 598                         struct mb_cache_entry *ce;
 599
 600                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
 601                                                 bh->b_blocknr);
 602                         if (ce) {
 603                                 ce->e_reusable = 1;
 604                                 mb_cache_entry_put(ext4_mb_cache, ce);
 605                         }
 606                 }
 607
 608                 /*
 609                  * Beware of this ugliness: Releasing of xattr block references
 610                  * from different inodes can race and so we have to protect
 611                  * from a race where someone else frees the block (and releases
 612                  * its journal_head) before we are done dirtying the buffer. In
 613                  * nojournal mode this race is harmless and we actually cannot
 614                  * call ext4_handle_dirty_xattr_block() with locked buffer as
 615                  * that function can call sync_dirty_buffer() so for that case
 616                  * we handle the dirtying after unlocking the buffer.
 617                  */
 618                 if (ext4_handle_valid(handle))
 619                         error = ext4_handle_dirty_xattr_block(handle, inode,
 620                                                               bh);
 621                 unlock_buffer(bh);
 622                 if (!ext4_handle_valid(handle))
 623                         error = ext4_handle_dirty_xattr_block(handle, inode,
 624                                                               bh);
 625                 if (IS_SYNC(inode))
 626                         ext4_handle_sync(handle);
 627                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
 628                 ea_bdebug(bh, "refcount now=%d; releasing",
 629                           le32_to_cpu(BHDR(bh)->h_refcount));
 630         }
 631 out:
 632         ext4_std_error(inode->i_sb, error);
 633         return;
 634 }
 635
 636 /*
 637  * Find the available free space for EAs. This also returns the total number of
 638  * bytes used by EA entries.
 639  */
 640 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
 641                                     size_t *min_offs, void *base, int *total)
 642 {
 643         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 644                 if (!last->e_value_block && last->e_value_size) {
 645                         size_t offs = le16_to_cpu(last->e_value_offs);
 646                         if (offs < *min_offs)
 647                                 *min_offs = offs;
 648                 }
 649                 if (total)
 650                         *total += EXT4_XATTR_LEN(last->e_name_len);
 651         }
 652         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 653 }
 654
 655 static int
 656 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 657 {
 658         struct ext4_xattr_entry *last;
 659         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 660
 661         /* Compute min_offs and last. */
 662         last = s->first;
 663         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 664                 if (!last->e_value_block && last->e_value_size) {
 665                         size_t offs = le16_to_cpu(last->e_value_offs);
 666                         if (offs < min_offs)
 667                                 min_offs = offs;
 668                 }
 669         }
 670         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 671         if (!s->not_found) {
 672                 if (!s->here->e_value_block && s->here->e_value_size) {
 673                         size_t size = le32_to_cpu(s->here->e_value_size);
 674                         free += EXT4_XATTR_SIZE(size);
 675                 }
 676                 free += EXT4_XATTR_LEN(name_len);
 677         }
 678         if (i->value) {
 679                 if (free < EXT4_XATTR_LEN(name_len) +
 680                            EXT4_XATTR_SIZE(i->value_len))
 681                         return -ENOSPC;
 682         }
 683
 684         if (i->value && s->not_found) {
 685                 /* Insert the new name. */
 686                 size_t size = EXT4_XATTR_LEN(name_len);
 687                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
 688                 memmove((void *)s->here + size, s->here, rest);
 689                 memset(s->here, 0, size);
 690                 s->here->e_name_index = i->name_index;
 691                 s->here->e_name_len = name_len;
 692                 memcpy(s->here->e_name, i->name, name_len);
 693         } else {
 694                 if (!s->here->e_value_block && s->here->e_value_size) {
 695                         void *first_val = s->base + min_offs;
 696                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 697                         void *val = s->base + offs;
 698                         size_t size = EXT4_XATTR_SIZE(
 699                                 le32_to_cpu(s->here->e_value_size));
 700
 701                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
 702                                 /* The old and the new value have the same
 703                                    size. Just replace. */
 704                                 s->here->e_value_size =
 705                                         cpu_to_le32(i->value_len);
 706                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
 707                                         memset(val, 0, size);
 708                                 } else {
 709                                         /* Clear pad bytes first. */
 710                                         memset(val + size - EXT4_XATTR_PAD, 0,
 711                                                EXT4_XATTR_PAD);
 712                                         memcpy(val, i->value, i->value_len);
 713                                 }
 714                                 return 0;
 715                         }
 716
 717                         /* Remove the old value. */
 718                         memmove(first_val + size, first_val, val - first_val);
 719                         memset(first_val, 0, size);
 720                         s->here->e_value_size = 0;
 721                         s->here->e_value_offs = 0;
 722                         min_offs += size;
 723
 724                         /* Adjust all value offsets. */
 725                         last = s->first;
 726                         while (!IS_LAST_ENTRY(last)) {
 727                                 size_t o = le16_to_cpu(last->e_value_offs);
 728                                 if (!last->e_value_block &&
 729                                     last->e_value_size && o < offs)
 730                                         last->e_value_offs =
 731                                                 cpu_to_le16(o + size);
 732                                 last = EXT4_XATTR_NEXT(last);
 733                         }
 734                 }
 735                 if (!i->value) {
 736                         /* Remove the old name. */
 737                         size_t size = EXT4_XATTR_LEN(name_len);
 738                         last = ENTRY((void *)last - size);
 739                         memmove(s->here, (void *)s->here + size,
 740                                 (void *)last - (void *)s->here + sizeof(__u32));
 741                         memset(last, 0, size);
 742                 }
 743         }
 744
 745         if (i->value) {
 746                 /* Insert the new value. */
 747                 s->here->e_value_size = cpu_to_le32(i->value_len);
 748                 if (i->value_len) {
 749                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 750                         void *val = s->base + min_offs - size;
 751                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 752                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 753                                 memset(val, 0, size);
 754                         } else {
 755                                 /* Clear the pad bytes first. */
 756                                 memset(val + size - EXT4_XATTR_PAD, 0,
 757                                        EXT4_XATTR_PAD);
 758                                 memcpy(val, i->value, i->value_len);
 759                         }
 760                 }
 761         }
 762         return 0;
 763 }
 764
 765 struct ext4_xattr_block_find {
 766         struct ext4_xattr_search s;
 767         struct buffer_head *bh;
 768 };
 769
 770 static int
 771 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
 772                       struct ext4_xattr_block_find *bs)
 773 {
 774         struct super_block *sb = inode->i_sb;
 775         int error;
 776
 777         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 778                   i->name_index, i->name, i->value, (long)i->value_len);
 779
 780         if (EXT4_I(inode)->i_file_acl) {
 781                 /* The inode already has an extended attribute block. */
 782                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
 783                 error = -EIO;
 784                 if (!bs->bh)
 785                         goto cleanup;
 786                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
 787                         atomic_read(&(bs->bh->b_count)),
 788                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
 789                 if (ext4_xattr_check_block(inode, bs->bh)) {
 790                         EXT4_ERROR_INODE(inode, "bad block %llu",
 791                                          EXT4_I(inode)->i_file_acl);
 792                         error = -EFSCORRUPTED;
 793                         goto cleanup;
 794                 }
 795                 /* Find the named attribute. */
 796                 bs->s.base = BHDR(bs->bh);
 797                 bs->s.first = BFIRST(bs->bh);
 798                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 799                 bs->s.here = bs->s.first;
 800                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 801                                               i->name, bs->bh->b_size, 1);
 802                 if (error && error != -ENODATA)
 803                         goto cleanup;
 804                 bs->s.not_found = error;
 805         }
 806         error = 0;
 807
 808 cleanup:
 809         return error;
 810 }
 811
 812 static int
 813 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 814                      struct ext4_xattr_info *i,
 815                      struct ext4_xattr_block_find *bs)
 816 {
 817         struct super_block *sb = inode->i_sb;
 818         struct buffer_head *new_bh = NULL;
 819         struct ext4_xattr_search *s = &bs->s;
 820         struct mb_cache_entry *ce = NULL;
 821         int error = 0;
 822         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 823
 824 #define header(x) ((struct ext4_xattr_header *)(x))
 825
 826         if (i->value && i->value_len > sb->s_blocksize)
 827                 return -ENOSPC;
 828         if (s->base) {
 829                 BUFFER_TRACE(bs->bh, "get_write_access");
 830                 error = ext4_journal_get_write_access(handle, bs->bh);
 831                 if (error)
 832                         goto cleanup;
 833                 lock_buffer(bs->bh);
 834
 835                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
 836                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
 837
 838                         /*
 839                          * This must happen under buffer lock for
 840                          * ext4_xattr_block_set() to reliably detect modified
 841                          * block
 842                          */
 843                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
 844                                                     bs->bh->b_blocknr);
 845                         ea_bdebug(bs->bh, "modifying in-place");
 846                         error = ext4_xattr_set_entry(i, s);
 847                         if (!error) {
 848                                 if (!IS_LAST_ENTRY(s->first))
 849                                         ext4_xattr_rehash(header(s->base),
 850                                                           s->here);
 851                                 ext4_xattr_cache_insert(ext4_mb_cache,
 852                                         bs->bh);
 853                         }
 854                         unlock_buffer(bs->bh);
 855                         if (error == -EFSCORRUPTED)
 856                                 goto bad_block;
 857                         if (!error)
 858                                 error = ext4_handle_dirty_xattr_block(handle,
 859                                                                       inode,
 860                                                                       bs->bh);
 861                         if (error)
 862                                 goto cleanup;
 863                         goto inserted;
 864                 } else {
 865                         int offset = (char *)s->here - bs->bh->b_data;
 866
 867                         unlock_buffer(bs->bh);
 868                         ea_bdebug(bs->bh, "cloning");
 869                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
 870                         error = -ENOMEM;
 871                         if (s->base == NULL)
 872                                 goto cleanup;
 873                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
 874                         s->first = ENTRY(header(s->base)+1);
 875                         header(s->base)->h_refcount = cpu_to_le32(1);
 876                         s->here = ENTRY(s->base + offset);
 877                         s->end = s->base + bs->bh->b_size;
 878                 }
 879         } else {
 880                 /* Allocate a buffer where we construct the new block. */
 881                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 882                 /* assert(header == s->base) */
 883                 error = -ENOMEM;
 884                 if (s->base == NULL)
 885                         goto cleanup;
 886                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
 887                 header(s->base)->h_blocks = cpu_to_le32(1);
 888                 header(s->base)->h_refcount = cpu_to_le32(1);
 889                 s->first = ENTRY(header(s->base)+1);
 890                 s->here = ENTRY(header(s->base)+1);
 891                 s->end = s->base + sb->s_blocksize;
 892         }
 893
 894         error = ext4_xattr_set_entry(i, s);
 895         if (error == -EFSCORRUPTED)
 896                 goto bad_block;
 897         if (error)
 898                 goto cleanup;
 899         if (!IS_LAST_ENTRY(s->first))
 900                 ext4_xattr_rehash(header(s->base), s->here);
 901
 902 inserted:
 903         if (!IS_LAST_ENTRY(s->first)) {
 904                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
 905                 if (new_bh) {
 906                         /* We found an identical block in the cache. */
 907                         if (new_bh == bs->bh)
 908                                 ea_bdebug(new_bh, "keeping");
 909                         else {
 910                                 u32 ref;
 911
 912                                 /* The old block is released after updating
 913                                    the inode. */
 914                                 error = dquot_alloc_block(inode,
 915                                                 EXT4_C2B(EXT4_SB(sb), 1));
 916                                 if (error)
 917                                         goto cleanup;
 918                                 BUFFER_TRACE(new_bh, "get_write_access");
 919                                 error = ext4_journal_get_write_access(handle,
 920                                                                       new_bh);
 921                                 if (error)
 922                                         goto cleanup_dquot;
 923                                 lock_buffer(new_bh);
 924                                 /*
 925                                  * We have to be careful about races with
 926                                  * freeing, rehashing or adding references to
 927                                  * xattr block. Once we hold buffer lock xattr
 928                                  * block's state is stable so we can check
 929                                  * whether the block got freed / rehashed or
 930                                  * not.  Since we unhash mbcache entry under
 931                                  * buffer lock when freeing / rehashing xattr
 932                                  * block, checking whether entry is still
 933                                  * hashed is reliable. Same rules hold for
 934                                  * e_reusable handling.
 935                                  */
 936                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
 937                                     !ce->e_reusable) {
 938                                         /*
 939                                          * Undo everything and check mbcache
 940                                          * again.
 941                                          */
 942                                         unlock_buffer(new_bh);
 943                                         dquot_free_block(inode,
 944                                                          EXT4_C2B(EXT4_SB(sb),
 945                                                                   1));
 946                                         brelse(new_bh);
 947                                         mb_cache_entry_put(ext4_mb_cache, ce);
 948                                         ce = NULL;
 949                                         new_bh = NULL;
 950                                         goto inserted;
 951                                 }
 952                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
 953                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
 954                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
 955                                         ce->e_reusable = 0;
 956                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
 957                                           ref);
 958                                 unlock_buffer(new_bh);
 959                                 error = ext4_handle_dirty_xattr_block(handle,
 960                                                                       inode,
 961                                                                       new_bh);
 962                                 if (error)
 963                                         goto cleanup_dquot;
 964                         }
 965                         mb_cache_entry_touch(ext4_mb_cache, ce);
 966                         mb_cache_entry_put(ext4_mb_cache, ce);
 967                         ce = NULL;
 968                 } else if (bs->bh && s->base == bs->bh->b_data) {
 969                         /* We were modifying this block in-place. */
 970                         ea_bdebug(bs->bh, "keeping this block");
 971                         new_bh = bs->bh;
 972                         get_bh(new_bh);
 973                 } else {
 974                         /* We need to allocate a new block */
 975                         ext4_fsblk_t goal, block;
 976
 977                         goal = ext4_group_first_block_no(sb,
 978                                                 EXT4_I(inode)->i_block_group);
 979
 980                         /* non-extent files can't have physical blocks past 2^32 */
 981                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 982                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 983
 984                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
 985                                                      NULL, &error);
 986                         if (error)
 987                                 goto cleanup;
 988
 989                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 990                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
 991
 992                         ea_idebug(inode, "creating block %llu",
 993                                   (unsigned long long)block);
 994
 995                         new_bh = sb_getblk(sb, block);
 996                         if (unlikely(!new_bh)) {
 997                                 error = -ENOMEM;
 998 getblk_failed:
 999                                 ext4_free_blocks(handle, inode, NULL, block, 1,
1000                                                  EXT4_FREE_BLOCKS_METADATA);
1001                                 goto cleanup;
1002                         }
1003                         lock_buffer(new_bh);
1004                         error = ext4_journal_get_create_access(handle, new_bh);
1005                         if (error) {
1006                                 unlock_buffer(new_bh);
1007                                 error = -EIO;
1008                                 goto getblk_failed;
1009                         }
1010                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1011                         set_buffer_uptodate(new_bh);
1012                         unlock_buffer(new_bh);
1013                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1014                         error = ext4_handle_dirty_xattr_block(handle,
1015                                                               inode, new_bh);
1016                         if (error)
1017                                 goto cleanup;
1018                 }
1019         }
1020
1021         /* Update the inode. */
1022         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1023
1024         /* Drop the previous xattr block. */
1025         if (bs->bh && bs->bh != new_bh)
1026                 ext4_xattr_release_block(handle, inode, bs->bh);
1027         error = 0;
1028
1029 cleanup:
1030         if (ce)
1031                 mb_cache_entry_put(ext4_mb_cache, ce);
1032         brelse(new_bh);
1033         if (!(bs->bh && s->base == bs->bh->b_data))
1034                 kfree(s->base);
1035
1036         return error;
1037
1038 cleanup_dquot:
1039         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1040         goto cleanup;
1041
1042 bad_block:
1043         EXT4_ERROR_INODE(inode, "bad block %llu",
1044                          EXT4_I(inode)->i_file_acl);
1045         goto cleanup;
1046
1047 #undef header
1048 }
1049
1050 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1051                           struct ext4_xattr_ibody_find *is)
1052 {
1053         struct ext4_xattr_ibody_header *header;
1054         struct ext4_inode *raw_inode;
1055         int error;
1056
1057         if (EXT4_I(inode)->i_extra_isize == 0)
1058                 return 0;
1059         raw_inode = ext4_raw_inode(&is->iloc);
1060         header = IHDR(inode, raw_inode);
1061         is->s.base = is->s.first = IFIRST(header);
1062         is->s.here = is->s.first;
1063         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1064         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1065                 error = xattr_check_inode(inode, header, is->s.end);
1066                 if (error)
1067                         return error;
1068                 /* Find the named attribute. */
1069                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1070                                               i->name, is->s.end -
1071                                               (void *)is->s.base, 0);
1072                 if (error && error != -ENODATA)
1073                         return error;
1074                 is->s.not_found = error;
1075         }
1076         return 0;
1077 }
1078
1079 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1080                                 struct ext4_xattr_info *i,
1081                                 struct ext4_xattr_ibody_find *is)
1082 {
1083         struct ext4_xattr_ibody_header *header;
1084         struct ext4_xattr_search *s = &is->s;
1085         int error;
1086
1087         if (EXT4_I(inode)->i_extra_isize == 0)
1088                 return -ENOSPC;
1089         error = ext4_xattr_set_entry(i, s);
1090         if (error) {
1091                 if (error == -ENOSPC &&
1092                     ext4_has_inline_data(inode)) {
1093                         error = ext4_try_to_evict_inline_data(handle, inode,
1094                                         EXT4_XATTR_LEN(strlen(i->name) +
1095                                         EXT4_XATTR_SIZE(i->value_len)));
1096                         if (error)
1097                                 return error;
1098                         error = ext4_xattr_ibody_find(inode, i, is);
1099                         if (error)
1100                                 return error;
1101                         error = ext4_xattr_set_entry(i, s);
1102                 }
1103                 if (error)
1104                         return error;
1105         }
1106         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1107         if (!IS_LAST_ENTRY(s->first)) {
1108                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1109                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1110         } else {
1111                 header->h_magic = cpu_to_le32(0);
1112                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1113         }
1114         return 0;
1115 }
1116
1117 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1118                                 struct ext4_xattr_info *i,
1119                                 struct ext4_xattr_ibody_find *is)
1120 {
1121         struct ext4_xattr_ibody_header *header;
1122         struct ext4_xattr_search *s = &is->s;
1123         int error;
1124
1125         if (EXT4_I(inode)->i_extra_isize == 0)
1126                 return -ENOSPC;
1127         error = ext4_xattr_set_entry(i, s);
1128         if (error)
1129                 return error;
1130         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1131         if (!IS_LAST_ENTRY(s->first)) {
1132                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1133                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1134         } else {
1135                 header->h_magic = cpu_to_le32(0);
1136                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1137         }
1138         return 0;
1139 }
1140
1141 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1142                                  struct ext4_xattr_info *i)
1143 {
1144         void *value;
1145
1146         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1147                 return 0;
1148         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1149         return !memcmp(value, i->value, i->value_len);
1150 }
1151
1152 /*
1153  * ext4_xattr_set_handle()
1154  *
1155  * Create, replace or remove an extended attribute for this inode.  Value
1156  * is NULL to remove an existing extended attribute, and non-NULL to
1157  * either replace an existing extended attribute, or create a new extended
1158  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1159  * specify that an extended attribute must exist and must not exist
1160  * previous to the call, respectively.
1161  *
1162  * Returns 0, or a negative error number on failure.
1163  */
1164 int
1165 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1166                       const char *name, const void *value, size_t value_len,
1167                       int flags)
1168 {
1169         struct ext4_xattr_info i = {
1170                 .name_index = name_index,
1171                 .name = name,
1172                 .value = value,
1173                 .value_len = value_len,
1174
1175         };
1176         struct ext4_xattr_ibody_find is = {
1177                 .s = { .not_found = -ENODATA, },
1178         };
1179         struct ext4_xattr_block_find bs = {
1180                 .s = { .not_found = -ENODATA, },
1181         };
1182         unsigned long no_expand;
1183         int error;
1184
1185         if (!name)
1186                 return -EINVAL;
1187         if (strlen(name) > 255)
1188                 return -ERANGE;
1189         down_write(&EXT4_I(inode)->xattr_sem);
1190         no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1191         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1192
1193         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1194         if (error)
1195                 goto cleanup;
1196
1197         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1198                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1199                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1200                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1201         }
1202
1203         error = ext4_xattr_ibody_find(inode, &i, &is);
1204         if (error)
1205                 goto cleanup;
1206         if (is.s.not_found)
1207                 error = ext4_xattr_block_find(inode, &i, &bs);
1208         if (error)
1209                 goto cleanup;
1210         if (is.s.not_found && bs.s.not_found) {
1211                 error = -ENODATA;
1212                 if (flags & XATTR_REPLACE)
1213                         goto cleanup;
1214                 error = 0;
1215                 if (!value)
1216                         goto cleanup;
1217         } else {
1218                 error = -EEXIST;
1219                 if (flags & XATTR_CREATE)
1220                         goto cleanup;
1221         }
1222         if (!value) {
1223                 if (!is.s.not_found)
1224                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1225                 else if (!bs.s.not_found)
1226                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1227         } else {
1228                 error = 0;
1229                 /* Xattr value did not change? Save us some work and bail out */
1230                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1231                         goto cleanup;
1232                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1233                         goto cleanup;
1234
1235                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1236                 if (!error && !bs.s.not_found) {
1237                         i.value = NULL;
1238                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1239                 } else if (error == -ENOSPC) {
1240                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1241                                 error = ext4_xattr_block_find(inode, &i, &bs);
1242                                 if (error)
1243                                         goto cleanup;
1244                         }
1245                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1246                         if (error)
1247                                 goto cleanup;
1248                         if (!is.s.not_found) {
1249                                 i.value = NULL;
1250                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1251                                                              &is);
1252                         }
1253                 }
1254         }
1255         if (!error) {
1256                 ext4_xattr_update_super_block(handle, inode->i_sb);
1257                 inode->i_ctime = ext4_current_time(inode);
1258                 if (!value)
1259                         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1260                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1261                 /*
1262                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1263                  * error != 0.
1264                  */
1265                 is.iloc.bh = NULL;
1266                 if (IS_SYNC(inode))
1267                         ext4_handle_sync(handle);
1268         }
1269
1270 cleanup:
1271         brelse(is.iloc.bh);
1272         brelse(bs.bh);
1273         if (no_expand == 0)
1274                 ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1275         up_write(&EXT4_I(inode)->xattr_sem);
1276         return error;
1277 }
1278
1279 /*
1280  * ext4_xattr_set()
1281  *
1282  * Like ext4_xattr_set_handle, but start from an inode. This extended
1283  * attribute modification is a filesystem transaction by itself.
1284  *
1285  * Returns 0, or a negative error number on failure.
1286  */
1287 int
1288 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1289                const void *value, size_t value_len, int flags)
1290 {
1291         handle_t *handle;
1292         int error, retries = 0;
1293         int credits = ext4_jbd2_credits_xattr(inode);
1294
1295 retry:
1296         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1297         if (IS_ERR(handle)) {
1298                 error = PTR_ERR(handle);
1299         } else {
1300                 int error2;
1301
1302                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1303                                               value, value_len, flags);
1304                 error2 = ext4_journal_stop(handle);
1305                 if (error == -ENOSPC &&
1306                     ext4_should_retry_alloc(inode->i_sb, &retries))
1307                         goto retry;
1308                 if (error == 0)
1309                         error = error2;
1310         }
1311
1312         return error;
1313 }
1314
1315 /*
1316  * Shift the EA entries in the inode to create space for the increased
1317  * i_extra_isize.
1318  */
1319 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1320                                      int value_offs_shift, void *to,
1321                                      void *from, size_t n, int blocksize)
1322 {
1323         struct ext4_xattr_entry *last = entry;
1324         int new_offs;
1325
1326         /* Adjust the value offsets of the entries */
1327         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1328                 if (!last->e_value_block && last->e_value_size) {
1329                         new_offs = le16_to_cpu(last->e_value_offs) +
1330                                                         value_offs_shift;
1331                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1332                                  > blocksize);
1333                         last->e_value_offs = cpu_to_le16(new_offs);
1334                 }
1335         }
1336         /* Shift the entries by n bytes */
1337         memmove(to, from, n);
1338 }
1339
1340 /*
1341  * Expand an inode by new_extra_isize bytes when EAs are present.
1342  * Returns 0 on success or negative error number on failure.
1343  */
1344 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1345                                struct ext4_inode *raw_inode, handle_t *handle)
1346 {
1347         struct ext4_xattr_ibody_header *header;
1348         struct ext4_xattr_entry *entry, *last, *first;
1349         struct buffer_head *bh = NULL;
1350         struct ext4_xattr_ibody_find *is = NULL;
1351         struct ext4_xattr_block_find *bs = NULL;
1352         char *buffer = NULL, *b_entry_name = NULL;
1353         size_t min_offs, free;
1354         int total_ino;
1355         void *base, *start, *end;
1356         int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1357         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1358
1359         down_write(&EXT4_I(inode)->xattr_sem);
1360 retry:
1361         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1362                 up_write(&EXT4_I(inode)->xattr_sem);
1363                 return 0;
1364         }
1365
1366         header = IHDR(inode, raw_inode);
1367         entry = IFIRST(header);
1368
1369         /*
1370          * Check if enough free space is available in the inode to shift the
1371          * entries ahead by new_extra_isize.
1372          */
1373
1374         base = start = entry;
1375         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1376         min_offs = end - base;
1377         last = entry;
1378         total_ino = sizeof(struct ext4_xattr_ibody_header);
1379
1380         error = xattr_check_inode(inode, header, end);
1381         if (error)
1382                 goto cleanup;
1383
1384         free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1385         if (free >= new_extra_isize) {
1386                 entry = IFIRST(header);
1387                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize
1388                                 - new_extra_isize, (void *)raw_inode +
1389                                 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1390                                 (void *)header, total_ino,
1391                                 inode->i_sb->s_blocksize);
1392                 EXT4_I(inode)->i_extra_isize = new_extra_isize;
1393                 error = 0;
1394                 goto cleanup;
1395         }
1396
1397         /*
1398          * Enough free space isn't available in the inode, check if
1399          * EA block can hold new_extra_isize bytes.
1400          */
1401         if (EXT4_I(inode)->i_file_acl) {
1402                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1403                 error = -EIO;
1404                 if (!bh)
1405                         goto cleanup;
1406                 if (ext4_xattr_check_block(inode, bh)) {
1407                         EXT4_ERROR_INODE(inode, "bad block %llu",
1408                                          EXT4_I(inode)->i_file_acl);
1409                         error = -EFSCORRUPTED;
1410                         goto cleanup;
1411                 }
1412                 base = BHDR(bh);
1413                 first = BFIRST(bh);
1414                 end = bh->b_data + bh->b_size;
1415                 min_offs = end - base;
1416                 free = ext4_xattr_free_space(first, &min_offs, base, NULL);
1417                 if (free < new_extra_isize) {
1418                         if (!tried_min_extra_isize && s_min_extra_isize) {
1419                                 tried_min_extra_isize++;
1420                                 new_extra_isize = s_min_extra_isize;
1421                                 brelse(bh);
1422                                 goto retry;
1423                         }
1424                         error = -1;
1425                         goto cleanup;
1426                 }
1427         } else {
1428                 free = inode->i_sb->s_blocksize;
1429         }
1430
1431         while (new_extra_isize > 0) {
1432                 size_t offs, size, entry_size;
1433                 struct ext4_xattr_entry *small_entry = NULL;
1434                 struct ext4_xattr_info i = {
1435                         .value = NULL,
1436                         .value_len = 0,
1437                 };
1438                 unsigned int total_size;  /* EA entry size + value size */
1439                 unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1440                 unsigned int min_total_size = ~0U;
1441
1442                 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1443                 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1444                 if (!is || !bs) {
1445                         error = -ENOMEM;
1446                         goto cleanup;
1447                 }
1448
1449                 is->s.not_found = -ENODATA;
1450                 bs->s.not_found = -ENODATA;
1451                 is->iloc.bh = NULL;
1452                 bs->bh = NULL;
1453
1454                 last = IFIRST(header);
1455                 /* Find the entry best suited to be pushed into EA block */
1456                 entry = NULL;
1457                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1458                         total_size =
1459                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1460                                         EXT4_XATTR_LEN(last->e_name_len);
1461                         if (total_size <= free && total_size < min_total_size) {
1462                                 if (total_size < new_extra_isize) {
1463                                         small_entry = last;
1464                                 } else {
1465                                         entry = last;
1466                                         min_total_size = total_size;
1467                                 }
1468                         }
1469                 }
1470
1471                 if (entry == NULL) {
1472                         if (small_entry) {
1473                                 entry = small_entry;
1474                         } else {
1475                                 if (!tried_min_extra_isize &&
1476                                     s_min_extra_isize) {
1477                                         tried_min_extra_isize++;
1478                                         new_extra_isize = s_min_extra_isize;
1479                                         kfree(is); is = NULL;
1480                                         kfree(bs); bs = NULL;
1481                                         brelse(bh);
1482                                         goto retry;
1483                                 }
1484                                 error = -1;
1485                                 goto cleanup;
1486                         }
1487                 }
1488                 offs = le16_to_cpu(entry->e_value_offs);
1489                 size = le32_to_cpu(entry->e_value_size);
1490                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1491                 i.name_index = entry->e_name_index,
1492                 buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1493                 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1494                 if (!buffer || !b_entry_name) {
1495                         error = -ENOMEM;
1496                         goto cleanup;
1497                 }
1498                 /* Save the entry name and the entry value */
1499                 memcpy(buffer, (void *)IFIRST(header) + offs,
1500                        EXT4_XATTR_SIZE(size));
1501                 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1502                 b_entry_name[entry->e_name_len] = '\0';
1503                 i.name = b_entry_name;
1504
1505                 error = ext4_get_inode_loc(inode, &is->iloc);
1506                 if (error)
1507                         goto cleanup;
1508
1509                 error = ext4_xattr_ibody_find(inode, &i, is);
1510                 if (error)
1511                         goto cleanup;
1512
1513                 /* Remove the chosen entry from the inode */
1514                 error = ext4_xattr_ibody_set(handle, inode, &i, is);
1515                 if (error)
1516                         goto cleanup;
1517
1518                 entry = IFIRST(header);
1519                 if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1520                         shift_bytes = new_extra_isize;
1521                 else
1522                         shift_bytes = entry_size + size;
1523                 /* Adjust the offsets and shift the remaining entries ahead */
1524                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1525                         shift_bytes, (void *)raw_inode +
1526                         EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1527                         (void *)header, total_ino - entry_size,
1528                         inode->i_sb->s_blocksize);
1529
1530                 extra_isize += shift_bytes;
1531                 new_extra_isize -= shift_bytes;
1532                 EXT4_I(inode)->i_extra_isize = extra_isize;
1533
1534                 i.name = b_entry_name;
1535                 i.value = buffer;
1536                 i.value_len = size;
1537                 error = ext4_xattr_block_find(inode, &i, bs);
1538                 if (error)
1539                         goto cleanup;
1540
1541                 /* Add entry which was removed from the inode into the block */
1542                 error = ext4_xattr_block_set(handle, inode, &i, bs);
1543                 if (error)
1544                         goto cleanup;
1545                 kfree(b_entry_name);
1546                 kfree(buffer);
1547                 b_entry_name = NULL;
1548                 buffer = NULL;
1549                 brelse(is->iloc.bh);
1550                 kfree(is);
1551                 kfree(bs);
1552         }
1553         brelse(bh);
1554         up_write(&EXT4_I(inode)->xattr_sem);
1555         return 0;
1556
1557 cleanup:
1558         kfree(b_entry_name);
1559         kfree(buffer);
1560         if (is)
1561                 brelse(is->iloc.bh);
1562         kfree(is);
1563         kfree(bs);
1564         brelse(bh);
1565         up_write(&EXT4_I(inode)->xattr_sem);
1566         return error;
1567 }
1568
1569
1570
1571 /*
1572  * ext4_xattr_delete_inode()
1573  *
1574  * Free extended attribute resources associated with this inode. This
1575  * is called immediately before an inode is freed. We have exclusive
1576  * access to the inode.
1577  */
1578 void
1579 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1580 {
1581         struct buffer_head *bh = NULL;
1582
1583         if (!EXT4_I(inode)->i_file_acl)
1584                 goto cleanup;
1585         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1586         if (!bh) {
1587                 EXT4_ERROR_INODE(inode, "block %llu read error",
1588                                  EXT4_I(inode)->i_file_acl);
1589                 goto cleanup;
1590         }
1591         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1592             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1593                 EXT4_ERROR_INODE(inode, "bad block %llu",
1594                                  EXT4_I(inode)->i_file_acl);
1595                 goto cleanup;
1596         }
1597         ext4_xattr_release_block(handle, inode, bh);
1598         EXT4_I(inode)->i_file_acl = 0;
1599
1600 cleanup:
1601         brelse(bh);
1602 }
1603
1604 /*
1605  * ext4_xattr_cache_insert()
1606  *
1607  * Create a new entry in the extended attribute cache, and insert
1608  * it unless such an entry is already in the cache.
1609  *
1610  * Returns 0, or a negative error number on failure.
1611  */
1612 static void
1613 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1614 {
1615         struct ext4_xattr_header *header = BHDR(bh);
1616         __u32 hash = le32_to_cpu(header->h_hash);
1617         int reusable = le32_to_cpu(header->h_refcount) <
1618                        EXT4_XATTR_REFCOUNT_MAX;
1619         int error;
1620
1621         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1622                                       bh->b_blocknr, reusable);
1623         if (error) {
1624                 if (error == -EBUSY)
1625                         ea_bdebug(bh, "already in cache");
1626         } else
1627                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1628 }
1629
1630 /*
1631  * ext4_xattr_cmp()
1632  *
1633  * Compare two extended attribute blocks for equality.
1634  *
1635  * Returns 0 if the blocks are equal, 1 if they differ, and
1636  * a negative error number on errors.
1637  */
1638 static int
1639 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1640                struct ext4_xattr_header *header2)
1641 {
1642         struct ext4_xattr_entry *entry1, *entry2;
1643
1644         entry1 = ENTRY(header1+1);
1645         entry2 = ENTRY(header2+1);
1646         while (!IS_LAST_ENTRY(entry1)) {
1647                 if (IS_LAST_ENTRY(entry2))
1648                         return 1;
1649                 if (entry1->e_hash != entry2->e_hash ||
1650                     entry1->e_name_index != entry2->e_name_index ||
1651                     entry1->e_name_len != entry2->e_name_len ||
1652                     entry1->e_value_size != entry2->e_value_size ||
1653                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1654                         return 1;
1655                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1656                         return -EFSCORRUPTED;
1657                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1658                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1659                            le32_to_cpu(entry1->e_value_size)))
1660                         return 1;
1661
1662                 entry1 = EXT4_XATTR_NEXT(entry1);
1663                 entry2 = EXT4_XATTR_NEXT(entry2);
1664         }
1665         if (!IS_LAST_ENTRY(entry2))
1666                 return 1;
1667         return 0;
1668 }
1669
1670 /*
1671  * ext4_xattr_cache_find()
1672  *
1673  * Find an identical extended attribute block.
1674  *
1675  * Returns a pointer to the block found, or NULL if such a block was
1676  * not found or an error occurred.
1677  */
1678 static struct buffer_head *
1679 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1680                       struct mb_cache_entry **pce)
1681 {
1682         __u32 hash = le32_to_cpu(header->h_hash);
1683         struct mb_cache_entry *ce;
1684         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1685
1686         if (!header->h_hash)
1687                 return NULL;  /* never share */
1688         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1689         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1690         while (ce) {
1691                 struct buffer_head *bh;
1692
1693                 bh = sb_bread(inode->i_sb, ce->e_block);
1694                 if (!bh) {
1695                         EXT4_ERROR_INODE(inode, "block %lu read error",
1696                                          (unsigned long) ce->e_block);
1697                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1698                         *pce = ce;
1699                         return bh;
1700                 }
1701                 brelse(bh);
1702                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1703         }
1704         return NULL;
1705 }
1706
1707 #define NAME_HASH_SHIFT 5
1708 #define VALUE_HASH_SHIFT 16
1709
1710 /*
1711  * ext4_xattr_hash_entry()
1712  *
1713  * Compute the hash of an extended attribute.
1714  */
1715 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1716                                          struct ext4_xattr_entry *entry)
1717 {
1718         __u32 hash = 0;
1719         char *name = entry->e_name;
1720         int n;
1721
1722         for (n = 0; n < entry->e_name_len; n++) {
1723                 hash = (hash << NAME_HASH_SHIFT) ^
1724                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1725                        *name++;
1726         }
1727
1728         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1729                 __le32 *value = (__le32 *)((char *)header +
1730                         le16_to_cpu(entry->e_value_offs));
1731                 for (n = (le32_to_cpu(entry->e_value_size) +
1732                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1733                         hash = (hash << VALUE_HASH_SHIFT) ^
1734                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1735                                le32_to_cpu(*value++);
1736                 }
1737         }
1738         entry->e_hash = cpu_to_le32(hash);
1739 }
1740
1741 #undef NAME_HASH_SHIFT
1742 #undef VALUE_HASH_SHIFT
1743
1744 #define BLOCK_HASH_SHIFT 16
1745
1746 /*
1747  * ext4_xattr_rehash()
1748  *
1749  * Re-compute the extended attribute hash value after an entry has changed.
1750  */
1751 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1752                               struct ext4_xattr_entry *entry)
1753 {
1754         struct ext4_xattr_entry *here;
1755         __u32 hash = 0;
1756
1757         ext4_xattr_hash_entry(header, entry);
1758         here = ENTRY(header+1);
1759         while (!IS_LAST_ENTRY(here)) {
1760                 if (!here->e_hash) {
1761                         /* Block is not shared if an entry's hash value == 0 */
1762                         hash = 0;
1763                         break;
1764                 }
1765                 hash = (hash << BLOCK_HASH_SHIFT) ^
1766                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1767                        le32_to_cpu(here->e_hash);
1768                 here = EXT4_XATTR_NEXT(here);
1769         }
1770         header->h_hash = cpu_to_le32(hash);
1771 }
1772
1773 #undef BLOCK_HASH_SHIFT
1774
1775 #define HASH_BUCKET_BITS        10
1776
1777 struct mb_cache *
1778 ext4_xattr_create_cache(void)
1779 {
1780         return mb_cache_create(HASH_BUCKET_BITS);
1781 }
1782
1783 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1784 {
1785         if (cache)
1786                 mb_cache_destroy(cache);
1787 }
1788