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                 char b[BDEVNAME_SIZE]; \
  72                 printk(KERN_DEBUG "block %s:%lu: ", \
  73                         bdevname(bh->b_bdev, b), \
  74                         (unsigned long) bh->b_blocknr); \
  75                 printk(f); \
  76                 printk("\n"); \
  77         } while (0)
  78 #else
  79 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  80 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  81 #endif
  82
  83 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  84 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
  85                                                  struct ext4_xattr_header *,
  86                                                  struct mb_cache_entry **);
  87 static void ext4_xattr_rehash(struct ext4_xattr_header *,
  88                               struct ext4_xattr_entry *);
  89 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
  90                            size_t buffer_size);
  91
  92 static const struct xattr_handler *ext4_xattr_handler_map[] = {
  93         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  94 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  95         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  96         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  97 #endif
  98         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  99 #ifdef CONFIG_EXT4_FS_SECURITY
 100         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
 101 #endif
 102 };
 103
 104 const struct xattr_handler *ext4_xattr_handlers[] = {
 105         &ext4_xattr_user_handler,
 106         &ext4_xattr_trusted_handler,
 107 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 108         &posix_acl_access_xattr_handler,
 109         &posix_acl_default_xattr_handler,
 110 #endif
 111 #ifdef CONFIG_EXT4_FS_SECURITY
 112         &ext4_xattr_security_handler,
 113 #endif
 114         NULL
 115 };
 116
 117 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
 118                                 inode->i_sb->s_fs_info)->s_mb_cache)
 119
 120 static __le32 ext4_xattr_block_csum(struct inode *inode,
 121                                     sector_t block_nr,
 122                                     struct ext4_xattr_header *hdr)
 123 {
 124         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 125         __u32 csum;
 126         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 127         __u32 dummy_csum = 0;
 128         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 129
 130         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 131                            sizeof(dsk_block_nr));
 132         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 133         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 134         offset += sizeof(dummy_csum);
 135         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 136                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 137
 138         return cpu_to_le32(csum);
 139 }
 140
 141 static int ext4_xattr_block_csum_verify(struct inode *inode,
 142                                         struct buffer_head *bh)
 143 {
 144         struct ext4_xattr_header *hdr = BHDR(bh);
 145         int ret = 1;
 146
 147         if (ext4_has_metadata_csum(inode->i_sb)) {
 148                 lock_buffer(bh);
 149                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 150                                                         bh->b_blocknr, hdr));
 151                 unlock_buffer(bh);
 152         }
 153         return ret;
 154 }
 155
 156 static void ext4_xattr_block_csum_set(struct inode *inode,
 157                                       struct buffer_head *bh)
 158 {
 159         if (ext4_has_metadata_csum(inode->i_sb))
 160                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 161                                                 bh->b_blocknr, BHDR(bh));
 162 }
 163
 164 static inline const struct xattr_handler *
 165 ext4_xattr_handler(int name_index)
 166 {
 167         const struct xattr_handler *handler = NULL;
 168
 169         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 170                 handler = ext4_xattr_handler_map[name_index];
 171         return handler;
 172 }
 173
 174 /*
 175  * Inode operation listxattr()
 176  *
 177  * d_inode(dentry)->i_mutex: don't care
 178  */
 179 ssize_t
 180 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
 181 {
 182         return ext4_xattr_list(dentry, buffer, size);
 183 }
 184
 185 static int
 186 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
 187                        void *value_start)
 188 {
 189         struct ext4_xattr_entry *e = entry;
 190
 191         while (!IS_LAST_ENTRY(e)) {
 192                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 193                 if ((void *)next >= end)
 194                         return -EFSCORRUPTED;
 195                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 196                         return -EFSCORRUPTED;
 197                 e = next;
 198         }
 199
 200         while (!IS_LAST_ENTRY(entry)) {
 201                 if (entry->e_value_size != 0 &&
 202                     (value_start + le16_to_cpu(entry->e_value_offs) <
 203                      (void *)e + sizeof(__u32) ||
 204                      value_start + le16_to_cpu(entry->e_value_offs) +
 205                     le32_to_cpu(entry->e_value_size) > end))
 206                         return -EFSCORRUPTED;
 207                 entry = EXT4_XATTR_NEXT(entry);
 208         }
 209
 210         return 0;
 211 }
 212
 213 static inline int
 214 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 215 {
 216         int error;
 217
 218         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 219             BHDR(bh)->h_blocks != cpu_to_le32(1))
 220                 return -EFSCORRUPTED;
 221         if (buffer_verified(bh))
 222                 return 0;
 223
 224         if (!ext4_xattr_block_csum_verify(inode, bh))
 225                 return -EFSBADCRC;
 226         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
 227                                        bh->b_data);
 228         if (!error)
 229                 set_buffer_verified(bh);
 230         return error;
 231 }
 232
 233 static int
 234 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 235                          void *end, const char *function, unsigned int line)
 236 {
 237         struct ext4_xattr_entry *entry = IFIRST(header);
 238         int error = -EFSCORRUPTED;
 239
 240         if (((void *) header >= end) ||
 241             (header->h_magic != le32_to_cpu(EXT4_XATTR_MAGIC)))
 242                 goto errout;
 243         error = ext4_xattr_check_names(entry, end, entry);
 244 errout:
 245         if (error)
 246                 __ext4_error_inode(inode, function, line, 0,
 247                                    "corrupted in-inode xattr");
 248         return error;
 249 }
 250
 251 #define xattr_check_inode(inode, header, end) \
 252         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 253
 254 static inline int
 255 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 256 {
 257         size_t value_size = le32_to_cpu(entry->e_value_size);
 258
 259         if (entry->e_value_block != 0 || value_size > size ||
 260             le16_to_cpu(entry->e_value_offs) + value_size > size)
 261                 return -EFSCORRUPTED;
 262         return 0;
 263 }
 264
 265 static int
 266 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 267                       const char *name, size_t size, int sorted)
 268 {
 269         struct ext4_xattr_entry *entry;
 270         size_t name_len;
 271         int cmp = 1;
 272
 273         if (name == NULL)
 274                 return -EINVAL;
 275         name_len = strlen(name);
 276         entry = *pentry;
 277         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 278                 cmp = name_index - entry->e_name_index;
 279                 if (!cmp)
 280                         cmp = name_len - entry->e_name_len;
 281                 if (!cmp)
 282                         cmp = memcmp(name, entry->e_name, name_len);
 283                 if (cmp <= 0 && (sorted || cmp == 0))
 284                         break;
 285         }
 286         *pentry = entry;
 287         if (!cmp && ext4_xattr_check_entry(entry, size))
 288                 return -EFSCORRUPTED;
 289         return cmp ? -ENODATA : 0;
 290 }
 291
 292 static int
 293 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 294                      void *buffer, size_t buffer_size)
 295 {
 296         struct buffer_head *bh = NULL;
 297         struct ext4_xattr_entry *entry;
 298         size_t size;
 299         int error;
 300         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 301
 302         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 303                   name_index, name, buffer, (long)buffer_size);
 304
 305         error = -ENODATA;
 306         if (!EXT4_I(inode)->i_file_acl)
 307                 goto cleanup;
 308         ea_idebug(inode, "reading block %llu",
 309                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 310         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 311         if (!bh)
 312                 goto cleanup;
 313         ea_bdebug(bh, "b_count=%d, refcount=%d",
 314                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 315         if (ext4_xattr_check_block(inode, bh)) {
 316 bad_block:
 317                 EXT4_ERROR_INODE(inode, "bad block %llu",
 318                                  EXT4_I(inode)->i_file_acl);
 319                 error = -EFSCORRUPTED;
 320                 goto cleanup;
 321         }
 322         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 323         entry = BFIRST(bh);
 324         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 325         if (error == -EFSCORRUPTED)
 326                 goto bad_block;
 327         if (error)
 328                 goto cleanup;
 329         size = le32_to_cpu(entry->e_value_size);
 330         if (buffer) {
 331                 error = -ERANGE;
 332                 if (size > buffer_size)
 333                         goto cleanup;
 334                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 335                        size);
 336         }
 337         error = size;
 338
 339 cleanup:
 340         brelse(bh);
 341         return error;
 342 }
 343
 344 int
 345 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 346                      void *buffer, size_t buffer_size)
 347 {
 348         struct ext4_xattr_ibody_header *header;
 349         struct ext4_xattr_entry *entry;
 350         struct ext4_inode *raw_inode;
 351         struct ext4_iloc iloc;
 352         size_t size;
 353         void *end;
 354         int error;
 355
 356         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 357                 return -ENODATA;
 358         error = ext4_get_inode_loc(inode, &iloc);
 359         if (error)
 360                 return error;
 361         raw_inode = ext4_raw_inode(&iloc);
 362         header = IHDR(inode, raw_inode);
 363         entry = IFIRST(header);
 364         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 365         error = xattr_check_inode(inode, header, end);
 366         if (error)
 367                 goto cleanup;
 368         error = ext4_xattr_find_entry(&entry, name_index, name,
 369                                       end - (void *)entry, 0);
 370         if (error)
 371                 goto cleanup;
 372         size = le32_to_cpu(entry->e_value_size);
 373         if (buffer) {
 374                 error = -ERANGE;
 375                 if (size > buffer_size)
 376                         goto cleanup;
 377                 memcpy(buffer, (void *)IFIRST(header) +
 378                        le16_to_cpu(entry->e_value_offs), size);
 379         }
 380         error = size;
 381
 382 cleanup:
 383         brelse(iloc.bh);
 384         return error;
 385 }
 386
 387 /*
 388  * ext4_xattr_get()
 389  *
 390  * Copy an extended attribute into the buffer
 391  * provided, or compute the buffer size required.
 392  * Buffer is NULL to compute the size of the buffer required.
 393  *
 394  * Returns a negative error number on failure, or the number of bytes
 395  * used / required on success.
 396  */
 397 int
 398 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 399                void *buffer, size_t buffer_size)
 400 {
 401         int error;
 402
 403         if (strlen(name) > 255)
 404                 return -ERANGE;
 405
 406         down_read(&EXT4_I(inode)->xattr_sem);
 407         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 408                                      buffer_size);
 409         if (error == -ENODATA)
 410                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 411                                              buffer_size);
 412         up_read(&EXT4_I(inode)->xattr_sem);
 413         return error;
 414 }
 415
 416 static int
 417 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 418                         char *buffer, size_t buffer_size)
 419 {
 420         size_t rest = buffer_size;
 421
 422         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 423                 const struct xattr_handler *handler =
 424                         ext4_xattr_handler(entry->e_name_index);
 425
 426                 if (handler) {
 427                         size_t size = handler->list(handler, dentry, buffer,
 428                                                     rest, entry->e_name,
 429                                                     entry->e_name_len);
 430                         if (buffer) {
 431                                 if (size > rest)
 432                                         return -ERANGE;
 433                                 buffer += size;
 434                         }
 435                         rest -= size;
 436                 }
 437         }
 438         return buffer_size - rest;
 439 }
 440
 441 static int
 442 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 443 {
 444         struct inode *inode = d_inode(dentry);
 445         struct buffer_head *bh = NULL;
 446         int error;
 447         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 448
 449         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 450                   buffer, (long)buffer_size);
 451
 452         error = 0;
 453         if (!EXT4_I(inode)->i_file_acl)
 454                 goto cleanup;
 455         ea_idebug(inode, "reading block %llu",
 456                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 457         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 458         error = -EIO;
 459         if (!bh)
 460                 goto cleanup;
 461         ea_bdebug(bh, "b_count=%d, refcount=%d",
 462                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 463         if (ext4_xattr_check_block(inode, bh)) {
 464                 EXT4_ERROR_INODE(inode, "bad block %llu",
 465                                  EXT4_I(inode)->i_file_acl);
 466                 error = -EFSCORRUPTED;
 467                 goto cleanup;
 468         }
 469         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 470         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 471
 472 cleanup:
 473         brelse(bh);
 474
 475         return error;
 476 }
 477
 478 static int
 479 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 480 {
 481         struct inode *inode = d_inode(dentry);
 482         struct ext4_xattr_ibody_header *header;
 483         struct ext4_inode *raw_inode;
 484         struct ext4_iloc iloc;
 485         void *end;
 486         int error;
 487
 488         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 489                 return 0;
 490         error = ext4_get_inode_loc(inode, &iloc);
 491         if (error)
 492                 return error;
 493         raw_inode = ext4_raw_inode(&iloc);
 494         header = IHDR(inode, raw_inode);
 495         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 496         error = xattr_check_inode(inode, header, end);
 497         if (error)
 498                 goto cleanup;
 499         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 500                                         buffer, buffer_size);
 501
 502 cleanup:
 503         brelse(iloc.bh);
 504         return error;
 505 }
 506
 507 /*
 508  * ext4_xattr_list()
 509  *
 510  * Copy a list of attribute names into the buffer
 511  * provided, or compute the buffer size required.
 512  * Buffer is NULL to compute the size of the buffer required.
 513  *
 514  * Returns a negative error number on failure, or the number of bytes
 515  * used / required on success.
 516  */
 517 static int
 518 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 519 {
 520         int ret, ret2;
 521
 522         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 523         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 524         if (ret < 0)
 525                 goto errout;
 526         if (buffer) {
 527                 buffer += ret;
 528                 buffer_size -= ret;
 529         }
 530         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 531         if (ret < 0)
 532                 goto errout;
 533         ret += ret2;
 534 errout:
 535         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 536         return ret;
 537 }
 538
 539 /*
 540  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 541  * not set, set it.
 542  */
 543 static void ext4_xattr_update_super_block(handle_t *handle,
 544                                           struct super_block *sb)
 545 {
 546         if (ext4_has_feature_xattr(sb))
 547                 return;
 548
 549         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 550         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 551                 ext4_set_feature_xattr(sb);
 552                 ext4_handle_dirty_super(handle, sb);
 553         }
 554 }
 555
 556 /*
 557  * Release the xattr block BH: If the reference count is > 1, decrement it;
 558  * otherwise free the block.
 559  */
 560 static void
 561 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 562                          struct buffer_head *bh)
 563 {
 564         struct mb_cache_entry *ce = NULL;
 565         int error = 0;
 566         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 567
 568         ce = mb_cache_entry_get(ext4_mb_cache, bh->b_bdev, bh->b_blocknr);
 569         BUFFER_TRACE(bh, "get_write_access");
 570         error = ext4_journal_get_write_access(handle, bh);
 571         if (error)
 572                 goto out;
 573
 574         lock_buffer(bh);
 575         if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
 576                 ea_bdebug(bh, "refcount now=0; freeing");
 577                 if (ce)
 578                         mb_cache_entry_free(ce);
 579                 get_bh(bh);
 580                 unlock_buffer(bh);
 581                 ext4_free_blocks(handle, inode, bh, 0, 1,
 582                                  EXT4_FREE_BLOCKS_METADATA |
 583                                  EXT4_FREE_BLOCKS_FORGET);
 584         } else {
 585                 le32_add_cpu(&BHDR(bh)->h_refcount, -1);
 586                 if (ce)
 587                         mb_cache_entry_release(ce);
 588
 589                 ext4_xattr_block_csum_set(inode, bh);
 590                 /*
 591                  * Beware of this ugliness: Releasing of xattr block references
 592                  * from different inodes can race and so we have to protect
 593                  * from a race where someone else frees the block (and releases
 594                  * its journal_head) before we are done dirtying the buffer. In
 595                  * nojournal mode this race is harmless and we actually cannot
 596                  * call ext4_handle_dirty_metadata() with locked buffer as
 597                  * that function can call sync_dirty_buffer() so for that case
 598                  * we handle the dirtying after unlocking the buffer.
 599                  */
 600                 if (ext4_handle_valid(handle))
 601                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 602                 unlock_buffer(bh);
 603                 if (!ext4_handle_valid(handle))
 604                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 605                 if (IS_SYNC(inode))
 606                         ext4_handle_sync(handle);
 607                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
 608                 ea_bdebug(bh, "refcount now=%d; releasing",
 609                           le32_to_cpu(BHDR(bh)->h_refcount));
 610         }
 611 out:
 612         ext4_std_error(inode->i_sb, error);
 613         return;
 614 }
 615
 616 /*
 617  * Find the available free space for EAs. This also returns the total number of
 618  * bytes used by EA entries.
 619  */
 620 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
 621                                     size_t *min_offs, void *base, int *total)
 622 {
 623         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 624                 if (!last->e_value_block && last->e_value_size) {
 625                         size_t offs = le16_to_cpu(last->e_value_offs);
 626                         if (offs < *min_offs)
 627                                 *min_offs = offs;
 628                 }
 629                 if (total)
 630                         *total += EXT4_XATTR_LEN(last->e_name_len);
 631         }
 632         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 633 }
 634
 635 static int
 636 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
 637                      struct inode *inode)
 638 {
 639         struct ext4_xattr_entry *last, *next;
 640         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 641
 642         /* Compute min_offs and last. */
 643         last = s->first;
 644         for (; !IS_LAST_ENTRY(last); last = next) {
 645                 next = EXT4_XATTR_NEXT(last);
 646                 if ((void *)next >= s->end) {
 647                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 648                         return -EFSCORRUPTED;
 649                 }
 650                 if (!last->e_value_block && last->e_value_size) {
 651                         size_t offs = le16_to_cpu(last->e_value_offs);
 652                         if (offs < min_offs)
 653                                 min_offs = offs;
 654                 }
 655         }
 656         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 657         if (!s->not_found) {
 658                 if (!s->here->e_value_block && s->here->e_value_size) {
 659                         size_t size = le32_to_cpu(s->here->e_value_size);
 660                         free += EXT4_XATTR_SIZE(size);
 661                 }
 662                 free += EXT4_XATTR_LEN(name_len);
 663         }
 664         if (i->value) {
 665                 if (free < EXT4_XATTR_LEN(name_len) +
 666                            EXT4_XATTR_SIZE(i->value_len))
 667                         return -ENOSPC;
 668         }
 669
 670         if (i->value && s->not_found) {
 671                 /* Insert the new name. */
 672                 size_t size = EXT4_XATTR_LEN(name_len);
 673                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
 674                 memmove((void *)s->here + size, s->here, rest);
 675                 memset(s->here, 0, size);
 676                 s->here->e_name_index = i->name_index;
 677                 s->here->e_name_len = name_len;
 678                 memcpy(s->here->e_name, i->name, name_len);
 679         } else {
 680                 if (!s->here->e_value_block && s->here->e_value_size) {
 681                         void *first_val = s->base + min_offs;
 682                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 683                         void *val = s->base + offs;
 684                         size_t size = EXT4_XATTR_SIZE(
 685                                 le32_to_cpu(s->here->e_value_size));
 686
 687                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
 688                                 /* The old and the new value have the same
 689                                    size. Just replace. */
 690                                 s->here->e_value_size =
 691                                         cpu_to_le32(i->value_len);
 692                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
 693                                         memset(val, 0, size);
 694                                 } else {
 695                                         /* Clear pad bytes first. */
 696                                         memset(val + size - EXT4_XATTR_PAD, 0,
 697                                                EXT4_XATTR_PAD);
 698                                         memcpy(val, i->value, i->value_len);
 699                                 }
 700                                 return 0;
 701                         }
 702
 703                         /* Remove the old value. */
 704                         memmove(first_val + size, first_val, val - first_val);
 705                         memset(first_val, 0, size);
 706                         s->here->e_value_size = 0;
 707                         s->here->e_value_offs = 0;
 708                         min_offs += size;
 709
 710                         /* Adjust all value offsets. */
 711                         last = s->first;
 712                         while (!IS_LAST_ENTRY(last)) {
 713                                 size_t o = le16_to_cpu(last->e_value_offs);
 714                                 if (!last->e_value_block &&
 715                                     last->e_value_size && o < offs)
 716                                         last->e_value_offs =
 717                                                 cpu_to_le16(o + size);
 718                                 last = EXT4_XATTR_NEXT(last);
 719                         }
 720                 }
 721                 if (!i->value) {
 722                         /* Remove the old name. */
 723                         size_t size = EXT4_XATTR_LEN(name_len);
 724                         last = ENTRY((void *)last - size);
 725                         memmove(s->here, (void *)s->here + size,
 726                                 (void *)last - (void *)s->here + sizeof(__u32));
 727                         memset(last, 0, size);
 728                 }
 729         }
 730
 731         if (i->value) {
 732                 /* Insert the new value. */
 733                 s->here->e_value_size = cpu_to_le32(i->value_len);
 734                 if (i->value_len) {
 735                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 736                         void *val = s->base + min_offs - size;
 737                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 738                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 739                                 memset(val, 0, size);
 740                         } else {
 741                                 /* Clear the pad bytes first. */
 742                                 memset(val + size - EXT4_XATTR_PAD, 0,
 743                                        EXT4_XATTR_PAD);
 744                                 memcpy(val, i->value, i->value_len);
 745                         }
 746                 }
 747         }
 748         return 0;
 749 }
 750
 751 struct ext4_xattr_block_find {
 752         struct ext4_xattr_search s;
 753         struct buffer_head *bh;
 754 };
 755
 756 static int
 757 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
 758                       struct ext4_xattr_block_find *bs)
 759 {
 760         struct super_block *sb = inode->i_sb;
 761         int error;
 762
 763         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 764                   i->name_index, i->name, i->value, (long)i->value_len);
 765
 766         if (EXT4_I(inode)->i_file_acl) {
 767                 /* The inode already has an extended attribute block. */
 768                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
 769                 error = -EIO;
 770                 if (!bs->bh)
 771                         goto cleanup;
 772                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
 773                         atomic_read(&(bs->bh->b_count)),
 774                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
 775                 if (ext4_xattr_check_block(inode, bs->bh)) {
 776                         EXT4_ERROR_INODE(inode, "bad block %llu",
 777                                          EXT4_I(inode)->i_file_acl);
 778                         error = -EFSCORRUPTED;
 779                         goto cleanup;
 780                 }
 781                 /* Find the named attribute. */
 782                 bs->s.base = BHDR(bs->bh);
 783                 bs->s.first = BFIRST(bs->bh);
 784                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 785                 bs->s.here = bs->s.first;
 786                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 787                                               i->name, bs->bh->b_size, 1);
 788                 if (error && error != -ENODATA)
 789                         goto cleanup;
 790                 bs->s.not_found = error;
 791         }
 792         error = 0;
 793
 794 cleanup:
 795         return error;
 796 }
 797
 798 static int
 799 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 800                      struct ext4_xattr_info *i,
 801                      struct ext4_xattr_block_find *bs)
 802 {
 803         struct super_block *sb = inode->i_sb;
 804         struct buffer_head *new_bh = NULL;
 805         struct ext4_xattr_search *s = &bs->s;
 806         struct mb_cache_entry *ce = NULL;
 807         int error = 0;
 808         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 809
 810 #define header(x) ((struct ext4_xattr_header *)(x))
 811
 812         if (i->value && i->value_len > sb->s_blocksize)
 813                 return -ENOSPC;
 814         if (s->base) {
 815                 ce = mb_cache_entry_get(ext4_mb_cache, bs->bh->b_bdev,
 816                                         bs->bh->b_blocknr);
 817                 BUFFER_TRACE(bs->bh, "get_write_access");
 818                 error = ext4_journal_get_write_access(handle, bs->bh);
 819                 if (error)
 820                         goto cleanup;
 821                 lock_buffer(bs->bh);
 822
 823                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
 824                         if (ce) {
 825                                 mb_cache_entry_free(ce);
 826                                 ce = NULL;
 827                         }
 828                         ea_bdebug(bs->bh, "modifying in-place");
 829                         error = ext4_xattr_set_entry(i, s, inode);
 830                         if (!error) {
 831                                 if (!IS_LAST_ENTRY(s->first))
 832                                         ext4_xattr_rehash(header(s->base),
 833                                                           s->here);
 834                         }
 835                         ext4_xattr_block_csum_set(inode, bs->bh);
 836                         unlock_buffer(bs->bh);
 837                         if (error == -EFSCORRUPTED)
 838                                 goto bad_block;
 839                         if (!error)
 840                                 error = ext4_handle_dirty_metadata(handle,
 841                                                                    inode,
 842                                                                    bs->bh);
 843                         if (error)
 844                                 goto cleanup;
 845                         goto inserted;
 846                 } else {
 847                         int offset = (char *)s->here - bs->bh->b_data;
 848
 849                         unlock_buffer(bs->bh);
 850                         if (ce) {
 851                                 mb_cache_entry_release(ce);
 852                                 ce = NULL;
 853                         }
 854                         ea_bdebug(bs->bh, "cloning");
 855                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
 856                         error = -ENOMEM;
 857                         if (s->base == NULL)
 858                                 goto cleanup;
 859                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
 860                         s->first = ENTRY(header(s->base)+1);
 861                         header(s->base)->h_refcount = cpu_to_le32(1);
 862                         s->here = ENTRY(s->base + offset);
 863                         s->end = s->base + bs->bh->b_size;
 864                 }
 865         } else {
 866                 /* Allocate a buffer where we construct the new block. */
 867                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 868                 /* assert(header == s->base) */
 869                 error = -ENOMEM;
 870                 if (s->base == NULL)
 871                         goto cleanup;
 872                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
 873                 header(s->base)->h_blocks = cpu_to_le32(1);
 874                 header(s->base)->h_refcount = cpu_to_le32(1);
 875                 s->first = ENTRY(header(s->base)+1);
 876                 s->here = ENTRY(header(s->base)+1);
 877                 s->end = s->base + sb->s_blocksize;
 878         }
 879
 880         error = ext4_xattr_set_entry(i, s, inode);
 881         if (error == -EFSCORRUPTED)
 882                 goto bad_block;
 883         if (error)
 884                 goto cleanup;
 885         if (!IS_LAST_ENTRY(s->first))
 886                 ext4_xattr_rehash(header(s->base), s->here);
 887
 888 inserted:
 889         if (!IS_LAST_ENTRY(s->first)) {
 890                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
 891                 if (new_bh) {
 892                         /* We found an identical block in the cache. */
 893                         if (new_bh == bs->bh)
 894                                 ea_bdebug(new_bh, "keeping");
 895                         else {
 896                                 /* The old block is released after updating
 897                                    the inode. */
 898                                 error = dquot_alloc_block(inode,
 899                                                 EXT4_C2B(EXT4_SB(sb), 1));
 900                                 if (error)
 901                                         goto cleanup;
 902                                 BUFFER_TRACE(new_bh, "get_write_access");
 903                                 error = ext4_journal_get_write_access(handle,
 904                                                                       new_bh);
 905                                 if (error)
 906                                         goto cleanup_dquot;
 907                                 lock_buffer(new_bh);
 908                                 le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
 909                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
 910                                         le32_to_cpu(BHDR(new_bh)->h_refcount));
 911                                 ext4_xattr_block_csum_set(inode, new_bh);
 912                                 unlock_buffer(new_bh);
 913                                 error = ext4_handle_dirty_metadata(handle,
 914                                                                    inode,
 915                                                                    new_bh);
 916                                 if (error)
 917                                         goto cleanup_dquot;
 918                         }
 919                         mb_cache_entry_release(ce);
 920                         ce = NULL;
 921                 } else if (bs->bh && s->base == bs->bh->b_data) {
 922                         /* We were modifying this block in-place. */
 923                         ea_bdebug(bs->bh, "keeping this block");
 924                         ext4_xattr_cache_insert(ext4_mb_cache, bs->bh);
 925                         new_bh = bs->bh;
 926                         get_bh(new_bh);
 927                 } else {
 928                         /* We need to allocate a new block */
 929                         ext4_fsblk_t goal, block;
 930
 931                         goal = ext4_group_first_block_no(sb,
 932                                                 EXT4_I(inode)->i_block_group);
 933
 934                         /* non-extent files can't have physical blocks past 2^32 */
 935                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 936                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 937
 938                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
 939                                                      NULL, &error);
 940                         if (error)
 941                                 goto cleanup;
 942
 943                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 944                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
 945
 946                         ea_idebug(inode, "creating block %llu",
 947                                   (unsigned long long)block);
 948
 949                         new_bh = sb_getblk(sb, block);
 950                         if (unlikely(!new_bh)) {
 951                                 error = -ENOMEM;
 952 getblk_failed:
 953                                 ext4_free_blocks(handle, inode, NULL, block, 1,
 954                                                  EXT4_FREE_BLOCKS_METADATA);
 955                                 goto cleanup;
 956                         }
 957                         lock_buffer(new_bh);
 958                         error = ext4_journal_get_create_access(handle, new_bh);
 959                         if (error) {
 960                                 unlock_buffer(new_bh);
 961                                 error = -EIO;
 962                                 goto getblk_failed;
 963                         }
 964                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
 965                         ext4_xattr_block_csum_set(inode, new_bh);
 966                         set_buffer_uptodate(new_bh);
 967                         unlock_buffer(new_bh);
 968                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
 969                         error = ext4_handle_dirty_metadata(handle, inode,
 970                                                            new_bh);
 971                         if (error)
 972                                 goto cleanup;
 973                 }
 974         }
 975
 976         /* Update the inode. */
 977         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 978
 979         /* Drop the previous xattr block. */
 980         if (bs->bh && bs->bh != new_bh)
 981                 ext4_xattr_release_block(handle, inode, bs->bh);
 982         error = 0;
 983
 984 cleanup:
 985         if (ce)
 986                 mb_cache_entry_release(ce);
 987         brelse(new_bh);
 988         if (!(bs->bh && s->base == bs->bh->b_data))
 989                 kfree(s->base);
 990
 991         return error;
 992
 993 cleanup_dquot:
 994         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
 995         goto cleanup;
 996
 997 bad_block:
 998         EXT4_ERROR_INODE(inode, "bad block %llu",
 999                          EXT4_I(inode)->i_file_acl);
1000         goto cleanup;
1001
1002 #undef header
1003 }
1004
1005 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1006                           struct ext4_xattr_ibody_find *is)
1007 {
1008         struct ext4_xattr_ibody_header *header;
1009         struct ext4_inode *raw_inode;
1010         int error;
1011
1012         if (EXT4_I(inode)->i_extra_isize == 0)
1013                 return 0;
1014         raw_inode = ext4_raw_inode(&is->iloc);
1015         header = IHDR(inode, raw_inode);
1016         is->s.base = is->s.first = IFIRST(header);
1017         is->s.here = is->s.first;
1018         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1019         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1020                 error = xattr_check_inode(inode, header, is->s.end);
1021                 if (error)
1022                         return error;
1023                 /* Find the named attribute. */
1024                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1025                                               i->name, is->s.end -
1026                                               (void *)is->s.base, 0);
1027                 if (error && error != -ENODATA)
1028                         return error;
1029                 is->s.not_found = error;
1030         }
1031         return 0;
1032 }
1033
1034 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1035                                 struct ext4_xattr_info *i,
1036                                 struct ext4_xattr_ibody_find *is)
1037 {
1038         struct ext4_xattr_ibody_header *header;
1039         struct ext4_xattr_search *s = &is->s;
1040         int error;
1041
1042         if (EXT4_I(inode)->i_extra_isize == 0)
1043                 return -ENOSPC;
1044         error = ext4_xattr_set_entry(i, s, inode);
1045         if (error)
1046                 return error;
1047         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1048         if (!IS_LAST_ENTRY(s->first)) {
1049                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1050                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1051         } else {
1052                 header->h_magic = cpu_to_le32(0);
1053                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1054         }
1055         return 0;
1056 }
1057
1058 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1059                                 struct ext4_xattr_info *i,
1060                                 struct ext4_xattr_ibody_find *is)
1061 {
1062         struct ext4_xattr_ibody_header *header;
1063         struct ext4_xattr_search *s = &is->s;
1064         int error;
1065
1066         if (EXT4_I(inode)->i_extra_isize == 0)
1067                 return -ENOSPC;
1068         error = ext4_xattr_set_entry(i, s, inode);
1069         if (error)
1070                 return error;
1071         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1072         if (!IS_LAST_ENTRY(s->first)) {
1073                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1074                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1075         } else {
1076                 header->h_magic = cpu_to_le32(0);
1077                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1078         }
1079         return 0;
1080 }
1081
1082 /*
1083  * ext4_xattr_set_handle()
1084  *
1085  * Create, replace or remove an extended attribute for this inode.  Value
1086  * is NULL to remove an existing extended attribute, and non-NULL to
1087  * either replace an existing extended attribute, or create a new extended
1088  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1089  * specify that an extended attribute must exist and must not exist
1090  * previous to the call, respectively.
1091  *
1092  * Returns 0, or a negative error number on failure.
1093  */
1094 int
1095 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1096                       const char *name, const void *value, size_t value_len,
1097                       int flags)
1098 {
1099         struct ext4_xattr_info i = {
1100                 .name_index = name_index,
1101                 .name = name,
1102                 .value = value,
1103                 .value_len = value_len,
1104
1105         };
1106         struct ext4_xattr_ibody_find is = {
1107                 .s = { .not_found = -ENODATA, },
1108         };
1109         struct ext4_xattr_block_find bs = {
1110                 .s = { .not_found = -ENODATA, },
1111         };
1112         int no_expand;
1113         int error;
1114
1115         if (!name)
1116                 return -EINVAL;
1117         if (strlen(name) > 255)
1118                 return -ERANGE;
1119         ext4_write_lock_xattr(inode, &no_expand);
1120
1121         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1122         if (error)
1123                 goto cleanup;
1124
1125         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1126                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1127                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1128                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1129         }
1130
1131         error = ext4_xattr_ibody_find(inode, &i, &is);
1132         if (error)
1133                 goto cleanup;
1134         if (is.s.not_found)
1135                 error = ext4_xattr_block_find(inode, &i, &bs);
1136         if (error)
1137                 goto cleanup;
1138         if (is.s.not_found && bs.s.not_found) {
1139                 error = -ENODATA;
1140                 if (flags & XATTR_REPLACE)
1141                         goto cleanup;
1142                 error = 0;
1143                 if (!value)
1144                         goto cleanup;
1145         } else {
1146                 error = -EEXIST;
1147                 if (flags & XATTR_CREATE)
1148                         goto cleanup;
1149         }
1150         if (!value) {
1151                 if (!is.s.not_found)
1152                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1153                 else if (!bs.s.not_found)
1154                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1155         } else {
1156                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1157                 if (!error && !bs.s.not_found) {
1158                         i.value = NULL;
1159                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1160                 } else if (error == -ENOSPC) {
1161                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1162                                 brelse(bs.bh);
1163                                 bs.bh = NULL;
1164                                 error = ext4_xattr_block_find(inode, &i, &bs);
1165                                 if (error)
1166                                         goto cleanup;
1167                         }
1168                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1169                         if (error)
1170                                 goto cleanup;
1171                         if (!is.s.not_found) {
1172                                 i.value = NULL;
1173                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1174                                                              &is);
1175                         }
1176                 }
1177         }
1178         if (!error) {
1179                 ext4_xattr_update_super_block(handle, inode->i_sb);
1180                 inode->i_ctime = ext4_current_time(inode);
1181                 if (!value)
1182                         no_expand = 0;
1183                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1184                 /*
1185                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1186                  * error != 0.
1187                  */
1188                 is.iloc.bh = NULL;
1189                 if (IS_SYNC(inode))
1190                         ext4_handle_sync(handle);
1191         }
1192
1193 cleanup:
1194         brelse(is.iloc.bh);
1195         brelse(bs.bh);
1196         ext4_write_unlock_xattr(inode, &no_expand);
1197         return error;
1198 }
1199
1200 /*
1201  * ext4_xattr_set()
1202  *
1203  * Like ext4_xattr_set_handle, but start from an inode. This extended
1204  * attribute modification is a filesystem transaction by itself.
1205  *
1206  * Returns 0, or a negative error number on failure.
1207  */
1208 int
1209 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1210                const void *value, size_t value_len, int flags)
1211 {
1212         handle_t *handle;
1213         int error, retries = 0;
1214         int credits = ext4_jbd2_credits_xattr(inode);
1215
1216 retry:
1217         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1218         if (IS_ERR(handle)) {
1219                 error = PTR_ERR(handle);
1220         } else {
1221                 int error2;
1222
1223                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1224                                               value, value_len, flags);
1225                 error2 = ext4_journal_stop(handle);
1226                 if (error == -ENOSPC &&
1227                     ext4_should_retry_alloc(inode->i_sb, &retries))
1228                         goto retry;
1229                 if (error == 0)
1230                         error = error2;
1231         }
1232
1233         return error;
1234 }
1235
1236 /*
1237  * Shift the EA entries in the inode to create space for the increased
1238  * i_extra_isize.
1239  */
1240 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1241                                      int value_offs_shift, void *to,
1242                                      void *from, size_t n, int blocksize)
1243 {
1244         struct ext4_xattr_entry *last = entry;
1245         int new_offs;
1246
1247         /* Adjust the value offsets of the entries */
1248         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1249                 if (!last->e_value_block && last->e_value_size) {
1250                         new_offs = le16_to_cpu(last->e_value_offs) +
1251                                                         value_offs_shift;
1252                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1253                                  > blocksize);
1254                         last->e_value_offs = cpu_to_le16(new_offs);
1255                 }
1256         }
1257         /* Shift the entries by n bytes */
1258         memmove(to, from, n);
1259 }
1260
1261 /*
1262  * Expand an inode by new_extra_isize bytes when EAs are present.
1263  * Returns 0 on success or negative error number on failure.
1264  */
1265 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1266                                struct ext4_inode *raw_inode, handle_t *handle)
1267 {
1268         struct ext4_xattr_ibody_header *header;
1269         struct ext4_xattr_entry *entry, *last, *first;
1270         struct buffer_head *bh = NULL;
1271         struct ext4_xattr_ibody_find *is = NULL;
1272         struct ext4_xattr_block_find *bs = NULL;
1273         char *buffer = NULL, *b_entry_name = NULL;
1274         size_t min_offs, free;
1275         int total_ino;
1276         void *base, *start, *end;
1277         int error = 0, tried_min_extra_isize = 0;
1278         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1279         int isize_diff; /* How much do we need to grow i_extra_isize */
1280         int no_expand;
1281
1282         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
1283                 return 0;
1284
1285 retry:
1286         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1287         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1288                 goto out;
1289
1290         header = IHDR(inode, raw_inode);
1291         entry = IFIRST(header);
1292
1293         /*
1294          * Check if enough free space is available in the inode to shift the
1295          * entries ahead by new_extra_isize.
1296          */
1297
1298         base = start = entry;
1299         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1300         min_offs = end - base;
1301         last = entry;
1302         total_ino = sizeof(struct ext4_xattr_ibody_header);
1303
1304         error = xattr_check_inode(inode, header, end);
1305         if (error)
1306                 goto cleanup;
1307
1308         free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1309         if (free >= isize_diff) {
1310                 entry = IFIRST(header);
1311                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize
1312                                 - new_extra_isize, (void *)raw_inode +
1313                                 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1314                                 (void *)header, total_ino,
1315                                 inode->i_sb->s_blocksize);
1316                 EXT4_I(inode)->i_extra_isize = new_extra_isize;
1317                 goto out;
1318         }
1319
1320         /*
1321          * Enough free space isn't available in the inode, check if
1322          * EA block can hold new_extra_isize bytes.
1323          */
1324         if (EXT4_I(inode)->i_file_acl) {
1325                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1326                 error = -EIO;
1327                 if (!bh)
1328                         goto cleanup;
1329                 if (ext4_xattr_check_block(inode, bh)) {
1330                         EXT4_ERROR_INODE(inode, "bad block %llu",
1331                                          EXT4_I(inode)->i_file_acl);
1332                         error = -EFSCORRUPTED;
1333                         goto cleanup;
1334                 }
1335                 base = BHDR(bh);
1336                 first = BFIRST(bh);
1337                 end = bh->b_data + bh->b_size;
1338                 min_offs = end - base;
1339                 free = ext4_xattr_free_space(first, &min_offs, base, NULL);
1340                 if (free < isize_diff) {
1341                         if (!tried_min_extra_isize && s_min_extra_isize) {
1342                                 tried_min_extra_isize++;
1343                                 new_extra_isize = s_min_extra_isize;
1344                                 brelse(bh);
1345                                 goto retry;
1346                         }
1347                         error = -1;
1348                         goto cleanup;
1349                 }
1350         } else {
1351                 free = inode->i_sb->s_blocksize;
1352         }
1353
1354         while (isize_diff > 0) {
1355                 size_t offs, size, entry_size;
1356                 struct ext4_xattr_entry *small_entry = NULL;
1357                 struct ext4_xattr_info i = {
1358                         .value = NULL,
1359                         .value_len = 0,
1360                 };
1361                 unsigned int total_size;  /* EA entry size + value size */
1362                 unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1363                 unsigned int min_total_size = ~0U;
1364
1365                 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1366                 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1367                 if (!is || !bs) {
1368                         error = -ENOMEM;
1369                         goto cleanup;
1370                 }
1371
1372                 is->s.not_found = -ENODATA;
1373                 bs->s.not_found = -ENODATA;
1374                 is->iloc.bh = NULL;
1375                 bs->bh = NULL;
1376
1377                 last = IFIRST(header);
1378                 /* Find the entry best suited to be pushed into EA block */
1379                 entry = NULL;
1380                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1381                         /* never move system.data out of the inode */
1382                         if ((last->e_name_len == 4) &&
1383                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
1384                             !memcmp(last->e_name, "data", 4))
1385                                 continue;
1386                         total_size =
1387                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1388                                         EXT4_XATTR_LEN(last->e_name_len);
1389                         if (total_size <= free && total_size < min_total_size) {
1390                                 if (total_size < isize_diff) {
1391                                         small_entry = last;
1392                                 } else {
1393                                         entry = last;
1394                                         min_total_size = total_size;
1395                                 }
1396                         }
1397                 }
1398
1399                 if (entry == NULL) {
1400                         if (small_entry) {
1401                                 entry = small_entry;
1402                         } else {
1403                                 if (!tried_min_extra_isize &&
1404                                     s_min_extra_isize) {
1405                                         tried_min_extra_isize++;
1406                                         new_extra_isize = s_min_extra_isize;
1407                                         kfree(is); is = NULL;
1408                                         kfree(bs); bs = NULL;
1409                                         brelse(bh);
1410                                         goto retry;
1411                                 }
1412                                 error = -1;
1413                                 goto cleanup;
1414                         }
1415                 }
1416                 offs = le16_to_cpu(entry->e_value_offs);
1417                 size = le32_to_cpu(entry->e_value_size);
1418                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1419                 i.name_index = entry->e_name_index,
1420                 buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1421                 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1422                 if (!buffer || !b_entry_name) {
1423                         error = -ENOMEM;
1424                         goto cleanup;
1425                 }
1426                 /* Save the entry name and the entry value */
1427                 memcpy(buffer, (void *)IFIRST(header) + offs,
1428                        EXT4_XATTR_SIZE(size));
1429                 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1430                 b_entry_name[entry->e_name_len] = '\0';
1431                 i.name = b_entry_name;
1432
1433                 error = ext4_get_inode_loc(inode, &is->iloc);
1434                 if (error)
1435                         goto cleanup;
1436
1437                 error = ext4_xattr_ibody_find(inode, &i, is);
1438                 if (error)
1439                         goto cleanup;
1440
1441                 /* Remove the chosen entry from the inode */
1442                 error = ext4_xattr_ibody_set(handle, inode, &i, is);
1443                 if (error)
1444                         goto cleanup;
1445                 total_ino -= entry_size;
1446
1447                 entry = IFIRST(header);
1448                 if (entry_size + EXT4_XATTR_SIZE(size) >= isize_diff)
1449                         shift_bytes = isize_diff;
1450                 else
1451                         shift_bytes = entry_size + EXT4_XATTR_SIZE(size);
1452                 /* Adjust the offsets and shift the remaining entries ahead */
1453                 ext4_xattr_shift_entries(entry, -shift_bytes,
1454                         (void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
1455                         EXT4_I(inode)->i_extra_isize + shift_bytes,
1456                         (void *)header, total_ino, inode->i_sb->s_blocksize);
1457
1458                 isize_diff -= shift_bytes;
1459                 EXT4_I(inode)->i_extra_isize += shift_bytes;
1460                 header = IHDR(inode, raw_inode);
1461
1462                 i.name = b_entry_name;
1463                 i.value = buffer;
1464                 i.value_len = size;
1465                 error = ext4_xattr_block_find(inode, &i, bs);
1466                 if (error)
1467                         goto cleanup;
1468
1469                 /* Add entry which was removed from the inode into the block */
1470                 error = ext4_xattr_block_set(handle, inode, &i, bs);
1471                 if (error)
1472                         goto cleanup;
1473                 kfree(b_entry_name);
1474                 kfree(buffer);
1475                 b_entry_name = NULL;
1476                 buffer = NULL;
1477                 brelse(is->iloc.bh);
1478                 kfree(is);
1479                 kfree(bs);
1480         }
1481         brelse(bh);
1482 out:
1483         ext4_write_unlock_xattr(inode, &no_expand);
1484         return 0;
1485
1486 cleanup:
1487         kfree(b_entry_name);
1488         kfree(buffer);
1489         if (is)
1490                 brelse(is->iloc.bh);
1491         if (bs)
1492                 brelse(bs->bh);
1493         kfree(is);
1494         kfree(bs);
1495         brelse(bh);
1496         /*
1497          * Inode size expansion failed; don't try again
1498          */
1499         no_expand = 1;
1500         ext4_write_unlock_xattr(inode, &no_expand);
1501         return error;
1502 }
1503
1504
1505
1506 /*
1507  * ext4_xattr_delete_inode()
1508  *
1509  * Free extended attribute resources associated with this inode. This
1510  * is called immediately before an inode is freed. We have exclusive
1511  * access to the inode.
1512  */
1513 void
1514 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1515 {
1516         struct buffer_head *bh = NULL;
1517
1518         if (!EXT4_I(inode)->i_file_acl)
1519                 goto cleanup;
1520         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1521         if (!bh) {
1522                 EXT4_ERROR_INODE(inode, "block %llu read error",
1523                                  EXT4_I(inode)->i_file_acl);
1524                 goto cleanup;
1525         }
1526         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1527             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1528                 EXT4_ERROR_INODE(inode, "bad block %llu",
1529                                  EXT4_I(inode)->i_file_acl);
1530                 goto cleanup;
1531         }
1532         ext4_xattr_release_block(handle, inode, bh);
1533         EXT4_I(inode)->i_file_acl = 0;
1534
1535 cleanup:
1536         brelse(bh);
1537 }
1538
1539 /*
1540  * ext4_xattr_put_super()
1541  *
1542  * This is called when a file system is unmounted.
1543  */
1544 void
1545 ext4_xattr_put_super(struct super_block *sb)
1546 {
1547         mb_cache_shrink(sb->s_bdev);
1548 }
1549
1550 /*
1551  * ext4_xattr_cache_insert()
1552  *
1553  * Create a new entry in the extended attribute cache, and insert
1554  * it unless such an entry is already in the cache.
1555  *
1556  * Returns 0, or a negative error number on failure.
1557  */
1558 static void
1559 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1560 {
1561         __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1562         struct mb_cache_entry *ce;
1563         int error;
1564
1565         ce = mb_cache_entry_alloc(ext4_mb_cache, GFP_NOFS);
1566         if (!ce) {
1567                 ea_bdebug(bh, "out of memory");
1568                 return;
1569         }
1570         error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1571         if (error) {
1572                 mb_cache_entry_free(ce);
1573                 if (error == -EBUSY) {
1574                         ea_bdebug(bh, "already in cache");
1575                         error = 0;
1576                 }
1577         } else {
1578                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1579                 mb_cache_entry_release(ce);
1580         }
1581 }
1582
1583 /*
1584  * ext4_xattr_cmp()
1585  *
1586  * Compare two extended attribute blocks for equality.
1587  *
1588  * Returns 0 if the blocks are equal, 1 if they differ, and
1589  * a negative error number on errors.
1590  */
1591 static int
1592 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1593                struct ext4_xattr_header *header2)
1594 {
1595         struct ext4_xattr_entry *entry1, *entry2;
1596
1597         entry1 = ENTRY(header1+1);
1598         entry2 = ENTRY(header2+1);
1599         while (!IS_LAST_ENTRY(entry1)) {
1600                 if (IS_LAST_ENTRY(entry2))
1601                         return 1;
1602                 if (entry1->e_hash != entry2->e_hash ||
1603                     entry1->e_name_index != entry2->e_name_index ||
1604                     entry1->e_name_len != entry2->e_name_len ||
1605                     entry1->e_value_size != entry2->e_value_size ||
1606                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1607                         return 1;
1608                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1609                         return -EFSCORRUPTED;
1610                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1611                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1612                            le32_to_cpu(entry1->e_value_size)))
1613                         return 1;
1614
1615                 entry1 = EXT4_XATTR_NEXT(entry1);
1616                 entry2 = EXT4_XATTR_NEXT(entry2);
1617         }
1618         if (!IS_LAST_ENTRY(entry2))
1619                 return 1;
1620         return 0;
1621 }
1622
1623 /*
1624  * ext4_xattr_cache_find()
1625  *
1626  * Find an identical extended attribute block.
1627  *
1628  * Returns a pointer to the block found, or NULL if such a block was
1629  * not found or an error occurred.
1630  */
1631 static struct buffer_head *
1632 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1633                       struct mb_cache_entry **pce)
1634 {
1635         __u32 hash = le32_to_cpu(header->h_hash);
1636         struct mb_cache_entry *ce;
1637         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1638
1639         if (!header->h_hash)
1640                 return NULL;  /* never share */
1641         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1642 again:
1643         ce = mb_cache_entry_find_first(ext4_mb_cache, inode->i_sb->s_bdev,
1644                                        hash);
1645         while (ce) {
1646                 struct buffer_head *bh;
1647
1648                 if (IS_ERR(ce)) {
1649                         if (PTR_ERR(ce) == -EAGAIN)
1650                                 goto again;
1651                         break;
1652                 }
1653                 bh = sb_bread(inode->i_sb, ce->e_block);
1654                 if (!bh) {
1655                         EXT4_ERROR_INODE(inode, "block %lu read error",
1656                                          (unsigned long) ce->e_block);
1657                 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1658                                 EXT4_XATTR_REFCOUNT_MAX) {
1659                         ea_idebug(inode, "block %lu refcount %d>=%d",
1660                                   (unsigned long) ce->e_block,
1661                                   le32_to_cpu(BHDR(bh)->h_refcount),
1662                                           EXT4_XATTR_REFCOUNT_MAX);
1663                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1664                         *pce = ce;
1665                         return bh;
1666                 }
1667                 brelse(bh);
1668                 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1669         }
1670         return NULL;
1671 }
1672
1673 #define NAME_HASH_SHIFT 5
1674 #define VALUE_HASH_SHIFT 16
1675
1676 /*
1677  * ext4_xattr_hash_entry()
1678  *
1679  * Compute the hash of an extended attribute.
1680  */
1681 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1682                                          struct ext4_xattr_entry *entry)
1683 {
1684         __u32 hash = 0;
1685         char *name = entry->e_name;
1686         int n;
1687
1688         for (n = 0; n < entry->e_name_len; n++) {
1689                 hash = (hash << NAME_HASH_SHIFT) ^
1690                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1691                        *name++;
1692         }
1693
1694         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1695                 __le32 *value = (__le32 *)((char *)header +
1696                         le16_to_cpu(entry->e_value_offs));
1697                 for (n = (le32_to_cpu(entry->e_value_size) +
1698                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1699                         hash = (hash << VALUE_HASH_SHIFT) ^
1700                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1701                                le32_to_cpu(*value++);
1702                 }
1703         }
1704         entry->e_hash = cpu_to_le32(hash);
1705 }
1706
1707 #undef NAME_HASH_SHIFT
1708 #undef VALUE_HASH_SHIFT
1709
1710 #define BLOCK_HASH_SHIFT 16
1711
1712 /*
1713  * ext4_xattr_rehash()
1714  *
1715  * Re-compute the extended attribute hash value after an entry has changed.
1716  */
1717 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1718                               struct ext4_xattr_entry *entry)
1719 {
1720         struct ext4_xattr_entry *here;
1721         __u32 hash = 0;
1722
1723         ext4_xattr_hash_entry(header, entry);
1724         here = ENTRY(header+1);
1725         while (!IS_LAST_ENTRY(here)) {
1726                 if (!here->e_hash) {
1727                         /* Block is not shared if an entry's hash value == 0 */
1728                         hash = 0;
1729                         break;
1730                 }
1731                 hash = (hash << BLOCK_HASH_SHIFT) ^
1732                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1733                        le32_to_cpu(here->e_hash);
1734                 here = EXT4_XATTR_NEXT(here);
1735         }
1736         header->h_hash = cpu_to_le32(hash);
1737 }
1738
1739 #undef BLOCK_HASH_SHIFT
1740
1741 #define HASH_BUCKET_BITS        10
1742
1743 struct mb_cache *
1744 ext4_xattr_create_cache(char *name)
1745 {
1746         return mb_cache_create(name, HASH_BUCKET_BITS);
1747 }
1748
1749 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1750 {
1751         if (cache)
1752                 mb_cache_destroy(cache);
1753 }
1754