fs/ext4/xattr.c

   1 /*
   2  * linux/fs/ext4/xattr.c
   3  *
   4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   5  *
   6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   7  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   8  * Extended attributes for symlinks and special files added per
   9  *  suggestion of Luka Renko <luka.renko@hermes.si>.
  10  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  11  *  Red Hat Inc.
  12  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  13  *  and Andreas Gruenbacher <agruen@suse.de>.
  14  */
  15
  16 /*
  17  * Extended attributes are stored directly in inodes (on file systems with
  18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  19  * field contains the block number if an inode uses an additional block. All
  20  * attributes must fit in the inode and one additional block. Blocks that
  21  * contain the identical set of attributes may be shared among several inodes.
  22  * Identical blocks are detected by keeping a cache of blocks that have
  23  * recently been accessed.
  24  *
  25  * The attributes in inodes and on blocks have a different header; the entries
  26  * are stored in the same format:
  27  *
  28  *   +------------------+
  29  *   | header           |
  30  *   | entry 1          | |
  31  *   | entry 2          | | growing downwards
  32  *   | entry 3          | v
  33  *   | four null bytes  |
  34  *   | . . .            |
  35  *   | value 1          | ^
  36  *   | value 3          | | growing upwards
  37  *   | value 2          | |
  38  *   +------------------+
  39  *
  40  * The header is followed by multiple entry descriptors. In disk blocks, the
  41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
  42  * attribute values are aligned to the end of the block in no specific order.
  43  *
  44  * Locking strategy
  45  * ----------------
  46  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  47  * EA blocks are only changed if they are exclusive to an inode, so
  48  * holding xattr_sem also means that nothing but the EA block's reference
  49  * count can change. Multiple writers to the same block are synchronized
  50  * by the buffer lock.
  51  */
  52
  53 #include <linux/init.h>
  54 #include <linux/fs.h>
  55 #include <linux/slab.h>
  56 #include <linux/mbcache.h>
  57 #include <linux/quotaops.h>
  58 #include "ext4_jbd2.h"
  59 #include "ext4.h"
  60 #include "xattr.h"
  61 #include "acl.h"
  62
  63 #ifdef EXT4_XATTR_DEBUG
  64 # define ea_idebug(inode, fmt, ...)                                     \
  65         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  66                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  67 # define ea_bdebug(bh, fmt, ...)                                        \
  68         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  69                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  70 #else
  71 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  72 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  73 #endif
  74
  75 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  76 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
  77                                                  struct ext4_xattr_header *,
  78                                                  struct mb_cache_entry **);
  79 static void ext4_xattr_rehash(struct ext4_xattr_header *,
  80                               struct ext4_xattr_entry *);
  81 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
  82                            size_t buffer_size);
  83
  84 static const struct xattr_handler *ext4_xattr_handler_map[] = {
  85         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  86 #ifdef CONFIG_EXT4_FS_POSIX_ACL
  87         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  88         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  89 #endif
  90         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  91 #ifdef CONFIG_EXT4_FS_SECURITY
  92         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  93 #endif
  94 };
  95
  96 const struct xattr_handler *ext4_xattr_handlers[] = {
  97         &ext4_xattr_user_handler,
  98         &ext4_xattr_trusted_handler,
  99 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 100         &posix_acl_access_xattr_handler,
 101         &posix_acl_default_xattr_handler,
 102 #endif
 103 #ifdef CONFIG_EXT4_FS_SECURITY
 104         &ext4_xattr_security_handler,
 105 #endif
 106         NULL
 107 };
 108
 109 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
 110                                 inode->i_sb->s_fs_info)->s_mb_cache)
 111
 112 static __le32 ext4_xattr_block_csum(struct inode *inode,
 113                                     sector_t block_nr,
 114                                     struct ext4_xattr_header *hdr)
 115 {
 116         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 117         __u32 csum;
 118         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 119         __u32 dummy_csum = 0;
 120         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 121
 122         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 123                            sizeof(dsk_block_nr));
 124         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 125         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 126         offset += sizeof(dummy_csum);
 127         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 128                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 129
 130         return cpu_to_le32(csum);
 131 }
 132
 133 static int ext4_xattr_block_csum_verify(struct inode *inode,
 134                                         struct buffer_head *bh)
 135 {
 136         struct ext4_xattr_header *hdr = BHDR(bh);
 137         int ret = 1;
 138
 139         if (ext4_has_metadata_csum(inode->i_sb)) {
 140                 lock_buffer(bh);
 141                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 142                                                         bh->b_blocknr, hdr));
 143                 unlock_buffer(bh);
 144         }
 145         return ret;
 146 }
 147
 148 static void ext4_xattr_block_csum_set(struct inode *inode,
 149                                       struct buffer_head *bh)
 150 {
 151         if (ext4_has_metadata_csum(inode->i_sb))
 152                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 153                                                 bh->b_blocknr, BHDR(bh));
 154 }
 155
 156 static inline const struct xattr_handler *
 157 ext4_xattr_handler(int name_index)
 158 {
 159         const struct xattr_handler *handler = NULL;
 160
 161         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 162                 handler = ext4_xattr_handler_map[name_index];
 163         return handler;
 164 }
 165
 166 /*
 167  * Inode operation listxattr()
 168  *
 169  * d_inode(dentry)->i_mutex: don't care
 170  */
 171 ssize_t
 172 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
 173 {
 174         return ext4_xattr_list(dentry, buffer, size);
 175 }
 176
 177 static int
 178 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
 179                        void *value_start)
 180 {
 181         struct ext4_xattr_entry *e = entry;
 182
 183         /* Find the end of the names list */
 184         while (!IS_LAST_ENTRY(e)) {
 185                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 186                 if ((void *)next >= end)
 187                         return -EFSCORRUPTED;
 188                 e = next;
 189         }
 190
 191         /* Check the values */
 192         while (!IS_LAST_ENTRY(entry)) {
 193                 if (entry->e_value_block != 0)
 194                         return -EFSCORRUPTED;
 195                 if (entry->e_value_size != 0) {
 196                         u16 offs = le16_to_cpu(entry->e_value_offs);
 197                         u32 size = le32_to_cpu(entry->e_value_size);
 198                         void *value;
 199
 200                         /*
 201                          * The value cannot overlap the names, and the value
 202                          * with padding cannot extend beyond 'end'.  Check both
 203                          * the padded and unpadded sizes, since the size may
 204                          * overflow to 0 when adding padding.
 205                          */
 206                         if (offs > end - value_start)
 207                                 return -EFSCORRUPTED;
 208                         value = value_start + offs;
 209                         if (value < (void *)e + sizeof(u32) ||
 210                             size > end - value ||
 211                             EXT4_XATTR_SIZE(size) > end - value)
 212                                 return -EFSCORRUPTED;
 213                 }
 214                 entry = EXT4_XATTR_NEXT(entry);
 215         }
 216
 217         return 0;
 218 }
 219
 220 static inline int
 221 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 222 {
 223         int error;
 224
 225         if (buffer_verified(bh))
 226                 return 0;
 227
 228         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 229             BHDR(bh)->h_blocks != cpu_to_le32(1))
 230                 return -EFSCORRUPTED;
 231         if (!ext4_xattr_block_csum_verify(inode, bh))
 232                 return -EFSBADCRC;
 233         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
 234                                        bh->b_data);
 235         if (!error)
 236                 set_buffer_verified(bh);
 237         return error;
 238 }
 239
 240 static int
 241 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 242                          void *end, const char *function, unsigned int line)
 243 {
 244         int error = -EFSCORRUPTED;
 245
 246         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 247             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 248                 goto errout;
 249         error = ext4_xattr_check_names(IFIRST(header), end, IFIRST(header));
 250 errout:
 251         if (error)
 252                 __ext4_error_inode(inode, function, line, 0,
 253                                    "corrupted in-inode xattr");
 254         return error;
 255 }
 256
 257 #define xattr_check_inode(inode, header, end) \
 258         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 259
 260 static inline int
 261 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 262 {
 263         size_t value_size = le32_to_cpu(entry->e_value_size);
 264
 265         if (entry->e_value_block != 0 || value_size > size ||
 266             le16_to_cpu(entry->e_value_offs) + value_size > size)
 267                 return -EFSCORRUPTED;
 268         return 0;
 269 }
 270
 271 static int
 272 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 273                       const char *name, size_t size, int sorted)
 274 {
 275         struct ext4_xattr_entry *entry;
 276         size_t name_len;
 277         int cmp = 1;
 278
 279         if (name == NULL)
 280                 return -EINVAL;
 281         name_len = strlen(name);
 282         entry = *pentry;
 283         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 284                 cmp = name_index - entry->e_name_index;
 285                 if (!cmp)
 286                         cmp = name_len - entry->e_name_len;
 287                 if (!cmp)
 288                         cmp = memcmp(name, entry->e_name, name_len);
 289                 if (cmp <= 0 && (sorted || cmp == 0))
 290                         break;
 291         }
 292         *pentry = entry;
 293         if (!cmp && ext4_xattr_check_entry(entry, size))
 294                 return -EFSCORRUPTED;
 295         return cmp ? -ENODATA : 0;
 296 }
 297
 298 static int
 299 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 300                      void *buffer, size_t buffer_size)
 301 {
 302         struct buffer_head *bh = NULL;
 303         struct ext4_xattr_entry *entry;
 304         size_t size;
 305         int error;
 306         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 307
 308         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 309                   name_index, name, buffer, (long)buffer_size);
 310
 311         error = -ENODATA;
 312         if (!EXT4_I(inode)->i_file_acl)
 313                 goto cleanup;
 314         ea_idebug(inode, "reading block %llu",
 315                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 316         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 317         if (!bh)
 318                 goto cleanup;
 319         ea_bdebug(bh, "b_count=%d, refcount=%d",
 320                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 321         if (ext4_xattr_check_block(inode, bh)) {
 322 bad_block:
 323                 EXT4_ERROR_INODE(inode, "bad block %llu",
 324                                  EXT4_I(inode)->i_file_acl);
 325                 error = -EFSCORRUPTED;
 326                 goto cleanup;
 327         }
 328         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 329         entry = BFIRST(bh);
 330         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 331         if (error == -EFSCORRUPTED)
 332                 goto bad_block;
 333         if (error)
 334                 goto cleanup;
 335         size = le32_to_cpu(entry->e_value_size);
 336         if (buffer) {
 337                 error = -ERANGE;
 338                 if (size > buffer_size)
 339                         goto cleanup;
 340                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 341                        size);
 342         }
 343         error = size;
 344
 345 cleanup:
 346         brelse(bh);
 347         return error;
 348 }
 349
 350 int
 351 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 352                      void *buffer, size_t buffer_size)
 353 {
 354         struct ext4_xattr_ibody_header *header;
 355         struct ext4_xattr_entry *entry;
 356         struct ext4_inode *raw_inode;
 357         struct ext4_iloc iloc;
 358         size_t size;
 359         void *end;
 360         int error;
 361
 362         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 363                 return -ENODATA;
 364         error = ext4_get_inode_loc(inode, &iloc);
 365         if (error)
 366                 return error;
 367         raw_inode = ext4_raw_inode(&iloc);
 368         header = IHDR(inode, raw_inode);
 369         entry = IFIRST(header);
 370         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 371         error = xattr_check_inode(inode, header, end);
 372         if (error)
 373                 goto cleanup;
 374         error = ext4_xattr_find_entry(&entry, name_index, name,
 375                                       end - (void *)entry, 0);
 376         if (error)
 377                 goto cleanup;
 378         size = le32_to_cpu(entry->e_value_size);
 379         if (buffer) {
 380                 error = -ERANGE;
 381                 if (size > buffer_size)
 382                         goto cleanup;
 383                 memcpy(buffer, (void *)IFIRST(header) +
 384                        le16_to_cpu(entry->e_value_offs), size);
 385         }
 386         error = size;
 387
 388 cleanup:
 389         brelse(iloc.bh);
 390         return error;
 391 }
 392
 393 /*
 394  * ext4_xattr_get()
 395  *
 396  * Copy an extended attribute into the buffer
 397  * provided, or compute the buffer size required.
 398  * Buffer is NULL to compute the size of the buffer required.
 399  *
 400  * Returns a negative error number on failure, or the number of bytes
 401  * used / required on success.
 402  */
 403 int
 404 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 405                void *buffer, size_t buffer_size)
 406 {
 407         int error;
 408
 409         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 410                 return -EIO;
 411
 412         if (strlen(name) > 255)
 413                 return -ERANGE;
 414
 415         down_read(&EXT4_I(inode)->xattr_sem);
 416         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 417                                      buffer_size);
 418         if (error == -ENODATA)
 419                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 420                                              buffer_size);
 421         up_read(&EXT4_I(inode)->xattr_sem);
 422         return error;
 423 }
 424
 425 static int
 426 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 427                         char *buffer, size_t buffer_size)
 428 {
 429         size_t rest = buffer_size;
 430
 431         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 432                 const struct xattr_handler *handler =
 433                         ext4_xattr_handler(entry->e_name_index);
 434
 435                 if (handler && (!handler->list || handler->list(dentry))) {
 436                         const char *prefix = handler->prefix ?: handler->name;
 437                         size_t prefix_len = strlen(prefix);
 438                         size_t size = prefix_len + entry->e_name_len + 1;
 439
 440                         if (buffer) {
 441                                 if (size > rest)
 442                                         return -ERANGE;
 443                                 memcpy(buffer, prefix, prefix_len);
 444                                 buffer += prefix_len;
 445                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 446                                 buffer += entry->e_name_len;
 447                                 *buffer++ = 0;
 448                         }
 449                         rest -= size;
 450                 }
 451         }
 452         return buffer_size - rest;  /* total size */
 453 }
 454
 455 static int
 456 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 457 {
 458         struct inode *inode = d_inode(dentry);
 459         struct buffer_head *bh = NULL;
 460         int error;
 461         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 462
 463         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 464                   buffer, (long)buffer_size);
 465
 466         error = 0;
 467         if (!EXT4_I(inode)->i_file_acl)
 468                 goto cleanup;
 469         ea_idebug(inode, "reading block %llu",
 470                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 471         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 472         error = -EIO;
 473         if (!bh)
 474                 goto cleanup;
 475         ea_bdebug(bh, "b_count=%d, refcount=%d",
 476                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 477         if (ext4_xattr_check_block(inode, bh)) {
 478                 EXT4_ERROR_INODE(inode, "bad block %llu",
 479                                  EXT4_I(inode)->i_file_acl);
 480                 error = -EFSCORRUPTED;
 481                 goto cleanup;
 482         }
 483         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 484         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 485
 486 cleanup:
 487         brelse(bh);
 488
 489         return error;
 490 }
 491
 492 static int
 493 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 494 {
 495         struct inode *inode = d_inode(dentry);
 496         struct ext4_xattr_ibody_header *header;
 497         struct ext4_inode *raw_inode;
 498         struct ext4_iloc iloc;
 499         void *end;
 500         int error;
 501
 502         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 503                 return 0;
 504         error = ext4_get_inode_loc(inode, &iloc);
 505         if (error)
 506                 return error;
 507         raw_inode = ext4_raw_inode(&iloc);
 508         header = IHDR(inode, raw_inode);
 509         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 510         error = xattr_check_inode(inode, header, end);
 511         if (error)
 512                 goto cleanup;
 513         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 514                                         buffer, buffer_size);
 515
 516 cleanup:
 517         brelse(iloc.bh);
 518         return error;
 519 }
 520
 521 /*
 522  * ext4_xattr_list()
 523  *
 524  * Copy a list of attribute names into the buffer
 525  * provided, or compute the buffer size required.
 526  * Buffer is NULL to compute the size of the buffer required.
 527  *
 528  * Returns a negative error number on failure, or the number of bytes
 529  * used / required on success.
 530  */
 531 static int
 532 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 533 {
 534         int ret, ret2;
 535
 536         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 537         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 538         if (ret < 0)
 539                 goto errout;
 540         if (buffer) {
 541                 buffer += ret;
 542                 buffer_size -= ret;
 543         }
 544         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 545         if (ret < 0)
 546                 goto errout;
 547         ret += ret2;
 548 errout:
 549         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 550         return ret;
 551 }
 552
 553 /*
 554  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 555  * not set, set it.
 556  */
 557 static void ext4_xattr_update_super_block(handle_t *handle,
 558                                           struct super_block *sb)
 559 {
 560         if (ext4_has_feature_xattr(sb))
 561                 return;
 562
 563         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 564         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 565                 ext4_set_feature_xattr(sb);
 566                 ext4_handle_dirty_super(handle, sb);
 567         }
 568 }
 569
 570 /*
 571  * Release the xattr block BH: If the reference count is > 1, decrement it;
 572  * otherwise free the block.
 573  */
 574 static void
 575 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 576                          struct buffer_head *bh)
 577 {
 578         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 579         u32 hash, ref;
 580         int error = 0;
 581
 582         BUFFER_TRACE(bh, "get_write_access");
 583         error = ext4_journal_get_write_access(handle, bh);
 584         if (error)
 585                 goto out;
 586
 587         lock_buffer(bh);
 588         hash = le32_to_cpu(BHDR(bh)->h_hash);
 589         ref = le32_to_cpu(BHDR(bh)->h_refcount);
 590         if (ref == 1) {
 591                 ea_bdebug(bh, "refcount now=0; freeing");
 592                 /*
 593                  * This must happen under buffer lock for
 594                  * ext4_xattr_block_set() to reliably detect freed block
 595                  */
 596                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
 597                 get_bh(bh);
 598                 unlock_buffer(bh);
 599                 ext4_free_blocks(handle, inode, bh, 0, 1,
 600                                  EXT4_FREE_BLOCKS_METADATA |
 601                                  EXT4_FREE_BLOCKS_FORGET);
 602         } else {
 603                 ref--;
 604                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
 605                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
 606                         struct mb_cache_entry *ce;
 607
 608                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
 609                                                 bh->b_blocknr);
 610                         if (ce) {
 611                                 ce->e_reusable = 1;
 612                                 mb_cache_entry_put(ext4_mb_cache, ce);
 613                         }
 614                 }
 615
 616                 ext4_xattr_block_csum_set(inode, bh);
 617                 /*
 618                  * Beware of this ugliness: Releasing of xattr block references
 619                  * from different inodes can race and so we have to protect
 620                  * from a race where someone else frees the block (and releases
 621                  * its journal_head) before we are done dirtying the buffer. In
 622                  * nojournal mode this race is harmless and we actually cannot
 623                  * call ext4_handle_dirty_metadata() with locked buffer as
 624                  * that function can call sync_dirty_buffer() so for that case
 625                  * we handle the dirtying after unlocking the buffer.
 626                  */
 627                 if (ext4_handle_valid(handle))
 628                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 629                 unlock_buffer(bh);
 630                 if (!ext4_handle_valid(handle))
 631                         error = ext4_handle_dirty_metadata(handle, inode, bh);
 632                 if (IS_SYNC(inode))
 633                         ext4_handle_sync(handle);
 634                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
 635                 ea_bdebug(bh, "refcount now=%d; releasing",
 636                           le32_to_cpu(BHDR(bh)->h_refcount));
 637         }
 638 out:
 639         ext4_std_error(inode->i_sb, error);
 640         return;
 641 }
 642
 643 /*
 644  * Find the available free space for EAs. This also returns the total number of
 645  * bytes used by EA entries.
 646  */
 647 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
 648                                     size_t *min_offs, void *base, int *total)
 649 {
 650         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 651                 if (last->e_value_size) {
 652                         size_t offs = le16_to_cpu(last->e_value_offs);
 653                         if (offs < *min_offs)
 654                                 *min_offs = offs;
 655                 }
 656                 if (total)
 657                         *total += EXT4_XATTR_LEN(last->e_name_len);
 658         }
 659         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 660 }
 661
 662 static int
 663 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 664 {
 665         struct ext4_xattr_entry *last;
 666         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 667
 668         /* Compute min_offs and last. */
 669         last = s->first;
 670         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 671                 if (last->e_value_size) {
 672                         size_t offs = le16_to_cpu(last->e_value_offs);
 673                         if (offs < min_offs)
 674                                 min_offs = offs;
 675                 }
 676         }
 677         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 678         if (!s->not_found) {
 679                 if (s->here->e_value_size) {
 680                         size_t size = le32_to_cpu(s->here->e_value_size);
 681                         free += EXT4_XATTR_SIZE(size);
 682                 }
 683                 free += EXT4_XATTR_LEN(name_len);
 684         }
 685         if (i->value) {
 686                 if (free < EXT4_XATTR_LEN(name_len) +
 687                            EXT4_XATTR_SIZE(i->value_len))
 688                         return -ENOSPC;
 689         }
 690
 691         if (i->value && s->not_found) {
 692                 /* Insert the new name. */
 693                 size_t size = EXT4_XATTR_LEN(name_len);
 694                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
 695                 memmove((void *)s->here + size, s->here, rest);
 696                 memset(s->here, 0, size);
 697                 s->here->e_name_index = i->name_index;
 698                 s->here->e_name_len = name_len;
 699                 memcpy(s->here->e_name, i->name, name_len);
 700         } else {
 701                 if (s->here->e_value_size) {
 702                         void *first_val = s->base + min_offs;
 703                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 704                         void *val = s->base + offs;
 705                         size_t size = EXT4_XATTR_SIZE(
 706                                 le32_to_cpu(s->here->e_value_size));
 707
 708                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
 709                                 /* The old and the new value have the same
 710                                    size. Just replace. */
 711                                 s->here->e_value_size =
 712                                         cpu_to_le32(i->value_len);
 713                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
 714                                         memset(val, 0, size);
 715                                 } else {
 716                                         /* Clear pad bytes first. */
 717                                         memset(val + size - EXT4_XATTR_PAD, 0,
 718                                                EXT4_XATTR_PAD);
 719                                         memcpy(val, i->value, i->value_len);
 720                                 }
 721                                 return 0;
 722                         }
 723
 724                         /* Remove the old value. */
 725                         memmove(first_val + size, first_val, val - first_val);
 726                         memset(first_val, 0, size);
 727                         s->here->e_value_size = 0;
 728                         s->here->e_value_offs = 0;
 729                         min_offs += size;
 730
 731                         /* Adjust all value offsets. */
 732                         last = s->first;
 733                         while (!IS_LAST_ENTRY(last)) {
 734                                 size_t o = le16_to_cpu(last->e_value_offs);
 735                                 if (last->e_value_size && o < offs)
 736                                         last->e_value_offs =
 737                                                 cpu_to_le16(o + size);
 738                                 last = EXT4_XATTR_NEXT(last);
 739                         }
 740                 }
 741                 if (!i->value) {
 742                         /* Remove the old name. */
 743                         size_t size = EXT4_XATTR_LEN(name_len);
 744                         last = ENTRY((void *)last - size);
 745                         memmove(s->here, (void *)s->here + size,
 746                                 (void *)last - (void *)s->here + sizeof(__u32));
 747                         memset(last, 0, size);
 748                 }
 749         }
 750
 751         if (i->value) {
 752                 /* Insert the new value. */
 753                 s->here->e_value_size = cpu_to_le32(i->value_len);
 754                 if (i->value_len) {
 755                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 756                         void *val = s->base + min_offs - size;
 757                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 758                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 759                                 memset(val, 0, size);
 760                         } else {
 761                                 /* Clear the pad bytes first. */
 762                                 memset(val + size - EXT4_XATTR_PAD, 0,
 763                                        EXT4_XATTR_PAD);
 764                                 memcpy(val, i->value, i->value_len);
 765                         }
 766                 }
 767         }
 768         return 0;
 769 }
 770
 771 struct ext4_xattr_block_find {
 772         struct ext4_xattr_search s;
 773         struct buffer_head *bh;
 774 };
 775
 776 static int
 777 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
 778                       struct ext4_xattr_block_find *bs)
 779 {
 780         struct super_block *sb = inode->i_sb;
 781         int error;
 782
 783         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 784                   i->name_index, i->name, i->value, (long)i->value_len);
 785
 786         if (EXT4_I(inode)->i_file_acl) {
 787                 /* The inode already has an extended attribute block. */
 788                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
 789                 error = -EIO;
 790                 if (!bs->bh)
 791                         goto cleanup;
 792                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
 793                         atomic_read(&(bs->bh->b_count)),
 794                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
 795                 if (ext4_xattr_check_block(inode, bs->bh)) {
 796                         EXT4_ERROR_INODE(inode, "bad block %llu",
 797                                          EXT4_I(inode)->i_file_acl);
 798                         error = -EFSCORRUPTED;
 799                         goto cleanup;
 800                 }
 801                 /* Find the named attribute. */
 802                 bs->s.base = BHDR(bs->bh);
 803                 bs->s.first = BFIRST(bs->bh);
 804                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 805                 bs->s.here = bs->s.first;
 806                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 807                                               i->name, bs->bh->b_size, 1);
 808                 if (error && error != -ENODATA)
 809                         goto cleanup;
 810                 bs->s.not_found = error;
 811         }
 812         error = 0;
 813
 814 cleanup:
 815         return error;
 816 }
 817
 818 static int
 819 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 820                      struct ext4_xattr_info *i,
 821                      struct ext4_xattr_block_find *bs)
 822 {
 823         struct super_block *sb = inode->i_sb;
 824         struct buffer_head *new_bh = NULL;
 825         struct ext4_xattr_search *s = &bs->s;
 826         struct mb_cache_entry *ce = NULL;
 827         int error = 0;
 828         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 829
 830 #define header(x) ((struct ext4_xattr_header *)(x))
 831
 832         if (i->value && i->value_len > sb->s_blocksize)
 833                 return -ENOSPC;
 834         if (s->base) {
 835                 BUFFER_TRACE(bs->bh, "get_write_access");
 836                 error = ext4_journal_get_write_access(handle, bs->bh);
 837                 if (error)
 838                         goto cleanup;
 839                 lock_buffer(bs->bh);
 840
 841                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
 842                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
 843
 844                         /*
 845                          * This must happen under buffer lock for
 846                          * ext4_xattr_block_set() to reliably detect modified
 847                          * block
 848                          */
 849                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
 850                                                     bs->bh->b_blocknr);
 851                         ea_bdebug(bs->bh, "modifying in-place");
 852                         error = ext4_xattr_set_entry(i, s);
 853                         if (!error) {
 854                                 if (!IS_LAST_ENTRY(s->first))
 855                                         ext4_xattr_rehash(header(s->base),
 856                                                           s->here);
 857                                 ext4_xattr_cache_insert(ext4_mb_cache,
 858                                         bs->bh);
 859                         }
 860                         ext4_xattr_block_csum_set(inode, bs->bh);
 861                         unlock_buffer(bs->bh);
 862                         if (error == -EFSCORRUPTED)
 863                                 goto bad_block;
 864                         if (!error)
 865                                 error = ext4_handle_dirty_metadata(handle,
 866                                                                    inode,
 867                                                                    bs->bh);
 868                         if (error)
 869                                 goto cleanup;
 870                         goto inserted;
 871                 } else {
 872                         int offset = (char *)s->here - bs->bh->b_data;
 873
 874                         unlock_buffer(bs->bh);
 875                         ea_bdebug(bs->bh, "cloning");
 876                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
 877                         error = -ENOMEM;
 878                         if (s->base == NULL)
 879                                 goto cleanup;
 880                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
 881                         s->first = ENTRY(header(s->base)+1);
 882                         header(s->base)->h_refcount = cpu_to_le32(1);
 883                         s->here = ENTRY(s->base + offset);
 884                         s->end = s->base + bs->bh->b_size;
 885                 }
 886         } else {
 887                 /* Allocate a buffer where we construct the new block. */
 888                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 889                 /* assert(header == s->base) */
 890                 error = -ENOMEM;
 891                 if (s->base == NULL)
 892                         goto cleanup;
 893                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
 894                 header(s->base)->h_blocks = cpu_to_le32(1);
 895                 header(s->base)->h_refcount = cpu_to_le32(1);
 896                 s->first = ENTRY(header(s->base)+1);
 897                 s->here = ENTRY(header(s->base)+1);
 898                 s->end = s->base + sb->s_blocksize;
 899         }
 900
 901         error = ext4_xattr_set_entry(i, s);
 902         if (error == -EFSCORRUPTED)
 903                 goto bad_block;
 904         if (error)
 905                 goto cleanup;
 906         if (!IS_LAST_ENTRY(s->first))
 907                 ext4_xattr_rehash(header(s->base), s->here);
 908
 909 inserted:
 910         if (!IS_LAST_ENTRY(s->first)) {
 911                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
 912                 if (new_bh) {
 913                         /* We found an identical block in the cache. */
 914                         if (new_bh == bs->bh)
 915                                 ea_bdebug(new_bh, "keeping");
 916                         else {
 917                                 u32 ref;
 918
 919                                 /* The old block is released after updating
 920                                    the inode. */
 921                                 error = dquot_alloc_block(inode,
 922                                                 EXT4_C2B(EXT4_SB(sb), 1));
 923                                 if (error)
 924                                         goto cleanup;
 925                                 BUFFER_TRACE(new_bh, "get_write_access");
 926                                 error = ext4_journal_get_write_access(handle,
 927                                                                       new_bh);
 928                                 if (error)
 929                                         goto cleanup_dquot;
 930                                 lock_buffer(new_bh);
 931                                 /*
 932                                  * We have to be careful about races with
 933                                  * freeing, rehashing or adding references to
 934                                  * xattr block. Once we hold buffer lock xattr
 935                                  * block's state is stable so we can check
 936                                  * whether the block got freed / rehashed or
 937                                  * not.  Since we unhash mbcache entry under
 938                                  * buffer lock when freeing / rehashing xattr
 939                                  * block, checking whether entry is still
 940                                  * hashed is reliable. Same rules hold for
 941                                  * e_reusable handling.
 942                                  */
 943                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
 944                                     !ce->e_reusable) {
 945                                         /*
 946                                          * Undo everything and check mbcache
 947                                          * again.
 948                                          */
 949                                         unlock_buffer(new_bh);
 950                                         dquot_free_block(inode,
 951                                                          EXT4_C2B(EXT4_SB(sb),
 952                                                                   1));
 953                                         brelse(new_bh);
 954                                         mb_cache_entry_put(ext4_mb_cache, ce);
 955                                         ce = NULL;
 956                                         new_bh = NULL;
 957                                         goto inserted;
 958                                 }
 959                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
 960                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
 961                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
 962                                         ce->e_reusable = 0;
 963                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
 964                                           ref);
 965                                 ext4_xattr_block_csum_set(inode, new_bh);
 966                                 unlock_buffer(new_bh);
 967                                 error = ext4_handle_dirty_metadata(handle,
 968                                                                    inode,
 969                                                                    new_bh);
 970                                 if (error)
 971                                         goto cleanup_dquot;
 972                         }
 973                         mb_cache_entry_touch(ext4_mb_cache, ce);
 974                         mb_cache_entry_put(ext4_mb_cache, ce);
 975                         ce = NULL;
 976                 } else if (bs->bh && s->base == bs->bh->b_data) {
 977                         /* We were modifying this block in-place. */
 978                         ea_bdebug(bs->bh, "keeping this block");
 979                         new_bh = bs->bh;
 980                         get_bh(new_bh);
 981                 } else {
 982                         /* We need to allocate a new block */
 983                         ext4_fsblk_t goal, block;
 984
 985                         goal = ext4_group_first_block_no(sb,
 986                                                 EXT4_I(inode)->i_block_group);
 987
 988                         /* non-extent files can't have physical blocks past 2^32 */
 989                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 990                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 991
 992                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
 993                                                      NULL, &error);
 994                         if (error)
 995                                 goto cleanup;
 996
 997                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 998                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
 999
1000                         ea_idebug(inode, "creating block %llu",
1001                                   (unsigned long long)block);
1002
1003                         new_bh = sb_getblk(sb, block);
1004                         if (unlikely(!new_bh)) {
1005                                 error = -ENOMEM;
1006 getblk_failed:
1007                                 ext4_free_blocks(handle, inode, NULL, block, 1,
1008                                                  EXT4_FREE_BLOCKS_METADATA);
1009                                 goto cleanup;
1010                         }
1011                         lock_buffer(new_bh);
1012                         error = ext4_journal_get_create_access(handle, new_bh);
1013                         if (error) {
1014                                 unlock_buffer(new_bh);
1015                                 error = -EIO;
1016                                 goto getblk_failed;
1017                         }
1018                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1019                         ext4_xattr_block_csum_set(inode, new_bh);
1020                         set_buffer_uptodate(new_bh);
1021                         unlock_buffer(new_bh);
1022                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1023                         error = ext4_handle_dirty_metadata(handle, inode,
1024                                                            new_bh);
1025                         if (error)
1026                                 goto cleanup;
1027                 }
1028         }
1029
1030         /* Update the inode. */
1031         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1032
1033         /* Drop the previous xattr block. */
1034         if (bs->bh && bs->bh != new_bh)
1035                 ext4_xattr_release_block(handle, inode, bs->bh);
1036         error = 0;
1037
1038 cleanup:
1039         if (ce)
1040                 mb_cache_entry_put(ext4_mb_cache, ce);
1041         brelse(new_bh);
1042         if (!(bs->bh && s->base == bs->bh->b_data))
1043                 kfree(s->base);
1044
1045         return error;
1046
1047 cleanup_dquot:
1048         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1049         goto cleanup;
1050
1051 bad_block:
1052         EXT4_ERROR_INODE(inode, "bad block %llu",
1053                          EXT4_I(inode)->i_file_acl);
1054         goto cleanup;
1055
1056 #undef header
1057 }
1058
1059 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1060                           struct ext4_xattr_ibody_find *is)
1061 {
1062         struct ext4_xattr_ibody_header *header;
1063         struct ext4_inode *raw_inode;
1064         int error;
1065
1066         if (EXT4_I(inode)->i_extra_isize == 0)
1067                 return 0;
1068         raw_inode = ext4_raw_inode(&is->iloc);
1069         header = IHDR(inode, raw_inode);
1070         is->s.base = is->s.first = IFIRST(header);
1071         is->s.here = is->s.first;
1072         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1073         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1074                 error = xattr_check_inode(inode, header, is->s.end);
1075                 if (error)
1076                         return error;
1077                 /* Find the named attribute. */
1078                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1079                                               i->name, is->s.end -
1080                                               (void *)is->s.base, 0);
1081                 if (error && error != -ENODATA)
1082                         return error;
1083                 is->s.not_found = error;
1084         }
1085         return 0;
1086 }
1087
1088 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1089                                 struct ext4_xattr_info *i,
1090                                 struct ext4_xattr_ibody_find *is)
1091 {
1092         struct ext4_xattr_ibody_header *header;
1093         struct ext4_xattr_search *s = &is->s;
1094         int error;
1095
1096         if (EXT4_I(inode)->i_extra_isize == 0)
1097                 return -ENOSPC;
1098         error = ext4_xattr_set_entry(i, s);
1099         if (error) {
1100                 if (error == -ENOSPC &&
1101                     ext4_has_inline_data(inode)) {
1102                         error = ext4_try_to_evict_inline_data(handle, inode,
1103                                         EXT4_XATTR_LEN(strlen(i->name) +
1104                                         EXT4_XATTR_SIZE(i->value_len)));
1105                         if (error)
1106                                 return error;
1107                         error = ext4_xattr_ibody_find(inode, i, is);
1108                         if (error)
1109                                 return error;
1110                         error = ext4_xattr_set_entry(i, s);
1111                 }
1112                 if (error)
1113                         return error;
1114         }
1115         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1116         if (!IS_LAST_ENTRY(s->first)) {
1117                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1118                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1119         } else {
1120                 header->h_magic = cpu_to_le32(0);
1121                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1122         }
1123         return 0;
1124 }
1125
1126 static int ext4_xattr_ibody_set(struct inode *inode,
1127                                 struct ext4_xattr_info *i,
1128                                 struct ext4_xattr_ibody_find *is)
1129 {
1130         struct ext4_xattr_ibody_header *header;
1131         struct ext4_xattr_search *s = &is->s;
1132         int error;
1133
1134         if (EXT4_I(inode)->i_extra_isize == 0)
1135                 return -ENOSPC;
1136         error = ext4_xattr_set_entry(i, s);
1137         if (error)
1138                 return error;
1139         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1140         if (!IS_LAST_ENTRY(s->first)) {
1141                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1142                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1143         } else {
1144                 header->h_magic = cpu_to_le32(0);
1145                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1146         }
1147         return 0;
1148 }
1149
1150 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1151                                  struct ext4_xattr_info *i)
1152 {
1153         void *value;
1154
1155         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1156                 return 0;
1157         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1158         return !memcmp(value, i->value, i->value_len);
1159 }
1160
1161 /*
1162  * ext4_xattr_set_handle()
1163  *
1164  * Create, replace or remove an extended attribute for this inode.  Value
1165  * is NULL to remove an existing extended attribute, and non-NULL to
1166  * either replace an existing extended attribute, or create a new extended
1167  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1168  * specify that an extended attribute must exist and must not exist
1169  * previous to the call, respectively.
1170  *
1171  * Returns 0, or a negative error number on failure.
1172  */
1173 int
1174 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1175                       const char *name, const void *value, size_t value_len,
1176                       int flags)
1177 {
1178         struct ext4_xattr_info i = {
1179                 .name_index = name_index,
1180                 .name = name,
1181                 .value = value,
1182                 .value_len = value_len,
1183
1184         };
1185         struct ext4_xattr_ibody_find is = {
1186                 .s = { .not_found = -ENODATA, },
1187         };
1188         struct ext4_xattr_block_find bs = {
1189                 .s = { .not_found = -ENODATA, },
1190         };
1191         int no_expand;
1192         int error;
1193
1194         if (!name)
1195                 return -EINVAL;
1196         if (strlen(name) > 255)
1197                 return -ERANGE;
1198         ext4_write_lock_xattr(inode, &no_expand);
1199
1200         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1201         if (error)
1202                 goto cleanup;
1203
1204         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1205                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1206                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1207                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1208         }
1209
1210         error = ext4_xattr_ibody_find(inode, &i, &is);
1211         if (error)
1212                 goto cleanup;
1213         if (is.s.not_found)
1214                 error = ext4_xattr_block_find(inode, &i, &bs);
1215         if (error)
1216                 goto cleanup;
1217         if (is.s.not_found && bs.s.not_found) {
1218                 error = -ENODATA;
1219                 if (flags & XATTR_REPLACE)
1220                         goto cleanup;
1221                 error = 0;
1222                 if (!value)
1223                         goto cleanup;
1224         } else {
1225                 error = -EEXIST;
1226                 if (flags & XATTR_CREATE)
1227                         goto cleanup;
1228         }
1229         if (!value) {
1230                 if (!is.s.not_found)
1231                         error = ext4_xattr_ibody_set(inode, &i, &is);
1232                 else if (!bs.s.not_found)
1233                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1234         } else {
1235                 error = 0;
1236                 /* Xattr value did not change? Save us some work and bail out */
1237                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1238                         goto cleanup;
1239                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1240                         goto cleanup;
1241
1242                 error = ext4_xattr_ibody_set(inode, &i, &is);
1243                 if (!error && !bs.s.not_found) {
1244                         i.value = NULL;
1245                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1246                 } else if (error == -ENOSPC) {
1247                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1248                                 error = ext4_xattr_block_find(inode, &i, &bs);
1249                                 if (error)
1250                                         goto cleanup;
1251                         }
1252                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1253                         if (error)
1254                                 goto cleanup;
1255                         if (!is.s.not_found) {
1256                                 i.value = NULL;
1257                                 error = ext4_xattr_ibody_set(inode, &i, &is);
1258                         }
1259                 }
1260         }
1261         if (!error) {
1262                 ext4_xattr_update_super_block(handle, inode->i_sb);
1263                 inode->i_ctime = current_time(inode);
1264                 if (!value)
1265                         no_expand = 0;
1266                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1267                 /*
1268                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1269                  * error != 0.
1270                  */
1271                 is.iloc.bh = NULL;
1272                 if (IS_SYNC(inode))
1273                         ext4_handle_sync(handle);
1274         }
1275
1276 cleanup:
1277         brelse(is.iloc.bh);
1278         brelse(bs.bh);
1279         ext4_write_unlock_xattr(inode, &no_expand);
1280         return error;
1281 }
1282
1283 /*
1284  * ext4_xattr_set()
1285  *
1286  * Like ext4_xattr_set_handle, but start from an inode. This extended
1287  * attribute modification is a filesystem transaction by itself.
1288  *
1289  * Returns 0, or a negative error number on failure.
1290  */
1291 int
1292 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1293                const void *value, size_t value_len, int flags)
1294 {
1295         handle_t *handle;
1296         int error, retries = 0;
1297         int credits = ext4_jbd2_credits_xattr(inode);
1298
1299 retry:
1300         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1301         if (IS_ERR(handle)) {
1302                 error = PTR_ERR(handle);
1303         } else {
1304                 int error2;
1305
1306                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1307                                               value, value_len, flags);
1308                 error2 = ext4_journal_stop(handle);
1309                 if (error == -ENOSPC &&
1310                     ext4_should_retry_alloc(inode->i_sb, &retries))
1311                         goto retry;
1312                 if (error == 0)
1313                         error = error2;
1314         }
1315
1316         return error;
1317 }
1318
1319 /*
1320  * Shift the EA entries in the inode to create space for the increased
1321  * i_extra_isize.
1322  */
1323 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1324                                      int value_offs_shift, void *to,
1325                                      void *from, size_t n)
1326 {
1327         struct ext4_xattr_entry *last = entry;
1328         int new_offs;
1329
1330         /* We always shift xattr headers further thus offsets get lower */
1331         BUG_ON(value_offs_shift > 0);
1332
1333         /* Adjust the value offsets of the entries */
1334         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1335                 if (last->e_value_size) {
1336                         new_offs = le16_to_cpu(last->e_value_offs) +
1337                                                         value_offs_shift;
1338                         last->e_value_offs = cpu_to_le16(new_offs);
1339                 }
1340         }
1341         /* Shift the entries by n bytes */
1342         memmove(to, from, n);
1343 }
1344
1345 /*
1346  * Move xattr pointed to by 'entry' from inode into external xattr block
1347  */
1348 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1349                                     struct ext4_inode *raw_inode,
1350                                     struct ext4_xattr_entry *entry)
1351 {
1352         struct ext4_xattr_ibody_find *is = NULL;
1353         struct ext4_xattr_block_find *bs = NULL;
1354         char *buffer = NULL, *b_entry_name = NULL;
1355         size_t value_offs, value_size;
1356         struct ext4_xattr_info i = {
1357                 .value = NULL,
1358                 .value_len = 0,
1359                 .name_index = entry->e_name_index,
1360         };
1361         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1362         int error;
1363
1364         value_offs = le16_to_cpu(entry->e_value_offs);
1365         value_size = le32_to_cpu(entry->e_value_size);
1366
1367         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1368         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1369         buffer = kmalloc(value_size, GFP_NOFS);
1370         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1371         if (!is || !bs || !buffer || !b_entry_name) {
1372                 error = -ENOMEM;
1373                 goto out;
1374         }
1375
1376         is->s.not_found = -ENODATA;
1377         bs->s.not_found = -ENODATA;
1378         is->iloc.bh = NULL;
1379         bs->bh = NULL;
1380
1381         /* Save the entry name and the entry value */
1382         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1383         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1384         b_entry_name[entry->e_name_len] = '\0';
1385         i.name = b_entry_name;
1386
1387         error = ext4_get_inode_loc(inode, &is->iloc);
1388         if (error)
1389                 goto out;
1390
1391         error = ext4_xattr_ibody_find(inode, &i, is);
1392         if (error)
1393                 goto out;
1394
1395         /* Remove the chosen entry from the inode */
1396         error = ext4_xattr_ibody_set(inode, &i, is);
1397         if (error)
1398                 goto out;
1399
1400         i.name = b_entry_name;
1401         i.value = buffer;
1402         i.value_len = value_size;
1403         error = ext4_xattr_block_find(inode, &i, bs);
1404         if (error)
1405                 goto out;
1406
1407         /* Add entry which was removed from the inode into the block */
1408         error = ext4_xattr_block_set(handle, inode, &i, bs);
1409         if (error)
1410                 goto out;
1411         error = 0;
1412 out:
1413         kfree(b_entry_name);
1414         kfree(buffer);
1415         if (is)
1416                 brelse(is->iloc.bh);
1417         kfree(is);
1418         kfree(bs);
1419
1420         return error;
1421 }
1422
1423 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1424                                        struct ext4_inode *raw_inode,
1425                                        int isize_diff, size_t ifree,
1426                                        size_t bfree, int *total_ino)
1427 {
1428         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1429         struct ext4_xattr_entry *small_entry;
1430         struct ext4_xattr_entry *entry;
1431         struct ext4_xattr_entry *last;
1432         unsigned int entry_size;        /* EA entry size */
1433         unsigned int total_size;        /* EA entry size + value size */
1434         unsigned int min_total_size;
1435         int error;
1436
1437         while (isize_diff > ifree) {
1438                 entry = NULL;
1439                 small_entry = NULL;
1440                 min_total_size = ~0U;
1441                 last = IFIRST(header);
1442                 /* Find the entry best suited to be pushed into EA block */
1443                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1444                         total_size =
1445                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1446                                         EXT4_XATTR_LEN(last->e_name_len);
1447                         if (total_size <= bfree &&
1448                             total_size < min_total_size) {
1449                                 if (total_size + ifree < isize_diff) {
1450                                         small_entry = last;
1451                                 } else {
1452                                         entry = last;
1453                                         min_total_size = total_size;
1454                                 }
1455                         }
1456                 }
1457
1458                 if (entry == NULL) {
1459                         if (small_entry == NULL)
1460                                 return -ENOSPC;
1461                         entry = small_entry;
1462                 }
1463
1464                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1465                 total_size = entry_size +
1466                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1467                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1468                                                  entry);
1469                 if (error)
1470                         return error;
1471
1472                 *total_ino -= entry_size;
1473                 ifree += total_size;
1474                 bfree -= total_size;
1475         }
1476
1477         return 0;
1478 }
1479
1480 /*
1481  * Expand an inode by new_extra_isize bytes when EAs are present.
1482  * Returns 0 on success or negative error number on failure.
1483  */
1484 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1485                                struct ext4_inode *raw_inode, handle_t *handle)
1486 {
1487         struct ext4_xattr_ibody_header *header;
1488         struct buffer_head *bh = NULL;
1489         size_t min_offs;
1490         size_t ifree, bfree;
1491         int total_ino;
1492         void *base, *end;
1493         int error = 0, tried_min_extra_isize = 0;
1494         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1495         int isize_diff; /* How much do we need to grow i_extra_isize */
1496         int no_expand;
1497
1498         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
1499                 return 0;
1500
1501 retry:
1502         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1503         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1504                 goto out;
1505
1506         header = IHDR(inode, raw_inode);
1507
1508         /*
1509          * Check if enough free space is available in the inode to shift the
1510          * entries ahead by new_extra_isize.
1511          */
1512
1513         base = IFIRST(header);
1514         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1515         min_offs = end - base;
1516         total_ino = sizeof(struct ext4_xattr_ibody_header);
1517
1518         error = xattr_check_inode(inode, header, end);
1519         if (error)
1520                 goto cleanup;
1521
1522         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1523         if (ifree >= isize_diff)
1524                 goto shift;
1525
1526         /*
1527          * Enough free space isn't available in the inode, check if
1528          * EA block can hold new_extra_isize bytes.
1529          */
1530         if (EXT4_I(inode)->i_file_acl) {
1531                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1532                 error = -EIO;
1533                 if (!bh)
1534                         goto cleanup;
1535                 if (ext4_xattr_check_block(inode, bh)) {
1536                         EXT4_ERROR_INODE(inode, "bad block %llu",
1537                                          EXT4_I(inode)->i_file_acl);
1538                         error = -EFSCORRUPTED;
1539                         goto cleanup;
1540                 }
1541                 base = BHDR(bh);
1542                 end = bh->b_data + bh->b_size;
1543                 min_offs = end - base;
1544                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1545                                               NULL);
1546                 if (bfree + ifree < isize_diff) {
1547                         if (!tried_min_extra_isize && s_min_extra_isize) {
1548                                 tried_min_extra_isize++;
1549                                 new_extra_isize = s_min_extra_isize;
1550                                 brelse(bh);
1551                                 goto retry;
1552                         }
1553                         error = -ENOSPC;
1554                         goto cleanup;
1555                 }
1556         } else {
1557                 bfree = inode->i_sb->s_blocksize;
1558         }
1559
1560         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1561                                             isize_diff, ifree, bfree,
1562                                             &total_ino);
1563         if (error) {
1564                 if (error == -ENOSPC && !tried_min_extra_isize &&
1565                     s_min_extra_isize) {
1566                         tried_min_extra_isize++;
1567                         new_extra_isize = s_min_extra_isize;
1568                         brelse(bh);
1569                         goto retry;
1570                 }
1571                 goto cleanup;
1572         }
1573 shift:
1574         /* Adjust the offsets and shift the remaining entries ahead */
1575         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1576                         - new_extra_isize, (void *)raw_inode +
1577                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1578                         (void *)header, total_ino);
1579         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1580         brelse(bh);
1581 out:
1582         ext4_write_unlock_xattr(inode, &no_expand);
1583         return 0;
1584
1585 cleanup:
1586         brelse(bh);
1587         /*
1588          * Inode size expansion failed; don't try again
1589          */
1590         no_expand = 1;
1591         ext4_write_unlock_xattr(inode, &no_expand);
1592         return error;
1593 }
1594
1595
1596
1597 /*
1598  * ext4_xattr_delete_inode()
1599  *
1600  * Free extended attribute resources associated with this inode. This
1601  * is called immediately before an inode is freed. We have exclusive
1602  * access to the inode.
1603  */
1604 void
1605 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1606 {
1607         struct buffer_head *bh = NULL;
1608
1609         if (!EXT4_I(inode)->i_file_acl)
1610                 goto cleanup;
1611         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1612         if (!bh) {
1613                 EXT4_ERROR_INODE(inode, "block %llu read error",
1614                                  EXT4_I(inode)->i_file_acl);
1615                 goto cleanup;
1616         }
1617         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1618             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1619                 EXT4_ERROR_INODE(inode, "bad block %llu",
1620                                  EXT4_I(inode)->i_file_acl);
1621                 goto cleanup;
1622         }
1623         ext4_xattr_release_block(handle, inode, bh);
1624         EXT4_I(inode)->i_file_acl = 0;
1625
1626 cleanup:
1627         brelse(bh);
1628 }
1629
1630 /*
1631  * ext4_xattr_cache_insert()
1632  *
1633  * Create a new entry in the extended attribute cache, and insert
1634  * it unless such an entry is already in the cache.
1635  *
1636  * Returns 0, or a negative error number on failure.
1637  */
1638 static void
1639 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1640 {
1641         struct ext4_xattr_header *header = BHDR(bh);
1642         __u32 hash = le32_to_cpu(header->h_hash);
1643         int reusable = le32_to_cpu(header->h_refcount) <
1644                        EXT4_XATTR_REFCOUNT_MAX;
1645         int error;
1646
1647         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1648                                       bh->b_blocknr, reusable);
1649         if (error) {
1650                 if (error == -EBUSY)
1651                         ea_bdebug(bh, "already in cache");
1652         } else
1653                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1654 }
1655
1656 /*
1657  * ext4_xattr_cmp()
1658  *
1659  * Compare two extended attribute blocks for equality.
1660  *
1661  * Returns 0 if the blocks are equal, 1 if they differ, and
1662  * a negative error number on errors.
1663  */
1664 static int
1665 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1666                struct ext4_xattr_header *header2)
1667 {
1668         struct ext4_xattr_entry *entry1, *entry2;
1669
1670         entry1 = ENTRY(header1+1);
1671         entry2 = ENTRY(header2+1);
1672         while (!IS_LAST_ENTRY(entry1)) {
1673                 if (IS_LAST_ENTRY(entry2))
1674                         return 1;
1675                 if (entry1->e_hash != entry2->e_hash ||
1676                     entry1->e_name_index != entry2->e_name_index ||
1677                     entry1->e_name_len != entry2->e_name_len ||
1678                     entry1->e_value_size != entry2->e_value_size ||
1679                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1680                         return 1;
1681                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1682                         return -EFSCORRUPTED;
1683                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1684                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1685                            le32_to_cpu(entry1->e_value_size)))
1686                         return 1;
1687
1688                 entry1 = EXT4_XATTR_NEXT(entry1);
1689                 entry2 = EXT4_XATTR_NEXT(entry2);
1690         }
1691         if (!IS_LAST_ENTRY(entry2))
1692                 return 1;
1693         return 0;
1694 }
1695
1696 /*
1697  * ext4_xattr_cache_find()
1698  *
1699  * Find an identical extended attribute block.
1700  *
1701  * Returns a pointer to the block found, or NULL if such a block was
1702  * not found or an error occurred.
1703  */
1704 static struct buffer_head *
1705 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1706                       struct mb_cache_entry **pce)
1707 {
1708         __u32 hash = le32_to_cpu(header->h_hash);
1709         struct mb_cache_entry *ce;
1710         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1711
1712         if (!header->h_hash)
1713                 return NULL;  /* never share */
1714         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1715         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1716         while (ce) {
1717                 struct buffer_head *bh;
1718
1719                 bh = sb_bread(inode->i_sb, ce->e_block);
1720                 if (!bh) {
1721                         EXT4_ERROR_INODE(inode, "block %lu read error",
1722                                          (unsigned long) ce->e_block);
1723                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1724                         *pce = ce;
1725                         return bh;
1726                 }
1727                 brelse(bh);
1728                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1729         }
1730         return NULL;
1731 }
1732
1733 #define NAME_HASH_SHIFT 5
1734 #define VALUE_HASH_SHIFT 16
1735
1736 /*
1737  * ext4_xattr_hash_entry()
1738  *
1739  * Compute the hash of an extended attribute.
1740  */
1741 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1742                                          struct ext4_xattr_entry *entry)
1743 {
1744         __u32 hash = 0;
1745         char *name = entry->e_name;
1746         int n;
1747
1748         for (n = 0; n < entry->e_name_len; n++) {
1749                 hash = (hash << NAME_HASH_SHIFT) ^
1750                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1751                        *name++;
1752         }
1753
1754         if (entry->e_value_size != 0) {
1755                 __le32 *value = (__le32 *)((char *)header +
1756                         le16_to_cpu(entry->e_value_offs));
1757                 for (n = (le32_to_cpu(entry->e_value_size) +
1758                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1759                         hash = (hash << VALUE_HASH_SHIFT) ^
1760                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1761                                le32_to_cpu(*value++);
1762                 }
1763         }
1764         entry->e_hash = cpu_to_le32(hash);
1765 }
1766
1767 #undef NAME_HASH_SHIFT
1768 #undef VALUE_HASH_SHIFT
1769
1770 #define BLOCK_HASH_SHIFT 16
1771
1772 /*
1773  * ext4_xattr_rehash()
1774  *
1775  * Re-compute the extended attribute hash value after an entry has changed.
1776  */
1777 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1778                               struct ext4_xattr_entry *entry)
1779 {
1780         struct ext4_xattr_entry *here;
1781         __u32 hash = 0;
1782
1783         ext4_xattr_hash_entry(header, entry);
1784         here = ENTRY(header+1);
1785         while (!IS_LAST_ENTRY(here)) {
1786                 if (!here->e_hash) {
1787                         /* Block is not shared if an entry's hash value == 0 */
1788                         hash = 0;
1789                         break;
1790                 }
1791                 hash = (hash << BLOCK_HASH_SHIFT) ^
1792                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1793                        le32_to_cpu(here->e_hash);
1794                 here = EXT4_XATTR_NEXT(here);
1795         }
1796         header->h_hash = cpu_to_le32(hash);
1797 }
1798
1799 #undef BLOCK_HASH_SHIFT
1800
1801 #define HASH_BUCKET_BITS        10
1802
1803 struct mb_cache *
1804 ext4_xattr_create_cache(void)
1805 {
1806         return mb_cache_create(HASH_BUCKET_BITS);
1807 }
1808
1809 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1810 {
1811         if (cache)
1812                 mb_cache_destroy(cache);
1813 }
1814