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                                         sector_t block_nr,
 135                                         struct ext4_xattr_header *hdr)
 136 {
 137         if (ext4_has_metadata_csum(inode->i_sb) &&
 138             (hdr->h_checksum != ext4_xattr_block_csum(inode, block_nr, hdr)))
 139                 return 0;
 140         return 1;
 141 }
 142
 143 static void ext4_xattr_block_csum_set(struct inode *inode,
 144                                       sector_t block_nr,
 145                                       struct ext4_xattr_header *hdr)
 146 {
 147         if (!ext4_has_metadata_csum(inode->i_sb))
 148                 return;
 149
 150         hdr->h_checksum = ext4_xattr_block_csum(inode, block_nr, hdr);
 151 }
 152
 153 static inline int ext4_handle_dirty_xattr_block(handle_t *handle,
 154                                                 struct inode *inode,
 155                                                 struct buffer_head *bh)
 156 {
 157         ext4_xattr_block_csum_set(inode, bh->b_blocknr, BHDR(bh));
 158         return ext4_handle_dirty_metadata(handle, inode, bh);
 159 }
 160
 161 static inline const struct xattr_handler *
 162 ext4_xattr_handler(int name_index)
 163 {
 164         const struct xattr_handler *handler = NULL;
 165
 166         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 167                 handler = ext4_xattr_handler_map[name_index];
 168         return handler;
 169 }
 170
 171 /*
 172  * Inode operation listxattr()
 173  *
 174  * d_inode(dentry)->i_mutex: don't care
 175  */
 176 ssize_t
 177 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
 178 {
 179         return ext4_xattr_list(dentry, buffer, size);
 180 }
 181
 182 static int
 183 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
 184                        void *value_start)
 185 {
 186         struct ext4_xattr_entry *e = entry;
 187
 188         while (!IS_LAST_ENTRY(e)) {
 189                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 190                 if ((void *)next >= end)
 191                         return -EFSCORRUPTED;
 192                 e = next;
 193         }
 194
 195         while (!IS_LAST_ENTRY(entry)) {
 196                 if (entry->e_value_block != 0)
 197                         return -EFSCORRUPTED;
 198                 if (entry->e_value_size != 0 &&
 199                     (value_start + le16_to_cpu(entry->e_value_offs) <
 200                      (void *)e + sizeof(__u32) ||
 201                      value_start + le16_to_cpu(entry->e_value_offs) +
 202                     le32_to_cpu(entry->e_value_size) > end))
 203                         return -EFSCORRUPTED;
 204                 entry = EXT4_XATTR_NEXT(entry);
 205         }
 206
 207         return 0;
 208 }
 209
 210 static inline int
 211 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 212 {
 213         int error;
 214
 215         if (buffer_verified(bh))
 216                 return 0;
 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 (!ext4_xattr_block_csum_verify(inode, bh->b_blocknr, BHDR(bh)))
 222                 return -EFSBADCRC;
 223         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
 224                                        bh->b_data);
 225         if (!error)
 226                 set_buffer_verified(bh);
 227         return error;
 228 }
 229
 230 static int
 231 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 232                          void *end, const char *function, unsigned int line)
 233 {
 234         struct ext4_xattr_entry *entry = IFIRST(header);
 235         int error = -EFSCORRUPTED;
 236
 237         if (((void *) header >= end) ||
 238             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 239                 goto errout;
 240         error = ext4_xattr_check_names(entry, end, entry);
 241 errout:
 242         if (error)
 243                 __ext4_error_inode(inode, function, line, 0,
 244                                    "corrupted in-inode xattr");
 245         return error;
 246 }
 247
 248 #define xattr_check_inode(inode, header, end) \
 249         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 250
 251 static inline int
 252 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 253 {
 254         size_t value_size = le32_to_cpu(entry->e_value_size);
 255
 256         if (entry->e_value_block != 0 || value_size > size ||
 257             le16_to_cpu(entry->e_value_offs) + value_size > size)
 258                 return -EFSCORRUPTED;
 259         return 0;
 260 }
 261
 262 static int
 263 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 264                       const char *name, size_t size, int sorted)
 265 {
 266         struct ext4_xattr_entry *entry;
 267         size_t name_len;
 268         int cmp = 1;
 269
 270         if (name == NULL)
 271                 return -EINVAL;
 272         name_len = strlen(name);
 273         entry = *pentry;
 274         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 275                 cmp = name_index - entry->e_name_index;
 276                 if (!cmp)
 277                         cmp = name_len - entry->e_name_len;
 278                 if (!cmp)
 279                         cmp = memcmp(name, entry->e_name, name_len);
 280                 if (cmp <= 0 && (sorted || cmp == 0))
 281                         break;
 282         }
 283         *pentry = entry;
 284         if (!cmp && ext4_xattr_check_entry(entry, size))
 285                 return -EFSCORRUPTED;
 286         return cmp ? -ENODATA : 0;
 287 }
 288
 289 static int
 290 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 291                      void *buffer, size_t buffer_size)
 292 {
 293         struct buffer_head *bh = NULL;
 294         struct ext4_xattr_entry *entry;
 295         size_t size;
 296         int error;
 297         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 298
 299         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 300                   name_index, name, buffer, (long)buffer_size);
 301
 302         error = -ENODATA;
 303         if (!EXT4_I(inode)->i_file_acl)
 304                 goto cleanup;
 305         ea_idebug(inode, "reading block %llu",
 306                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 307         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 308         if (!bh)
 309                 goto cleanup;
 310         ea_bdebug(bh, "b_count=%d, refcount=%d",
 311                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 312         if (ext4_xattr_check_block(inode, bh)) {
 313 bad_block:
 314                 EXT4_ERROR_INODE(inode, "bad block %llu",
 315                                  EXT4_I(inode)->i_file_acl);
 316                 error = -EFSCORRUPTED;
 317                 goto cleanup;
 318         }
 319         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 320         entry = BFIRST(bh);
 321         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 322         if (error == -EFSCORRUPTED)
 323                 goto bad_block;
 324         if (error)
 325                 goto cleanup;
 326         size = le32_to_cpu(entry->e_value_size);
 327         if (buffer) {
 328                 error = -ERANGE;
 329                 if (size > buffer_size)
 330                         goto cleanup;
 331                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 332                        size);
 333         }
 334         error = size;
 335
 336 cleanup:
 337         brelse(bh);
 338         return error;
 339 }
 340
 341 int
 342 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 343                      void *buffer, size_t buffer_size)
 344 {
 345         struct ext4_xattr_ibody_header *header;
 346         struct ext4_xattr_entry *entry;
 347         struct ext4_inode *raw_inode;
 348         struct ext4_iloc iloc;
 349         size_t size;
 350         void *end;
 351         int error;
 352
 353         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 354                 return -ENODATA;
 355         error = ext4_get_inode_loc(inode, &iloc);
 356         if (error)
 357                 return error;
 358         raw_inode = ext4_raw_inode(&iloc);
 359         header = IHDR(inode, raw_inode);
 360         entry = IFIRST(header);
 361         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 362         error = xattr_check_inode(inode, header, end);
 363         if (error)
 364                 goto cleanup;
 365         error = ext4_xattr_find_entry(&entry, name_index, name,
 366                                       end - (void *)entry, 0);
 367         if (error)
 368                 goto cleanup;
 369         size = le32_to_cpu(entry->e_value_size);
 370         if (buffer) {
 371                 error = -ERANGE;
 372                 if (size > buffer_size)
 373                         goto cleanup;
 374                 memcpy(buffer, (void *)IFIRST(header) +
 375                        le16_to_cpu(entry->e_value_offs), size);
 376         }
 377         error = size;
 378
 379 cleanup:
 380         brelse(iloc.bh);
 381         return error;
 382 }
 383
 384 /*
 385  * ext4_xattr_get()
 386  *
 387  * Copy an extended attribute into the buffer
 388  * provided, or compute the buffer size required.
 389  * Buffer is NULL to compute the size of the buffer required.
 390  *
 391  * Returns a negative error number on failure, or the number of bytes
 392  * used / required on success.
 393  */
 394 int
 395 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 396                void *buffer, size_t buffer_size)
 397 {
 398         int error;
 399
 400         if (strlen(name) > 255)
 401                 return -ERANGE;
 402
 403         down_read(&EXT4_I(inode)->xattr_sem);
 404         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 405                                      buffer_size);
 406         if (error == -ENODATA)
 407                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 408                                              buffer_size);
 409         up_read(&EXT4_I(inode)->xattr_sem);
 410         return error;
 411 }
 412
 413 static int
 414 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 415                         char *buffer, size_t buffer_size)
 416 {
 417         size_t rest = buffer_size;
 418
 419         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 420                 const struct xattr_handler *handler =
 421                         ext4_xattr_handler(entry->e_name_index);
 422
 423                 if (handler && (!handler->list || handler->list(dentry))) {
 424                         const char *prefix = handler->prefix ?: handler->name;
 425                         size_t prefix_len = strlen(prefix);
 426                         size_t size = prefix_len + entry->e_name_len + 1;
 427
 428                         if (buffer) {
 429                                 if (size > rest)
 430                                         return -ERANGE;
 431                                 memcpy(buffer, prefix, prefix_len);
 432                                 buffer += prefix_len;
 433                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 434                                 buffer += entry->e_name_len;
 435                                 *buffer++ = 0;
 436                         }
 437                         rest -= size;
 438                 }
 439         }
 440         return buffer_size - rest;  /* total size */
 441 }
 442
 443 static int
 444 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 445 {
 446         struct inode *inode = d_inode(dentry);
 447         struct buffer_head *bh = NULL;
 448         int error;
 449         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 450
 451         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 452                   buffer, (long)buffer_size);
 453
 454         error = 0;
 455         if (!EXT4_I(inode)->i_file_acl)
 456                 goto cleanup;
 457         ea_idebug(inode, "reading block %llu",
 458                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 459         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 460         error = -EIO;
 461         if (!bh)
 462                 goto cleanup;
 463         ea_bdebug(bh, "b_count=%d, refcount=%d",
 464                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 465         if (ext4_xattr_check_block(inode, bh)) {
 466                 EXT4_ERROR_INODE(inode, "bad block %llu",
 467                                  EXT4_I(inode)->i_file_acl);
 468                 error = -EFSCORRUPTED;
 469                 goto cleanup;
 470         }
 471         ext4_xattr_cache_insert(ext4_mb_cache, bh);
 472         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 473
 474 cleanup:
 475         brelse(bh);
 476
 477         return error;
 478 }
 479
 480 static int
 481 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 482 {
 483         struct inode *inode = d_inode(dentry);
 484         struct ext4_xattr_ibody_header *header;
 485         struct ext4_inode *raw_inode;
 486         struct ext4_iloc iloc;
 487         void *end;
 488         int error;
 489
 490         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 491                 return 0;
 492         error = ext4_get_inode_loc(inode, &iloc);
 493         if (error)
 494                 return error;
 495         raw_inode = ext4_raw_inode(&iloc);
 496         header = IHDR(inode, raw_inode);
 497         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 498         error = xattr_check_inode(inode, header, end);
 499         if (error)
 500                 goto cleanup;
 501         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 502                                         buffer, buffer_size);
 503
 504 cleanup:
 505         brelse(iloc.bh);
 506         return error;
 507 }
 508
 509 /*
 510  * ext4_xattr_list()
 511  *
 512  * Copy a list of attribute names into the buffer
 513  * provided, or compute the buffer size required.
 514  * Buffer is NULL to compute the size of the buffer required.
 515  *
 516  * Returns a negative error number on failure, or the number of bytes
 517  * used / required on success.
 518  */
 519 static int
 520 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 521 {
 522         int ret, ret2;
 523
 524         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 525         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 526         if (ret < 0)
 527                 goto errout;
 528         if (buffer) {
 529                 buffer += ret;
 530                 buffer_size -= ret;
 531         }
 532         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 533         if (ret < 0)
 534                 goto errout;
 535         ret += ret2;
 536 errout:
 537         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 538         return ret;
 539 }
 540
 541 /*
 542  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 543  * not set, set it.
 544  */
 545 static void ext4_xattr_update_super_block(handle_t *handle,
 546                                           struct super_block *sb)
 547 {
 548         if (ext4_has_feature_xattr(sb))
 549                 return;
 550
 551         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 552         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 553                 ext4_set_feature_xattr(sb);
 554                 ext4_handle_dirty_super(handle, sb);
 555         }
 556 }
 557
 558 /*
 559  * Release the xattr block BH: If the reference count is > 1, decrement it;
 560  * otherwise free the block.
 561  */
 562 static void
 563 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 564                          struct buffer_head *bh)
 565 {
 566         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 567         u32 hash, ref;
 568         int error = 0;
 569
 570         BUFFER_TRACE(bh, "get_write_access");
 571         error = ext4_journal_get_write_access(handle, bh);
 572         if (error)
 573                 goto out;
 574
 575         lock_buffer(bh);
 576         hash = le32_to_cpu(BHDR(bh)->h_hash);
 577         ref = le32_to_cpu(BHDR(bh)->h_refcount);
 578         if (ref == 1) {
 579                 ea_bdebug(bh, "refcount now=0; freeing");
 580                 /*
 581                  * This must happen under buffer lock for
 582                  * ext4_xattr_block_set() to reliably detect freed block
 583                  */
 584                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
 585                 get_bh(bh);
 586                 unlock_buffer(bh);
 587                 ext4_free_blocks(handle, inode, bh, 0, 1,
 588                                  EXT4_FREE_BLOCKS_METADATA |
 589                                  EXT4_FREE_BLOCKS_FORGET);
 590         } else {
 591                 ref--;
 592                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
 593                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
 594                         struct mb_cache_entry *ce;
 595
 596                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
 597                                                 bh->b_blocknr);
 598                         if (ce) {
 599                                 ce->e_reusable = 1;
 600                                 mb_cache_entry_put(ext4_mb_cache, ce);
 601                         }
 602                 }
 603
 604                 /*
 605                  * Beware of this ugliness: Releasing of xattr block references
 606                  * from different inodes can race and so we have to protect
 607                  * from a race where someone else frees the block (and releases
 608                  * its journal_head) before we are done dirtying the buffer. In
 609                  * nojournal mode this race is harmless and we actually cannot
 610                  * call ext4_handle_dirty_xattr_block() with locked buffer as
 611                  * that function can call sync_dirty_buffer() so for that case
 612                  * we handle the dirtying after unlocking the buffer.
 613                  */
 614                 if (ext4_handle_valid(handle))
 615                         error = ext4_handle_dirty_xattr_block(handle, inode,
 616                                                               bh);
 617                 unlock_buffer(bh);
 618                 if (!ext4_handle_valid(handle))
 619                         error = ext4_handle_dirty_xattr_block(handle, inode,
 620                                                               bh);
 621                 if (IS_SYNC(inode))
 622                         ext4_handle_sync(handle);
 623                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
 624                 ea_bdebug(bh, "refcount now=%d; releasing",
 625                           le32_to_cpu(BHDR(bh)->h_refcount));
 626         }
 627 out:
 628         ext4_std_error(inode->i_sb, error);
 629         return;
 630 }
 631
 632 /*
 633  * Find the available free space for EAs. This also returns the total number of
 634  * bytes used by EA entries.
 635  */
 636 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
 637                                     size_t *min_offs, void *base, int *total)
 638 {
 639         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 640                 if (last->e_value_size) {
 641                         size_t offs = le16_to_cpu(last->e_value_offs);
 642                         if (offs < *min_offs)
 643                                 *min_offs = offs;
 644                 }
 645                 if (total)
 646                         *total += EXT4_XATTR_LEN(last->e_name_len);
 647         }
 648         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 649 }
 650
 651 static int
 652 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 653 {
 654         struct ext4_xattr_entry *last;
 655         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 656
 657         /* Compute min_offs and last. */
 658         last = s->first;
 659         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 660                 if (last->e_value_size) {
 661                         size_t offs = le16_to_cpu(last->e_value_offs);
 662                         if (offs < min_offs)
 663                                 min_offs = offs;
 664                 }
 665         }
 666         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 667         if (!s->not_found) {
 668                 if (s->here->e_value_size) {
 669                         size_t size = le32_to_cpu(s->here->e_value_size);
 670                         free += EXT4_XATTR_SIZE(size);
 671                 }
 672                 free += EXT4_XATTR_LEN(name_len);
 673         }
 674         if (i->value) {
 675                 if (free < EXT4_XATTR_LEN(name_len) +
 676                            EXT4_XATTR_SIZE(i->value_len))
 677                         return -ENOSPC;
 678         }
 679
 680         if (i->value && s->not_found) {
 681                 /* Insert the new name. */
 682                 size_t size = EXT4_XATTR_LEN(name_len);
 683                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
 684                 memmove((void *)s->here + size, s->here, rest);
 685                 memset(s->here, 0, size);
 686                 s->here->e_name_index = i->name_index;
 687                 s->here->e_name_len = name_len;
 688                 memcpy(s->here->e_name, i->name, name_len);
 689         } else {
 690                 if (s->here->e_value_size) {
 691                         void *first_val = s->base + min_offs;
 692                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 693                         void *val = s->base + offs;
 694                         size_t size = EXT4_XATTR_SIZE(
 695                                 le32_to_cpu(s->here->e_value_size));
 696
 697                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
 698                                 /* The old and the new value have the same
 699                                    size. Just replace. */
 700                                 s->here->e_value_size =
 701                                         cpu_to_le32(i->value_len);
 702                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
 703                                         memset(val, 0, size);
 704                                 } else {
 705                                         /* Clear pad bytes first. */
 706                                         memset(val + size - EXT4_XATTR_PAD, 0,
 707                                                EXT4_XATTR_PAD);
 708                                         memcpy(val, i->value, i->value_len);
 709                                 }
 710                                 return 0;
 711                         }
 712
 713                         /* Remove the old value. */
 714                         memmove(first_val + size, first_val, val - first_val);
 715                         memset(first_val, 0, size);
 716                         s->here->e_value_size = 0;
 717                         s->here->e_value_offs = 0;
 718                         min_offs += size;
 719
 720                         /* Adjust all value offsets. */
 721                         last = s->first;
 722                         while (!IS_LAST_ENTRY(last)) {
 723                                 size_t o = le16_to_cpu(last->e_value_offs);
 724                                 if (last->e_value_size && o < offs)
 725                                         last->e_value_offs =
 726                                                 cpu_to_le16(o + size);
 727                                 last = EXT4_XATTR_NEXT(last);
 728                         }
 729                 }
 730                 if (!i->value) {
 731                         /* Remove the old name. */
 732                         size_t size = EXT4_XATTR_LEN(name_len);
 733                         last = ENTRY((void *)last - size);
 734                         memmove(s->here, (void *)s->here + size,
 735                                 (void *)last - (void *)s->here + sizeof(__u32));
 736                         memset(last, 0, size);
 737                 }
 738         }
 739
 740         if (i->value) {
 741                 /* Insert the new value. */
 742                 s->here->e_value_size = cpu_to_le32(i->value_len);
 743                 if (i->value_len) {
 744                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 745                         void *val = s->base + min_offs - size;
 746                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 747                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 748                                 memset(val, 0, size);
 749                         } else {
 750                                 /* Clear the pad bytes first. */
 751                                 memset(val + size - EXT4_XATTR_PAD, 0,
 752                                        EXT4_XATTR_PAD);
 753                                 memcpy(val, i->value, i->value_len);
 754                         }
 755                 }
 756         }
 757         return 0;
 758 }
 759
 760 struct ext4_xattr_block_find {
 761         struct ext4_xattr_search s;
 762         struct buffer_head *bh;
 763 };
 764
 765 static int
 766 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
 767                       struct ext4_xattr_block_find *bs)
 768 {
 769         struct super_block *sb = inode->i_sb;
 770         int error;
 771
 772         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 773                   i->name_index, i->name, i->value, (long)i->value_len);
 774
 775         if (EXT4_I(inode)->i_file_acl) {
 776                 /* The inode already has an extended attribute block. */
 777                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
 778                 error = -EIO;
 779                 if (!bs->bh)
 780                         goto cleanup;
 781                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
 782                         atomic_read(&(bs->bh->b_count)),
 783                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
 784                 if (ext4_xattr_check_block(inode, bs->bh)) {
 785                         EXT4_ERROR_INODE(inode, "bad block %llu",
 786                                          EXT4_I(inode)->i_file_acl);
 787                         error = -EFSCORRUPTED;
 788                         goto cleanup;
 789                 }
 790                 /* Find the named attribute. */
 791                 bs->s.base = BHDR(bs->bh);
 792                 bs->s.first = BFIRST(bs->bh);
 793                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 794                 bs->s.here = bs->s.first;
 795                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 796                                               i->name, bs->bh->b_size, 1);
 797                 if (error && error != -ENODATA)
 798                         goto cleanup;
 799                 bs->s.not_found = error;
 800         }
 801         error = 0;
 802
 803 cleanup:
 804         return error;
 805 }
 806
 807 static int
 808 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 809                      struct ext4_xattr_info *i,
 810                      struct ext4_xattr_block_find *bs)
 811 {
 812         struct super_block *sb = inode->i_sb;
 813         struct buffer_head *new_bh = NULL;
 814         struct ext4_xattr_search *s = &bs->s;
 815         struct mb_cache_entry *ce = NULL;
 816         int error = 0;
 817         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
 818
 819 #define header(x) ((struct ext4_xattr_header *)(x))
 820
 821         if (i->value && i->value_len > sb->s_blocksize)
 822                 return -ENOSPC;
 823         if (s->base) {
 824                 BUFFER_TRACE(bs->bh, "get_write_access");
 825                 error = ext4_journal_get_write_access(handle, bs->bh);
 826                 if (error)
 827                         goto cleanup;
 828                 lock_buffer(bs->bh);
 829
 830                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
 831                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
 832
 833                         /*
 834                          * This must happen under buffer lock for
 835                          * ext4_xattr_block_set() to reliably detect modified
 836                          * block
 837                          */
 838                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
 839                                                     bs->bh->b_blocknr);
 840                         ea_bdebug(bs->bh, "modifying in-place");
 841                         error = ext4_xattr_set_entry(i, s);
 842                         if (!error) {
 843                                 if (!IS_LAST_ENTRY(s->first))
 844                                         ext4_xattr_rehash(header(s->base),
 845                                                           s->here);
 846                                 ext4_xattr_cache_insert(ext4_mb_cache,
 847                                         bs->bh);
 848                         }
 849                         unlock_buffer(bs->bh);
 850                         if (error == -EFSCORRUPTED)
 851                                 goto bad_block;
 852                         if (!error)
 853                                 error = ext4_handle_dirty_xattr_block(handle,
 854                                                                       inode,
 855                                                                       bs->bh);
 856                         if (error)
 857                                 goto cleanup;
 858                         goto inserted;
 859                 } else {
 860                         int offset = (char *)s->here - bs->bh->b_data;
 861
 862                         unlock_buffer(bs->bh);
 863                         ea_bdebug(bs->bh, "cloning");
 864                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
 865                         error = -ENOMEM;
 866                         if (s->base == NULL)
 867                                 goto cleanup;
 868                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
 869                         s->first = ENTRY(header(s->base)+1);
 870                         header(s->base)->h_refcount = cpu_to_le32(1);
 871                         s->here = ENTRY(s->base + offset);
 872                         s->end = s->base + bs->bh->b_size;
 873                 }
 874         } else {
 875                 /* Allocate a buffer where we construct the new block. */
 876                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 877                 /* assert(header == s->base) */
 878                 error = -ENOMEM;
 879                 if (s->base == NULL)
 880                         goto cleanup;
 881                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
 882                 header(s->base)->h_blocks = cpu_to_le32(1);
 883                 header(s->base)->h_refcount = cpu_to_le32(1);
 884                 s->first = ENTRY(header(s->base)+1);
 885                 s->here = ENTRY(header(s->base)+1);
 886                 s->end = s->base + sb->s_blocksize;
 887         }
 888
 889         error = ext4_xattr_set_entry(i, s);
 890         if (error == -EFSCORRUPTED)
 891                 goto bad_block;
 892         if (error)
 893                 goto cleanup;
 894         if (!IS_LAST_ENTRY(s->first))
 895                 ext4_xattr_rehash(header(s->base), s->here);
 896
 897 inserted:
 898         if (!IS_LAST_ENTRY(s->first)) {
 899                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
 900                 if (new_bh) {
 901                         /* We found an identical block in the cache. */
 902                         if (new_bh == bs->bh)
 903                                 ea_bdebug(new_bh, "keeping");
 904                         else {
 905                                 u32 ref;
 906
 907                                 /* The old block is released after updating
 908                                    the inode. */
 909                                 error = dquot_alloc_block(inode,
 910                                                 EXT4_C2B(EXT4_SB(sb), 1));
 911                                 if (error)
 912                                         goto cleanup;
 913                                 BUFFER_TRACE(new_bh, "get_write_access");
 914                                 error = ext4_journal_get_write_access(handle,
 915                                                                       new_bh);
 916                                 if (error)
 917                                         goto cleanup_dquot;
 918                                 lock_buffer(new_bh);
 919                                 /*
 920                                  * We have to be careful about races with
 921                                  * freeing, rehashing or adding references to
 922                                  * xattr block. Once we hold buffer lock xattr
 923                                  * block's state is stable so we can check
 924                                  * whether the block got freed / rehashed or
 925                                  * not.  Since we unhash mbcache entry under
 926                                  * buffer lock when freeing / rehashing xattr
 927                                  * block, checking whether entry is still
 928                                  * hashed is reliable. Same rules hold for
 929                                  * e_reusable handling.
 930                                  */
 931                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
 932                                     !ce->e_reusable) {
 933                                         /*
 934                                          * Undo everything and check mbcache
 935                                          * again.
 936                                          */
 937                                         unlock_buffer(new_bh);
 938                                         dquot_free_block(inode,
 939                                                          EXT4_C2B(EXT4_SB(sb),
 940                                                                   1));
 941                                         brelse(new_bh);
 942                                         mb_cache_entry_put(ext4_mb_cache, ce);
 943                                         ce = NULL;
 944                                         new_bh = NULL;
 945                                         goto inserted;
 946                                 }
 947                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
 948                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
 949                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
 950                                         ce->e_reusable = 0;
 951                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
 952                                           ref);
 953                                 unlock_buffer(new_bh);
 954                                 error = ext4_handle_dirty_xattr_block(handle,
 955                                                                       inode,
 956                                                                       new_bh);
 957                                 if (error)
 958                                         goto cleanup_dquot;
 959                         }
 960                         mb_cache_entry_touch(ext4_mb_cache, ce);
 961                         mb_cache_entry_put(ext4_mb_cache, ce);
 962                         ce = NULL;
 963                 } else if (bs->bh && s->base == bs->bh->b_data) {
 964                         /* We were modifying this block in-place. */
 965                         ea_bdebug(bs->bh, "keeping this block");
 966                         new_bh = bs->bh;
 967                         get_bh(new_bh);
 968                 } else {
 969                         /* We need to allocate a new block */
 970                         ext4_fsblk_t goal, block;
 971
 972                         goal = ext4_group_first_block_no(sb,
 973                                                 EXT4_I(inode)->i_block_group);
 974
 975                         /* non-extent files can't have physical blocks past 2^32 */
 976                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 977                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
 978
 979                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
 980                                                      NULL, &error);
 981                         if (error)
 982                                 goto cleanup;
 983
 984                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 985                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
 986
 987                         ea_idebug(inode, "creating block %llu",
 988                                   (unsigned long long)block);
 989
 990                         new_bh = sb_getblk(sb, block);
 991                         if (unlikely(!new_bh)) {
 992                                 error = -ENOMEM;
 993 getblk_failed:
 994                                 ext4_free_blocks(handle, inode, NULL, block, 1,
 995                                                  EXT4_FREE_BLOCKS_METADATA);
 996                                 goto cleanup;
 997                         }
 998                         lock_buffer(new_bh);
 999                         error = ext4_journal_get_create_access(handle, new_bh);
1000                         if (error) {
1001                                 unlock_buffer(new_bh);
1002                                 error = -EIO;
1003                                 goto getblk_failed;
1004                         }
1005                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1006                         set_buffer_uptodate(new_bh);
1007                         unlock_buffer(new_bh);
1008                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1009                         error = ext4_handle_dirty_xattr_block(handle,
1010                                                               inode, new_bh);
1011                         if (error)
1012                                 goto cleanup;
1013                 }
1014         }
1015
1016         /* Update the inode. */
1017         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1018
1019         /* Drop the previous xattr block. */
1020         if (bs->bh && bs->bh != new_bh)
1021                 ext4_xattr_release_block(handle, inode, bs->bh);
1022         error = 0;
1023
1024 cleanup:
1025         if (ce)
1026                 mb_cache_entry_put(ext4_mb_cache, ce);
1027         brelse(new_bh);
1028         if (!(bs->bh && s->base == bs->bh->b_data))
1029                 kfree(s->base);
1030
1031         return error;
1032
1033 cleanup_dquot:
1034         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1035         goto cleanup;
1036
1037 bad_block:
1038         EXT4_ERROR_INODE(inode, "bad block %llu",
1039                          EXT4_I(inode)->i_file_acl);
1040         goto cleanup;
1041
1042 #undef header
1043 }
1044
1045 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1046                           struct ext4_xattr_ibody_find *is)
1047 {
1048         struct ext4_xattr_ibody_header *header;
1049         struct ext4_inode *raw_inode;
1050         int error;
1051
1052         if (EXT4_I(inode)->i_extra_isize == 0)
1053                 return 0;
1054         raw_inode = ext4_raw_inode(&is->iloc);
1055         header = IHDR(inode, raw_inode);
1056         is->s.base = is->s.first = IFIRST(header);
1057         is->s.here = is->s.first;
1058         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1059         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1060                 error = xattr_check_inode(inode, header, is->s.end);
1061                 if (error)
1062                         return error;
1063                 /* Find the named attribute. */
1064                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1065                                               i->name, is->s.end -
1066                                               (void *)is->s.base, 0);
1067                 if (error && error != -ENODATA)
1068                         return error;
1069                 is->s.not_found = error;
1070         }
1071         return 0;
1072 }
1073
1074 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1075                                 struct ext4_xattr_info *i,
1076                                 struct ext4_xattr_ibody_find *is)
1077 {
1078         struct ext4_xattr_ibody_header *header;
1079         struct ext4_xattr_search *s = &is->s;
1080         int error;
1081
1082         if (EXT4_I(inode)->i_extra_isize == 0)
1083                 return -ENOSPC;
1084         error = ext4_xattr_set_entry(i, s);
1085         if (error) {
1086                 if (error == -ENOSPC &&
1087                     ext4_has_inline_data(inode)) {
1088                         error = ext4_try_to_evict_inline_data(handle, inode,
1089                                         EXT4_XATTR_LEN(strlen(i->name) +
1090                                         EXT4_XATTR_SIZE(i->value_len)));
1091                         if (error)
1092                                 return error;
1093                         error = ext4_xattr_ibody_find(inode, i, is);
1094                         if (error)
1095                                 return error;
1096                         error = ext4_xattr_set_entry(i, s);
1097                 }
1098                 if (error)
1099                         return error;
1100         }
1101         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1102         if (!IS_LAST_ENTRY(s->first)) {
1103                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1104                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1105         } else {
1106                 header->h_magic = cpu_to_le32(0);
1107                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1108         }
1109         return 0;
1110 }
1111
1112 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1113                                 struct ext4_xattr_info *i,
1114                                 struct ext4_xattr_ibody_find *is)
1115 {
1116         struct ext4_xattr_ibody_header *header;
1117         struct ext4_xattr_search *s = &is->s;
1118         int error;
1119
1120         if (EXT4_I(inode)->i_extra_isize == 0)
1121                 return -ENOSPC;
1122         error = ext4_xattr_set_entry(i, s);
1123         if (error)
1124                 return error;
1125         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1126         if (!IS_LAST_ENTRY(s->first)) {
1127                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1128                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1129         } else {
1130                 header->h_magic = cpu_to_le32(0);
1131                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1132         }
1133         return 0;
1134 }
1135
1136 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1137                                  struct ext4_xattr_info *i)
1138 {
1139         void *value;
1140
1141         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1142                 return 0;
1143         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1144         return !memcmp(value, i->value, i->value_len);
1145 }
1146
1147 /*
1148  * ext4_xattr_set_handle()
1149  *
1150  * Create, replace or remove an extended attribute for this inode.  Value
1151  * is NULL to remove an existing extended attribute, and non-NULL to
1152  * either replace an existing extended attribute, or create a new extended
1153  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1154  * specify that an extended attribute must exist and must not exist
1155  * previous to the call, respectively.
1156  *
1157  * Returns 0, or a negative error number on failure.
1158  */
1159 int
1160 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1161                       const char *name, const void *value, size_t value_len,
1162                       int flags)
1163 {
1164         struct ext4_xattr_info i = {
1165                 .name_index = name_index,
1166                 .name = name,
1167                 .value = value,
1168                 .value_len = value_len,
1169
1170         };
1171         struct ext4_xattr_ibody_find is = {
1172                 .s = { .not_found = -ENODATA, },
1173         };
1174         struct ext4_xattr_block_find bs = {
1175                 .s = { .not_found = -ENODATA, },
1176         };
1177         unsigned long no_expand;
1178         int error;
1179
1180         if (!name)
1181                 return -EINVAL;
1182         if (strlen(name) > 255)
1183                 return -ERANGE;
1184         down_write(&EXT4_I(inode)->xattr_sem);
1185         no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1186         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1187
1188         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1189         if (error)
1190                 goto cleanup;
1191
1192         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1193                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1194                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1195                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1196         }
1197
1198         error = ext4_xattr_ibody_find(inode, &i, &is);
1199         if (error)
1200                 goto cleanup;
1201         if (is.s.not_found)
1202                 error = ext4_xattr_block_find(inode, &i, &bs);
1203         if (error)
1204                 goto cleanup;
1205         if (is.s.not_found && bs.s.not_found) {
1206                 error = -ENODATA;
1207                 if (flags & XATTR_REPLACE)
1208                         goto cleanup;
1209                 error = 0;
1210                 if (!value)
1211                         goto cleanup;
1212         } else {
1213                 error = -EEXIST;
1214                 if (flags & XATTR_CREATE)
1215                         goto cleanup;
1216         }
1217         if (!value) {
1218                 if (!is.s.not_found)
1219                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1220                 else if (!bs.s.not_found)
1221                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1222         } else {
1223                 error = 0;
1224                 /* Xattr value did not change? Save us some work and bail out */
1225                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1226                         goto cleanup;
1227                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1228                         goto cleanup;
1229
1230                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1231                 if (!error && !bs.s.not_found) {
1232                         i.value = NULL;
1233                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1234                 } else if (error == -ENOSPC) {
1235                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1236                                 error = ext4_xattr_block_find(inode, &i, &bs);
1237                                 if (error)
1238                                         goto cleanup;
1239                         }
1240                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1241                         if (error)
1242                                 goto cleanup;
1243                         if (!is.s.not_found) {
1244                                 i.value = NULL;
1245                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1246                                                              &is);
1247                         }
1248                 }
1249         }
1250         if (!error) {
1251                 ext4_xattr_update_super_block(handle, inode->i_sb);
1252                 inode->i_ctime = ext4_current_time(inode);
1253                 if (!value)
1254                         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1255                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1256                 /*
1257                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1258                  * error != 0.
1259                  */
1260                 is.iloc.bh = NULL;
1261                 if (IS_SYNC(inode))
1262                         ext4_handle_sync(handle);
1263         }
1264
1265 cleanup:
1266         brelse(is.iloc.bh);
1267         brelse(bs.bh);
1268         if (no_expand == 0)
1269                 ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1270         up_write(&EXT4_I(inode)->xattr_sem);
1271         return error;
1272 }
1273
1274 /*
1275  * ext4_xattr_set()
1276  *
1277  * Like ext4_xattr_set_handle, but start from an inode. This extended
1278  * attribute modification is a filesystem transaction by itself.
1279  *
1280  * Returns 0, or a negative error number on failure.
1281  */
1282 int
1283 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1284                const void *value, size_t value_len, int flags)
1285 {
1286         handle_t *handle;
1287         int error, retries = 0;
1288         int credits = ext4_jbd2_credits_xattr(inode);
1289
1290 retry:
1291         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1292         if (IS_ERR(handle)) {
1293                 error = PTR_ERR(handle);
1294         } else {
1295                 int error2;
1296
1297                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1298                                               value, value_len, flags);
1299                 error2 = ext4_journal_stop(handle);
1300                 if (error == -ENOSPC &&
1301                     ext4_should_retry_alloc(inode->i_sb, &retries))
1302                         goto retry;
1303                 if (error == 0)
1304                         error = error2;
1305         }
1306
1307         return error;
1308 }
1309
1310 /*
1311  * Shift the EA entries in the inode to create space for the increased
1312  * i_extra_isize.
1313  */
1314 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1315                                      int value_offs_shift, void *to,
1316                                      void *from, size_t n)
1317 {
1318         struct ext4_xattr_entry *last = entry;
1319         int new_offs;
1320
1321         /* We always shift xattr headers further thus offsets get lower */
1322         BUG_ON(value_offs_shift > 0);
1323
1324         /* Adjust the value offsets of the entries */
1325         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1326                 if (last->e_value_size) {
1327                         new_offs = le16_to_cpu(last->e_value_offs) +
1328                                                         value_offs_shift;
1329                         last->e_value_offs = cpu_to_le16(new_offs);
1330                 }
1331         }
1332         /* Shift the entries by n bytes */
1333         memmove(to, from, n);
1334 }
1335
1336 /*
1337  * Move xattr pointed to by 'entry' from inode into external xattr block
1338  */
1339 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1340                                     struct ext4_inode *raw_inode,
1341                                     struct ext4_xattr_entry *entry)
1342 {
1343         struct ext4_xattr_ibody_find *is = NULL;
1344         struct ext4_xattr_block_find *bs = NULL;
1345         char *buffer = NULL, *b_entry_name = NULL;
1346         size_t value_offs, value_size;
1347         struct ext4_xattr_info i = {
1348                 .value = NULL,
1349                 .value_len = 0,
1350                 .name_index = entry->e_name_index,
1351         };
1352         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1353         int error;
1354
1355         value_offs = le16_to_cpu(entry->e_value_offs);
1356         value_size = le32_to_cpu(entry->e_value_size);
1357
1358         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1359         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1360         buffer = kmalloc(value_size, GFP_NOFS);
1361         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1362         if (!is || !bs || !buffer || !b_entry_name) {
1363                 error = -ENOMEM;
1364                 goto out;
1365         }
1366
1367         is->s.not_found = -ENODATA;
1368         bs->s.not_found = -ENODATA;
1369         is->iloc.bh = NULL;
1370         bs->bh = NULL;
1371
1372         /* Save the entry name and the entry value */
1373         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1374         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1375         b_entry_name[entry->e_name_len] = '\0';
1376         i.name = b_entry_name;
1377
1378         error = ext4_get_inode_loc(inode, &is->iloc);
1379         if (error)
1380                 goto out;
1381
1382         error = ext4_xattr_ibody_find(inode, &i, is);
1383         if (error)
1384                 goto out;
1385
1386         /* Remove the chosen entry from the inode */
1387         error = ext4_xattr_ibody_set(handle, inode, &i, is);
1388         if (error)
1389                 goto out;
1390
1391         i.name = b_entry_name;
1392         i.value = buffer;
1393         i.value_len = value_size;
1394         error = ext4_xattr_block_find(inode, &i, bs);
1395         if (error)
1396                 goto out;
1397
1398         /* Add entry which was removed from the inode into the block */
1399         error = ext4_xattr_block_set(handle, inode, &i, bs);
1400         if (error)
1401                 goto out;
1402         error = 0;
1403 out:
1404         kfree(b_entry_name);
1405         kfree(buffer);
1406         if (is)
1407                 brelse(is->iloc.bh);
1408         kfree(is);
1409         kfree(bs);
1410
1411         return error;
1412 }
1413
1414 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1415                                        struct ext4_inode *raw_inode,
1416                                        int isize_diff, size_t ifree,
1417                                        size_t bfree, int *total_ino)
1418 {
1419         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1420         struct ext4_xattr_entry *small_entry;
1421         struct ext4_xattr_entry *entry;
1422         struct ext4_xattr_entry *last;
1423         unsigned int entry_size;        /* EA entry size */
1424         unsigned int total_size;        /* EA entry size + value size */
1425         unsigned int min_total_size;
1426         int error;
1427
1428         while (isize_diff > ifree) {
1429                 entry = NULL;
1430                 small_entry = NULL;
1431                 min_total_size = ~0U;
1432                 last = IFIRST(header);
1433                 /* Find the entry best suited to be pushed into EA block */
1434                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1435                         total_size =
1436                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1437                                         EXT4_XATTR_LEN(last->e_name_len);
1438                         if (total_size <= bfree &&
1439                             total_size < min_total_size) {
1440                                 if (total_size + ifree < isize_diff) {
1441                                         small_entry = last;
1442                                 } else {
1443                                         entry = last;
1444                                         min_total_size = total_size;
1445                                 }
1446                         }
1447                 }
1448
1449                 if (entry == NULL) {
1450                         if (small_entry == NULL)
1451                                 return -ENOSPC;
1452                         entry = small_entry;
1453                 }
1454
1455                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1456                 total_size = entry_size +
1457                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1458                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1459                                                  entry);
1460                 if (error)
1461                         return error;
1462
1463                 *total_ino -= entry_size;
1464                 ifree += total_size;
1465                 bfree -= total_size;
1466         }
1467
1468         return 0;
1469 }
1470
1471 /*
1472  * Expand an inode by new_extra_isize bytes when EAs are present.
1473  * Returns 0 on success or negative error number on failure.
1474  */
1475 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1476                                struct ext4_inode *raw_inode, handle_t *handle)
1477 {
1478         struct ext4_xattr_ibody_header *header;
1479         struct buffer_head *bh = NULL;
1480         size_t min_offs;
1481         size_t ifree, bfree;
1482         int total_ino;
1483         void *base, *end;
1484         int error = 0, tried_min_extra_isize = 0;
1485         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1486         int isize_diff; /* How much do we need to grow i_extra_isize */
1487
1488         down_write(&EXT4_I(inode)->xattr_sem);
1489         /*
1490          * Set EXT4_STATE_NO_EXPAND to avoid recursion when marking inode dirty
1491          */
1492         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1493 retry:
1494         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1495         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1496                 goto out;
1497
1498         header = IHDR(inode, raw_inode);
1499
1500         /*
1501          * Check if enough free space is available in the inode to shift the
1502          * entries ahead by new_extra_isize.
1503          */
1504
1505         base = IFIRST(header);
1506         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1507         min_offs = end - base;
1508         total_ino = sizeof(struct ext4_xattr_ibody_header);
1509
1510         error = xattr_check_inode(inode, header, end);
1511         if (error)
1512                 goto cleanup;
1513
1514         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1515         if (ifree >= isize_diff)
1516                 goto shift;
1517
1518         /*
1519          * Enough free space isn't available in the inode, check if
1520          * EA block can hold new_extra_isize bytes.
1521          */
1522         if (EXT4_I(inode)->i_file_acl) {
1523                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1524                 error = -EIO;
1525                 if (!bh)
1526                         goto cleanup;
1527                 if (ext4_xattr_check_block(inode, bh)) {
1528                         EXT4_ERROR_INODE(inode, "bad block %llu",
1529                                          EXT4_I(inode)->i_file_acl);
1530                         error = -EFSCORRUPTED;
1531                         goto cleanup;
1532                 }
1533                 base = BHDR(bh);
1534                 end = bh->b_data + bh->b_size;
1535                 min_offs = end - base;
1536                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1537                                               NULL);
1538                 if (bfree + ifree < isize_diff) {
1539                         if (!tried_min_extra_isize && s_min_extra_isize) {
1540                                 tried_min_extra_isize++;
1541                                 new_extra_isize = s_min_extra_isize;
1542                                 brelse(bh);
1543                                 goto retry;
1544                         }
1545                         error = -ENOSPC;
1546                         goto cleanup;
1547                 }
1548         } else {
1549                 bfree = inode->i_sb->s_blocksize;
1550         }
1551
1552         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1553                                             isize_diff, ifree, bfree,
1554                                             &total_ino);
1555         if (error) {
1556                 if (error == -ENOSPC && !tried_min_extra_isize &&
1557                     s_min_extra_isize) {
1558                         tried_min_extra_isize++;
1559                         new_extra_isize = s_min_extra_isize;
1560                         brelse(bh);
1561                         goto retry;
1562                 }
1563                 goto cleanup;
1564         }
1565 shift:
1566         /* Adjust the offsets and shift the remaining entries ahead */
1567         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1568                         - new_extra_isize, (void *)raw_inode +
1569                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1570                         (void *)header, total_ino);
1571         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1572         brelse(bh);
1573 out:
1574         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1575         up_write(&EXT4_I(inode)->xattr_sem);
1576         return 0;
1577
1578 cleanup:
1579         brelse(bh);
1580         /*
1581          * We deliberately leave EXT4_STATE_NO_EXPAND set here since inode
1582          * size expansion failed.
1583          */
1584         up_write(&EXT4_I(inode)->xattr_sem);
1585         return error;
1586 }
1587
1588
1589
1590 /*
1591  * ext4_xattr_delete_inode()
1592  *
1593  * Free extended attribute resources associated with this inode. This
1594  * is called immediately before an inode is freed. We have exclusive
1595  * access to the inode.
1596  */
1597 void
1598 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1599 {
1600         struct buffer_head *bh = NULL;
1601
1602         if (!EXT4_I(inode)->i_file_acl)
1603                 goto cleanup;
1604         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1605         if (!bh) {
1606                 EXT4_ERROR_INODE(inode, "block %llu read error",
1607                                  EXT4_I(inode)->i_file_acl);
1608                 goto cleanup;
1609         }
1610         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1611             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1612                 EXT4_ERROR_INODE(inode, "bad block %llu",
1613                                  EXT4_I(inode)->i_file_acl);
1614                 goto cleanup;
1615         }
1616         ext4_xattr_release_block(handle, inode, bh);
1617         EXT4_I(inode)->i_file_acl = 0;
1618
1619 cleanup:
1620         brelse(bh);
1621 }
1622
1623 /*
1624  * ext4_xattr_cache_insert()
1625  *
1626  * Create a new entry in the extended attribute cache, and insert
1627  * it unless such an entry is already in the cache.
1628  *
1629  * Returns 0, or a negative error number on failure.
1630  */
1631 static void
1632 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1633 {
1634         struct ext4_xattr_header *header = BHDR(bh);
1635         __u32 hash = le32_to_cpu(header->h_hash);
1636         int reusable = le32_to_cpu(header->h_refcount) <
1637                        EXT4_XATTR_REFCOUNT_MAX;
1638         int error;
1639
1640         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1641                                       bh->b_blocknr, reusable);
1642         if (error) {
1643                 if (error == -EBUSY)
1644                         ea_bdebug(bh, "already in cache");
1645         } else
1646                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1647 }
1648
1649 /*
1650  * ext4_xattr_cmp()
1651  *
1652  * Compare two extended attribute blocks for equality.
1653  *
1654  * Returns 0 if the blocks are equal, 1 if they differ, and
1655  * a negative error number on errors.
1656  */
1657 static int
1658 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1659                struct ext4_xattr_header *header2)
1660 {
1661         struct ext4_xattr_entry *entry1, *entry2;
1662
1663         entry1 = ENTRY(header1+1);
1664         entry2 = ENTRY(header2+1);
1665         while (!IS_LAST_ENTRY(entry1)) {
1666                 if (IS_LAST_ENTRY(entry2))
1667                         return 1;
1668                 if (entry1->e_hash != entry2->e_hash ||
1669                     entry1->e_name_index != entry2->e_name_index ||
1670                     entry1->e_name_len != entry2->e_name_len ||
1671                     entry1->e_value_size != entry2->e_value_size ||
1672                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1673                         return 1;
1674                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1675                         return -EFSCORRUPTED;
1676                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1677                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1678                            le32_to_cpu(entry1->e_value_size)))
1679                         return 1;
1680
1681                 entry1 = EXT4_XATTR_NEXT(entry1);
1682                 entry2 = EXT4_XATTR_NEXT(entry2);
1683         }
1684         if (!IS_LAST_ENTRY(entry2))
1685                 return 1;
1686         return 0;
1687 }
1688
1689 /*
1690  * ext4_xattr_cache_find()
1691  *
1692  * Find an identical extended attribute block.
1693  *
1694  * Returns a pointer to the block found, or NULL if such a block was
1695  * not found or an error occurred.
1696  */
1697 static struct buffer_head *
1698 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1699                       struct mb_cache_entry **pce)
1700 {
1701         __u32 hash = le32_to_cpu(header->h_hash);
1702         struct mb_cache_entry *ce;
1703         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1704
1705         if (!header->h_hash)
1706                 return NULL;  /* never share */
1707         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1708         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1709         while (ce) {
1710                 struct buffer_head *bh;
1711
1712                 bh = sb_bread(inode->i_sb, ce->e_block);
1713                 if (!bh) {
1714                         EXT4_ERROR_INODE(inode, "block %lu read error",
1715                                          (unsigned long) ce->e_block);
1716                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1717                         *pce = ce;
1718                         return bh;
1719                 }
1720                 brelse(bh);
1721                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1722         }
1723         return NULL;
1724 }
1725
1726 #define NAME_HASH_SHIFT 5
1727 #define VALUE_HASH_SHIFT 16
1728
1729 /*
1730  * ext4_xattr_hash_entry()
1731  *
1732  * Compute the hash of an extended attribute.
1733  */
1734 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1735                                          struct ext4_xattr_entry *entry)
1736 {
1737         __u32 hash = 0;
1738         char *name = entry->e_name;
1739         int n;
1740
1741         for (n = 0; n < entry->e_name_len; n++) {
1742                 hash = (hash << NAME_HASH_SHIFT) ^
1743                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1744                        *name++;
1745         }
1746
1747         if (entry->e_value_size != 0) {
1748                 __le32 *value = (__le32 *)((char *)header +
1749                         le16_to_cpu(entry->e_value_offs));
1750                 for (n = (le32_to_cpu(entry->e_value_size) +
1751                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1752                         hash = (hash << VALUE_HASH_SHIFT) ^
1753                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1754                                le32_to_cpu(*value++);
1755                 }
1756         }
1757         entry->e_hash = cpu_to_le32(hash);
1758 }
1759
1760 #undef NAME_HASH_SHIFT
1761 #undef VALUE_HASH_SHIFT
1762
1763 #define BLOCK_HASH_SHIFT 16
1764
1765 /*
1766  * ext4_xattr_rehash()
1767  *
1768  * Re-compute the extended attribute hash value after an entry has changed.
1769  */
1770 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1771                               struct ext4_xattr_entry *entry)
1772 {
1773         struct ext4_xattr_entry *here;
1774         __u32 hash = 0;
1775
1776         ext4_xattr_hash_entry(header, entry);
1777         here = ENTRY(header+1);
1778         while (!IS_LAST_ENTRY(here)) {
1779                 if (!here->e_hash) {
1780                         /* Block is not shared if an entry's hash value == 0 */
1781                         hash = 0;
1782                         break;
1783                 }
1784                 hash = (hash << BLOCK_HASH_SHIFT) ^
1785                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1786                        le32_to_cpu(here->e_hash);
1787                 here = EXT4_XATTR_NEXT(here);
1788         }
1789         header->h_hash = cpu_to_le32(hash);
1790 }
1791
1792 #undef BLOCK_HASH_SHIFT
1793
1794 #define HASH_BUCKET_BITS        10
1795
1796 struct mb_cache *
1797 ext4_xattr_create_cache(void)
1798 {
1799         return mb_cache_create(HASH_BUCKET_BITS);
1800 }
1801
1802 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1803 {
1804         if (cache)
1805                 mb_cache_destroy(cache);
1806 }
1807