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_block_cache_insert(struct mb_cache *,
  76                                           struct buffer_head *);
  77 static struct buffer_head *
  78 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  79                             struct mb_cache_entry **);
  80 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  81                                     size_t value_count);
  82 static void ext4_xattr_rehash(struct ext4_xattr_header *);
  83
  84 static const struct xattr_handler * const 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 EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 110                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
 111
 112 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 113                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
 114
 115 static int
 116 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 117                         struct inode *inode);
 118
 119 #ifdef CONFIG_LOCKDEP
 120 void ext4_xattr_inode_set_class(struct inode *ea_inode)
 121 {
 122         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 123 }
 124 #endif
 125
 126 static __le32 ext4_xattr_block_csum(struct inode *inode,
 127                                     sector_t block_nr,
 128                                     struct ext4_xattr_header *hdr)
 129 {
 130         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 131         __u32 csum;
 132         __le64 dsk_block_nr = cpu_to_le64(block_nr);
 133         __u32 dummy_csum = 0;
 134         int offset = offsetof(struct ext4_xattr_header, h_checksum);
 135
 136         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 137                            sizeof(dsk_block_nr));
 138         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 139         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 140         offset += sizeof(dummy_csum);
 141         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 142                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 143
 144         return cpu_to_le32(csum);
 145 }
 146
 147 static int ext4_xattr_block_csum_verify(struct inode *inode,
 148                                         struct buffer_head *bh)
 149 {
 150         struct ext4_xattr_header *hdr = BHDR(bh);
 151         int ret = 1;
 152
 153         if (ext4_has_metadata_csum(inode->i_sb)) {
 154                 lock_buffer(bh);
 155                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 156                                                         bh->b_blocknr, hdr));
 157                 unlock_buffer(bh);
 158         }
 159         return ret;
 160 }
 161
 162 static void ext4_xattr_block_csum_set(struct inode *inode,
 163                                       struct buffer_head *bh)
 164 {
 165         if (ext4_has_metadata_csum(inode->i_sb))
 166                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 167                                                 bh->b_blocknr, BHDR(bh));
 168 }
 169
 170 static inline const struct xattr_handler *
 171 ext4_xattr_handler(int name_index)
 172 {
 173         const struct xattr_handler *handler = NULL;
 174
 175         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 176                 handler = ext4_xattr_handler_map[name_index];
 177         return handler;
 178 }
 179
 180 static int
 181 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 182                          void *value_start)
 183 {
 184         struct ext4_xattr_entry *e = entry;
 185
 186         /* Find the end of the names list */
 187         while (!IS_LAST_ENTRY(e)) {
 188                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 189                 if ((void *)next >= end)
 190                         return -EFSCORRUPTED;
 191                 e = next;
 192         }
 193
 194         /* Check the values */
 195         while (!IS_LAST_ENTRY(entry)) {
 196                 if (entry->e_value_size != 0 &&
 197                     entry->e_value_inum == 0) {
 198                         u16 offs = le16_to_cpu(entry->e_value_offs);
 199                         u32 size = le32_to_cpu(entry->e_value_size);
 200                         void *value;
 201
 202                         /*
 203                          * The value cannot overlap the names, and the value
 204                          * with padding cannot extend beyond 'end'.  Check both
 205                          * the padded and unpadded sizes, since the size may
 206                          * overflow to 0 when adding padding.
 207                          */
 208                         if (offs > end - value_start)
 209                                 return -EFSCORRUPTED;
 210                         value = value_start + offs;
 211                         if (value < (void *)e + sizeof(u32) ||
 212                             size > end - value ||
 213                             EXT4_XATTR_SIZE(size) > end - value)
 214                                 return -EFSCORRUPTED;
 215                 }
 216                 entry = EXT4_XATTR_NEXT(entry);
 217         }
 218
 219         return 0;
 220 }
 221
 222 static inline int
 223 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 224 {
 225         int error;
 226
 227         if (buffer_verified(bh))
 228                 return 0;
 229
 230         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 231             BHDR(bh)->h_blocks != cpu_to_le32(1))
 232                 return -EFSCORRUPTED;
 233         if (!ext4_xattr_block_csum_verify(inode, bh))
 234                 return -EFSBADCRC;
 235         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 236                                          bh->b_data);
 237         if (!error)
 238                 set_buffer_verified(bh);
 239         return error;
 240 }
 241
 242 static int
 243 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 244                          void *end, const char *function, unsigned int line)
 245 {
 246         int error = -EFSCORRUPTED;
 247
 248         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 249             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 250                 goto errout;
 251         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 252 errout:
 253         if (error)
 254                 __ext4_error_inode(inode, function, line, 0,
 255                                    "corrupted in-inode xattr");
 256         return error;
 257 }
 258
 259 #define xattr_check_inode(inode, header, end) \
 260         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 261
 262 static int
 263 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 264                       const char *name, 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         return cmp ? -ENODATA : 0;
 285 }
 286
 287 static u32
 288 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 289 {
 290         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 291 }
 292
 293 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 294 {
 295         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 296                ((u32)ea_inode->i_version);
 297 }
 298
 299 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 300 {
 301         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 302         ea_inode->i_version = (u32)ref_count;
 303 }
 304
 305 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 306 {
 307         return (u32)ea_inode->i_atime.tv_sec;
 308 }
 309
 310 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 311 {
 312         ea_inode->i_atime.tv_sec = hash;
 313 }
 314
 315 /*
 316  * Read the EA value from an inode.
 317  */
 318 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 319 {
 320         unsigned long block = 0;
 321         struct buffer_head *bh;
 322         int blocksize = ea_inode->i_sb->s_blocksize;
 323         size_t csize, copied = 0;
 324         void *copy_pos = buf;
 325
 326         while (copied < size) {
 327                 csize = (size - copied) > blocksize ? blocksize : size - copied;
 328                 bh = ext4_bread(NULL, ea_inode, block, 0);
 329                 if (IS_ERR(bh))
 330                         return PTR_ERR(bh);
 331                 if (!bh)
 332                         return -EFSCORRUPTED;
 333
 334                 memcpy(copy_pos, bh->b_data, csize);
 335                 brelse(bh);
 336
 337                 copy_pos += csize;
 338                 block += 1;
 339                 copied += csize;
 340         }
 341         return 0;
 342 }
 343
 344 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 345                                  struct inode **ea_inode)
 346 {
 347         struct inode *inode;
 348         int err;
 349
 350         inode = ext4_iget(parent->i_sb, ea_ino);
 351         if (IS_ERR(inode)) {
 352                 err = PTR_ERR(inode);
 353                 ext4_error(parent->i_sb,
 354                            "error while reading EA inode %lu err=%d", ea_ino,
 355                            err);
 356                 return err;
 357         }
 358
 359         if (is_bad_inode(inode)) {
 360                 ext4_error(parent->i_sb,
 361                            "error while reading EA inode %lu is_bad_inode",
 362                            ea_ino);
 363                 err = -EIO;
 364                 goto error;
 365         }
 366
 367         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 368                 ext4_error(parent->i_sb,
 369                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 370                             ea_ino);
 371                 err = -EINVAL;
 372                 goto error;
 373         }
 374
 375         *ea_inode = inode;
 376         return 0;
 377 error:
 378         iput(inode);
 379         return err;
 380 }
 381
 382 static int
 383 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 384                                struct ext4_xattr_entry *entry, void *buffer,
 385                                size_t size)
 386 {
 387         u32 hash;
 388
 389         /* Verify stored hash matches calculated hash. */
 390         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 391         if (hash != ext4_xattr_inode_get_hash(ea_inode))
 392                 return -EFSCORRUPTED;
 393
 394         if (entry) {
 395                 __le32 e_hash, tmp_data;
 396
 397                 /* Verify entry hash. */
 398                 tmp_data = cpu_to_le32(hash);
 399                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 400                                                &tmp_data, 1);
 401                 if (e_hash != entry->e_hash)
 402                         return -EFSCORRUPTED;
 403         }
 404         return 0;
 405 }
 406
 407 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 408
 409 /*
 410  * Read xattr value from the EA inode.
 411  */
 412 static int
 413 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 414                      void *buffer, size_t size)
 415 {
 416         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 417         struct inode *ea_inode;
 418         int err;
 419
 420         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 421                                     &ea_inode);
 422         if (err) {
 423                 ea_inode = NULL;
 424                 goto out;
 425         }
 426
 427         if (i_size_read(ea_inode) != size) {
 428                 ext4_warning_inode(ea_inode,
 429                                    "ea_inode file size=%llu entry size=%zu",
 430                                    i_size_read(ea_inode), size);
 431                 err = -EFSCORRUPTED;
 432                 goto out;
 433         }
 434
 435         err = ext4_xattr_inode_read(ea_inode, buffer, size);
 436         if (err)
 437                 goto out;
 438
 439         err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer, size);
 440         /*
 441          * Compatibility check for old Lustre ea_inode implementation. Old
 442          * version does not have hash validation, but it has a backpointer
 443          * from ea_inode to the parent inode.
 444          */
 445         if (err == -EFSCORRUPTED) {
 446                 if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != inode->i_ino ||
 447                     ea_inode->i_generation != inode->i_generation) {
 448                         ext4_warning_inode(ea_inode,
 449                                            "EA inode hash validation failed");
 450                         goto out;
 451                 }
 452                 /* Do not add ea_inode to the cache. */
 453                 ea_inode_cache = NULL;
 454                 err = 0;
 455         } else if (err)
 456                 goto out;
 457
 458         if (ea_inode_cache)
 459                 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 460                                       ext4_xattr_inode_get_hash(ea_inode),
 461                                       ea_inode->i_ino, true /* reusable */);
 462 out:
 463         iput(ea_inode);
 464         return err;
 465 }
 466
 467 static int
 468 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 469                      void *buffer, size_t buffer_size)
 470 {
 471         struct buffer_head *bh = NULL;
 472         struct ext4_xattr_entry *entry;
 473         size_t size;
 474         int error;
 475         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 476
 477         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 478                   name_index, name, buffer, (long)buffer_size);
 479
 480         error = -ENODATA;
 481         if (!EXT4_I(inode)->i_file_acl)
 482                 goto cleanup;
 483         ea_idebug(inode, "reading block %llu",
 484                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 485         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 486         if (!bh)
 487                 goto cleanup;
 488         ea_bdebug(bh, "b_count=%d, refcount=%d",
 489                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 490         if (ext4_xattr_check_block(inode, bh)) {
 491                 EXT4_ERROR_INODE(inode, "bad block %llu",
 492                                  EXT4_I(inode)->i_file_acl);
 493                 error = -EFSCORRUPTED;
 494                 goto cleanup;
 495         }
 496         ext4_xattr_block_cache_insert(ea_block_cache, bh);
 497         entry = BFIRST(bh);
 498         error = ext4_xattr_find_entry(&entry, name_index, name, 1);
 499         if (error)
 500                 goto cleanup;
 501         size = le32_to_cpu(entry->e_value_size);
 502         if (buffer) {
 503                 error = -ERANGE;
 504                 if (size > buffer_size)
 505                         goto cleanup;
 506                 if (entry->e_value_inum) {
 507                         error = ext4_xattr_inode_get(inode, entry, buffer,
 508                                                      size);
 509                         if (error)
 510                                 goto cleanup;
 511                 } else {
 512                         memcpy(buffer, bh->b_data +
 513                                le16_to_cpu(entry->e_value_offs), size);
 514                 }
 515         }
 516         error = size;
 517
 518 cleanup:
 519         brelse(bh);
 520         return error;
 521 }
 522
 523 int
 524 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 525                      void *buffer, size_t buffer_size)
 526 {
 527         struct ext4_xattr_ibody_header *header;
 528         struct ext4_xattr_entry *entry;
 529         struct ext4_inode *raw_inode;
 530         struct ext4_iloc iloc;
 531         size_t size;
 532         void *end;
 533         int error;
 534
 535         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 536                 return -ENODATA;
 537         error = ext4_get_inode_loc(inode, &iloc);
 538         if (error)
 539                 return error;
 540         raw_inode = ext4_raw_inode(&iloc);
 541         header = IHDR(inode, raw_inode);
 542         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 543         error = xattr_check_inode(inode, header, end);
 544         if (error)
 545                 goto cleanup;
 546         entry = IFIRST(header);
 547         error = ext4_xattr_find_entry(&entry, name_index, name, 0);
 548         if (error)
 549                 goto cleanup;
 550         size = le32_to_cpu(entry->e_value_size);
 551         if (buffer) {
 552                 error = -ERANGE;
 553                 if (size > buffer_size)
 554                         goto cleanup;
 555                 if (entry->e_value_inum) {
 556                         error = ext4_xattr_inode_get(inode, entry, buffer,
 557                                                      size);
 558                         if (error)
 559                                 goto cleanup;
 560                 } else {
 561                         memcpy(buffer, (void *)IFIRST(header) +
 562                                le16_to_cpu(entry->e_value_offs), size);
 563                 }
 564         }
 565         error = size;
 566
 567 cleanup:
 568         brelse(iloc.bh);
 569         return error;
 570 }
 571
 572 /*
 573  * ext4_xattr_get()
 574  *
 575  * Copy an extended attribute into the buffer
 576  * provided, or compute the buffer size required.
 577  * Buffer is NULL to compute the size of the buffer required.
 578  *
 579  * Returns a negative error number on failure, or the number of bytes
 580  * used / required on success.
 581  */
 582 int
 583 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 584                void *buffer, size_t buffer_size)
 585 {
 586         int error;
 587
 588         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 589                 return -EIO;
 590
 591         if (strlen(name) > 255)
 592                 return -ERANGE;
 593
 594         down_read(&EXT4_I(inode)->xattr_sem);
 595         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 596                                      buffer_size);
 597         if (error == -ENODATA)
 598                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
 599                                              buffer_size);
 600         up_read(&EXT4_I(inode)->xattr_sem);
 601         return error;
 602 }
 603
 604 static int
 605 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 606                         char *buffer, size_t buffer_size)
 607 {
 608         size_t rest = buffer_size;
 609
 610         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 611                 const struct xattr_handler *handler =
 612                         ext4_xattr_handler(entry->e_name_index);
 613
 614                 if (handler && (!handler->list || handler->list(dentry))) {
 615                         const char *prefix = handler->prefix ?: handler->name;
 616                         size_t prefix_len = strlen(prefix);
 617                         size_t size = prefix_len + entry->e_name_len + 1;
 618
 619                         if (buffer) {
 620                                 if (size > rest)
 621                                         return -ERANGE;
 622                                 memcpy(buffer, prefix, prefix_len);
 623                                 buffer += prefix_len;
 624                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 625                                 buffer += entry->e_name_len;
 626                                 *buffer++ = 0;
 627                         }
 628                         rest -= size;
 629                 }
 630         }
 631         return buffer_size - rest;  /* total size */
 632 }
 633
 634 static int
 635 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 636 {
 637         struct inode *inode = d_inode(dentry);
 638         struct buffer_head *bh = NULL;
 639         int error;
 640
 641         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 642                   buffer, (long)buffer_size);
 643
 644         error = 0;
 645         if (!EXT4_I(inode)->i_file_acl)
 646                 goto cleanup;
 647         ea_idebug(inode, "reading block %llu",
 648                   (unsigned long long)EXT4_I(inode)->i_file_acl);
 649         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 650         error = -EIO;
 651         if (!bh)
 652                 goto cleanup;
 653         ea_bdebug(bh, "b_count=%d, refcount=%d",
 654                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 655         if (ext4_xattr_check_block(inode, bh)) {
 656                 EXT4_ERROR_INODE(inode, "bad block %llu",
 657                                  EXT4_I(inode)->i_file_acl);
 658                 error = -EFSCORRUPTED;
 659                 goto cleanup;
 660         }
 661         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 662         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 663
 664 cleanup:
 665         brelse(bh);
 666
 667         return error;
 668 }
 669
 670 static int
 671 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 672 {
 673         struct inode *inode = d_inode(dentry);
 674         struct ext4_xattr_ibody_header *header;
 675         struct ext4_inode *raw_inode;
 676         struct ext4_iloc iloc;
 677         void *end;
 678         int error;
 679
 680         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 681                 return 0;
 682         error = ext4_get_inode_loc(inode, &iloc);
 683         if (error)
 684                 return error;
 685         raw_inode = ext4_raw_inode(&iloc);
 686         header = IHDR(inode, raw_inode);
 687         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 688         error = xattr_check_inode(inode, header, end);
 689         if (error)
 690                 goto cleanup;
 691         error = ext4_xattr_list_entries(dentry, IFIRST(header),
 692                                         buffer, buffer_size);
 693
 694 cleanup:
 695         brelse(iloc.bh);
 696         return error;
 697 }
 698
 699 /*
 700  * Inode operation listxattr()
 701  *
 702  * d_inode(dentry)->i_rwsem: don't care
 703  *
 704  * Copy a list of attribute names into the buffer
 705  * provided, or compute the buffer size required.
 706  * Buffer is NULL to compute the size of the buffer required.
 707  *
 708  * Returns a negative error number on failure, or the number of bytes
 709  * used / required on success.
 710  */
 711 ssize_t
 712 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 713 {
 714         int ret, ret2;
 715
 716         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 717         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 718         if (ret < 0)
 719                 goto errout;
 720         if (buffer) {
 721                 buffer += ret;
 722                 buffer_size -= ret;
 723         }
 724         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 725         if (ret < 0)
 726                 goto errout;
 727         ret += ret2;
 728 errout:
 729         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 730         return ret;
 731 }
 732
 733 /*
 734  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 735  * not set, set it.
 736  */
 737 static void ext4_xattr_update_super_block(handle_t *handle,
 738                                           struct super_block *sb)
 739 {
 740         if (ext4_has_feature_xattr(sb))
 741                 return;
 742
 743         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 744         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 745                 ext4_set_feature_xattr(sb);
 746                 ext4_handle_dirty_super(handle, sb);
 747         }
 748 }
 749
 750 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 751 {
 752         struct ext4_iloc iloc = { .bh = NULL };
 753         struct buffer_head *bh = NULL;
 754         struct ext4_inode *raw_inode;
 755         struct ext4_xattr_ibody_header *header;
 756         struct ext4_xattr_entry *entry;
 757         qsize_t ea_inode_refs = 0;
 758         void *end;
 759         int ret;
 760
 761         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 762
 763         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 764                 ret = ext4_get_inode_loc(inode, &iloc);
 765                 if (ret)
 766                         goto out;
 767                 raw_inode = ext4_raw_inode(&iloc);
 768                 header = IHDR(inode, raw_inode);
 769                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 770                 ret = xattr_check_inode(inode, header, end);
 771                 if (ret)
 772                         goto out;
 773
 774                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 775                      entry = EXT4_XATTR_NEXT(entry))
 776                         if (entry->e_value_inum)
 777                                 ea_inode_refs++;
 778         }
 779
 780         if (EXT4_I(inode)->i_file_acl) {
 781                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 782                 if (!bh) {
 783                         ret = -EIO;
 784                         goto out;
 785                 }
 786
 787                 if (ext4_xattr_check_block(inode, bh)) {
 788                         ret = -EFSCORRUPTED;
 789                         goto out;
 790                 }
 791
 792                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 793                      entry = EXT4_XATTR_NEXT(entry))
 794                         if (entry->e_value_inum)
 795                                 ea_inode_refs++;
 796         }
 797         *usage = ea_inode_refs + 1;
 798         ret = 0;
 799 out:
 800         brelse(iloc.bh);
 801         brelse(bh);
 802         return ret;
 803 }
 804
 805 static inline size_t round_up_cluster(struct inode *inode, size_t length)
 806 {
 807         struct super_block *sb = inode->i_sb;
 808         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 809                                     inode->i_blkbits);
 810         size_t mask = ~(cluster_size - 1);
 811
 812         return (length + cluster_size - 1) & mask;
 813 }
 814
 815 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 816 {
 817         int err;
 818
 819         err = dquot_alloc_inode(inode);
 820         if (err)
 821                 return err;
 822         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 823         if (err)
 824                 dquot_free_inode(inode);
 825         return err;
 826 }
 827
 828 static void ext4_xattr_inode_free_quota(struct inode *inode, size_t len)
 829 {
 830         dquot_free_space_nodirty(inode, round_up_cluster(inode, len));
 831         dquot_free_inode(inode);
 832 }
 833
 834 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 835                              struct buffer_head *block_bh, size_t value_len,
 836                              bool is_create)
 837 {
 838         int credits;
 839         int blocks;
 840
 841         /*
 842          * 1) Owner inode update
 843          * 2) Ref count update on old xattr block
 844          * 3) new xattr block
 845          * 4) block bitmap update for new xattr block
 846          * 5) group descriptor for new xattr block
 847          * 6) block bitmap update for old xattr block
 848          * 7) group descriptor for old block
 849          *
 850          * 6 & 7 can happen if we have two racing threads T_a and T_b
 851          * which are each trying to set an xattr on inodes I_a and I_b
 852          * which were both initially sharing an xattr block.
 853          */
 854         credits = 7;
 855
 856         /* Quota updates. */
 857         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 858
 859         /*
 860          * In case of inline data, we may push out the data to a block,
 861          * so we need to reserve credits for this eventuality
 862          */
 863         if (inode && ext4_has_inline_data(inode))
 864                 credits += ext4_writepage_trans_blocks(inode) + 1;
 865
 866         /* We are done if ea_inode feature is not enabled. */
 867         if (!ext4_has_feature_ea_inode(sb))
 868                 return credits;
 869
 870         /* New ea_inode, inode map, block bitmap, group descriptor. */
 871         credits += 4;
 872
 873         /* Data blocks. */
 874         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 875
 876         /* Indirection block or one level of extent tree. */
 877         blocks += 1;
 878
 879         /* Block bitmap and group descriptor updates for each block. */
 880         credits += blocks * 2;
 881
 882         /* Blocks themselves. */
 883         credits += blocks;
 884
 885         if (!is_create) {
 886                 /* Dereference ea_inode holding old xattr value.
 887                  * Old ea_inode, inode map, block bitmap, group descriptor.
 888                  */
 889                 credits += 4;
 890
 891                 /* Data blocks for old ea_inode. */
 892                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 893
 894                 /* Indirection block or one level of extent tree for old
 895                  * ea_inode.
 896                  */
 897                 blocks += 1;
 898
 899                 /* Block bitmap and group descriptor updates for each block. */
 900                 credits += blocks * 2;
 901         }
 902
 903         /* We may need to clone the existing xattr block in which case we need
 904          * to increment ref counts for existing ea_inodes referenced by it.
 905          */
 906         if (block_bh) {
 907                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
 908
 909                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 910                         if (entry->e_value_inum)
 911                                 /* Ref count update on ea_inode. */
 912                                 credits += 1;
 913         }
 914         return credits;
 915 }
 916
 917 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 918                                      int credits, struct buffer_head *bh,
 919                                      bool dirty, bool block_csum)
 920 {
 921         int error;
 922
 923         if (!ext4_handle_valid(handle))
 924                 return 0;
 925
 926         if (handle->h_buffer_credits >= credits)
 927                 return 0;
 928
 929         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 930         if (!error)
 931                 return 0;
 932         if (error < 0) {
 933                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 934                 return error;
 935         }
 936
 937         if (bh && dirty) {
 938                 if (block_csum)
 939                         ext4_xattr_block_csum_set(inode, bh);
 940                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
 941                 if (error) {
 942                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 943                                      error);
 944                         return error;
 945                 }
 946         }
 947
 948         error = ext4_journal_restart(handle, credits);
 949         if (error) {
 950                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
 951                 return error;
 952         }
 953
 954         if (bh) {
 955                 error = ext4_journal_get_write_access(handle, bh);
 956                 if (error) {
 957                         ext4_warning(inode->i_sb,
 958                                      "Get write access failed (error %d)",
 959                                      error);
 960                         return error;
 961                 }
 962         }
 963         return 0;
 964 }
 965
 966 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
 967                                        int ref_change)
 968 {
 969         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
 970         struct ext4_iloc iloc;
 971         s64 ref_count;
 972         u32 hash;
 973         int ret;
 974
 975         inode_lock(ea_inode);
 976
 977         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
 978         if (ret) {
 979                 iloc.bh = NULL;
 980                 goto out;
 981         }
 982
 983         ref_count = ext4_xattr_inode_get_ref(ea_inode);
 984         ref_count += ref_change;
 985         ext4_xattr_inode_set_ref(ea_inode, ref_count);
 986
 987         if (ref_change > 0) {
 988                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
 989                           ea_inode->i_ino, ref_count);
 990
 991                 if (ref_count == 1) {
 992                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
 993                                   ea_inode->i_ino, ea_inode->i_nlink);
 994
 995                         set_nlink(ea_inode, 1);
 996                         ext4_orphan_del(handle, ea_inode);
 997
 998                         if (ea_inode_cache) {
 999                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1000                                 mb_cache_entry_create(ea_inode_cache,
1001                                                       GFP_NOFS, hash,
1002                                                       ea_inode->i_ino,
1003                                                       true /* reusable */);
1004                         }
1005                 }
1006         } else {
1007                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1008                           ea_inode->i_ino, ref_count);
1009
1010                 if (ref_count == 0) {
1011                         WARN_ONCE(ea_inode->i_nlink != 1,
1012                                   "EA inode %lu i_nlink=%u",
1013                                   ea_inode->i_ino, ea_inode->i_nlink);
1014
1015                         clear_nlink(ea_inode);
1016                         ext4_orphan_add(handle, ea_inode);
1017
1018                         if (ea_inode_cache) {
1019                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1020                                 mb_cache_entry_delete(ea_inode_cache, hash,
1021                                                       ea_inode->i_ino);
1022                         }
1023                 }
1024         }
1025
1026         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1027         iloc.bh = NULL;
1028         if (ret)
1029                 ext4_warning_inode(ea_inode,
1030                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1031 out:
1032         brelse(iloc.bh);
1033         inode_unlock(ea_inode);
1034         return ret;
1035 }
1036
1037 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1038 {
1039         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1040 }
1041
1042 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1043 {
1044         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1045 }
1046
1047 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1048                                         struct ext4_xattr_entry *first)
1049 {
1050         struct inode *ea_inode;
1051         struct ext4_xattr_entry *entry;
1052         struct ext4_xattr_entry *failed_entry;
1053         unsigned int ea_ino;
1054         int err, saved_err;
1055
1056         for (entry = first; !IS_LAST_ENTRY(entry);
1057              entry = EXT4_XATTR_NEXT(entry)) {
1058                 if (!entry->e_value_inum)
1059                         continue;
1060                 ea_ino = le32_to_cpu(entry->e_value_inum);
1061                 err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1062                 if (err)
1063                         goto cleanup;
1064                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1065                 if (err) {
1066                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1067                         iput(ea_inode);
1068                         goto cleanup;
1069                 }
1070                 iput(ea_inode);
1071         }
1072         return 0;
1073
1074 cleanup:
1075         saved_err = err;
1076         failed_entry = entry;
1077
1078         for (entry = first; entry != failed_entry;
1079              entry = EXT4_XATTR_NEXT(entry)) {
1080                 if (!entry->e_value_inum)
1081                         continue;
1082                 ea_ino = le32_to_cpu(entry->e_value_inum);
1083                 err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1084                 if (err) {
1085                         ext4_warning(parent->i_sb,
1086                                      "cleanup ea_ino %u iget error %d", ea_ino,
1087                                      err);
1088                         continue;
1089                 }
1090                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1091                 if (err)
1092                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1093                                            err);
1094                 iput(ea_inode);
1095         }
1096         return saved_err;
1097 }
1098
1099 static void
1100 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1101                              struct buffer_head *bh,
1102                              struct ext4_xattr_entry *first, bool block_csum,
1103                              struct ext4_xattr_inode_array **ea_inode_array,
1104                              int extra_credits, bool skip_quota)
1105 {
1106         struct inode *ea_inode;
1107         struct ext4_xattr_entry *entry;
1108         bool dirty = false;
1109         unsigned int ea_ino;
1110         int err;
1111         int credits;
1112
1113         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1114         credits = 2 + extra_credits;
1115
1116         for (entry = first; !IS_LAST_ENTRY(entry);
1117              entry = EXT4_XATTR_NEXT(entry)) {
1118                 if (!entry->e_value_inum)
1119                         continue;
1120                 ea_ino = le32_to_cpu(entry->e_value_inum);
1121                 err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1122                 if (err)
1123                         continue;
1124
1125                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1126                 if (err) {
1127                         ext4_warning_inode(ea_inode,
1128                                            "Expand inode array err=%d", err);
1129                         iput(ea_inode);
1130                         continue;
1131                 }
1132
1133                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1134                                                 dirty, block_csum);
1135                 if (err) {
1136                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1137                                            err);
1138                         continue;
1139                 }
1140
1141                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1142                 if (err) {
1143                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1144                                            err);
1145                         continue;
1146                 }
1147
1148                 if (!skip_quota)
1149                         ext4_xattr_inode_free_quota(parent,
1150                                               le32_to_cpu(entry->e_value_size));
1151
1152                 /*
1153                  * Forget about ea_inode within the same transaction that
1154                  * decrements the ref count. This avoids duplicate decrements in
1155                  * case the rest of the work spills over to subsequent
1156                  * transactions.
1157                  */
1158                 entry->e_value_inum = 0;
1159                 entry->e_value_size = 0;
1160
1161                 dirty = true;
1162         }
1163
1164         if (dirty) {
1165                 /*
1166                  * Note that we are deliberately skipping csum calculation for
1167                  * the final update because we do not expect any journal
1168                  * restarts until xattr block is freed.
1169                  */
1170
1171                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1172                 if (err)
1173                         ext4_warning_inode(parent,
1174                                            "handle dirty metadata err=%d", err);
1175         }
1176 }
1177
1178 /*
1179  * Release the xattr block BH: If the reference count is > 1, decrement it;
1180  * otherwise free the block.
1181  */
1182 static void
1183 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1184                          struct buffer_head *bh,
1185                          struct ext4_xattr_inode_array **ea_inode_array,
1186                          int extra_credits)
1187 {
1188         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1189         u32 hash, ref;
1190         int error = 0;
1191
1192         BUFFER_TRACE(bh, "get_write_access");
1193         error = ext4_journal_get_write_access(handle, bh);
1194         if (error)
1195                 goto out;
1196
1197         lock_buffer(bh);
1198         hash = le32_to_cpu(BHDR(bh)->h_hash);
1199         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1200         if (ref == 1) {
1201                 ea_bdebug(bh, "refcount now=0; freeing");
1202                 /*
1203                  * This must happen under buffer lock for
1204                  * ext4_xattr_block_set() to reliably detect freed block
1205                  */
1206                 if (ea_block_cache)
1207                         mb_cache_entry_delete(ea_block_cache, hash,
1208                                               bh->b_blocknr);
1209                 get_bh(bh);
1210                 unlock_buffer(bh);
1211
1212                 if (ext4_has_feature_ea_inode(inode->i_sb))
1213                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1214                                                      BFIRST(bh),
1215                                                      true /* block_csum */,
1216                                                      ea_inode_array,
1217                                                      extra_credits,
1218                                                      true /* skip_quota */);
1219                 ext4_free_blocks(handle, inode, bh, 0, 1,
1220                                  EXT4_FREE_BLOCKS_METADATA |
1221                                  EXT4_FREE_BLOCKS_FORGET);
1222         } else {
1223                 ref--;
1224                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1225                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1226                         struct mb_cache_entry *ce;
1227
1228                         if (ea_block_cache) {
1229                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1230                                                         bh->b_blocknr);
1231                                 if (ce) {
1232                                         ce->e_reusable = 1;
1233                                         mb_cache_entry_put(ea_block_cache, ce);
1234                                 }
1235                         }
1236                 }
1237
1238                 ext4_xattr_block_csum_set(inode, bh);
1239                 /*
1240                  * Beware of this ugliness: Releasing of xattr block references
1241                  * from different inodes can race and so we have to protect
1242                  * from a race where someone else frees the block (and releases
1243                  * its journal_head) before we are done dirtying the buffer. In
1244                  * nojournal mode this race is harmless and we actually cannot
1245                  * call ext4_handle_dirty_metadata() with locked buffer as
1246                  * that function can call sync_dirty_buffer() so for that case
1247                  * we handle the dirtying after unlocking the buffer.
1248                  */
1249                 if (ext4_handle_valid(handle))
1250                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1251                 unlock_buffer(bh);
1252                 if (!ext4_handle_valid(handle))
1253                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1254                 if (IS_SYNC(inode))
1255                         ext4_handle_sync(handle);
1256                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1257                 ea_bdebug(bh, "refcount now=%d; releasing",
1258                           le32_to_cpu(BHDR(bh)->h_refcount));
1259         }
1260 out:
1261         ext4_std_error(inode->i_sb, error);
1262         return;
1263 }
1264
1265 /*
1266  * Find the available free space for EAs. This also returns the total number of
1267  * bytes used by EA entries.
1268  */
1269 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1270                                     size_t *min_offs, void *base, int *total)
1271 {
1272         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1273                 if (!last->e_value_inum && last->e_value_size) {
1274                         size_t offs = le16_to_cpu(last->e_value_offs);
1275                         if (offs < *min_offs)
1276                                 *min_offs = offs;
1277                 }
1278                 if (total)
1279                         *total += EXT4_XATTR_LEN(last->e_name_len);
1280         }
1281         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1282 }
1283
1284 /*
1285  * Write the value of the EA in an inode.
1286  */
1287 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1288                                   const void *buf, int bufsize)
1289 {
1290         struct buffer_head *bh = NULL;
1291         unsigned long block = 0;
1292         int blocksize = ea_inode->i_sb->s_blocksize;
1293         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1294         int csize, wsize = 0;
1295         int ret = 0;
1296         int retries = 0;
1297
1298 retry:
1299         while (ret >= 0 && ret < max_blocks) {
1300                 struct ext4_map_blocks map;
1301                 map.m_lblk = block += ret;
1302                 map.m_len = max_blocks -= ret;
1303
1304                 ret = ext4_map_blocks(handle, ea_inode, &map,
1305                                       EXT4_GET_BLOCKS_CREATE);
1306                 if (ret <= 0) {
1307                         ext4_mark_inode_dirty(handle, ea_inode);
1308                         if (ret == -ENOSPC &&
1309                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1310                                 ret = 0;
1311                                 goto retry;
1312                         }
1313                         break;
1314                 }
1315         }
1316
1317         if (ret < 0)
1318                 return ret;
1319
1320         block = 0;
1321         while (wsize < bufsize) {
1322                 if (bh != NULL)
1323                         brelse(bh);
1324                 csize = (bufsize - wsize) > blocksize ? blocksize :
1325                                                                 bufsize - wsize;
1326                 bh = ext4_getblk(handle, ea_inode, block, 0);
1327                 if (IS_ERR(bh))
1328                         return PTR_ERR(bh);
1329                 ret = ext4_journal_get_write_access(handle, bh);
1330                 if (ret)
1331                         goto out;
1332
1333                 memcpy(bh->b_data, buf, csize);
1334                 set_buffer_uptodate(bh);
1335                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1336
1337                 buf += csize;
1338                 wsize += csize;
1339                 block += 1;
1340         }
1341
1342         inode_lock(ea_inode);
1343         i_size_write(ea_inode, wsize);
1344         ext4_update_i_disksize(ea_inode, wsize);
1345         inode_unlock(ea_inode);
1346
1347         ext4_mark_inode_dirty(handle, ea_inode);
1348
1349 out:
1350         brelse(bh);
1351
1352         return ret;
1353 }
1354
1355 /*
1356  * Create an inode to store the value of a large EA.
1357  */
1358 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1359                                              struct inode *inode, u32 hash)
1360 {
1361         struct inode *ea_inode = NULL;
1362         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1363         int err;
1364
1365         /*
1366          * Let the next inode be the goal, so we try and allocate the EA inode
1367          * in the same group, or nearby one.
1368          */
1369         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1370                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1371                                   EXT4_EA_INODE_FL);
1372         if (!IS_ERR(ea_inode)) {
1373                 ea_inode->i_op = &ext4_file_inode_operations;
1374                 ea_inode->i_fop = &ext4_file_operations;
1375                 ext4_set_aops(ea_inode);
1376                 ext4_xattr_inode_set_class(ea_inode);
1377                 unlock_new_inode(ea_inode);
1378                 ext4_xattr_inode_set_ref(ea_inode, 1);
1379                 ext4_xattr_inode_set_hash(ea_inode, hash);
1380                 err = ext4_mark_inode_dirty(handle, ea_inode);
1381                 if (!err)
1382                         err = ext4_inode_attach_jinode(ea_inode);
1383                 if (err) {
1384                         iput(ea_inode);
1385                         return ERR_PTR(err);
1386                 }
1387
1388                 /*
1389                  * Xattr inodes are shared therefore quota charging is performed
1390                  * at a higher level.
1391                  */
1392                 dquot_free_inode(ea_inode);
1393                 dquot_drop(ea_inode);
1394                 inode_lock(ea_inode);
1395                 ea_inode->i_flags |= S_NOQUOTA;
1396                 inode_unlock(ea_inode);
1397         }
1398
1399         return ea_inode;
1400 }
1401
1402 static struct inode *
1403 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1404                             size_t value_len, u32 hash)
1405 {
1406         struct inode *ea_inode;
1407         struct mb_cache_entry *ce;
1408         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1409         void *ea_data;
1410
1411         if (!ea_inode_cache)
1412                 return NULL;
1413
1414         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1415         if (!ce)
1416                 return NULL;
1417
1418         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1419         if (!ea_data) {
1420                 mb_cache_entry_put(ea_inode_cache, ce);
1421                 return NULL;
1422         }
1423
1424         while (ce) {
1425                 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1426                 if (!IS_ERR(ea_inode) &&
1427                     !is_bad_inode(ea_inode) &&
1428                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1429                     i_size_read(ea_inode) == value_len &&
1430                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1431                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1432                                                     value_len) &&
1433                     !memcmp(value, ea_data, value_len)) {
1434                         mb_cache_entry_touch(ea_inode_cache, ce);
1435                         mb_cache_entry_put(ea_inode_cache, ce);
1436                         kvfree(ea_data);
1437                         return ea_inode;
1438                 }
1439
1440                 if (!IS_ERR(ea_inode))
1441                         iput(ea_inode);
1442                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1443         }
1444         kvfree(ea_data);
1445         return NULL;
1446 }
1447
1448 /*
1449  * Add value of the EA in an inode.
1450  */
1451 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1452                                           const void *value, size_t value_len,
1453                                           struct inode **ret_inode)
1454 {
1455         struct inode *ea_inode;
1456         u32 hash;
1457         int err;
1458
1459         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1460         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1461         if (ea_inode) {
1462                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1463                 if (err) {
1464                         iput(ea_inode);
1465                         return err;
1466                 }
1467
1468                 *ret_inode = ea_inode;
1469                 return 0;
1470         }
1471
1472         /* Create an inode for the EA value */
1473         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1474         if (IS_ERR(ea_inode))
1475                 return PTR_ERR(ea_inode);
1476
1477         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1478         if (err) {
1479                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1480                 iput(ea_inode);
1481                 return err;
1482         }
1483
1484         if (EA_INODE_CACHE(inode))
1485                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1486                                       ea_inode->i_ino, true /* reusable */);
1487
1488         *ret_inode = ea_inode;
1489         return 0;
1490 }
1491
1492 /*
1493  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1494  * feature is enabled.
1495  */
1496 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1497
1498 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1499                                 struct ext4_xattr_search *s,
1500                                 handle_t *handle, struct inode *inode,
1501                                 bool is_block)
1502 {
1503         struct ext4_xattr_entry *last;
1504         struct ext4_xattr_entry *here = s->here;
1505         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1506         int in_inode = i->in_inode;
1507         struct inode *old_ea_inode = NULL;
1508         struct inode *new_ea_inode = NULL;
1509         size_t old_size, new_size;
1510         int ret;
1511
1512         /* Space used by old and new values. */
1513         old_size = (!s->not_found && !here->e_value_inum) ?
1514                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1515         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1516
1517         /*
1518          * Optimization for the simple case when old and new values have the
1519          * same padded sizes. Not applicable if external inodes are involved.
1520          */
1521         if (new_size && new_size == old_size) {
1522                 size_t offs = le16_to_cpu(here->e_value_offs);
1523                 void *val = s->base + offs;
1524
1525                 here->e_value_size = cpu_to_le32(i->value_len);
1526                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1527                         memset(val, 0, new_size);
1528                 } else {
1529                         memcpy(val, i->value, i->value_len);
1530                         /* Clear padding bytes. */
1531                         memset(val + i->value_len, 0, new_size - i->value_len);
1532                 }
1533                 return 0;
1534         }
1535
1536         /* Compute min_offs and last. */
1537         last = s->first;
1538         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1539                 if (!last->e_value_inum && last->e_value_size) {
1540                         size_t offs = le16_to_cpu(last->e_value_offs);
1541                         if (offs < min_offs)
1542                                 min_offs = offs;
1543                 }
1544         }
1545
1546         /* Check whether we have enough space. */
1547         if (i->value) {
1548                 size_t free;
1549
1550                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1551                 if (!s->not_found)
1552                         free += EXT4_XATTR_LEN(name_len) + old_size;
1553
1554                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1555                         ret = -ENOSPC;
1556                         goto out;
1557                 }
1558
1559                 /*
1560                  * If storing the value in an external inode is an option,
1561                  * reserve space for xattr entries/names in the external
1562                  * attribute block so that a long value does not occupy the
1563                  * whole space and prevent futher entries being added.
1564                  */
1565                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1566                     new_size && is_block &&
1567                     (min_offs + old_size - new_size) <
1568                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1569                         ret = -ENOSPC;
1570                         goto out;
1571                 }
1572         }
1573
1574         /*
1575          * Getting access to old and new ea inodes is subject to failures.
1576          * Finish that work before doing any modifications to the xattr data.
1577          */
1578         if (!s->not_found && here->e_value_inum) {
1579                 ret = ext4_xattr_inode_iget(inode,
1580                                             le32_to_cpu(here->e_value_inum),
1581                                             &old_ea_inode);
1582                 if (ret) {
1583                         old_ea_inode = NULL;
1584                         goto out;
1585                 }
1586         }
1587         if (i->value && in_inode) {
1588                 WARN_ON_ONCE(!i->value_len);
1589
1590                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1591                 if (ret)
1592                         goto out;
1593
1594                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1595                                                      i->value_len,
1596                                                      &new_ea_inode);
1597                 if (ret) {
1598                         new_ea_inode = NULL;
1599                         ext4_xattr_inode_free_quota(inode, i->value_len);
1600                         goto out;
1601                 }
1602         }
1603
1604         if (old_ea_inode) {
1605                 /* We are ready to release ref count on the old_ea_inode. */
1606                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1607                 if (ret) {
1608                         /* Release newly required ref count on new_ea_inode. */
1609                         if (new_ea_inode) {
1610                                 int err;
1611
1612                                 err = ext4_xattr_inode_dec_ref(handle,
1613                                                                new_ea_inode);
1614                                 if (err)
1615                                         ext4_warning_inode(new_ea_inode,
1616                                                   "dec ref new_ea_inode err=%d",
1617                                                   err);
1618                                 ext4_xattr_inode_free_quota(inode,
1619                                                             i->value_len);
1620                         }
1621                         goto out;
1622                 }
1623
1624                 ext4_xattr_inode_free_quota(inode,
1625                                             le32_to_cpu(here->e_value_size));
1626         }
1627
1628         /* No failures allowed past this point. */
1629
1630         if (!s->not_found && here->e_value_offs) {
1631                 /* Remove the old value. */
1632                 void *first_val = s->base + min_offs;
1633                 size_t offs = le16_to_cpu(here->e_value_offs);
1634                 void *val = s->base + offs;
1635
1636                 memmove(first_val + old_size, first_val, val - first_val);
1637                 memset(first_val, 0, old_size);
1638                 min_offs += old_size;
1639
1640                 /* Adjust all value offsets. */
1641                 last = s->first;
1642                 while (!IS_LAST_ENTRY(last)) {
1643                         size_t o = le16_to_cpu(last->e_value_offs);
1644
1645                         if (!last->e_value_inum &&
1646                             last->e_value_size && o < offs)
1647                                 last->e_value_offs = cpu_to_le16(o + old_size);
1648                         last = EXT4_XATTR_NEXT(last);
1649                 }
1650         }
1651
1652         if (!i->value) {
1653                 /* Remove old name. */
1654                 size_t size = EXT4_XATTR_LEN(name_len);
1655
1656                 last = ENTRY((void *)last - size);
1657                 memmove(here, (void *)here + size,
1658                         (void *)last - (void *)here + sizeof(__u32));
1659                 memset(last, 0, size);
1660         } else if (s->not_found) {
1661                 /* Insert new name. */
1662                 size_t size = EXT4_XATTR_LEN(name_len);
1663                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1664
1665                 memmove((void *)here + size, here, rest);
1666                 memset(here, 0, size);
1667                 here->e_name_index = i->name_index;
1668                 here->e_name_len = name_len;
1669                 memcpy(here->e_name, i->name, name_len);
1670         } else {
1671                 /* This is an update, reset value info. */
1672                 here->e_value_inum = 0;
1673                 here->e_value_offs = 0;
1674                 here->e_value_size = 0;
1675         }
1676
1677         if (i->value) {
1678                 /* Insert new value. */
1679                 if (in_inode) {
1680                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1681                 } else if (i->value_len) {
1682                         void *val = s->base + min_offs - new_size;
1683
1684                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1685                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1686                                 memset(val, 0, new_size);
1687                         } else {
1688                                 memcpy(val, i->value, i->value_len);
1689                                 /* Clear padding bytes. */
1690                                 memset(val + i->value_len, 0,
1691                                        new_size - i->value_len);
1692                         }
1693                 }
1694                 here->e_value_size = cpu_to_le32(i->value_len);
1695         }
1696
1697         if (i->value) {
1698                 __le32 hash = 0;
1699
1700                 /* Entry hash calculation. */
1701                 if (in_inode) {
1702                         __le32 crc32c_hash;
1703
1704                         /*
1705                          * Feed crc32c hash instead of the raw value for entry
1706                          * hash calculation. This is to avoid walking
1707                          * potentially long value buffer again.
1708                          */
1709                         crc32c_hash = cpu_to_le32(
1710                                        ext4_xattr_inode_get_hash(new_ea_inode));
1711                         hash = ext4_xattr_hash_entry(here->e_name,
1712                                                      here->e_name_len,
1713                                                      &crc32c_hash, 1);
1714                 } else if (is_block) {
1715                         __le32 *value = s->base + min_offs - new_size;
1716
1717                         hash = ext4_xattr_hash_entry(here->e_name,
1718                                                      here->e_name_len, value,
1719                                                      new_size >> 2);
1720                 }
1721                 here->e_hash = hash;
1722         }
1723
1724         if (is_block)
1725                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1726
1727         ret = 0;
1728 out:
1729         iput(old_ea_inode);
1730         iput(new_ea_inode);
1731         return ret;
1732 }
1733
1734 struct ext4_xattr_block_find {
1735         struct ext4_xattr_search s;
1736         struct buffer_head *bh;
1737 };
1738
1739 static int
1740 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1741                       struct ext4_xattr_block_find *bs)
1742 {
1743         struct super_block *sb = inode->i_sb;
1744         int error;
1745
1746         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1747                   i->name_index, i->name, i->value, (long)i->value_len);
1748
1749         if (EXT4_I(inode)->i_file_acl) {
1750                 /* The inode already has an extended attribute block. */
1751                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1752                 error = -EIO;
1753                 if (!bs->bh)
1754                         goto cleanup;
1755                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1756                         atomic_read(&(bs->bh->b_count)),
1757                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1758                 if (ext4_xattr_check_block(inode, bs->bh)) {
1759                         EXT4_ERROR_INODE(inode, "bad block %llu",
1760                                          EXT4_I(inode)->i_file_acl);
1761                         error = -EFSCORRUPTED;
1762                         goto cleanup;
1763                 }
1764                 /* Find the named attribute. */
1765                 bs->s.base = BHDR(bs->bh);
1766                 bs->s.first = BFIRST(bs->bh);
1767                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1768                 bs->s.here = bs->s.first;
1769                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
1770                                               i->name, 1);
1771                 if (error && error != -ENODATA)
1772                         goto cleanup;
1773                 bs->s.not_found = error;
1774         }
1775         error = 0;
1776
1777 cleanup:
1778         return error;
1779 }
1780
1781 static int
1782 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1783                      struct ext4_xattr_info *i,
1784                      struct ext4_xattr_block_find *bs)
1785 {
1786         struct super_block *sb = inode->i_sb;
1787         struct buffer_head *new_bh = NULL;
1788         struct ext4_xattr_search s_copy = bs->s;
1789         struct ext4_xattr_search *s = &s_copy;
1790         struct mb_cache_entry *ce = NULL;
1791         int error = 0;
1792         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1793         struct inode *ea_inode = NULL;
1794         size_t old_ea_inode_size = 0;
1795
1796 #define header(x) ((struct ext4_xattr_header *)(x))
1797
1798         if (s->base) {
1799                 BUFFER_TRACE(bs->bh, "get_write_access");
1800                 error = ext4_journal_get_write_access(handle, bs->bh);
1801                 if (error)
1802                         goto cleanup;
1803                 lock_buffer(bs->bh);
1804
1805                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1806                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1807
1808                         /*
1809                          * This must happen under buffer lock for
1810                          * ext4_xattr_block_set() to reliably detect modified
1811                          * block
1812                          */
1813                         if (ea_block_cache)
1814                                 mb_cache_entry_delete(ea_block_cache, hash,
1815                                                       bs->bh->b_blocknr);
1816                         ea_bdebug(bs->bh, "modifying in-place");
1817                         error = ext4_xattr_set_entry(i, s, handle, inode,
1818                                                      true /* is_block */);
1819                         ext4_xattr_block_csum_set(inode, bs->bh);
1820                         unlock_buffer(bs->bh);
1821                         if (error == -EFSCORRUPTED)
1822                                 goto bad_block;
1823                         if (!error)
1824                                 error = ext4_handle_dirty_metadata(handle,
1825                                                                    inode,
1826                                                                    bs->bh);
1827                         if (error)
1828                                 goto cleanup;
1829                         goto inserted;
1830                 } else {
1831                         int offset = (char *)s->here - bs->bh->b_data;
1832
1833                         unlock_buffer(bs->bh);
1834                         ea_bdebug(bs->bh, "cloning");
1835                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1836                         error = -ENOMEM;
1837                         if (s->base == NULL)
1838                                 goto cleanup;
1839                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1840                         s->first = ENTRY(header(s->base)+1);
1841                         header(s->base)->h_refcount = cpu_to_le32(1);
1842                         s->here = ENTRY(s->base + offset);
1843                         s->end = s->base + bs->bh->b_size;
1844
1845                         /*
1846                          * If existing entry points to an xattr inode, we need
1847                          * to prevent ext4_xattr_set_entry() from decrementing
1848                          * ref count on it because the reference belongs to the
1849                          * original block. In this case, make the entry look
1850                          * like it has an empty value.
1851                          */
1852                         if (!s->not_found && s->here->e_value_inum) {
1853                                 /*
1854                                  * Defer quota free call for previous inode
1855                                  * until success is guaranteed.
1856                                  */
1857                                 old_ea_inode_size = le32_to_cpu(
1858                                                         s->here->e_value_size);
1859                                 s->here->e_value_inum = 0;
1860                                 s->here->e_value_size = 0;
1861                         }
1862                 }
1863         } else {
1864                 /* Allocate a buffer where we construct the new block. */
1865                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1866                 /* assert(header == s->base) */
1867                 error = -ENOMEM;
1868                 if (s->base == NULL)
1869                         goto cleanup;
1870                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1871                 header(s->base)->h_blocks = cpu_to_le32(1);
1872                 header(s->base)->h_refcount = cpu_to_le32(1);
1873                 s->first = ENTRY(header(s->base)+1);
1874                 s->here = ENTRY(header(s->base)+1);
1875                 s->end = s->base + sb->s_blocksize;
1876         }
1877
1878         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1879         if (error == -EFSCORRUPTED)
1880                 goto bad_block;
1881         if (error)
1882                 goto cleanup;
1883
1884         if (i->value && s->here->e_value_inum) {
1885                 unsigned int ea_ino;
1886
1887                 /*
1888                  * A ref count on ea_inode has been taken as part of the call to
1889                  * ext4_xattr_set_entry() above. We would like to drop this
1890                  * extra ref but we have to wait until the xattr block is
1891                  * initialized and has its own ref count on the ea_inode.
1892                  */
1893                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1894                 error = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
1895                 if (error) {
1896                         ea_inode = NULL;
1897                         goto cleanup;
1898                 }
1899         }
1900
1901 inserted:
1902         if (!IS_LAST_ENTRY(s->first)) {
1903                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1904                                                      &ce);
1905                 if (new_bh) {
1906                         /* We found an identical block in the cache. */
1907                         if (new_bh == bs->bh)
1908                                 ea_bdebug(new_bh, "keeping");
1909                         else {
1910                                 u32 ref;
1911
1912                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1913
1914                                 /* The old block is released after updating
1915                                    the inode. */
1916                                 error = dquot_alloc_block(inode,
1917                                                 EXT4_C2B(EXT4_SB(sb), 1));
1918                                 if (error)
1919                                         goto cleanup;
1920                                 BUFFER_TRACE(new_bh, "get_write_access");
1921                                 error = ext4_journal_get_write_access(handle,
1922                                                                       new_bh);
1923                                 if (error)
1924                                         goto cleanup_dquot;
1925                                 lock_buffer(new_bh);
1926                                 /*
1927                                  * We have to be careful about races with
1928                                  * freeing, rehashing or adding references to
1929                                  * xattr block. Once we hold buffer lock xattr
1930                                  * block's state is stable so we can check
1931                                  * whether the block got freed / rehashed or
1932                                  * not.  Since we unhash mbcache entry under
1933                                  * buffer lock when freeing / rehashing xattr
1934                                  * block, checking whether entry is still
1935                                  * hashed is reliable. Same rules hold for
1936                                  * e_reusable handling.
1937                                  */
1938                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
1939                                     !ce->e_reusable) {
1940                                         /*
1941                                          * Undo everything and check mbcache
1942                                          * again.
1943                                          */
1944                                         unlock_buffer(new_bh);
1945                                         dquot_free_block(inode,
1946                                                          EXT4_C2B(EXT4_SB(sb),
1947                                                                   1));
1948                                         brelse(new_bh);
1949                                         mb_cache_entry_put(ea_block_cache, ce);
1950                                         ce = NULL;
1951                                         new_bh = NULL;
1952                                         goto inserted;
1953                                 }
1954                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
1955                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
1956                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
1957                                         ce->e_reusable = 0;
1958                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
1959                                           ref);
1960                                 ext4_xattr_block_csum_set(inode, new_bh);
1961                                 unlock_buffer(new_bh);
1962                                 error = ext4_handle_dirty_metadata(handle,
1963                                                                    inode,
1964                                                                    new_bh);
1965                                 if (error)
1966                                         goto cleanup_dquot;
1967                         }
1968                         mb_cache_entry_touch(ea_block_cache, ce);
1969                         mb_cache_entry_put(ea_block_cache, ce);
1970                         ce = NULL;
1971                 } else if (bs->bh && s->base == bs->bh->b_data) {
1972                         /* We were modifying this block in-place. */
1973                         ea_bdebug(bs->bh, "keeping this block");
1974                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
1975                         new_bh = bs->bh;
1976                         get_bh(new_bh);
1977                 } else {
1978                         /* We need to allocate a new block */
1979                         ext4_fsblk_t goal, block;
1980
1981                         WARN_ON_ONCE(dquot_initialize_needed(inode));
1982
1983                         goal = ext4_group_first_block_no(sb,
1984                                                 EXT4_I(inode)->i_block_group);
1985
1986                         /* non-extent files can't have physical blocks past 2^32 */
1987                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
1988                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
1989
1990                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
1991                                                      NULL, &error);
1992                         if (error)
1993                                 goto cleanup;
1994
1995                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
1996                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
1997
1998                         ea_idebug(inode, "creating block %llu",
1999                                   (unsigned long long)block);
2000
2001                         new_bh = sb_getblk(sb, block);
2002                         if (unlikely(!new_bh)) {
2003                                 error = -ENOMEM;
2004 getblk_failed:
2005                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2006                                                  EXT4_FREE_BLOCKS_METADATA);
2007                                 goto cleanup;
2008                         }
2009                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2010                                                       ENTRY(header(s->base)+1));
2011                         if (error)
2012                                 goto getblk_failed;
2013                         if (ea_inode) {
2014                                 /* Drop the extra ref on ea_inode. */
2015                                 error = ext4_xattr_inode_dec_ref(handle,
2016                                                                  ea_inode);
2017                                 if (error)
2018                                         ext4_warning_inode(ea_inode,
2019                                                            "dec ref error=%d",
2020                                                            error);
2021                                 iput(ea_inode);
2022                                 ea_inode = NULL;
2023                         }
2024
2025                         lock_buffer(new_bh);
2026                         error = ext4_journal_get_create_access(handle, new_bh);
2027                         if (error) {
2028                                 unlock_buffer(new_bh);
2029                                 error = -EIO;
2030                                 goto getblk_failed;
2031                         }
2032                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2033                         ext4_xattr_block_csum_set(inode, new_bh);
2034                         set_buffer_uptodate(new_bh);
2035                         unlock_buffer(new_bh);
2036                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2037                         error = ext4_handle_dirty_metadata(handle, inode,
2038                                                            new_bh);
2039                         if (error)
2040                                 goto cleanup;
2041                 }
2042         }
2043
2044         if (old_ea_inode_size)
2045                 ext4_xattr_inode_free_quota(inode, old_ea_inode_size);
2046
2047         /* Update the inode. */
2048         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2049
2050         /* Drop the previous xattr block. */
2051         if (bs->bh && bs->bh != new_bh) {
2052                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2053
2054                 ext4_xattr_release_block(handle, inode, bs->bh,
2055                                          &ea_inode_array,
2056                                          0 /* extra_credits */);
2057                 ext4_xattr_inode_array_free(ea_inode_array);
2058         }
2059         error = 0;
2060
2061 cleanup:
2062         if (ea_inode) {
2063                 int error2;
2064
2065                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2066                 if (error2)
2067                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2068                                            error2);
2069
2070                 /* If there was an error, revert the quota charge. */
2071                 if (error)
2072                         ext4_xattr_inode_free_quota(inode,
2073                                                     i_size_read(ea_inode));
2074                 iput(ea_inode);
2075         }
2076         if (ce)
2077                 mb_cache_entry_put(ea_block_cache, ce);
2078         brelse(new_bh);
2079         if (!(bs->bh && s->base == bs->bh->b_data))
2080                 kfree(s->base);
2081
2082         return error;
2083
2084 cleanup_dquot:
2085         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2086         goto cleanup;
2087
2088 bad_block:
2089         EXT4_ERROR_INODE(inode, "bad block %llu",
2090                          EXT4_I(inode)->i_file_acl);
2091         goto cleanup;
2092
2093 #undef header
2094 }
2095
2096 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2097                           struct ext4_xattr_ibody_find *is)
2098 {
2099         struct ext4_xattr_ibody_header *header;
2100         struct ext4_inode *raw_inode;
2101         int error;
2102
2103         if (EXT4_I(inode)->i_extra_isize == 0)
2104                 return 0;
2105         raw_inode = ext4_raw_inode(&is->iloc);
2106         header = IHDR(inode, raw_inode);
2107         is->s.base = is->s.first = IFIRST(header);
2108         is->s.here = is->s.first;
2109         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2110         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2111                 error = xattr_check_inode(inode, header, is->s.end);
2112                 if (error)
2113                         return error;
2114                 /* Find the named attribute. */
2115                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
2116                                               i->name, 0);
2117                 if (error && error != -ENODATA)
2118                         return error;
2119                 is->s.not_found = error;
2120         }
2121         return 0;
2122 }
2123
2124 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2125                                 struct ext4_xattr_info *i,
2126                                 struct ext4_xattr_ibody_find *is)
2127 {
2128         struct ext4_xattr_ibody_header *header;
2129         struct ext4_xattr_search *s = &is->s;
2130         int error;
2131
2132         if (EXT4_I(inode)->i_extra_isize == 0)
2133                 return -ENOSPC;
2134         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2135         if (error) {
2136                 if (error == -ENOSPC &&
2137                     ext4_has_inline_data(inode)) {
2138                         error = ext4_try_to_evict_inline_data(handle, inode,
2139                                         EXT4_XATTR_LEN(strlen(i->name) +
2140                                         EXT4_XATTR_SIZE(i->value_len)));
2141                         if (error)
2142                                 return error;
2143                         error = ext4_xattr_ibody_find(inode, i, is);
2144                         if (error)
2145                                 return error;
2146                         error = ext4_xattr_set_entry(i, s, handle, inode,
2147                                                      false /* is_block */);
2148                 }
2149                 if (error)
2150                         return error;
2151         }
2152         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2153         if (!IS_LAST_ENTRY(s->first)) {
2154                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2155                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2156         } else {
2157                 header->h_magic = cpu_to_le32(0);
2158                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2159         }
2160         return 0;
2161 }
2162
2163 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2164                                 struct ext4_xattr_info *i,
2165                                 struct ext4_xattr_ibody_find *is)
2166 {
2167         struct ext4_xattr_ibody_header *header;
2168         struct ext4_xattr_search *s = &is->s;
2169         int error;
2170
2171         if (EXT4_I(inode)->i_extra_isize == 0)
2172                 return -ENOSPC;
2173         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2174         if (error)
2175                 return error;
2176         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2177         if (!IS_LAST_ENTRY(s->first)) {
2178                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2179                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2180         } else {
2181                 header->h_magic = cpu_to_le32(0);
2182                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2183         }
2184         return 0;
2185 }
2186
2187 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2188                                  struct ext4_xattr_info *i)
2189 {
2190         void *value;
2191
2192         /* When e_value_inum is set the value is stored externally. */
2193         if (s->here->e_value_inum)
2194                 return 0;
2195         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2196                 return 0;
2197         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2198         return !memcmp(value, i->value, i->value_len);
2199 }
2200
2201 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2202 {
2203         struct buffer_head *bh;
2204         int error;
2205
2206         if (!EXT4_I(inode)->i_file_acl)
2207                 return NULL;
2208         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2209         if (!bh)
2210                 return ERR_PTR(-EIO);
2211         error = ext4_xattr_check_block(inode, bh);
2212         if (error)
2213                 return ERR_PTR(error);
2214         return bh;
2215 }
2216
2217 /*
2218  * ext4_xattr_set_handle()
2219  *
2220  * Create, replace or remove an extended attribute for this inode.  Value
2221  * is NULL to remove an existing extended attribute, and non-NULL to
2222  * either replace an existing extended attribute, or create a new extended
2223  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2224  * specify that an extended attribute must exist and must not exist
2225  * previous to the call, respectively.
2226  *
2227  * Returns 0, or a negative error number on failure.
2228  */
2229 int
2230 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2231                       const char *name, const void *value, size_t value_len,
2232                       int flags)
2233 {
2234         struct ext4_xattr_info i = {
2235                 .name_index = name_index,
2236                 .name = name,
2237                 .value = value,
2238                 .value_len = value_len,
2239                 .in_inode = 0,
2240         };
2241         struct ext4_xattr_ibody_find is = {
2242                 .s = { .not_found = -ENODATA, },
2243         };
2244         struct ext4_xattr_block_find bs = {
2245                 .s = { .not_found = -ENODATA, },
2246         };
2247         int no_expand;
2248         int error;
2249
2250         if (!name)
2251                 return -EINVAL;
2252         if (strlen(name) > 255)
2253                 return -ERANGE;
2254
2255         ext4_write_lock_xattr(inode, &no_expand);
2256
2257         /* Check journal credits under write lock. */
2258         if (ext4_handle_valid(handle)) {
2259                 struct buffer_head *bh;
2260                 int credits;
2261
2262                 bh = ext4_xattr_get_block(inode);
2263                 if (IS_ERR(bh)) {
2264                         error = PTR_ERR(bh);
2265                         goto cleanup;
2266                 }
2267
2268                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2269                                                    value_len,
2270                                                    flags & XATTR_CREATE);
2271                 brelse(bh);
2272
2273                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2274                         error = -ENOSPC;
2275                         goto cleanup;
2276                 }
2277         }
2278
2279         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2280         if (error)
2281                 goto cleanup;
2282
2283         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2284                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2285                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2286                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2287         }
2288
2289         error = ext4_xattr_ibody_find(inode, &i, &is);
2290         if (error)
2291                 goto cleanup;
2292         if (is.s.not_found)
2293                 error = ext4_xattr_block_find(inode, &i, &bs);
2294         if (error)
2295                 goto cleanup;
2296         if (is.s.not_found && bs.s.not_found) {
2297                 error = -ENODATA;
2298                 if (flags & XATTR_REPLACE)
2299                         goto cleanup;
2300                 error = 0;
2301                 if (!value)
2302                         goto cleanup;
2303         } else {
2304                 error = -EEXIST;
2305                 if (flags & XATTR_CREATE)
2306                         goto cleanup;
2307         }
2308
2309         if (!value) {
2310                 if (!is.s.not_found)
2311                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2312                 else if (!bs.s.not_found)
2313                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2314         } else {
2315                 error = 0;
2316                 /* Xattr value did not change? Save us some work and bail out */
2317                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2318                         goto cleanup;
2319                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2320                         goto cleanup;
2321
2322                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2323                     (EXT4_XATTR_SIZE(i.value_len) >
2324                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2325                         i.in_inode = 1;
2326 retry_inode:
2327                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2328                 if (!error && !bs.s.not_found) {
2329                         i.value = NULL;
2330                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2331                 } else if (error == -ENOSPC) {
2332                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2333                                 error = ext4_xattr_block_find(inode, &i, &bs);
2334                                 if (error)
2335                                         goto cleanup;
2336                         }
2337                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2338                         if (!error && !is.s.not_found) {
2339                                 i.value = NULL;
2340                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2341                                                              &is);
2342                         } else if (error == -ENOSPC) {
2343                                 /*
2344                                  * Xattr does not fit in the block, store at
2345                                  * external inode if possible.
2346                                  */
2347                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2348                                     !i.in_inode) {
2349                                         i.in_inode = 1;
2350                                         goto retry_inode;
2351                                 }
2352                         }
2353                 }
2354         }
2355         if (!error) {
2356                 ext4_xattr_update_super_block(handle, inode->i_sb);
2357                 inode->i_ctime = current_time(inode);
2358                 if (!value)
2359                         no_expand = 0;
2360                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2361                 /*
2362                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2363                  * error != 0.
2364                  */
2365                 is.iloc.bh = NULL;
2366                 if (IS_SYNC(inode))
2367                         ext4_handle_sync(handle);
2368         }
2369
2370 cleanup:
2371         brelse(is.iloc.bh);
2372         brelse(bs.bh);
2373         ext4_write_unlock_xattr(inode, &no_expand);
2374         return error;
2375 }
2376
2377 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2378                            bool is_create, int *credits)
2379 {
2380         struct buffer_head *bh;
2381         int err;
2382
2383         *credits = 0;
2384
2385         if (!EXT4_SB(inode->i_sb)->s_journal)
2386                 return 0;
2387
2388         down_read(&EXT4_I(inode)->xattr_sem);
2389
2390         bh = ext4_xattr_get_block(inode);
2391         if (IS_ERR(bh)) {
2392                 err = PTR_ERR(bh);
2393         } else {
2394                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2395                                                     value_len, is_create);
2396                 brelse(bh);
2397                 err = 0;
2398         }
2399
2400         up_read(&EXT4_I(inode)->xattr_sem);
2401         return err;
2402 }
2403
2404 /*
2405  * ext4_xattr_set()
2406  *
2407  * Like ext4_xattr_set_handle, but start from an inode. This extended
2408  * attribute modification is a filesystem transaction by itself.
2409  *
2410  * Returns 0, or a negative error number on failure.
2411  */
2412 int
2413 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2414                const void *value, size_t value_len, int flags)
2415 {
2416         handle_t *handle;
2417         struct super_block *sb = inode->i_sb;
2418         int error, retries = 0;
2419         int credits;
2420
2421         error = dquot_initialize(inode);
2422         if (error)
2423                 return error;
2424
2425 retry:
2426         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2427                                        &credits);
2428         if (error)
2429                 return error;
2430
2431         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2432         if (IS_ERR(handle)) {
2433                 error = PTR_ERR(handle);
2434         } else {
2435                 int error2;
2436
2437                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2438                                               value, value_len, flags);
2439                 error2 = ext4_journal_stop(handle);
2440                 if (error == -ENOSPC &&
2441                     ext4_should_retry_alloc(sb, &retries))
2442                         goto retry;
2443                 if (error == 0)
2444                         error = error2;
2445         }
2446
2447         return error;
2448 }
2449
2450 /*
2451  * Shift the EA entries in the inode to create space for the increased
2452  * i_extra_isize.
2453  */
2454 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2455                                      int value_offs_shift, void *to,
2456                                      void *from, size_t n)
2457 {
2458         struct ext4_xattr_entry *last = entry;
2459         int new_offs;
2460
2461         /* We always shift xattr headers further thus offsets get lower */
2462         BUG_ON(value_offs_shift > 0);
2463
2464         /* Adjust the value offsets of the entries */
2465         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2466                 if (!last->e_value_inum && last->e_value_size) {
2467                         new_offs = le16_to_cpu(last->e_value_offs) +
2468                                                         value_offs_shift;
2469                         last->e_value_offs = cpu_to_le16(new_offs);
2470                 }
2471         }
2472         /* Shift the entries by n bytes */
2473         memmove(to, from, n);
2474 }
2475
2476 /*
2477  * Move xattr pointed to by 'entry' from inode into external xattr block
2478  */
2479 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2480                                     struct ext4_inode *raw_inode,
2481                                     struct ext4_xattr_entry *entry)
2482 {
2483         struct ext4_xattr_ibody_find *is = NULL;
2484         struct ext4_xattr_block_find *bs = NULL;
2485         char *buffer = NULL, *b_entry_name = NULL;
2486         size_t value_size = le32_to_cpu(entry->e_value_size);
2487         struct ext4_xattr_info i = {
2488                 .value = NULL,
2489                 .value_len = 0,
2490                 .name_index = entry->e_name_index,
2491                 .in_inode = !!entry->e_value_inum,
2492         };
2493         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2494         int error;
2495
2496         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2497         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2498         buffer = kmalloc(value_size, GFP_NOFS);
2499         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2500         if (!is || !bs || !buffer || !b_entry_name) {
2501                 error = -ENOMEM;
2502                 goto out;
2503         }
2504
2505         is->s.not_found = -ENODATA;
2506         bs->s.not_found = -ENODATA;
2507         is->iloc.bh = NULL;
2508         bs->bh = NULL;
2509
2510         /* Save the entry name and the entry value */
2511         if (entry->e_value_inum) {
2512                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2513                 if (error)
2514                         goto out;
2515         } else {
2516                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2517                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2518         }
2519
2520         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2521         b_entry_name[entry->e_name_len] = '\0';
2522         i.name = b_entry_name;
2523
2524         error = ext4_get_inode_loc(inode, &is->iloc);
2525         if (error)
2526                 goto out;
2527
2528         error = ext4_xattr_ibody_find(inode, &i, is);
2529         if (error)
2530                 goto out;
2531
2532         /* Remove the chosen entry from the inode */
2533         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2534         if (error)
2535                 goto out;
2536
2537         i.value = buffer;
2538         i.value_len = value_size;
2539         error = ext4_xattr_block_find(inode, &i, bs);
2540         if (error)
2541                 goto out;
2542
2543         /* Add entry which was removed from the inode into the block */
2544         error = ext4_xattr_block_set(handle, inode, &i, bs);
2545         if (error)
2546                 goto out;
2547         error = 0;
2548 out:
2549         kfree(b_entry_name);
2550         kfree(buffer);
2551         if (is)
2552                 brelse(is->iloc.bh);
2553         kfree(is);
2554         kfree(bs);
2555
2556         return error;
2557 }
2558
2559 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2560                                        struct ext4_inode *raw_inode,
2561                                        int isize_diff, size_t ifree,
2562                                        size_t bfree, int *total_ino)
2563 {
2564         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2565         struct ext4_xattr_entry *small_entry;
2566         struct ext4_xattr_entry *entry;
2567         struct ext4_xattr_entry *last;
2568         unsigned int entry_size;        /* EA entry size */
2569         unsigned int total_size;        /* EA entry size + value size */
2570         unsigned int min_total_size;
2571         int error;
2572
2573         while (isize_diff > ifree) {
2574                 entry = NULL;
2575                 small_entry = NULL;
2576                 min_total_size = ~0U;
2577                 last = IFIRST(header);
2578                 /* Find the entry best suited to be pushed into EA block */
2579                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2580                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2581                         if (!last->e_value_inum)
2582                                 total_size += EXT4_XATTR_SIZE(
2583                                                le32_to_cpu(last->e_value_size));
2584                         if (total_size <= bfree &&
2585                             total_size < min_total_size) {
2586                                 if (total_size + ifree < isize_diff) {
2587                                         small_entry = last;
2588                                 } else {
2589                                         entry = last;
2590                                         min_total_size = total_size;
2591                                 }
2592                         }
2593                 }
2594
2595                 if (entry == NULL) {
2596                         if (small_entry == NULL)
2597                                 return -ENOSPC;
2598                         entry = small_entry;
2599                 }
2600
2601                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2602                 total_size = entry_size;
2603                 if (!entry->e_value_inum)
2604                         total_size += EXT4_XATTR_SIZE(
2605                                               le32_to_cpu(entry->e_value_size));
2606                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2607                                                  entry);
2608                 if (error)
2609                         return error;
2610
2611                 *total_ino -= entry_size;
2612                 ifree += total_size;
2613                 bfree -= total_size;
2614         }
2615
2616         return 0;
2617 }
2618
2619 /*
2620  * Expand an inode by new_extra_isize bytes when EAs are present.
2621  * Returns 0 on success or negative error number on failure.
2622  */
2623 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2624                                struct ext4_inode *raw_inode, handle_t *handle)
2625 {
2626         struct ext4_xattr_ibody_header *header;
2627         struct buffer_head *bh = NULL;
2628         size_t min_offs;
2629         size_t ifree, bfree;
2630         int total_ino;
2631         void *base, *end;
2632         int error = 0, tried_min_extra_isize = 0;
2633         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
2634         int isize_diff; /* How much do we need to grow i_extra_isize */
2635         int no_expand;
2636
2637         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
2638                 return 0;
2639
2640 retry:
2641         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2642         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2643                 goto out;
2644
2645         header = IHDR(inode, raw_inode);
2646
2647         /*
2648          * Check if enough free space is available in the inode to shift the
2649          * entries ahead by new_extra_isize.
2650          */
2651
2652         base = IFIRST(header);
2653         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2654         min_offs = end - base;
2655         total_ino = sizeof(struct ext4_xattr_ibody_header);
2656
2657         error = xattr_check_inode(inode, header, end);
2658         if (error)
2659                 goto cleanup;
2660
2661         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2662         if (ifree >= isize_diff)
2663                 goto shift;
2664
2665         /*
2666          * Enough free space isn't available in the inode, check if
2667          * EA block can hold new_extra_isize bytes.
2668          */
2669         if (EXT4_I(inode)->i_file_acl) {
2670                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2671                 error = -EIO;
2672                 if (!bh)
2673                         goto cleanup;
2674                 if (ext4_xattr_check_block(inode, bh)) {
2675                         EXT4_ERROR_INODE(inode, "bad block %llu",
2676                                          EXT4_I(inode)->i_file_acl);
2677                         error = -EFSCORRUPTED;
2678                         goto cleanup;
2679                 }
2680                 base = BHDR(bh);
2681                 end = bh->b_data + bh->b_size;
2682                 min_offs = end - base;
2683                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2684                                               NULL);
2685                 if (bfree + ifree < isize_diff) {
2686                         if (!tried_min_extra_isize && s_min_extra_isize) {
2687                                 tried_min_extra_isize++;
2688                                 new_extra_isize = s_min_extra_isize;
2689                                 brelse(bh);
2690                                 goto retry;
2691                         }
2692                         error = -ENOSPC;
2693                         goto cleanup;
2694                 }
2695         } else {
2696                 bfree = inode->i_sb->s_blocksize;
2697         }
2698
2699         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2700                                             isize_diff, ifree, bfree,
2701                                             &total_ino);
2702         if (error) {
2703                 if (error == -ENOSPC && !tried_min_extra_isize &&
2704                     s_min_extra_isize) {
2705                         tried_min_extra_isize++;
2706                         new_extra_isize = s_min_extra_isize;
2707                         brelse(bh);
2708                         goto retry;
2709                 }
2710                 goto cleanup;
2711         }
2712 shift:
2713         /* Adjust the offsets and shift the remaining entries ahead */
2714         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2715                         - new_extra_isize, (void *)raw_inode +
2716                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2717                         (void *)header, total_ino);
2718         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2719         brelse(bh);
2720 out:
2721         ext4_write_unlock_xattr(inode, &no_expand);
2722         return 0;
2723
2724 cleanup:
2725         brelse(bh);
2726         /*
2727          * Inode size expansion failed; don't try again
2728          */
2729         no_expand = 1;
2730         ext4_write_unlock_xattr(inode, &no_expand);
2731         return error;
2732 }
2733
2734 #define EIA_INCR 16 /* must be 2^n */
2735 #define EIA_MASK (EIA_INCR - 1)
2736
2737 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2738  * If @ea_inode_array is new or full it will be grown and the old
2739  * contents copied over.
2740  */
2741 static int
2742 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2743                         struct inode *inode)
2744 {
2745         if (*ea_inode_array == NULL) {
2746                 /*
2747                  * Start with 15 inodes, so it fits into a power-of-two size.
2748                  * If *ea_inode_array is NULL, this is essentially offsetof()
2749                  */
2750                 (*ea_inode_array) =
2751                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2752                                          inodes[EIA_MASK]),
2753                                 GFP_NOFS);
2754                 if (*ea_inode_array == NULL)
2755                         return -ENOMEM;
2756                 (*ea_inode_array)->count = 0;
2757         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2758                 /* expand the array once all 15 + n * 16 slots are full */
2759                 struct ext4_xattr_inode_array *new_array = NULL;
2760                 int count = (*ea_inode_array)->count;
2761
2762                 /* if new_array is NULL, this is essentially offsetof() */
2763                 new_array = kmalloc(
2764                                 offsetof(struct ext4_xattr_inode_array,
2765                                          inodes[count + EIA_INCR]),
2766                                 GFP_NOFS);
2767                 if (new_array == NULL)
2768                         return -ENOMEM;
2769                 memcpy(new_array, *ea_inode_array,
2770                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2771                 kfree(*ea_inode_array);
2772                 *ea_inode_array = new_array;
2773         }
2774         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2775         return 0;
2776 }
2777
2778 /*
2779  * ext4_xattr_delete_inode()
2780  *
2781  * Free extended attribute resources associated with this inode. Traverse
2782  * all entries and decrement reference on any xattr inodes associated with this
2783  * inode. This is called immediately before an inode is freed. We have exclusive
2784  * access to the inode. If an orphan inode is deleted it will also release its
2785  * references on xattr block and xattr inodes.
2786  */
2787 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2788                             struct ext4_xattr_inode_array **ea_inode_array,
2789                             int extra_credits)
2790 {
2791         struct buffer_head *bh = NULL;
2792         struct ext4_xattr_ibody_header *header;
2793         struct ext4_iloc iloc = { .bh = NULL };
2794         struct ext4_xattr_entry *entry;
2795         int error;
2796
2797         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2798                                           NULL /* bh */,
2799                                           false /* dirty */,
2800                                           false /* block_csum */);
2801         if (error) {
2802                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2803                 goto cleanup;
2804         }
2805
2806         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2807             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2808
2809                 error = ext4_get_inode_loc(inode, &iloc);
2810                 if (error) {
2811                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2812                         goto cleanup;
2813                 }
2814
2815                 error = ext4_journal_get_write_access(handle, iloc.bh);
2816                 if (error) {
2817                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2818                                          error);
2819                         goto cleanup;
2820                 }
2821
2822                 header = IHDR(inode, ext4_raw_inode(&iloc));
2823                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2824                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2825                                                      IFIRST(header),
2826                                                      false /* block_csum */,
2827                                                      ea_inode_array,
2828                                                      extra_credits,
2829                                                      false /* skip_quota */);
2830         }
2831
2832         if (EXT4_I(inode)->i_file_acl) {
2833                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2834                 if (!bh) {
2835                         EXT4_ERROR_INODE(inode, "block %llu read error",
2836                                          EXT4_I(inode)->i_file_acl);
2837                         error = -EIO;
2838                         goto cleanup;
2839                 }
2840                 error = ext4_xattr_check_block(inode, bh);
2841                 if (error) {
2842                         EXT4_ERROR_INODE(inode, "bad block %llu (error %d)",
2843                                          EXT4_I(inode)->i_file_acl, error);
2844                         goto cleanup;
2845                 }
2846
2847                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2848                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2849                              entry = EXT4_XATTR_NEXT(entry))
2850                                 if (entry->e_value_inum)
2851                                         ext4_xattr_inode_free_quota(inode,
2852                                               le32_to_cpu(entry->e_value_size));
2853
2854                 }
2855
2856                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2857                                          extra_credits);
2858                 /*
2859                  * Update i_file_acl value in the same transaction that releases
2860                  * block.
2861                  */
2862                 EXT4_I(inode)->i_file_acl = 0;
2863                 error = ext4_mark_inode_dirty(handle, inode);
2864                 if (error) {
2865                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2866                                          error);
2867                         goto cleanup;
2868                 }
2869         }
2870         error = 0;
2871 cleanup:
2872         brelse(iloc.bh);
2873         brelse(bh);
2874         return error;
2875 }
2876
2877 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2878 {
2879         int idx;
2880
2881         if (ea_inode_array == NULL)
2882                 return;
2883
2884         for (idx = 0; idx < ea_inode_array->count; ++idx)
2885                 iput(ea_inode_array->inodes[idx]);
2886         kfree(ea_inode_array);
2887 }
2888
2889 /*
2890  * ext4_xattr_block_cache_insert()
2891  *
2892  * Create a new entry in the extended attribute block cache, and insert
2893  * it unless such an entry is already in the cache.
2894  *
2895  * Returns 0, or a negative error number on failure.
2896  */
2897 static void
2898 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2899                               struct buffer_head *bh)
2900 {
2901         struct ext4_xattr_header *header = BHDR(bh);
2902         __u32 hash = le32_to_cpu(header->h_hash);
2903         int reusable = le32_to_cpu(header->h_refcount) <
2904                        EXT4_XATTR_REFCOUNT_MAX;
2905         int error;
2906
2907         if (!ea_block_cache)
2908                 return;
2909         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2910                                       bh->b_blocknr, reusable);
2911         if (error) {
2912                 if (error == -EBUSY)
2913                         ea_bdebug(bh, "already in cache");
2914         } else
2915                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2916 }
2917
2918 /*
2919  * ext4_xattr_cmp()
2920  *
2921  * Compare two extended attribute blocks for equality.
2922  *
2923  * Returns 0 if the blocks are equal, 1 if they differ, and
2924  * a negative error number on errors.
2925  */
2926 static int
2927 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2928                struct ext4_xattr_header *header2)
2929 {
2930         struct ext4_xattr_entry *entry1, *entry2;
2931
2932         entry1 = ENTRY(header1+1);
2933         entry2 = ENTRY(header2+1);
2934         while (!IS_LAST_ENTRY(entry1)) {
2935                 if (IS_LAST_ENTRY(entry2))
2936                         return 1;
2937                 if (entry1->e_hash != entry2->e_hash ||
2938                     entry1->e_name_index != entry2->e_name_index ||
2939                     entry1->e_name_len != entry2->e_name_len ||
2940                     entry1->e_value_size != entry2->e_value_size ||
2941                     entry1->e_value_inum != entry2->e_value_inum ||
2942                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2943                         return 1;
2944                 if (!entry1->e_value_inum &&
2945                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
2946                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
2947                            le32_to_cpu(entry1->e_value_size)))
2948                         return 1;
2949
2950                 entry1 = EXT4_XATTR_NEXT(entry1);
2951                 entry2 = EXT4_XATTR_NEXT(entry2);
2952         }
2953         if (!IS_LAST_ENTRY(entry2))
2954                 return 1;
2955         return 0;
2956 }
2957
2958 /*
2959  * ext4_xattr_block_cache_find()
2960  *
2961  * Find an identical extended attribute block.
2962  *
2963  * Returns a pointer to the block found, or NULL if such a block was
2964  * not found or an error occurred.
2965  */
2966 static struct buffer_head *
2967 ext4_xattr_block_cache_find(struct inode *inode,
2968                             struct ext4_xattr_header *header,
2969                             struct mb_cache_entry **pce)
2970 {
2971         __u32 hash = le32_to_cpu(header->h_hash);
2972         struct mb_cache_entry *ce;
2973         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
2974
2975         if (!ea_block_cache)
2976                 return NULL;
2977         if (!header->h_hash)
2978                 return NULL;  /* never share */
2979         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
2980         ce = mb_cache_entry_find_first(ea_block_cache, hash);
2981         while (ce) {
2982                 struct buffer_head *bh;
2983
2984                 bh = sb_bread(inode->i_sb, ce->e_value);
2985                 if (!bh) {
2986                         EXT4_ERROR_INODE(inode, "block %lu read error",
2987                                          (unsigned long)ce->e_value);
2988                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
2989                         *pce = ce;
2990                         return bh;
2991                 }
2992                 brelse(bh);
2993                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
2994         }
2995         return NULL;
2996 }
2997
2998 #define NAME_HASH_SHIFT 5
2999 #define VALUE_HASH_SHIFT 16
3000
3001 /*
3002  * ext4_xattr_hash_entry()
3003  *
3004  * Compute the hash of an extended attribute.
3005  */
3006 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3007                                     size_t value_count)
3008 {
3009         __u32 hash = 0;
3010
3011         while (name_len--) {
3012                 hash = (hash << NAME_HASH_SHIFT) ^
3013                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3014                        *name++;
3015         }
3016         while (value_count--) {
3017                 hash = (hash << VALUE_HASH_SHIFT) ^
3018                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3019                        le32_to_cpu(*value++);
3020         }
3021         return cpu_to_le32(hash);
3022 }
3023
3024 #undef NAME_HASH_SHIFT
3025 #undef VALUE_HASH_SHIFT
3026
3027 #define BLOCK_HASH_SHIFT 16
3028
3029 /*
3030  * ext4_xattr_rehash()
3031  *
3032  * Re-compute the extended attribute hash value after an entry has changed.
3033  */
3034 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3035 {
3036         struct ext4_xattr_entry *here;
3037         __u32 hash = 0;
3038
3039         here = ENTRY(header+1);
3040         while (!IS_LAST_ENTRY(here)) {
3041                 if (!here->e_hash) {
3042                         /* Block is not shared if an entry's hash value == 0 */
3043                         hash = 0;
3044                         break;
3045                 }
3046                 hash = (hash << BLOCK_HASH_SHIFT) ^
3047                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3048                        le32_to_cpu(here->e_hash);
3049                 here = EXT4_XATTR_NEXT(here);
3050         }
3051         header->h_hash = cpu_to_le32(hash);
3052 }
3053
3054 #undef BLOCK_HASH_SHIFT
3055
3056 #define HASH_BUCKET_BITS        10
3057
3058 struct mb_cache *
3059 ext4_xattr_create_cache(void)
3060 {
3061         return mb_cache_create(HASH_BUCKET_BITS);
3062 }
3063
3064 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3065 {
3066         if (cache)
3067                 mb_cache_destroy(cache);
3068 }
3069