fs/ext2/xattr.c

   1 /*
   2  * linux/fs/ext2/xattr.c
   3  *
   4  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
   5  *
   6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   7  * Extended attributes for symlinks and special files added per
   8  *  suggestion of Luka Renko <luka.renko@hermes.si>.
   9  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  10  *  Red Hat Inc.
  11  *
  12  */
  13
  14 /*
  15  * Extended attributes are stored on disk blocks allocated outside of
  16  * any inode. The i_file_acl field is then made to point to this allocated
  17  * block. If all extended attributes of an inode are identical, these
  18  * inodes may share the same extended attribute block. Such situations
  19  * are automatically detected by keeping a cache of recent attribute block
  20  * numbers and hashes over the block's contents in memory.
  21  *
  22  *
  23  * Extended attribute block layout:
  24  *
  25  *   +------------------+
  26  *   | header           |
  27  *   | entry 1          | |
  28  *   | entry 2          | | growing downwards
  29  *   | entry 3          | v
  30  *   | four null bytes  |
  31  *   | . . .            |
  32  *   | value 1          | ^
  33  *   | value 3          | | growing upwards
  34  *   | value 2          | |
  35  *   +------------------+
  36  *
  37  * The block header is followed by multiple entry descriptors. These entry
  38  * descriptors are variable in size, and alligned to EXT2_XATTR_PAD
  39  * byte boundaries. The entry descriptors are sorted by attribute name,
  40  * so that two extended attribute blocks can be compared efficiently.
  41  *
  42  * Attribute values are aligned to the end of the block, stored in
  43  * no specific order. They are also padded to EXT2_XATTR_PAD byte
  44  * boundaries. No additional gaps are left between them.
  45  *
  46  * Locking strategy
  47  * ----------------
  48  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
  49  * EA blocks are only changed if they are exclusive to an inode, so
  50  * holding xattr_sem also means that nothing but the EA block's reference
  51  * count will change. Multiple writers to an EA block are synchronized
  52  * by the bh lock. No more than a single bh lock is held at any time
  53  * to avoid deadlocks.
  54  */
  55
  56 #include <linux/buffer_head.h>
  57 #include <linux/module.h>
  58 #include <linux/init.h>
  59 #include <linux/slab.h>
  60 #include <linux/mbcache.h>
  61 #include <linux/quotaops.h>
  62 #include <linux/rwsem.h>
  63 #include <linux/vs_dlimit.h>
  64 #include "ext2.h"
  65 #include "xattr.h"
  66 #include "acl.h"
  67
  68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
  69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
  70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
  71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
  72
  73 #ifdef EXT2_XATTR_DEBUG
  74 # define ea_idebug(inode, f...) do { \
  75                 printk(KERN_DEBUG "inode %s:%ld: ", \
  76                         inode->i_sb->s_id, inode->i_ino); \
  77                 printk(f); \
  78                 printk("\n"); \
  79         } while (0)
  80 # define ea_bdebug(bh, f...) do { \
  81                 char b[BDEVNAME_SIZE]; \
  82                 printk(KERN_DEBUG "block %s:%lu: ", \
  83                         bdevname(bh->b_bdev, b), \
  84                         (unsigned long) bh->b_blocknr); \
  85                 printk(f); \
  86                 printk("\n"); \
  87         } while (0)
  88 #else
  89 # define ea_idebug(f...)
  90 # define ea_bdebug(f...)
  91 #endif
  92
  93 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  94                            struct ext2_xattr_header *);
  95
  96 static int ext2_xattr_cache_insert(struct buffer_head *);
  97 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  98                                                  struct ext2_xattr_header *);
  99 static void ext2_xattr_rehash(struct ext2_xattr_header *,
 100                               struct ext2_xattr_entry *);
 101
 102 static struct mb_cache *ext2_xattr_cache;
 103
 104 static struct xattr_handler *ext2_xattr_handler_map[] = {
 105         [EXT2_XATTR_INDEX_USER]              = &ext2_xattr_user_handler,
 106 #ifdef CONFIG_EXT2_FS_POSIX_ACL
 107         [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext2_xattr_acl_access_handler,
 108         [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
 109 #endif
 110         [EXT2_XATTR_INDEX_TRUSTED]           = &ext2_xattr_trusted_handler,
 111 #ifdef CONFIG_EXT2_FS_SECURITY
 112         [EXT2_XATTR_INDEX_SECURITY]          = &ext2_xattr_security_handler,
 113 #endif
 114 };
 115
 116 struct xattr_handler *ext2_xattr_handlers[] = {
 117         &ext2_xattr_user_handler,
 118         &ext2_xattr_trusted_handler,
 119 #ifdef CONFIG_EXT2_FS_POSIX_ACL
 120         &ext2_xattr_acl_access_handler,
 121         &ext2_xattr_acl_default_handler,
 122 #endif
 123 #ifdef CONFIG_EXT2_FS_SECURITY
 124         &ext2_xattr_security_handler,
 125 #endif
 126         NULL
 127 };
 128
 129 static inline struct xattr_handler *
 130 ext2_xattr_handler(int name_index)
 131 {
 132         struct xattr_handler *handler = NULL;
 133
 134         if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 135                 handler = ext2_xattr_handler_map[name_index];
 136         return handler;
 137 }
 138
 139 /*
 140  * ext2_xattr_get()
 141  *
 142  * Copy an extended attribute into the buffer
 143  * provided, or compute the buffer size required.
 144  * Buffer is NULL to compute the size of the buffer required.
 145  *
 146  * Returns a negative error number on failure, or the number of bytes
 147  * used / required on success.
 148  */
 149 int
 150 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 151                void *buffer, size_t buffer_size)
 152 {
 153         struct buffer_head *bh = NULL;
 154         struct ext2_xattr_entry *entry;
 155         size_t name_len, size;
 156         char *end;
 157         int error;
 158
 159         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 160                   name_index, name, buffer, (long)buffer_size);
 161
 162         if (name == NULL)
 163                 return -EINVAL;
 164         down_read(&EXT2_I(inode)->xattr_sem);
 165         error = -ENODATA;
 166         if (!EXT2_I(inode)->i_file_acl)
 167                 goto cleanup;
 168         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 169         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 170         error = -EIO;
 171         if (!bh)
 172                 goto cleanup;
 173         ea_bdebug(bh, "b_count=%d, refcount=%d",
 174                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 175         end = bh->b_data + bh->b_size;
 176         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 177             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 178 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_get",
 179                         "inode %ld: bad block %d", inode->i_ino,
 180                         EXT2_I(inode)->i_file_acl);
 181                 error = -EIO;
 182                 goto cleanup;
 183         }
 184         /* find named attribute */
 185         name_len = strlen(name);
 186
 187         error = -ERANGE;
 188         if (name_len > 255)
 189                 goto cleanup;
 190         entry = FIRST_ENTRY(bh);
 191         while (!IS_LAST_ENTRY(entry)) {
 192                 struct ext2_xattr_entry *next =
 193                         EXT2_XATTR_NEXT(entry);
 194                 if ((char *)next >= end)
 195                         goto bad_block;
 196                 if (name_index == entry->e_name_index &&
 197                     name_len == entry->e_name_len &&
 198                     memcmp(name, entry->e_name, name_len) == 0)
 199                         goto found;
 200                 entry = next;
 201         }
 202         /* Check the remaining name entries */
 203         while (!IS_LAST_ENTRY(entry)) {
 204                 struct ext2_xattr_entry *next =
 205                         EXT2_XATTR_NEXT(entry);
 206                 if ((char *)next >= end)
 207                         goto bad_block;
 208                 entry = next;
 209         }
 210         if (ext2_xattr_cache_insert(bh))
 211                 ea_idebug(inode, "cache insert failed");
 212         error = -ENODATA;
 213         goto cleanup;
 214 found:
 215         /* check the buffer size */
 216         if (entry->e_value_block != 0)
 217                 goto bad_block;
 218         size = le32_to_cpu(entry->e_value_size);
 219         if (size > inode->i_sb->s_blocksize ||
 220             le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
 221                 goto bad_block;
 222
 223         if (ext2_xattr_cache_insert(bh))
 224                 ea_idebug(inode, "cache insert failed");
 225         if (buffer) {
 226                 error = -ERANGE;
 227                 if (size > buffer_size)
 228                         goto cleanup;
 229                 /* return value of attribute */
 230                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 231                         size);
 232         }
 233         error = size;
 234
 235 cleanup:
 236         brelse(bh);
 237         up_read(&EXT2_I(inode)->xattr_sem);
 238
 239         return error;
 240 }
 241
 242 /*
 243  * ext2_xattr_list()
 244  *
 245  * Copy a list of attribute names into the buffer
 246  * provided, or compute the buffer size required.
 247  * Buffer is NULL to compute the size of the buffer required.
 248  *
 249  * Returns a negative error number on failure, or the number of bytes
 250  * used / required on success.
 251  */
 252 static int
 253 ext2_xattr_list(struct inode *inode, char *buffer, size_t buffer_size)
 254 {
 255         struct buffer_head *bh = NULL;
 256         struct ext2_xattr_entry *entry;
 257         char *end;
 258         size_t rest = buffer_size;
 259         int error;
 260
 261         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 262                   buffer, (long)buffer_size);
 263
 264         down_read(&EXT2_I(inode)->xattr_sem);
 265         error = 0;
 266         if (!EXT2_I(inode)->i_file_acl)
 267                 goto cleanup;
 268         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 269         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 270         error = -EIO;
 271         if (!bh)
 272                 goto cleanup;
 273         ea_bdebug(bh, "b_count=%d, refcount=%d",
 274                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 275         end = bh->b_data + bh->b_size;
 276         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 277             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 278 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_list",
 279                         "inode %ld: bad block %d", inode->i_ino,
 280                         EXT2_I(inode)->i_file_acl);
 281                 error = -EIO;
 282                 goto cleanup;
 283         }
 284
 285         /* check the on-disk data structure */
 286         entry = FIRST_ENTRY(bh);
 287         while (!IS_LAST_ENTRY(entry)) {
 288                 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
 289
 290                 if ((char *)next >= end)
 291                         goto bad_block;
 292                 entry = next;
 293         }
 294         if (ext2_xattr_cache_insert(bh))
 295                 ea_idebug(inode, "cache insert failed");
 296
 297         /* list the attribute names */
 298         for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 299              entry = EXT2_XATTR_NEXT(entry)) {
 300                 struct xattr_handler *handler =
 301                         ext2_xattr_handler(entry->e_name_index);
 302
 303                 if (handler) {
 304                         size_t size = handler->list(inode, buffer, rest,
 305                                                     entry->e_name,
 306                                                     entry->e_name_len);
 307                         if (buffer) {
 308                                 if (size > rest) {
 309                                         error = -ERANGE;
 310                                         goto cleanup;
 311                                 }
 312                                 buffer += size;
 313                         }
 314                         rest -= size;
 315                 }
 316         }
 317         error = buffer_size - rest;  /* total size */
 318
 319 cleanup:
 320         brelse(bh);
 321         up_read(&EXT2_I(inode)->xattr_sem);
 322
 323         return error;
 324 }
 325
 326 /*
 327  * Inode operation listxattr()
 328  *
 329  * dentry->d_inode->i_mutex: don't care
 330  */
 331 ssize_t
 332 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 333 {
 334         return ext2_xattr_list(dentry->d_inode, buffer, size);
 335 }
 336
 337 /*
 338  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 339  * not set, set it.
 340  */
 341 static void ext2_xattr_update_super_block(struct super_block *sb)
 342 {
 343         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 344                 return;
 345
 346         EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 347         sb->s_dirt = 1;
 348         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 349 }
 350
 351 /*
 352  * ext2_xattr_set()
 353  *
 354  * Create, replace or remove an extended attribute for this inode. Buffer
 355  * is NULL to remove an existing extended attribute, and non-NULL to
 356  * either replace an existing extended attribute, or create a new extended
 357  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 358  * specify that an extended attribute must exist and must not exist
 359  * previous to the call, respectively.
 360  *
 361  * Returns 0, or a negative error number on failure.
 362  */
 363 int
 364 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 365                const void *value, size_t value_len, int flags)
 366 {
 367         struct super_block *sb = inode->i_sb;
 368         struct buffer_head *bh = NULL;
 369         struct ext2_xattr_header *header = NULL;
 370         struct ext2_xattr_entry *here, *last;
 371         size_t name_len, free, min_offs = sb->s_blocksize;
 372         int not_found = 1, error;
 373         char *end;
 374
 375         /*
 376          * header -- Points either into bh, or to a temporarily
 377          *           allocated buffer.
 378          * here -- The named entry found, or the place for inserting, within
 379          *         the block pointed to by header.
 380          * last -- Points right after the last named entry within the block
 381          *         pointed to by header.
 382          * min_offs -- The offset of the first value (values are aligned
 383          *             towards the end of the block).
 384          * end -- Points right after the block pointed to by header.
 385          */
 386
 387         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 388                   name_index, name, value, (long)value_len);
 389
 390         if (value == NULL)
 391                 value_len = 0;
 392         if (name == NULL)
 393                 return -EINVAL;
 394         name_len = strlen(name);
 395         if (name_len > 255 || value_len > sb->s_blocksize)
 396                 return -ERANGE;
 397         down_write(&EXT2_I(inode)->xattr_sem);
 398         if (EXT2_I(inode)->i_file_acl) {
 399                 /* The inode already has an extended attribute block. */
 400                 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 401                 error = -EIO;
 402                 if (!bh)
 403                         goto cleanup;
 404                 ea_bdebug(bh, "b_count=%d, refcount=%d",
 405                         atomic_read(&(bh->b_count)),
 406                         le32_to_cpu(HDR(bh)->h_refcount));
 407                 header = HDR(bh);
 408                 end = bh->b_data + bh->b_size;
 409                 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 410                     header->h_blocks != cpu_to_le32(1)) {
 411 bad_block:              ext2_error(sb, "ext2_xattr_set",
 412                                 "inode %ld: bad block %d", inode->i_ino,
 413                                    EXT2_I(inode)->i_file_acl);
 414                         error = -EIO;
 415                         goto cleanup;
 416                 }
 417                 /* Find the named attribute. */
 418                 here = FIRST_ENTRY(bh);
 419                 while (!IS_LAST_ENTRY(here)) {
 420                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
 421                         if ((char *)next >= end)
 422                                 goto bad_block;
 423                         if (!here->e_value_block && here->e_value_size) {
 424                                 size_t offs = le16_to_cpu(here->e_value_offs);
 425                                 if (offs < min_offs)
 426                                         min_offs = offs;
 427                         }
 428                         not_found = name_index - here->e_name_index;
 429                         if (!not_found)
 430                                 not_found = name_len - here->e_name_len;
 431                         if (!not_found)
 432                                 not_found = memcmp(name, here->e_name,name_len);
 433                         if (not_found <= 0)
 434                                 break;
 435                         here = next;
 436                 }
 437                 last = here;
 438                 /* We still need to compute min_offs and last. */
 439                 while (!IS_LAST_ENTRY(last)) {
 440                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
 441                         if ((char *)next >= end)
 442                                 goto bad_block;
 443                         if (!last->e_value_block && last->e_value_size) {
 444                                 size_t offs = le16_to_cpu(last->e_value_offs);
 445                                 if (offs < min_offs)
 446                                         min_offs = offs;
 447                         }
 448                         last = next;
 449                 }
 450
 451                 /* Check whether we have enough space left. */
 452                 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 453         } else {
 454                 /* We will use a new extended attribute block. */
 455                 free = sb->s_blocksize -
 456                         sizeof(struct ext2_xattr_header) - sizeof(__u32);
 457                 here = last = NULL;  /* avoid gcc uninitialized warning. */
 458         }
 459
 460         if (not_found) {
 461                 /* Request to remove a nonexistent attribute? */
 462                 error = -ENODATA;
 463                 if (flags & XATTR_REPLACE)
 464                         goto cleanup;
 465                 error = 0;
 466                 if (value == NULL)
 467                         goto cleanup;
 468         } else {
 469                 /* Request to create an existing attribute? */
 470                 error = -EEXIST;
 471                 if (flags & XATTR_CREATE)
 472                         goto cleanup;
 473                 if (!here->e_value_block && here->e_value_size) {
 474                         size_t size = le32_to_cpu(here->e_value_size);
 475
 476                         if (le16_to_cpu(here->e_value_offs) + size >
 477                             sb->s_blocksize || size > sb->s_blocksize)
 478                                 goto bad_block;
 479                         free += EXT2_XATTR_SIZE(size);
 480                 }
 481                 free += EXT2_XATTR_LEN(name_len);
 482         }
 483         error = -ENOSPC;
 484         if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 485                 goto cleanup;
 486
 487         /* Here we know that we can set the new attribute. */
 488
 489         if (header) {
 490                 struct mb_cache_entry *ce;
 491
 492                 /* assert(header == HDR(bh)); */
 493                 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
 494                                         bh->b_blocknr);
 495                 lock_buffer(bh);
 496                 if (header->h_refcount == cpu_to_le32(1)) {
 497                         ea_bdebug(bh, "modifying in-place");
 498                         if (ce)
 499                                 mb_cache_entry_free(ce);
 500                         /* keep the buffer locked while modifying it. */
 501                 } else {
 502                         int offset;
 503
 504                         if (ce)
 505                                 mb_cache_entry_release(ce);
 506                         unlock_buffer(bh);
 507                         ea_bdebug(bh, "cloning");
 508                         header = kmalloc(bh->b_size, GFP_KERNEL);
 509                         error = -ENOMEM;
 510                         if (header == NULL)
 511                                 goto cleanup;
 512                         memcpy(header, HDR(bh), bh->b_size);
 513                         header->h_refcount = cpu_to_le32(1);
 514
 515                         offset = (char *)here - bh->b_data;
 516                         here = ENTRY((char *)header + offset);
 517                         offset = (char *)last - bh->b_data;
 518                         last = ENTRY((char *)header + offset);
 519                 }
 520         } else {
 521                 /* Allocate a buffer where we construct the new block. */
 522                 header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 523                 error = -ENOMEM;
 524                 if (header == NULL)
 525                         goto cleanup;
 526                 end = (char *)header + sb->s_blocksize;
 527                 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 528                 header->h_blocks = header->h_refcount = cpu_to_le32(1);
 529                 last = here = ENTRY(header+1);
 530         }
 531
 532         /* Iff we are modifying the block in-place, bh is locked here. */
 533
 534         if (not_found) {
 535                 /* Insert the new name. */
 536                 size_t size = EXT2_XATTR_LEN(name_len);
 537                 size_t rest = (char *)last - (char *)here;
 538                 memmove((char *)here + size, here, rest);
 539                 memset(here, 0, size);
 540                 here->e_name_index = name_index;
 541                 here->e_name_len = name_len;
 542                 memcpy(here->e_name, name, name_len);
 543         } else {
 544                 if (!here->e_value_block && here->e_value_size) {
 545                         char *first_val = (char *)header + min_offs;
 546                         size_t offs = le16_to_cpu(here->e_value_offs);
 547                         char *val = (char *)header + offs;
 548                         size_t size = EXT2_XATTR_SIZE(
 549                                 le32_to_cpu(here->e_value_size));
 550
 551                         if (size == EXT2_XATTR_SIZE(value_len)) {
 552                                 /* The old and the new value have the same
 553                                    size. Just replace. */
 554                                 here->e_value_size = cpu_to_le32(value_len);
 555                                 memset(val + size - EXT2_XATTR_PAD, 0,
 556                                        EXT2_XATTR_PAD); /* Clear pad bytes. */
 557                                 memcpy(val, value, value_len);
 558                                 goto skip_replace;
 559                         }
 560
 561                         /* Remove the old value. */
 562                         memmove(first_val + size, first_val, val - first_val);
 563                         memset(first_val, 0, size);
 564                         here->e_value_offs = 0;
 565                         min_offs += size;
 566
 567                         /* Adjust all value offsets. */
 568                         last = ENTRY(header+1);
 569                         while (!IS_LAST_ENTRY(last)) {
 570                                 size_t o = le16_to_cpu(last->e_value_offs);
 571                                 if (!last->e_value_block && o < offs)
 572                                         last->e_value_offs =
 573                                                 cpu_to_le16(o + size);
 574                                 last = EXT2_XATTR_NEXT(last);
 575                         }
 576                 }
 577                 if (value == NULL) {
 578                         /* Remove the old name. */
 579                         size_t size = EXT2_XATTR_LEN(name_len);
 580                         last = ENTRY((char *)last - size);
 581                         memmove(here, (char*)here + size,
 582                                 (char*)last - (char*)here);
 583                         memset(last, 0, size);
 584                 }
 585         }
 586
 587         if (value != NULL) {
 588                 /* Insert the new value. */
 589                 here->e_value_size = cpu_to_le32(value_len);
 590                 if (value_len) {
 591                         size_t size = EXT2_XATTR_SIZE(value_len);
 592                         char *val = (char *)header + min_offs - size;
 593                         here->e_value_offs =
 594                                 cpu_to_le16((char *)val - (char *)header);
 595                         memset(val + size - EXT2_XATTR_PAD, 0,
 596                                EXT2_XATTR_PAD); /* Clear the pad bytes. */
 597                         memcpy(val, value, value_len);
 598                 }
 599         }
 600
 601 skip_replace:
 602         if (IS_LAST_ENTRY(ENTRY(header+1))) {
 603                 /* This block is now empty. */
 604                 if (bh && header == HDR(bh))
 605                         unlock_buffer(bh);  /* we were modifying in-place. */
 606                 error = ext2_xattr_set2(inode, bh, NULL);
 607         } else {
 608                 ext2_xattr_rehash(header, here);
 609                 if (bh && header == HDR(bh))
 610                         unlock_buffer(bh);  /* we were modifying in-place. */
 611                 error = ext2_xattr_set2(inode, bh, header);
 612         }
 613
 614 cleanup:
 615         brelse(bh);
 616         if (!(bh && header == HDR(bh)))
 617                 kfree(header);
 618         up_write(&EXT2_I(inode)->xattr_sem);
 619
 620         return error;
 621 }
 622
 623 /*
 624  * Second half of ext2_xattr_set(): Update the file system.
 625  */
 626 static int
 627 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 628                 struct ext2_xattr_header *header)
 629 {
 630         struct super_block *sb = inode->i_sb;
 631         struct buffer_head *new_bh = NULL;
 632         int error;
 633
 634         if (header) {
 635                 new_bh = ext2_xattr_cache_find(inode, header);
 636                 if (new_bh) {
 637                         /* We found an identical block in the cache. */
 638                         if (new_bh == old_bh) {
 639                                 ea_bdebug(new_bh, "keeping this block");
 640                         } else {
 641                                 /* The old block is released after updating
 642                                    the inode.  */
 643                                 ea_bdebug(new_bh, "reusing block");
 644
 645                                 error = -ENOSPC;
 646                                 if (DLIMIT_ALLOC_BLOCK(inode, 1))
 647                                         goto cleanup;
 648                                 error = -EDQUOT;
 649                                 if (DQUOT_ALLOC_BLOCK(inode, 1)) {
 650                                         DLIMIT_FREE_BLOCK(inode, 1);
 651                                         unlock_buffer(new_bh);
 652                                         goto cleanup;
 653                                 }
 654                                 HDR(new_bh)->h_refcount = cpu_to_le32(1 +
 655                                         le32_to_cpu(HDR(new_bh)->h_refcount));
 656                                 ea_bdebug(new_bh, "refcount now=%d",
 657                                         le32_to_cpu(HDR(new_bh)->h_refcount));
 658                         }
 659                         unlock_buffer(new_bh);
 660                 } else if (old_bh && header == HDR(old_bh)) {
 661                         /* Keep this block. No need to lock the block as we
 662                            don't need to change the reference count. */
 663                         new_bh = old_bh;
 664                         get_bh(new_bh);
 665                         ext2_xattr_cache_insert(new_bh);
 666                 } else {
 667                         /* We need to allocate a new block */
 668                         int goal = le32_to_cpu(EXT2_SB(sb)->s_es->
 669                                                            s_first_data_block) +
 670                                    EXT2_I(inode)->i_block_group *
 671                                    EXT2_BLOCKS_PER_GROUP(sb);
 672                         int block = ext2_new_block(inode, goal,
 673                                                    NULL, NULL, &error);
 674                         if (error)
 675                                 goto cleanup;
 676                         ea_idebug(inode, "creating block %d", block);
 677
 678                         new_bh = sb_getblk(sb, block);
 679                         if (!new_bh) {
 680                                 ext2_free_blocks(inode, block, 1);
 681                                 error = -EIO;
 682                                 goto cleanup;
 683                         }
 684                         lock_buffer(new_bh);
 685                         memcpy(new_bh->b_data, header, new_bh->b_size);
 686                         set_buffer_uptodate(new_bh);
 687                         unlock_buffer(new_bh);
 688                         ext2_xattr_cache_insert(new_bh);
 689
 690                         ext2_xattr_update_super_block(sb);
 691                 }
 692                 mark_buffer_dirty(new_bh);
 693                 if (IS_SYNC(inode)) {
 694                         sync_dirty_buffer(new_bh);
 695                         error = -EIO;
 696                         if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 697                                 goto cleanup;
 698                 }
 699         }
 700
 701         /* Update the inode. */
 702         EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 703         inode->i_ctime = CURRENT_TIME_SEC;
 704         if (IS_SYNC(inode)) {
 705                 error = ext2_sync_inode (inode);
 706                 /* In case sync failed due to ENOSPC the inode was actually
 707                  * written (only some dirty data were not) so we just proceed
 708                  * as if nothing happened and cleanup the unused block */
 709                 if (error && error != -ENOSPC) {
 710                         if (new_bh && new_bh != old_bh)
 711                                 DQUOT_FREE_BLOCK(inode, 1);
 712                         goto cleanup;
 713                 }
 714         } else
 715                 mark_inode_dirty(inode);
 716
 717         error = 0;
 718         if (old_bh && old_bh != new_bh) {
 719                 struct mb_cache_entry *ce;
 720
 721                 /*
 722                  * If there was an old block and we are no longer using it,
 723                  * release the old block.
 724                  */
 725                 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
 726                                         old_bh->b_blocknr);
 727                 lock_buffer(old_bh);
 728                 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 729                         /* Free the old block. */
 730                         if (ce)
 731                                 mb_cache_entry_free(ce);
 732                         ea_bdebug(old_bh, "freeing");
 733                         ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 734                         /* We let our caller release old_bh, so we
 735                          * need to duplicate the buffer before. */
 736                         get_bh(old_bh);
 737                         bforget(old_bh);
 738                 } else {
 739                         /* Decrement the refcount only. */
 740                         HDR(old_bh)->h_refcount = cpu_to_le32(
 741                                 le32_to_cpu(HDR(old_bh)->h_refcount) - 1);
 742                         if (ce)
 743                                 mb_cache_entry_release(ce);
 744                         DLIMIT_FREE_BLOCK(inode, 1);
 745                         DQUOT_FREE_BLOCK(inode, 1);
 746                         mark_buffer_dirty(old_bh);
 747                         ea_bdebug(old_bh, "refcount now=%d",
 748                                 le32_to_cpu(HDR(old_bh)->h_refcount));
 749                 }
 750                 unlock_buffer(old_bh);
 751         }
 752
 753 cleanup:
 754         brelse(new_bh);
 755
 756         return error;
 757 }
 758
 759 /*
 760  * ext2_xattr_delete_inode()
 761  *
 762  * Free extended attribute resources associated with this inode. This
 763  * is called immediately before an inode is freed.
 764  */
 765 void
 766 ext2_xattr_delete_inode(struct inode *inode)
 767 {
 768         struct buffer_head *bh = NULL;
 769         struct mb_cache_entry *ce;
 770
 771         down_write(&EXT2_I(inode)->xattr_sem);
 772         if (!EXT2_I(inode)->i_file_acl)
 773                 goto cleanup;
 774         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 775         if (!bh) {
 776                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 777                         "inode %ld: block %d read error", inode->i_ino,
 778                         EXT2_I(inode)->i_file_acl);
 779                 goto cleanup;
 780         }
 781         ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 782         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 783             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 784                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 785                         "inode %ld: bad block %d", inode->i_ino,
 786                         EXT2_I(inode)->i_file_acl);
 787                 goto cleanup;
 788         }
 789         ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
 790         lock_buffer(bh);
 791         if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 792                 if (ce)
 793                         mb_cache_entry_free(ce);
 794                 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 795                 get_bh(bh);
 796                 bforget(bh);
 797                 unlock_buffer(bh);
 798         } else {
 799                 HDR(bh)->h_refcount = cpu_to_le32(
 800                         le32_to_cpu(HDR(bh)->h_refcount) - 1);
 801                 if (ce)
 802                         mb_cache_entry_release(ce);
 803                 ea_bdebug(bh, "refcount now=%d",
 804                         le32_to_cpu(HDR(bh)->h_refcount));
 805                 unlock_buffer(bh);
 806                 mark_buffer_dirty(bh);
 807                 if (IS_SYNC(inode))
 808                         sync_dirty_buffer(bh);
 809                 DLIMIT_FREE_BLOCK(inode, 1);
 810                 DQUOT_FREE_BLOCK(inode, 1);
 811         }
 812         EXT2_I(inode)->i_file_acl = 0;
 813
 814 cleanup:
 815         brelse(bh);
 816         up_write(&EXT2_I(inode)->xattr_sem);
 817 }
 818
 819 /*
 820  * ext2_xattr_put_super()
 821  *
 822  * This is called when a file system is unmounted.
 823  */
 824 void
 825 ext2_xattr_put_super(struct super_block *sb)
 826 {
 827         mb_cache_shrink(sb->s_bdev);
 828 }
 829
 830
 831 /*
 832  * ext2_xattr_cache_insert()
 833  *
 834  * Create a new entry in the extended attribute cache, and insert
 835  * it unless such an entry is already in the cache.
 836  *
 837  * Returns 0, or a negative error number on failure.
 838  */
 839 static int
 840 ext2_xattr_cache_insert(struct buffer_head *bh)
 841 {
 842         __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 843         struct mb_cache_entry *ce;
 844         int error;
 845
 846         ce = mb_cache_entry_alloc(ext2_xattr_cache);
 847         if (!ce)
 848                 return -ENOMEM;
 849         error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
 850         if (error) {
 851                 mb_cache_entry_free(ce);
 852                 if (error == -EBUSY) {
 853                         ea_bdebug(bh, "already in cache (%d cache entries)",
 854                                 atomic_read(&ext2_xattr_cache->c_entry_count));
 855                         error = 0;
 856                 }
 857         } else {
 858                 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
 859                           atomic_read(&ext2_xattr_cache->c_entry_count));
 860                 mb_cache_entry_release(ce);
 861         }
 862         return error;
 863 }
 864
 865 /*
 866  * ext2_xattr_cmp()
 867  *
 868  * Compare two extended attribute blocks for equality.
 869  *
 870  * Returns 0 if the blocks are equal, 1 if they differ, and
 871  * a negative error number on errors.
 872  */
 873 static int
 874 ext2_xattr_cmp(struct ext2_xattr_header *header1,
 875                struct ext2_xattr_header *header2)
 876 {
 877         struct ext2_xattr_entry *entry1, *entry2;
 878
 879         entry1 = ENTRY(header1+1);
 880         entry2 = ENTRY(header2+1);
 881         while (!IS_LAST_ENTRY(entry1)) {
 882                 if (IS_LAST_ENTRY(entry2))
 883                         return 1;
 884                 if (entry1->e_hash != entry2->e_hash ||
 885                     entry1->e_name_index != entry2->e_name_index ||
 886                     entry1->e_name_len != entry2->e_name_len ||
 887                     entry1->e_value_size != entry2->e_value_size ||
 888                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 889                         return 1;
 890                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 891                         return -EIO;
 892                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 893                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 894                            le32_to_cpu(entry1->e_value_size)))
 895                         return 1;
 896
 897                 entry1 = EXT2_XATTR_NEXT(entry1);
 898                 entry2 = EXT2_XATTR_NEXT(entry2);
 899         }
 900         if (!IS_LAST_ENTRY(entry2))
 901                 return 1;
 902         return 0;
 903 }
 904
 905 /*
 906  * ext2_xattr_cache_find()
 907  *
 908  * Find an identical extended attribute block.
 909  *
 910  * Returns a locked buffer head to the block found, or NULL if such
 911  * a block was not found or an error occurred.
 912  */
 913 static struct buffer_head *
 914 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 915 {
 916         __u32 hash = le32_to_cpu(header->h_hash);
 917         struct mb_cache_entry *ce;
 918
 919         if (!header->h_hash)
 920                 return NULL;  /* never share */
 921         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 922 again:
 923         ce = mb_cache_entry_find_first(ext2_xattr_cache, 0,
 924                                        inode->i_sb->s_bdev, hash);
 925         while (ce) {
 926                 struct buffer_head *bh;
 927
 928                 if (IS_ERR(ce)) {
 929                         if (PTR_ERR(ce) == -EAGAIN)
 930                                 goto again;
 931                         break;
 932                 }
 933
 934                 bh = sb_bread(inode->i_sb, ce->e_block);
 935                 if (!bh) {
 936                         ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 937                                 "inode %ld: block %ld read error",
 938                                 inode->i_ino, (unsigned long) ce->e_block);
 939                 } else {
 940                         lock_buffer(bh);
 941                         if (le32_to_cpu(HDR(bh)->h_refcount) >
 942                                    EXT2_XATTR_REFCOUNT_MAX) {
 943                                 ea_idebug(inode, "block %ld refcount %d>%d",
 944                                           (unsigned long) ce->e_block,
 945                                           le32_to_cpu(HDR(bh)->h_refcount),
 946                                           EXT2_XATTR_REFCOUNT_MAX);
 947                         } else if (!ext2_xattr_cmp(header, HDR(bh))) {
 948                                 ea_bdebug(bh, "b_count=%d",
 949                                           atomic_read(&(bh->b_count)));
 950                                 mb_cache_entry_release(ce);
 951                                 return bh;
 952                         }
 953                         unlock_buffer(bh);
 954                         brelse(bh);
 955                 }
 956                 ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
 957         }
 958         return NULL;
 959 }
 960
 961 #define NAME_HASH_SHIFT 5
 962 #define VALUE_HASH_SHIFT 16
 963
 964 /*
 965  * ext2_xattr_hash_entry()
 966  *
 967  * Compute the hash of an extended attribute.
 968  */
 969 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 970                                          struct ext2_xattr_entry *entry)
 971 {
 972         __u32 hash = 0;
 973         char *name = entry->e_name;
 974         int n;
 975
 976         for (n=0; n < entry->e_name_len; n++) {
 977                 hash = (hash << NAME_HASH_SHIFT) ^
 978                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 979                        *name++;
 980         }
 981
 982         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 983                 __le32 *value = (__le32 *)((char *)header +
 984                         le16_to_cpu(entry->e_value_offs));
 985                 for (n = (le32_to_cpu(entry->e_value_size) +
 986                      EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 987                         hash = (hash << VALUE_HASH_SHIFT) ^
 988                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 989                                le32_to_cpu(*value++);
 990                 }
 991         }
 992         entry->e_hash = cpu_to_le32(hash);
 993 }
 994
 995 #undef NAME_HASH_SHIFT
 996 #undef VALUE_HASH_SHIFT
 997
 998 #define BLOCK_HASH_SHIFT 16
 999
1000 /*
1001  * ext2_xattr_rehash()
1002  *
1003  * Re-compute the extended attribute hash value after an entry has changed.
1004  */
1005 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1006                               struct ext2_xattr_entry *entry)
1007 {
1008         struct ext2_xattr_entry *here;
1009         __u32 hash = 0;
1010
1011         ext2_xattr_hash_entry(header, entry);
1012         here = ENTRY(header+1);
1013         while (!IS_LAST_ENTRY(here)) {
1014                 if (!here->e_hash) {
1015                         /* Block is not shared if an entry's hash value == 0 */
1016                         hash = 0;
1017                         break;
1018                 }
1019                 hash = (hash << BLOCK_HASH_SHIFT) ^
1020                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1021                        le32_to_cpu(here->e_hash);
1022                 here = EXT2_XATTR_NEXT(here);
1023         }
1024         header->h_hash = cpu_to_le32(hash);
1025 }
1026
1027 #undef BLOCK_HASH_SHIFT
1028
1029 int __init
1030 init_ext2_xattr(void)
1031 {
1032         ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL,
1033                 sizeof(struct mb_cache_entry) +
1034                 sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1035         if (!ext2_xattr_cache)
1036                 return -ENOMEM;
1037         return 0;
1038 }
1039
1040 void
1041 exit_ext2_xattr(void)
1042 {
1043         mb_cache_destroy(ext2_xattr_cache);
1044 }