fs/ext2/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/fs/ext2/xattr.c
   4  *
   5  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
   6  *
   7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   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  *
  13  */
  14
  15 /*
  16  * Extended attributes are stored on disk blocks allocated outside of
  17  * any inode. The i_file_acl field is then made to point to this allocated
  18  * block. If all extended attributes of an inode are identical, these
  19  * inodes may share the same extended attribute block. Such situations
  20  * are automatically detected by keeping a cache of recent attribute block
  21  * numbers and hashes over the block's contents in memory.
  22  *
  23  *
  24  * Extended attribute block layout:
  25  *
  26  *   +------------------+
  27  *   | header           |
  28  *   | entry 1          | |
  29  *   | entry 2          | | growing downwards
  30  *   | entry 3          | v
  31  *   | four null bytes  |
  32  *   | . . .            |
  33  *   | value 1          | ^
  34  *   | value 3          | | growing upwards
  35  *   | value 2          | |
  36  *   +------------------+
  37  *
  38  * The block header is followed by multiple entry descriptors. These entry
  39  * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
  40  * byte boundaries. The entry descriptors are sorted by attribute name,
  41  * so that two extended attribute blocks can be compared efficiently.
  42  *
  43  * Attribute values are aligned to the end of the block, stored in
  44  * no specific order. They are also padded to EXT2_XATTR_PAD byte
  45  * boundaries. No additional gaps are left between them.
  46  *
  47  * Locking strategy
  48  * ----------------
  49  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
  50  * EA blocks are only changed if they are exclusive to an inode, so
  51  * holding xattr_sem also means that nothing but the EA block's reference
  52  * count will change. Multiple writers to an EA block are synchronized
  53  * by the bh lock. No more than a single bh lock is held at any time
  54  * to avoid deadlocks.
  55  */
  56
  57 #include <linux/buffer_head.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/security.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                 printk(KERN_DEBUG "block %pg:%lu: ", \
  82                         bh->b_bdev, (unsigned long) bh->b_blocknr); \
  83                 printk(f); \
  84                 printk("\n"); \
  85         } while (0)
  86 #else
  87 # define ea_idebug(f...)
  88 # define ea_bdebug(f...)
  89 #endif
  90
  91 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  92                            struct ext2_xattr_header *);
  93
  94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  95 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  96                                                  struct ext2_xattr_header *);
  97 static void ext2_xattr_rehash(struct ext2_xattr_header *,
  98                               struct ext2_xattr_entry *);
  99
 100 static const struct xattr_handler *ext2_xattr_handler_map[] = {
 101         [EXT2_XATTR_INDEX_USER]              = &ext2_xattr_user_handler,
 102 #ifdef CONFIG_EXT2_FS_POSIX_ACL
 103         [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
 104         [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
 105 #endif
 106         [EXT2_XATTR_INDEX_TRUSTED]           = &ext2_xattr_trusted_handler,
 107 #ifdef CONFIG_EXT2_FS_SECURITY
 108         [EXT2_XATTR_INDEX_SECURITY]          = &ext2_xattr_security_handler,
 109 #endif
 110 };
 111
 112 const struct xattr_handler *ext2_xattr_handlers[] = {
 113         &ext2_xattr_user_handler,
 114         &ext2_xattr_trusted_handler,
 115 #ifdef CONFIG_EXT2_FS_POSIX_ACL
 116         &posix_acl_access_xattr_handler,
 117         &posix_acl_default_xattr_handler,
 118 #endif
 119 #ifdef CONFIG_EXT2_FS_SECURITY
 120         &ext2_xattr_security_handler,
 121 #endif
 122         NULL
 123 };
 124
 125 #define EA_BLOCK_CACHE(inode)   (EXT2_SB(inode->i_sb)->s_ea_block_cache)
 126
 127 static inline const struct xattr_handler *
 128 ext2_xattr_handler(int name_index)
 129 {
 130         const struct xattr_handler *handler = NULL;
 131
 132         if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 133                 handler = ext2_xattr_handler_map[name_index];
 134         return handler;
 135 }
 136
 137 /*
 138  * ext2_xattr_get()
 139  *
 140  * Copy an extended attribute into the buffer
 141  * provided, or compute the buffer size required.
 142  * Buffer is NULL to compute the size of the buffer required.
 143  *
 144  * Returns a negative error number on failure, or the number of bytes
 145  * used / required on success.
 146  */
 147 int
 148 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 149                void *buffer, size_t buffer_size)
 150 {
 151         struct buffer_head *bh = NULL;
 152         struct ext2_xattr_entry *entry;
 153         size_t name_len, size;
 154         char *end;
 155         int error;
 156         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 157
 158         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 159                   name_index, name, buffer, (long)buffer_size);
 160
 161         if (name == NULL)
 162                 return -EINVAL;
 163         name_len = strlen(name);
 164         if (name_len > 255)
 165                 return -ERANGE;
 166
 167         down_read(&EXT2_I(inode)->xattr_sem);
 168         error = -ENODATA;
 169         if (!EXT2_I(inode)->i_file_acl)
 170                 goto cleanup;
 171         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 172         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 173         error = -EIO;
 174         if (!bh)
 175                 goto cleanup;
 176         ea_bdebug(bh, "b_count=%d, refcount=%d",
 177                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 178         end = bh->b_data + bh->b_size;
 179         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 180             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 181 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_get",
 182                         "inode %ld: bad block %d", inode->i_ino,
 183                         EXT2_I(inode)->i_file_acl);
 184                 error = -EIO;
 185                 goto cleanup;
 186         }
 187
 188         /* find named attribute */
 189         entry = FIRST_ENTRY(bh);
 190         while (!IS_LAST_ENTRY(entry)) {
 191                 struct ext2_xattr_entry *next =
 192                         EXT2_XATTR_NEXT(entry);
 193                 if ((char *)next >= end)
 194                         goto bad_block;
 195                 if (name_index == entry->e_name_index &&
 196                     name_len == entry->e_name_len &&
 197                     memcmp(name, entry->e_name, name_len) == 0)
 198                         goto found;
 199                 entry = next;
 200         }
 201         if (ext2_xattr_cache_insert(ea_block_cache, bh))
 202                 ea_idebug(inode, "cache insert failed");
 203         error = -ENODATA;
 204         goto cleanup;
 205 found:
 206         /* check the buffer size */
 207         if (entry->e_value_block != 0)
 208                 goto bad_block;
 209         size = le32_to_cpu(entry->e_value_size);
 210         if (size > inode->i_sb->s_blocksize ||
 211             le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
 212                 goto bad_block;
 213
 214         if (ext2_xattr_cache_insert(ea_block_cache, bh))
 215                 ea_idebug(inode, "cache insert failed");
 216         if (buffer) {
 217                 error = -ERANGE;
 218                 if (size > buffer_size)
 219                         goto cleanup;
 220                 /* return value of attribute */
 221                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 222                         size);
 223         }
 224         error = size;
 225
 226 cleanup:
 227         brelse(bh);
 228         up_read(&EXT2_I(inode)->xattr_sem);
 229
 230         return error;
 231 }
 232
 233 /*
 234  * ext2_xattr_list()
 235  *
 236  * Copy a list of attribute names into the buffer
 237  * provided, or compute the buffer size required.
 238  * Buffer is NULL to compute the size of the buffer required.
 239  *
 240  * Returns a negative error number on failure, or the number of bytes
 241  * used / required on success.
 242  */
 243 static int
 244 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 245 {
 246         struct inode *inode = d_inode(dentry);
 247         struct buffer_head *bh = NULL;
 248         struct ext2_xattr_entry *entry;
 249         char *end;
 250         size_t rest = buffer_size;
 251         int error;
 252         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 253
 254         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 255                   buffer, (long)buffer_size);
 256
 257         down_read(&EXT2_I(inode)->xattr_sem);
 258         error = 0;
 259         if (!EXT2_I(inode)->i_file_acl)
 260                 goto cleanup;
 261         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 262         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 263         error = -EIO;
 264         if (!bh)
 265                 goto cleanup;
 266         ea_bdebug(bh, "b_count=%d, refcount=%d",
 267                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 268         end = bh->b_data + bh->b_size;
 269         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 270             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 271 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_list",
 272                         "inode %ld: bad block %d", inode->i_ino,
 273                         EXT2_I(inode)->i_file_acl);
 274                 error = -EIO;
 275                 goto cleanup;
 276         }
 277
 278         /* check the on-disk data structure */
 279         entry = FIRST_ENTRY(bh);
 280         while (!IS_LAST_ENTRY(entry)) {
 281                 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
 282
 283                 if ((char *)next >= end)
 284                         goto bad_block;
 285                 entry = next;
 286         }
 287         if (ext2_xattr_cache_insert(ea_block_cache, bh))
 288                 ea_idebug(inode, "cache insert failed");
 289
 290         /* list the attribute names */
 291         for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 292              entry = EXT2_XATTR_NEXT(entry)) {
 293                 const struct xattr_handler *handler =
 294                         ext2_xattr_handler(entry->e_name_index);
 295
 296                 if (handler && (!handler->list || handler->list(dentry))) {
 297                         const char *prefix = handler->prefix ?: handler->name;
 298                         size_t prefix_len = strlen(prefix);
 299                         size_t size = prefix_len + entry->e_name_len + 1;
 300
 301                         if (buffer) {
 302                                 if (size > rest) {
 303                                         error = -ERANGE;
 304                                         goto cleanup;
 305                                 }
 306                                 memcpy(buffer, prefix, prefix_len);
 307                                 buffer += prefix_len;
 308                                 memcpy(buffer, entry->e_name, entry->e_name_len);
 309                                 buffer += entry->e_name_len;
 310                                 *buffer++ = 0;
 311                         }
 312                         rest -= size;
 313                 }
 314         }
 315         error = buffer_size - rest;  /* total size */
 316
 317 cleanup:
 318         brelse(bh);
 319         up_read(&EXT2_I(inode)->xattr_sem);
 320
 321         return error;
 322 }
 323
 324 /*
 325  * Inode operation listxattr()
 326  *
 327  * d_inode(dentry)->i_mutex: don't care
 328  */
 329 ssize_t
 330 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 331 {
 332         return ext2_xattr_list(dentry, buffer, size);
 333 }
 334
 335 /*
 336  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 337  * not set, set it.
 338  */
 339 static void ext2_xattr_update_super_block(struct super_block *sb)
 340 {
 341         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 342                 return;
 343
 344         spin_lock(&EXT2_SB(sb)->s_lock);
 345         EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 346         spin_unlock(&EXT2_SB(sb)->s_lock);
 347         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 348 }
 349
 350 /*
 351  * ext2_xattr_set()
 352  *
 353  * Create, replace or remove an extended attribute for this inode.  Value
 354  * is NULL to remove an existing extended attribute, and non-NULL to
 355  * either replace an existing extended attribute, or create a new extended
 356  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 357  * specify that an extended attribute must exist and must not exist
 358  * previous to the call, respectively.
 359  *
 360  * Returns 0, or a negative error number on failure.
 361  */
 362 int
 363 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 364                const void *value, size_t value_len, int flags)
 365 {
 366         struct super_block *sb = inode->i_sb;
 367         struct buffer_head *bh = NULL;
 368         struct ext2_xattr_header *header = NULL;
 369         struct ext2_xattr_entry *here, *last;
 370         size_t name_len, free, min_offs = sb->s_blocksize;
 371         int not_found = 1, error;
 372         char *end;
 373
 374         /*
 375          * header -- Points either into bh, or to a temporarily
 376          *           allocated buffer.
 377          * here -- The named entry found, or the place for inserting, within
 378          *         the block pointed to by header.
 379          * last -- Points right after the last named entry within the block
 380          *         pointed to by header.
 381          * min_offs -- The offset of the first value (values are aligned
 382          *             towards the end of the block).
 383          * end -- Points right after the block pointed to by header.
 384          */
 385
 386         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 387                   name_index, name, value, (long)value_len);
 388
 389         if (value == NULL)
 390                 value_len = 0;
 391         if (name == NULL)
 392                 return -EINVAL;
 393         name_len = strlen(name);
 394         if (name_len > 255 || value_len > sb->s_blocksize)
 395                 return -ERANGE;
 396         down_write(&EXT2_I(inode)->xattr_sem);
 397         if (EXT2_I(inode)->i_file_acl) {
 398                 /* The inode already has an extended attribute block. */
 399                 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 400                 error = -EIO;
 401                 if (!bh)
 402                         goto cleanup;
 403                 ea_bdebug(bh, "b_count=%d, refcount=%d",
 404                         atomic_read(&(bh->b_count)),
 405                         le32_to_cpu(HDR(bh)->h_refcount));
 406                 header = HDR(bh);
 407                 end = bh->b_data + bh->b_size;
 408                 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 409                     header->h_blocks != cpu_to_le32(1)) {
 410 bad_block:              ext2_error(sb, "ext2_xattr_set",
 411                                 "inode %ld: bad block %d", inode->i_ino,
 412                                    EXT2_I(inode)->i_file_acl);
 413                         error = -EIO;
 414                         goto cleanup;
 415                 }
 416                 /* Find the named attribute. */
 417                 here = FIRST_ENTRY(bh);
 418                 while (!IS_LAST_ENTRY(here)) {
 419                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
 420                         if ((char *)next >= end)
 421                                 goto bad_block;
 422                         if (!here->e_value_block && here->e_value_size) {
 423                                 size_t offs = le16_to_cpu(here->e_value_offs);
 424                                 if (offs < min_offs)
 425                                         min_offs = offs;
 426                         }
 427                         not_found = name_index - here->e_name_index;
 428                         if (!not_found)
 429                                 not_found = name_len - here->e_name_len;
 430                         if (!not_found)
 431                                 not_found = memcmp(name, here->e_name,name_len);
 432                         if (not_found <= 0)
 433                                 break;
 434                         here = next;
 435                 }
 436                 last = here;
 437                 /* We still need to compute min_offs and last. */
 438                 while (!IS_LAST_ENTRY(last)) {
 439                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
 440                         if ((char *)next >= end)
 441                                 goto bad_block;
 442                         if (!last->e_value_block && last->e_value_size) {
 443                                 size_t offs = le16_to_cpu(last->e_value_offs);
 444                                 if (offs < min_offs)
 445                                         min_offs = offs;
 446                         }
 447                         last = next;
 448                 }
 449
 450                 /* Check whether we have enough space left. */
 451                 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 452         } else {
 453                 /* We will use a new extended attribute block. */
 454                 free = sb->s_blocksize -
 455                         sizeof(struct ext2_xattr_header) - sizeof(__u32);
 456                 here = last = NULL;  /* avoid gcc uninitialized warning. */
 457         }
 458
 459         if (not_found) {
 460                 /* Request to remove a nonexistent attribute? */
 461                 error = -ENODATA;
 462                 if (flags & XATTR_REPLACE)
 463                         goto cleanup;
 464                 error = 0;
 465                 if (value == NULL)
 466                         goto cleanup;
 467         } else {
 468                 /* Request to create an existing attribute? */
 469                 error = -EEXIST;
 470                 if (flags & XATTR_CREATE)
 471                         goto cleanup;
 472                 if (!here->e_value_block && here->e_value_size) {
 473                         size_t size = le32_to_cpu(here->e_value_size);
 474
 475                         if (le16_to_cpu(here->e_value_offs) + size >
 476                             sb->s_blocksize || size > sb->s_blocksize)
 477                                 goto bad_block;
 478                         free += EXT2_XATTR_SIZE(size);
 479                 }
 480                 free += EXT2_XATTR_LEN(name_len);
 481         }
 482         error = -ENOSPC;
 483         if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 484                 goto cleanup;
 485
 486         /* Here we know that we can set the new attribute. */
 487
 488         if (header) {
 489                 /* assert(header == HDR(bh)); */
 490                 lock_buffer(bh);
 491                 if (header->h_refcount == cpu_to_le32(1)) {
 492                         __u32 hash = le32_to_cpu(header->h_hash);
 493
 494                         ea_bdebug(bh, "modifying in-place");
 495                         /*
 496                          * This must happen under buffer lock for
 497                          * ext2_xattr_set2() to reliably detect modified block
 498                          */
 499                         mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 500                                               bh->b_blocknr);
 501
 502                         /* keep the buffer locked while modifying it. */
 503                 } else {
 504                         int offset;
 505
 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         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 634
 635         if (header) {
 636                 new_bh = ext2_xattr_cache_find(inode, header);
 637                 if (new_bh) {
 638                         /* We found an identical block in the cache. */
 639                         if (new_bh == old_bh) {
 640                                 ea_bdebug(new_bh, "keeping this block");
 641                         } else {
 642                                 /* The old block is released after updating
 643                                    the inode.  */
 644                                 ea_bdebug(new_bh, "reusing block");
 645
 646                                 error = dquot_alloc_block(inode, 1);
 647                                 if (error) {
 648                                         unlock_buffer(new_bh);
 649                                         goto cleanup;
 650                                 }
 651                                 le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 652                                 ea_bdebug(new_bh, "refcount now=%d",
 653                                         le32_to_cpu(HDR(new_bh)->h_refcount));
 654                         }
 655                         unlock_buffer(new_bh);
 656                 } else if (old_bh && header == HDR(old_bh)) {
 657                         /* Keep this block. No need to lock the block as we
 658                            don't need to change the reference count. */
 659                         new_bh = old_bh;
 660                         get_bh(new_bh);
 661                         ext2_xattr_cache_insert(ea_block_cache, new_bh);
 662                 } else {
 663                         /* We need to allocate a new block */
 664                         ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 665                                                 EXT2_I(inode)->i_block_group);
 666                         int block = ext2_new_block(inode, goal, &error);
 667                         if (error)
 668                                 goto cleanup;
 669                         ea_idebug(inode, "creating block %d", block);
 670
 671                         new_bh = sb_getblk(sb, block);
 672                         if (unlikely(!new_bh)) {
 673                                 ext2_free_blocks(inode, block, 1);
 674                                 mark_inode_dirty(inode);
 675                                 error = -ENOMEM;
 676                                 goto cleanup;
 677                         }
 678                         lock_buffer(new_bh);
 679                         memcpy(new_bh->b_data, header, new_bh->b_size);
 680                         set_buffer_uptodate(new_bh);
 681                         unlock_buffer(new_bh);
 682                         ext2_xattr_cache_insert(ea_block_cache, new_bh);
 683
 684                         ext2_xattr_update_super_block(sb);
 685                 }
 686                 mark_buffer_dirty(new_bh);
 687                 if (IS_SYNC(inode)) {
 688                         sync_dirty_buffer(new_bh);
 689                         error = -EIO;
 690                         if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 691                                 goto cleanup;
 692                 }
 693         }
 694
 695         /* Update the inode. */
 696         EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 697         inode->i_ctime = current_time(inode);
 698         if (IS_SYNC(inode)) {
 699                 error = sync_inode_metadata(inode, 1);
 700                 /* In case sync failed due to ENOSPC the inode was actually
 701                  * written (only some dirty data were not) so we just proceed
 702                  * as if nothing happened and cleanup the unused block */
 703                 if (error && error != -ENOSPC) {
 704                         if (new_bh && new_bh != old_bh) {
 705                                 dquot_free_block_nodirty(inode, 1);
 706                                 mark_inode_dirty(inode);
 707                         }
 708                         goto cleanup;
 709                 }
 710         } else
 711                 mark_inode_dirty(inode);
 712
 713         error = 0;
 714         if (old_bh && old_bh != new_bh) {
 715                 /*
 716                  * If there was an old block and we are no longer using it,
 717                  * release the old block.
 718                  */
 719                 lock_buffer(old_bh);
 720                 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 721                         __u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
 722
 723                         /*
 724                          * This must happen under buffer lock for
 725                          * ext2_xattr_set2() to reliably detect freed block
 726                          */
 727                         mb_cache_entry_delete(ea_block_cache, hash,
 728                                               old_bh->b_blocknr);
 729                         /* Free the old block. */
 730                         ea_bdebug(old_bh, "freeing");
 731                         ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 732                         mark_inode_dirty(inode);
 733                         /* We let our caller release old_bh, so we
 734                          * need to duplicate the buffer before. */
 735                         get_bh(old_bh);
 736                         bforget(old_bh);
 737                 } else {
 738                         /* Decrement the refcount only. */
 739                         le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 740                         dquot_free_block_nodirty(inode, 1);
 741                         mark_inode_dirty(inode);
 742                         mark_buffer_dirty(old_bh);
 743                         ea_bdebug(old_bh, "refcount now=%d",
 744                                 le32_to_cpu(HDR(old_bh)->h_refcount));
 745                 }
 746                 unlock_buffer(old_bh);
 747         }
 748
 749 cleanup:
 750         brelse(new_bh);
 751
 752         return error;
 753 }
 754
 755 /*
 756  * ext2_xattr_delete_inode()
 757  *
 758  * Free extended attribute resources associated with this inode. This
 759  * is called immediately before an inode is freed.
 760  */
 761 void
 762 ext2_xattr_delete_inode(struct inode *inode)
 763 {
 764         struct buffer_head *bh = NULL;
 765         struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 766
 767         down_write(&EXT2_I(inode)->xattr_sem);
 768         if (!EXT2_I(inode)->i_file_acl)
 769                 goto cleanup;
 770
 771         if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
 772                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 773                         "inode %ld: xattr block %d is out of data blocks range",
 774                         inode->i_ino, EXT2_I(inode)->i_file_acl);
 775                 goto cleanup;
 776         }
 777
 778         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 779         if (!bh) {
 780                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 781                         "inode %ld: block %d read error", inode->i_ino,
 782                         EXT2_I(inode)->i_file_acl);
 783                 goto cleanup;
 784         }
 785         ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 786         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 787             HDR(bh)->h_blocks != cpu_to_le32(1)) {
 788                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 789                         "inode %ld: bad block %d", inode->i_ino,
 790                         EXT2_I(inode)->i_file_acl);
 791                 goto cleanup;
 792         }
 793         lock_buffer(bh);
 794         if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 795                 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 796
 797                 /*
 798                  * This must happen under buffer lock for ext2_xattr_set2() to
 799                  * reliably detect freed block
 800                  */
 801                 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 802                                       bh->b_blocknr);
 803                 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 804                 get_bh(bh);
 805                 bforget(bh);
 806                 unlock_buffer(bh);
 807         } else {
 808                 le32_add_cpu(&HDR(bh)->h_refcount, -1);
 809                 ea_bdebug(bh, "refcount now=%d",
 810                         le32_to_cpu(HDR(bh)->h_refcount));
 811                 unlock_buffer(bh);
 812                 mark_buffer_dirty(bh);
 813                 if (IS_SYNC(inode))
 814                         sync_dirty_buffer(bh);
 815                 dquot_free_block_nodirty(inode, 1);
 816         }
 817         EXT2_I(inode)->i_file_acl = 0;
 818
 819 cleanup:
 820         brelse(bh);
 821         up_write(&EXT2_I(inode)->xattr_sem);
 822 }
 823
 824 /*
 825  * ext2_xattr_cache_insert()
 826  *
 827  * Create a new entry in the extended attribute cache, and insert
 828  * it unless such an entry is already in the cache.
 829  *
 830  * Returns 0, or a negative error number on failure.
 831  */
 832 static int
 833 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 834 {
 835         __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 836         int error;
 837
 838         error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 1);
 839         if (error) {
 840                 if (error == -EBUSY) {
 841                         ea_bdebug(bh, "already in cache (%d cache entries)",
 842                                 atomic_read(&ext2_xattr_cache->c_entry_count));
 843                         error = 0;
 844                 }
 845         } else
 846                 ea_bdebug(bh, "inserting [%x]", (int)hash);
 847         return error;
 848 }
 849
 850 /*
 851  * ext2_xattr_cmp()
 852  *
 853  * Compare two extended attribute blocks for equality.
 854  *
 855  * Returns 0 if the blocks are equal, 1 if they differ, and
 856  * a negative error number on errors.
 857  */
 858 static int
 859 ext2_xattr_cmp(struct ext2_xattr_header *header1,
 860                struct ext2_xattr_header *header2)
 861 {
 862         struct ext2_xattr_entry *entry1, *entry2;
 863
 864         entry1 = ENTRY(header1+1);
 865         entry2 = ENTRY(header2+1);
 866         while (!IS_LAST_ENTRY(entry1)) {
 867                 if (IS_LAST_ENTRY(entry2))
 868                         return 1;
 869                 if (entry1->e_hash != entry2->e_hash ||
 870                     entry1->e_name_index != entry2->e_name_index ||
 871                     entry1->e_name_len != entry2->e_name_len ||
 872                     entry1->e_value_size != entry2->e_value_size ||
 873                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 874                         return 1;
 875                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 876                         return -EIO;
 877                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 878                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 879                            le32_to_cpu(entry1->e_value_size)))
 880                         return 1;
 881
 882                 entry1 = EXT2_XATTR_NEXT(entry1);
 883                 entry2 = EXT2_XATTR_NEXT(entry2);
 884         }
 885         if (!IS_LAST_ENTRY(entry2))
 886                 return 1;
 887         return 0;
 888 }
 889
 890 /*
 891  * ext2_xattr_cache_find()
 892  *
 893  * Find an identical extended attribute block.
 894  *
 895  * Returns a locked buffer head to the block found, or NULL if such
 896  * a block was not found or an error occurred.
 897  */
 898 static struct buffer_head *
 899 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 900 {
 901         __u32 hash = le32_to_cpu(header->h_hash);
 902         struct mb_cache_entry *ce;
 903         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 904
 905         if (!header->h_hash)
 906                 return NULL;  /* never share */
 907         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 908 again:
 909         ce = mb_cache_entry_find_first(ea_block_cache, hash);
 910         while (ce) {
 911                 struct buffer_head *bh;
 912
 913                 bh = sb_bread(inode->i_sb, ce->e_value);
 914                 if (!bh) {
 915                         ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 916                                 "inode %ld: block %ld read error",
 917                                 inode->i_ino, (unsigned long) ce->e_value);
 918                 } else {
 919                         lock_buffer(bh);
 920                         /*
 921                          * We have to be careful about races with freeing or
 922                          * rehashing of xattr block. Once we hold buffer lock
 923                          * xattr block's state is stable so we can check
 924                          * whether the block got freed / rehashed or not.
 925                          * Since we unhash mbcache entry under buffer lock when
 926                          * freeing / rehashing xattr block, checking whether
 927                          * entry is still hashed is reliable.
 928                          */
 929                         if (hlist_bl_unhashed(&ce->e_hash_list)) {
 930                                 mb_cache_entry_put(ea_block_cache, ce);
 931                                 unlock_buffer(bh);
 932                                 brelse(bh);
 933                                 goto again;
 934                         } else if (le32_to_cpu(HDR(bh)->h_refcount) >
 935                                    EXT2_XATTR_REFCOUNT_MAX) {
 936                                 ea_idebug(inode, "block %ld refcount %d>%d",
 937                                           (unsigned long) ce->e_value,
 938                                           le32_to_cpu(HDR(bh)->h_refcount),
 939                                           EXT2_XATTR_REFCOUNT_MAX);
 940                         } else if (!ext2_xattr_cmp(header, HDR(bh))) {
 941                                 ea_bdebug(bh, "b_count=%d",
 942                                           atomic_read(&(bh->b_count)));
 943                                 mb_cache_entry_touch(ea_block_cache, ce);
 944                                 mb_cache_entry_put(ea_block_cache, ce);
 945                                 return bh;
 946                         }
 947                         unlock_buffer(bh);
 948                         brelse(bh);
 949                 }
 950                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
 951         }
 952         return NULL;
 953 }
 954
 955 #define NAME_HASH_SHIFT 5
 956 #define VALUE_HASH_SHIFT 16
 957
 958 /*
 959  * ext2_xattr_hash_entry()
 960  *
 961  * Compute the hash of an extended attribute.
 962  */
 963 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 964                                          struct ext2_xattr_entry *entry)
 965 {
 966         __u32 hash = 0;
 967         char *name = entry->e_name;
 968         int n;
 969
 970         for (n=0; n < entry->e_name_len; n++) {
 971                 hash = (hash << NAME_HASH_SHIFT) ^
 972                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 973                        *name++;
 974         }
 975
 976         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 977                 __le32 *value = (__le32 *)((char *)header +
 978                         le16_to_cpu(entry->e_value_offs));
 979                 for (n = (le32_to_cpu(entry->e_value_size) +
 980                      EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 981                         hash = (hash << VALUE_HASH_SHIFT) ^
 982                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 983                                le32_to_cpu(*value++);
 984                 }
 985         }
 986         entry->e_hash = cpu_to_le32(hash);
 987 }
 988
 989 #undef NAME_HASH_SHIFT
 990 #undef VALUE_HASH_SHIFT
 991
 992 #define BLOCK_HASH_SHIFT 16
 993
 994 /*
 995  * ext2_xattr_rehash()
 996  *
 997  * Re-compute the extended attribute hash value after an entry has changed.
 998  */
 999 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1000                               struct ext2_xattr_entry *entry)
1001 {
1002         struct ext2_xattr_entry *here;
1003         __u32 hash = 0;
1004
1005         ext2_xattr_hash_entry(header, entry);
1006         here = ENTRY(header+1);
1007         while (!IS_LAST_ENTRY(here)) {
1008                 if (!here->e_hash) {
1009                         /* Block is not shared if an entry's hash value == 0 */
1010                         hash = 0;
1011                         break;
1012                 }
1013                 hash = (hash << BLOCK_HASH_SHIFT) ^
1014                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1015                        le32_to_cpu(here->e_hash);
1016                 here = EXT2_XATTR_NEXT(here);
1017         }
1018         header->h_hash = cpu_to_le32(hash);
1019 }
1020
1021 #undef BLOCK_HASH_SHIFT
1022
1023 #define HASH_BUCKET_BITS 10
1024
1025 struct mb_cache *ext2_xattr_create_cache(void)
1026 {
1027         return mb_cache_create(HASH_BUCKET_BITS);
1028 }
1029
1030 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1031 {
1032         if (cache)
1033                 mb_cache_destroy(cache);
1034 }