Linux Audio

Check our new training course

Loading...
v6.2
   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/printk.h>
  60#include <linux/slab.h>
  61#include <linux/mbcache.h>
  62#include <linux/quotaops.h>
  63#include <linux/rwsem.h>
  64#include <linux/security.h>
  65#include "ext2.h"
  66#include "xattr.h"
  67#include "acl.h"
  68
  69#define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
  70#define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
  71#define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
  72#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
  73
  74#ifdef EXT2_XATTR_DEBUG
  75# define ea_idebug(inode, f...) do { \
  76		printk(KERN_DEBUG "inode %s:%ld: ", \
  77			inode->i_sb->s_id, inode->i_ino); \
  78		printk(f); \
  79		printk("\n"); \
  80	} while (0)
  81# define ea_bdebug(bh, f...) do { \
  82		printk(KERN_DEBUG "block %pg:%lu: ", \
  83			bh->b_bdev, (unsigned long) bh->b_blocknr); \
  84		printk(f); \
  85		printk("\n"); \
  86	} while (0)
  87#else
  88# define ea_idebug(inode, f...)	no_printk(f)
  89# define ea_bdebug(bh, f...)	no_printk(f)
  90#endif
  91
  92static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  93			   struct ext2_xattr_header *);
  94
  95static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  96static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  97						 struct ext2_xattr_header *);
  98static void ext2_xattr_rehash(struct ext2_xattr_header *,
  99			      struct ext2_xattr_entry *);
 100
 101static const struct xattr_handler *ext2_xattr_handler_map[] = {
 102	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
 103#ifdef CONFIG_EXT2_FS_POSIX_ACL
 104	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
 105	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
 106#endif
 107	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
 108#ifdef CONFIG_EXT2_FS_SECURITY
 109	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
 110#endif
 111};
 112
 113const struct xattr_handler *ext2_xattr_handlers[] = {
 114	&ext2_xattr_user_handler,
 115	&ext2_xattr_trusted_handler,
 116#ifdef CONFIG_EXT2_FS_POSIX_ACL
 117	&posix_acl_access_xattr_handler,
 118	&posix_acl_default_xattr_handler,
 119#endif
 120#ifdef CONFIG_EXT2_FS_SECURITY
 121	&ext2_xattr_security_handler,
 122#endif
 123	NULL
 124};
 125
 126#define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
 127
 128static inline const struct xattr_handler *
 129ext2_xattr_handler(int name_index)
 130{
 131	const struct xattr_handler *handler = NULL;
 132
 133	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 134		handler = ext2_xattr_handler_map[name_index];
 135	return handler;
 136}
 137
 138static bool
 139ext2_xattr_header_valid(struct ext2_xattr_header *header)
 140{
 141	if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 142	    header->h_blocks != cpu_to_le32(1))
 143		return false;
 144
 145	return true;
 146}
 147
 148static bool
 149ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
 150		       char *end, size_t end_offs)
 151{
 152	struct ext2_xattr_entry *next;
 153	size_t size;
 154
 155	next = EXT2_XATTR_NEXT(entry);
 156	if ((char *)next >= end)
 157		return false;
 158
 159	if (entry->e_value_block != 0)
 160		return false;
 161
 162	size = le32_to_cpu(entry->e_value_size);
 163	if (size > end_offs ||
 164	    le16_to_cpu(entry->e_value_offs) + size > end_offs)
 165		return false;
 166
 167	return true;
 168}
 169
 170static int
 171ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
 172		     struct ext2_xattr_entry *entry)
 173{
 174	int cmp;
 175
 176	cmp = name_index - entry->e_name_index;
 177	if (!cmp)
 178		cmp = name_len - entry->e_name_len;
 179	if (!cmp)
 180		cmp = memcmp(name, entry->e_name, name_len);
 181
 182	return cmp;
 183}
 184
 185/*
 186 * ext2_xattr_get()
 187 *
 188 * Copy an extended attribute into the buffer
 189 * provided, or compute the buffer size required.
 190 * Buffer is NULL to compute the size of the buffer required.
 191 *
 192 * Returns a negative error number on failure, or the number of bytes
 193 * used / required on success.
 194 */
 195int
 196ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 197	       void *buffer, size_t buffer_size)
 198{
 199	struct buffer_head *bh = NULL;
 200	struct ext2_xattr_entry *entry;
 201	size_t name_len, size;
 202	char *end;
 203	int error, not_found;
 204	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 205
 206	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 207		  name_index, name, buffer, (long)buffer_size);
 208
 209	if (name == NULL)
 210		return -EINVAL;
 211	name_len = strlen(name);
 212	if (name_len > 255)
 213		return -ERANGE;
 214
 215	down_read(&EXT2_I(inode)->xattr_sem);
 216	error = -ENODATA;
 217	if (!EXT2_I(inode)->i_file_acl)
 218		goto cleanup;
 219	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 220	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 221	error = -EIO;
 222	if (!bh)
 223		goto cleanup;
 224	ea_bdebug(bh, "b_count=%d, refcount=%d",
 225		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 226	end = bh->b_data + bh->b_size;
 227	if (!ext2_xattr_header_valid(HDR(bh))) {
 228bad_block:
 229		ext2_error(inode->i_sb, "ext2_xattr_get",
 230			"inode %ld: bad block %d", inode->i_ino,
 231			EXT2_I(inode)->i_file_acl);
 232		error = -EIO;
 233		goto cleanup;
 234	}
 235
 236	/* find named attribute */
 237	entry = FIRST_ENTRY(bh);
 238	while (!IS_LAST_ENTRY(entry)) {
 239		if (!ext2_xattr_entry_valid(entry, end,
 240		    inode->i_sb->s_blocksize))
 
 241			goto bad_block;
 242
 243		not_found = ext2_xattr_cmp_entry(name_index, name_len, name,
 244						 entry);
 245		if (!not_found)
 246			goto found;
 247		if (not_found < 0)
 248			break;
 249
 250		entry = EXT2_XATTR_NEXT(entry);
 251	}
 252	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 253		ea_idebug(inode, "cache insert failed");
 254	error = -ENODATA;
 255	goto cleanup;
 256found:
 
 
 
 257	size = le32_to_cpu(entry->e_value_size);
 258	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 
 
 
 
 259		ea_idebug(inode, "cache insert failed");
 260	if (buffer) {
 261		error = -ERANGE;
 262		if (size > buffer_size)
 263			goto cleanup;
 264		/* return value of attribute */
 265		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 266			size);
 267	}
 268	error = size;
 269
 270cleanup:
 271	brelse(bh);
 272	up_read(&EXT2_I(inode)->xattr_sem);
 273
 274	return error;
 275}
 276
 277/*
 278 * ext2_xattr_list()
 279 *
 280 * Copy a list of attribute names into the buffer
 281 * provided, or compute the buffer size required.
 282 * Buffer is NULL to compute the size of the buffer required.
 283 *
 284 * Returns a negative error number on failure, or the number of bytes
 285 * used / required on success.
 286 */
 287static int
 288ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 289{
 290	struct inode *inode = d_inode(dentry);
 291	struct buffer_head *bh = NULL;
 292	struct ext2_xattr_entry *entry;
 293	char *end;
 294	size_t rest = buffer_size;
 295	int error;
 296	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 297
 298	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 299		  buffer, (long)buffer_size);
 300
 301	down_read(&EXT2_I(inode)->xattr_sem);
 302	error = 0;
 303	if (!EXT2_I(inode)->i_file_acl)
 304		goto cleanup;
 305	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 306	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 307	error = -EIO;
 308	if (!bh)
 309		goto cleanup;
 310	ea_bdebug(bh, "b_count=%d, refcount=%d",
 311		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 312	end = bh->b_data + bh->b_size;
 313	if (!ext2_xattr_header_valid(HDR(bh))) {
 314bad_block:
 315		ext2_error(inode->i_sb, "ext2_xattr_list",
 316			"inode %ld: bad block %d", inode->i_ino,
 317			EXT2_I(inode)->i_file_acl);
 318		error = -EIO;
 319		goto cleanup;
 320	}
 321
 322	/* check the on-disk data structure */
 323	entry = FIRST_ENTRY(bh);
 324	while (!IS_LAST_ENTRY(entry)) {
 325		if (!ext2_xattr_entry_valid(entry, end,
 326		    inode->i_sb->s_blocksize))
 
 327			goto bad_block;
 328		entry = EXT2_XATTR_NEXT(entry);
 329	}
 330	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 331		ea_idebug(inode, "cache insert failed");
 332
 333	/* list the attribute names */
 334	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 335	     entry = EXT2_XATTR_NEXT(entry)) {
 336		const struct xattr_handler *handler =
 337			ext2_xattr_handler(entry->e_name_index);
 338
 339		if (handler && (!handler->list || handler->list(dentry))) {
 340			const char *prefix = handler->prefix ?: handler->name;
 341			size_t prefix_len = strlen(prefix);
 342			size_t size = prefix_len + entry->e_name_len + 1;
 343
 344			if (buffer) {
 345				if (size > rest) {
 346					error = -ERANGE;
 347					goto cleanup;
 348				}
 349				memcpy(buffer, prefix, prefix_len);
 350				buffer += prefix_len;
 351				memcpy(buffer, entry->e_name, entry->e_name_len);
 352				buffer += entry->e_name_len;
 353				*buffer++ = 0;
 354			}
 355			rest -= size;
 356		}
 357	}
 358	error = buffer_size - rest;  /* total size */
 359
 360cleanup:
 361	brelse(bh);
 362	up_read(&EXT2_I(inode)->xattr_sem);
 363
 364	return error;
 365}
 366
 367/*
 368 * Inode operation listxattr()
 369 *
 370 * d_inode(dentry)->i_mutex: don't care
 371 */
 372ssize_t
 373ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 374{
 375	return ext2_xattr_list(dentry, buffer, size);
 376}
 377
 378/*
 379 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 380 * not set, set it.
 381 */
 382static void ext2_xattr_update_super_block(struct super_block *sb)
 383{
 384	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 385		return;
 386
 387	spin_lock(&EXT2_SB(sb)->s_lock);
 388	ext2_update_dynamic_rev(sb);
 389	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 390	spin_unlock(&EXT2_SB(sb)->s_lock);
 391	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 392}
 393
 394/*
 395 * ext2_xattr_set()
 396 *
 397 * Create, replace or remove an extended attribute for this inode.  Value
 398 * is NULL to remove an existing extended attribute, and non-NULL to
 399 * either replace an existing extended attribute, or create a new extended
 400 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 401 * specify that an extended attribute must exist and must not exist
 402 * previous to the call, respectively.
 403 *
 404 * Returns 0, or a negative error number on failure.
 405 */
 406int
 407ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 408	       const void *value, size_t value_len, int flags)
 409{
 410	struct super_block *sb = inode->i_sb;
 411	struct buffer_head *bh = NULL;
 412	struct ext2_xattr_header *header = NULL;
 413	struct ext2_xattr_entry *here = NULL, *last = NULL;
 414	size_t name_len, free, min_offs = sb->s_blocksize;
 415	int not_found = 1, error;
 416	char *end;
 417	
 418	/*
 419	 * header -- Points either into bh, or to a temporarily
 420	 *           allocated buffer.
 421	 * here -- The named entry found, or the place for inserting, within
 422	 *         the block pointed to by header.
 423	 * last -- Points right after the last named entry within the block
 424	 *         pointed to by header.
 425	 * min_offs -- The offset of the first value (values are aligned
 426	 *             towards the end of the block).
 427	 * end -- Points right after the block pointed to by header.
 428	 */
 429	
 430	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 431		  name_index, name, value, (long)value_len);
 432
 433	if (value == NULL)
 434		value_len = 0;
 435	if (name == NULL)
 436		return -EINVAL;
 437	name_len = strlen(name);
 438	if (name_len > 255 || value_len > sb->s_blocksize)
 439		return -ERANGE;
 440	error = dquot_initialize(inode);
 441	if (error)
 442		return error;
 443	down_write(&EXT2_I(inode)->xattr_sem);
 444	if (EXT2_I(inode)->i_file_acl) {
 445		/* The inode already has an extended attribute block. */
 446		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 447		error = -EIO;
 448		if (!bh)
 449			goto cleanup;
 450		ea_bdebug(bh, "b_count=%d, refcount=%d",
 451			atomic_read(&(bh->b_count)),
 452			le32_to_cpu(HDR(bh)->h_refcount));
 453		header = HDR(bh);
 454		end = bh->b_data + bh->b_size;
 455		if (!ext2_xattr_header_valid(header)) {
 456bad_block:
 457			ext2_error(sb, "ext2_xattr_set",
 458				"inode %ld: bad block %d", inode->i_ino, 
 459				   EXT2_I(inode)->i_file_acl);
 460			error = -EIO;
 461			goto cleanup;
 462		}
 463		/*
 464		 * Find the named attribute. If not found, 'here' will point
 465		 * to entry where the new attribute should be inserted to
 466		 * maintain sorting.
 467		 */
 468		last = FIRST_ENTRY(bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 469		while (!IS_LAST_ENTRY(last)) {
 470			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
 
 471				goto bad_block;
 472			if (last->e_value_size) {
 473				size_t offs = le16_to_cpu(last->e_value_offs);
 474				if (offs < min_offs)
 475					min_offs = offs;
 476			}
 477			if (not_found > 0) {
 478				not_found = ext2_xattr_cmp_entry(name_index,
 479								 name_len,
 480								 name, last);
 481				if (not_found <= 0)
 482					here = last;
 483			}
 484			last = EXT2_XATTR_NEXT(last);
 485		}
 486		if (not_found > 0)
 487			here = last;
 488
 489		/* Check whether we have enough space left. */
 490		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 491	} else {
 492		/* We will use a new extended attribute block. */
 493		free = sb->s_blocksize -
 494			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 
 495	}
 496
 497	if (not_found) {
 498		/* Request to remove a nonexistent attribute? */
 499		error = -ENODATA;
 500		if (flags & XATTR_REPLACE)
 501			goto cleanup;
 502		error = 0;
 503		if (value == NULL)
 504			goto cleanup;
 505	} else {
 506		/* Request to create an existing attribute? */
 507		error = -EEXIST;
 508		if (flags & XATTR_CREATE)
 509			goto cleanup;
 510		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
 
 
 
 
 
 
 
 511		free += EXT2_XATTR_LEN(name_len);
 512	}
 513	error = -ENOSPC;
 514	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 515		goto cleanup;
 516
 517	/* Here we know that we can set the new attribute. */
 518
 519	if (header) {
 520		int offset;
 521
 522		lock_buffer(bh);
 523		if (header->h_refcount == cpu_to_le32(1)) {
 524			__u32 hash = le32_to_cpu(header->h_hash);
 525			struct mb_cache_entry *oe;
 526
 527			oe = mb_cache_entry_delete_or_get(EA_BLOCK_CACHE(inode),
 528					hash, bh->b_blocknr);
 529			if (!oe) {
 530				ea_bdebug(bh, "modifying in-place");
 531				goto update_block;
 532			}
 533			/*
 534			 * Someone is trying to reuse the block, leave it alone
 
 535			 */
 536			mb_cache_entry_put(EA_BLOCK_CACHE(inode), oe);
 537		}
 538		unlock_buffer(bh);
 539		ea_bdebug(bh, "cloning");
 540		header = kmemdup(HDR(bh), bh->b_size, GFP_KERNEL);
 541		error = -ENOMEM;
 542		if (header == NULL)
 543			goto cleanup;
 544		header->h_refcount = cpu_to_le32(1);
 545
 546		offset = (char *)here - bh->b_data;
 547		here = ENTRY((char *)header + offset);
 548		offset = (char *)last - bh->b_data;
 549		last = ENTRY((char *)header + offset);
 
 
 
 
 
 
 
 
 
 
 550	} else {
 551		/* Allocate a buffer where we construct the new block. */
 552		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 553		error = -ENOMEM;
 554		if (header == NULL)
 555			goto cleanup;
 556		end = (char *)header + sb->s_blocksize;
 557		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 558		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 559		last = here = ENTRY(header+1);
 560	}
 561
 562update_block:
 563	/* Iff we are modifying the block in-place, bh is locked here. */
 564
 565	if (not_found) {
 566		/* Insert the new name. */
 567		size_t size = EXT2_XATTR_LEN(name_len);
 568		size_t rest = (char *)last - (char *)here;
 569		memmove((char *)here + size, here, rest);
 570		memset(here, 0, size);
 571		here->e_name_index = name_index;
 572		here->e_name_len = name_len;
 573		memcpy(here->e_name, name, name_len);
 574	} else {
 575		if (here->e_value_size) {
 576			char *first_val = (char *)header + min_offs;
 577			size_t offs = le16_to_cpu(here->e_value_offs);
 578			char *val = (char *)header + offs;
 579			size_t size = EXT2_XATTR_SIZE(
 580				le32_to_cpu(here->e_value_size));
 581
 582			if (size == EXT2_XATTR_SIZE(value_len)) {
 583				/* The old and the new value have the same
 584				   size. Just replace. */
 585				here->e_value_size = cpu_to_le32(value_len);
 586				memset(val + size - EXT2_XATTR_PAD, 0,
 587				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 588				memcpy(val, value, value_len);
 589				goto skip_replace;
 590			}
 591
 592			/* Remove the old value. */
 593			memmove(first_val + size, first_val, val - first_val);
 594			memset(first_val, 0, size);
 
 595			min_offs += size;
 596
 597			/* Adjust all value offsets. */
 598			last = ENTRY(header+1);
 599			while (!IS_LAST_ENTRY(last)) {
 600				size_t o = le16_to_cpu(last->e_value_offs);
 601				if (o < offs)
 602					last->e_value_offs =
 603						cpu_to_le16(o + size);
 604				last = EXT2_XATTR_NEXT(last);
 605			}
 606
 607			here->e_value_offs = 0;
 608		}
 609		if (value == NULL) {
 610			/* Remove the old name. */
 611			size_t size = EXT2_XATTR_LEN(name_len);
 612			last = ENTRY((char *)last - size);
 613			memmove(here, (char*)here + size,
 614				(char*)last - (char*)here);
 615			memset(last, 0, size);
 616		}
 617	}
 618
 619	if (value != NULL) {
 620		/* Insert the new value. */
 621		here->e_value_size = cpu_to_le32(value_len);
 622		if (value_len) {
 623			size_t size = EXT2_XATTR_SIZE(value_len);
 624			char *val = (char *)header + min_offs - size;
 625			here->e_value_offs =
 626				cpu_to_le16((char *)val - (char *)header);
 627			memset(val + size - EXT2_XATTR_PAD, 0,
 628			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 629			memcpy(val, value, value_len);
 630		}
 631	}
 632
 633skip_replace:
 634	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 635		/* This block is now empty. */
 636		if (bh && header == HDR(bh))
 637			unlock_buffer(bh);  /* we were modifying in-place. */
 638		error = ext2_xattr_set2(inode, bh, NULL);
 639	} else {
 640		ext2_xattr_rehash(header, here);
 641		if (bh && header == HDR(bh))
 642			unlock_buffer(bh);  /* we were modifying in-place. */
 643		error = ext2_xattr_set2(inode, bh, header);
 644	}
 645
 646cleanup:
 
 647	if (!(bh && header == HDR(bh)))
 648		kfree(header);
 649	brelse(bh);
 650	up_write(&EXT2_I(inode)->xattr_sem);
 651
 652	return error;
 653}
 654
 655static void ext2_xattr_release_block(struct inode *inode,
 656				     struct buffer_head *bh)
 657{
 658	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 659
 660retry_ref:
 661	lock_buffer(bh);
 662	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 663		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 664		struct mb_cache_entry *oe;
 665
 666		/*
 667		 * This must happen under buffer lock to properly
 668		 * serialize with ext2_xattr_set() reusing the block.
 669		 */
 670		oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
 671						  bh->b_blocknr);
 672		if (oe) {
 673			/*
 674			 * Someone is trying to reuse the block. Wait
 675			 * and retry.
 676			 */
 677			unlock_buffer(bh);
 678			mb_cache_entry_wait_unused(oe);
 679			mb_cache_entry_put(ea_block_cache, oe);
 680			goto retry_ref;
 681		}
 682
 683		/* Free the old block. */
 684		ea_bdebug(bh, "freeing");
 685		ext2_free_blocks(inode, bh->b_blocknr, 1);
 686		/* We let our caller release bh, so we
 687		 * need to duplicate the buffer before. */
 688		get_bh(bh);
 689		bforget(bh);
 690		unlock_buffer(bh);
 691	} else {
 692		/* Decrement the refcount only. */
 693		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 694		dquot_free_block(inode, 1);
 695		mark_buffer_dirty(bh);
 696		unlock_buffer(bh);
 697		ea_bdebug(bh, "refcount now=%d",
 698			le32_to_cpu(HDR(bh)->h_refcount));
 699		if (IS_SYNC(inode))
 700			sync_dirty_buffer(bh);
 701	}
 702}
 703
 704/*
 705 * Second half of ext2_xattr_set(): Update the file system.
 706 */
 707static int
 708ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 709		struct ext2_xattr_header *header)
 710{
 711	struct super_block *sb = inode->i_sb;
 712	struct buffer_head *new_bh = NULL;
 713	int error;
 714	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 715
 716	if (header) {
 717		new_bh = ext2_xattr_cache_find(inode, header);
 718		if (new_bh) {
 719			/* We found an identical block in the cache. */
 720			if (new_bh == old_bh) {
 721				ea_bdebug(new_bh, "keeping this block");
 722			} else {
 723				/* The old block is released after updating
 724				   the inode.  */
 725				ea_bdebug(new_bh, "reusing block");
 726
 727				error = dquot_alloc_block(inode, 1);
 728				if (error) {
 729					unlock_buffer(new_bh);
 730					goto cleanup;
 731				}
 732				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 733				ea_bdebug(new_bh, "refcount now=%d",
 734					le32_to_cpu(HDR(new_bh)->h_refcount));
 735			}
 736			unlock_buffer(new_bh);
 737		} else if (old_bh && header == HDR(old_bh)) {
 738			/* Keep this block. No need to lock the block as we
 739			   don't need to change the reference count. */
 740			new_bh = old_bh;
 741			get_bh(new_bh);
 742			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 743		} else {
 744			/* We need to allocate a new block */
 745			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 746						EXT2_I(inode)->i_block_group);
 747			int block = ext2_new_block(inode, goal, &error);
 748			if (error)
 749				goto cleanup;
 750			ea_idebug(inode, "creating block %d", block);
 751
 752			new_bh = sb_getblk(sb, block);
 753			if (unlikely(!new_bh)) {
 754				ext2_free_blocks(inode, block, 1);
 755				mark_inode_dirty(inode);
 756				error = -ENOMEM;
 757				goto cleanup;
 758			}
 759			lock_buffer(new_bh);
 760			memcpy(new_bh->b_data, header, new_bh->b_size);
 761			set_buffer_uptodate(new_bh);
 762			unlock_buffer(new_bh);
 763			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 764			
 765			ext2_xattr_update_super_block(sb);
 766		}
 767		mark_buffer_dirty(new_bh);
 768		if (IS_SYNC(inode)) {
 769			sync_dirty_buffer(new_bh);
 770			error = -EIO;
 771			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 772				goto cleanup;
 773		}
 774	}
 775
 776	/* Update the inode. */
 777	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 778	inode->i_ctime = current_time(inode);
 779	if (IS_SYNC(inode)) {
 780		error = sync_inode_metadata(inode, 1);
 781		/* In case sync failed due to ENOSPC the inode was actually
 782		 * written (only some dirty data were not) so we just proceed
 783		 * as if nothing happened and cleanup the unused block */
 784		if (error && error != -ENOSPC) {
 785			if (new_bh && new_bh != old_bh) {
 786				dquot_free_block_nodirty(inode, 1);
 787				mark_inode_dirty(inode);
 788			}
 789			goto cleanup;
 790		}
 791	} else
 792		mark_inode_dirty(inode);
 793
 794	error = 0;
 795	if (old_bh && old_bh != new_bh) {
 796		/*
 797		 * If there was an old block and we are no longer using it,
 798		 * release the old block.
 799		 */
 800		ext2_xattr_release_block(inode, old_bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 801	}
 802
 803cleanup:
 804	brelse(new_bh);
 805
 806	return error;
 807}
 808
 809/*
 810 * ext2_xattr_delete_inode()
 811 *
 812 * Free extended attribute resources associated with this inode. This
 813 * is called immediately before an inode is freed.
 814 */
 815void
 816ext2_xattr_delete_inode(struct inode *inode)
 817{
 818	struct buffer_head *bh = NULL;
 819	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 820
 821	/*
 822	 * We are the only ones holding inode reference. The xattr_sem should
 823	 * better be unlocked! We could as well just not acquire xattr_sem at
 824	 * all but this makes the code more futureproof. OTOH we need trylock
 825	 * here to avoid false-positive warning from lockdep about reclaim
 826	 * circular dependency.
 827	 */
 828	if (WARN_ON_ONCE(!down_write_trylock(&EXT2_I(inode)->xattr_sem)))
 829		return;
 830	if (!EXT2_I(inode)->i_file_acl)
 831		goto cleanup;
 832
 833	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
 834		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 835			"inode %ld: xattr block %d is out of data blocks range",
 836			inode->i_ino, EXT2_I(inode)->i_file_acl);
 837		goto cleanup;
 838	}
 839
 840	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 841	if (!bh) {
 842		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 843			"inode %ld: block %d read error", inode->i_ino,
 844			EXT2_I(inode)->i_file_acl);
 845		goto cleanup;
 846	}
 847	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 848	if (!ext2_xattr_header_valid(HDR(bh))) {
 
 849		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 850			"inode %ld: bad block %d", inode->i_ino,
 851			EXT2_I(inode)->i_file_acl);
 852		goto cleanup;
 853	}
 854	ext2_xattr_release_block(inode, bh);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 855	EXT2_I(inode)->i_file_acl = 0;
 856
 857cleanup:
 858	brelse(bh);
 859	up_write(&EXT2_I(inode)->xattr_sem);
 860}
 861
 862/*
 863 * ext2_xattr_cache_insert()
 864 *
 865 * Create a new entry in the extended attribute cache, and insert
 866 * it unless such an entry is already in the cache.
 867 *
 868 * Returns 0, or a negative error number on failure.
 869 */
 870static int
 871ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 872{
 873	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 874	int error;
 875
 876	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
 877				      true);
 878	if (error) {
 879		if (error == -EBUSY) {
 880			ea_bdebug(bh, "already in cache");
 
 881			error = 0;
 882		}
 883	} else
 884		ea_bdebug(bh, "inserting [%x]", (int)hash);
 885	return error;
 886}
 887
 888/*
 889 * ext2_xattr_cmp()
 890 *
 891 * Compare two extended attribute blocks for equality.
 892 *
 893 * Returns 0 if the blocks are equal, 1 if they differ, and
 894 * a negative error number on errors.
 895 */
 896static int
 897ext2_xattr_cmp(struct ext2_xattr_header *header1,
 898	       struct ext2_xattr_header *header2)
 899{
 900	struct ext2_xattr_entry *entry1, *entry2;
 901
 902	entry1 = ENTRY(header1+1);
 903	entry2 = ENTRY(header2+1);
 904	while (!IS_LAST_ENTRY(entry1)) {
 905		if (IS_LAST_ENTRY(entry2))
 906			return 1;
 907		if (entry1->e_hash != entry2->e_hash ||
 908		    entry1->e_name_index != entry2->e_name_index ||
 909		    entry1->e_name_len != entry2->e_name_len ||
 910		    entry1->e_value_size != entry2->e_value_size ||
 911		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 912			return 1;
 913		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 914			return -EIO;
 915		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 916			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 917			   le32_to_cpu(entry1->e_value_size)))
 918			return 1;
 919
 920		entry1 = EXT2_XATTR_NEXT(entry1);
 921		entry2 = EXT2_XATTR_NEXT(entry2);
 922	}
 923	if (!IS_LAST_ENTRY(entry2))
 924		return 1;
 925	return 0;
 926}
 927
 928/*
 929 * ext2_xattr_cache_find()
 930 *
 931 * Find an identical extended attribute block.
 932 *
 933 * Returns a locked buffer head to the block found, or NULL if such
 934 * a block was not found or an error occurred.
 935 */
 936static struct buffer_head *
 937ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 938{
 939	__u32 hash = le32_to_cpu(header->h_hash);
 940	struct mb_cache_entry *ce;
 941	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 942
 943	if (!header->h_hash)
 944		return NULL;  /* never share */
 945	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 946
 947	ce = mb_cache_entry_find_first(ea_block_cache, hash);
 948	while (ce) {
 949		struct buffer_head *bh;
 950
 951		bh = sb_bread(inode->i_sb, ce->e_value);
 952		if (!bh) {
 953			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 954				"inode %ld: block %ld read error",
 955				inode->i_ino, (unsigned long) ce->e_value);
 956		} else {
 957			lock_buffer(bh);
 958			if (le32_to_cpu(HDR(bh)->h_refcount) >
 959			    EXT2_XATTR_REFCOUNT_MAX) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 960				ea_idebug(inode, "block %ld refcount %d>%d",
 961					  (unsigned long) ce->e_value,
 962					  le32_to_cpu(HDR(bh)->h_refcount),
 963					  EXT2_XATTR_REFCOUNT_MAX);
 964			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 965				ea_bdebug(bh, "b_count=%d",
 966					  atomic_read(&(bh->b_count)));
 967				mb_cache_entry_touch(ea_block_cache, ce);
 968				mb_cache_entry_put(ea_block_cache, ce);
 969				return bh;
 970			}
 971			unlock_buffer(bh);
 972			brelse(bh);
 973		}
 974		ce = mb_cache_entry_find_next(ea_block_cache, ce);
 975	}
 976	return NULL;
 977}
 978
 979#define NAME_HASH_SHIFT 5
 980#define VALUE_HASH_SHIFT 16
 981
 982/*
 983 * ext2_xattr_hash_entry()
 984 *
 985 * Compute the hash of an extended attribute.
 986 */
 987static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 988					 struct ext2_xattr_entry *entry)
 989{
 990	__u32 hash = 0;
 991	char *name = entry->e_name;
 992	int n;
 993
 994	for (n=0; n < entry->e_name_len; n++) {
 995		hash = (hash << NAME_HASH_SHIFT) ^
 996		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 997		       *name++;
 998	}
 999
1000	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1001		__le32 *value = (__le32 *)((char *)header +
1002			le16_to_cpu(entry->e_value_offs));
1003		for (n = (le32_to_cpu(entry->e_value_size) +
1004		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1005			hash = (hash << VALUE_HASH_SHIFT) ^
1006			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1007			       le32_to_cpu(*value++);
1008		}
1009	}
1010	entry->e_hash = cpu_to_le32(hash);
1011}
1012
1013#undef NAME_HASH_SHIFT
1014#undef VALUE_HASH_SHIFT
1015
1016#define BLOCK_HASH_SHIFT 16
1017
1018/*
1019 * ext2_xattr_rehash()
1020 *
1021 * Re-compute the extended attribute hash value after an entry has changed.
1022 */
1023static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1024			      struct ext2_xattr_entry *entry)
1025{
1026	struct ext2_xattr_entry *here;
1027	__u32 hash = 0;
1028	
1029	ext2_xattr_hash_entry(header, entry);
1030	here = ENTRY(header+1);
1031	while (!IS_LAST_ENTRY(here)) {
1032		if (!here->e_hash) {
1033			/* Block is not shared if an entry's hash value == 0 */
1034			hash = 0;
1035			break;
1036		}
1037		hash = (hash << BLOCK_HASH_SHIFT) ^
1038		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1039		       le32_to_cpu(here->e_hash);
1040		here = EXT2_XATTR_NEXT(here);
1041	}
1042	header->h_hash = cpu_to_le32(hash);
1043}
1044
1045#undef BLOCK_HASH_SHIFT
1046
1047#define HASH_BUCKET_BITS 10
1048
1049struct mb_cache *ext2_xattr_create_cache(void)
1050{
1051	return mb_cache_create(HASH_BUCKET_BITS);
1052}
1053
1054void ext2_xattr_destroy_cache(struct mb_cache *cache)
1055{
1056	if (cache)
1057		mb_cache_destroy(cache);
1058}
v4.6
 
   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 aligned 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/init.h>
 
  58#include <linux/slab.h>
  59#include <linux/mbcache.h>
  60#include <linux/quotaops.h>
  61#include <linux/rwsem.h>
  62#include <linux/security.h>
  63#include "ext2.h"
  64#include "xattr.h"
  65#include "acl.h"
  66
  67#define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
  68#define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
  69#define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
  70#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
  71
  72#ifdef EXT2_XATTR_DEBUG
  73# define ea_idebug(inode, f...) do { \
  74		printk(KERN_DEBUG "inode %s:%ld: ", \
  75			inode->i_sb->s_id, inode->i_ino); \
  76		printk(f); \
  77		printk("\n"); \
  78	} while (0)
  79# define ea_bdebug(bh, f...) do { \
  80		printk(KERN_DEBUG "block %pg:%lu: ", \
  81			bh->b_bdev, (unsigned long) bh->b_blocknr); \
  82		printk(f); \
  83		printk("\n"); \
  84	} while (0)
  85#else
  86# define ea_idebug(f...)
  87# define ea_bdebug(f...)
  88#endif
  89
  90static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  91			   struct ext2_xattr_header *);
  92
  93static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  94static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  95						 struct ext2_xattr_header *);
  96static void ext2_xattr_rehash(struct ext2_xattr_header *,
  97			      struct ext2_xattr_entry *);
  98
  99static const struct xattr_handler *ext2_xattr_handler_map[] = {
 100	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
 101#ifdef CONFIG_EXT2_FS_POSIX_ACL
 102	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
 103	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
 104#endif
 105	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
 106#ifdef CONFIG_EXT2_FS_SECURITY
 107	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
 108#endif
 109};
 110
 111const struct xattr_handler *ext2_xattr_handlers[] = {
 112	&ext2_xattr_user_handler,
 113	&ext2_xattr_trusted_handler,
 114#ifdef CONFIG_EXT2_FS_POSIX_ACL
 115	&posix_acl_access_xattr_handler,
 116	&posix_acl_default_xattr_handler,
 117#endif
 118#ifdef CONFIG_EXT2_FS_SECURITY
 119	&ext2_xattr_security_handler,
 120#endif
 121	NULL
 122};
 123
 
 
 124static inline const struct xattr_handler *
 125ext2_xattr_handler(int name_index)
 126{
 127	const struct xattr_handler *handler = NULL;
 128
 129	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 130		handler = ext2_xattr_handler_map[name_index];
 131	return handler;
 132}
 133
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 134/*
 135 * ext2_xattr_get()
 136 *
 137 * Copy an extended attribute into the buffer
 138 * provided, or compute the buffer size required.
 139 * Buffer is NULL to compute the size of the buffer required.
 140 *
 141 * Returns a negative error number on failure, or the number of bytes
 142 * used / required on success.
 143 */
 144int
 145ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 146	       void *buffer, size_t buffer_size)
 147{
 148	struct buffer_head *bh = NULL;
 149	struct ext2_xattr_entry *entry;
 150	size_t name_len, size;
 151	char *end;
 152	int error;
 153	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
 154
 155	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 156		  name_index, name, buffer, (long)buffer_size);
 157
 158	if (name == NULL)
 159		return -EINVAL;
 160	name_len = strlen(name);
 161	if (name_len > 255)
 162		return -ERANGE;
 163
 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)) {
 178bad_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
 185	/* find named attribute */
 186	entry = FIRST_ENTRY(bh);
 187	while (!IS_LAST_ENTRY(entry)) {
 188		struct ext2_xattr_entry *next =
 189			EXT2_XATTR_NEXT(entry);
 190		if ((char *)next >= end)
 191			goto bad_block;
 192		if (name_index == entry->e_name_index &&
 193		    name_len == entry->e_name_len &&
 194		    memcmp(name, entry->e_name, name_len) == 0)
 
 195			goto found;
 196		entry = next;
 
 
 
 197	}
 198	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
 199		ea_idebug(inode, "cache insert failed");
 200	error = -ENODATA;
 201	goto cleanup;
 202found:
 203	/* check the buffer size */
 204	if (entry->e_value_block != 0)
 205		goto bad_block;
 206	size = le32_to_cpu(entry->e_value_size);
 207	if (size > inode->i_sb->s_blocksize ||
 208	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
 209		goto bad_block;
 210
 211	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
 212		ea_idebug(inode, "cache insert failed");
 213	if (buffer) {
 214		error = -ERANGE;
 215		if (size > buffer_size)
 216			goto cleanup;
 217		/* return value of attribute */
 218		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 219			size);
 220	}
 221	error = size;
 222
 223cleanup:
 224	brelse(bh);
 225	up_read(&EXT2_I(inode)->xattr_sem);
 226
 227	return error;
 228}
 229
 230/*
 231 * ext2_xattr_list()
 232 *
 233 * Copy a list of attribute names into the buffer
 234 * provided, or compute the buffer size required.
 235 * Buffer is NULL to compute the size of the buffer required.
 236 *
 237 * Returns a negative error number on failure, or the number of bytes
 238 * used / required on success.
 239 */
 240static int
 241ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 242{
 243	struct inode *inode = d_inode(dentry);
 244	struct buffer_head *bh = NULL;
 245	struct ext2_xattr_entry *entry;
 246	char *end;
 247	size_t rest = buffer_size;
 248	int error;
 249	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
 250
 251	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 252		  buffer, (long)buffer_size);
 253
 254	down_read(&EXT2_I(inode)->xattr_sem);
 255	error = 0;
 256	if (!EXT2_I(inode)->i_file_acl)
 257		goto cleanup;
 258	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 259	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 260	error = -EIO;
 261	if (!bh)
 262		goto cleanup;
 263	ea_bdebug(bh, "b_count=%d, refcount=%d",
 264		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 265	end = bh->b_data + bh->b_size;
 266	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 267	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 268bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
 269			"inode %ld: bad block %d", inode->i_ino,
 270			EXT2_I(inode)->i_file_acl);
 271		error = -EIO;
 272		goto cleanup;
 273	}
 274
 275	/* check the on-disk data structure */
 276	entry = FIRST_ENTRY(bh);
 277	while (!IS_LAST_ENTRY(entry)) {
 278		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
 279
 280		if ((char *)next >= end)
 281			goto bad_block;
 282		entry = next;
 283	}
 284	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
 285		ea_idebug(inode, "cache insert failed");
 286
 287	/* list the attribute names */
 288	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 289	     entry = EXT2_XATTR_NEXT(entry)) {
 290		const struct xattr_handler *handler =
 291			ext2_xattr_handler(entry->e_name_index);
 292
 293		if (handler && (!handler->list || handler->list(dentry))) {
 294			const char *prefix = handler->prefix ?: handler->name;
 295			size_t prefix_len = strlen(prefix);
 296			size_t size = prefix_len + entry->e_name_len + 1;
 297
 298			if (buffer) {
 299				if (size > rest) {
 300					error = -ERANGE;
 301					goto cleanup;
 302				}
 303				memcpy(buffer, prefix, prefix_len);
 304				buffer += prefix_len;
 305				memcpy(buffer, entry->e_name, entry->e_name_len);
 306				buffer += entry->e_name_len;
 307				*buffer++ = 0;
 308			}
 309			rest -= size;
 310		}
 311	}
 312	error = buffer_size - rest;  /* total size */
 313
 314cleanup:
 315	brelse(bh);
 316	up_read(&EXT2_I(inode)->xattr_sem);
 317
 318	return error;
 319}
 320
 321/*
 322 * Inode operation listxattr()
 323 *
 324 * d_inode(dentry)->i_mutex: don't care
 325 */
 326ssize_t
 327ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 328{
 329	return ext2_xattr_list(dentry, buffer, size);
 330}
 331
 332/*
 333 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 334 * not set, set it.
 335 */
 336static void ext2_xattr_update_super_block(struct super_block *sb)
 337{
 338	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 339		return;
 340
 341	spin_lock(&EXT2_SB(sb)->s_lock);
 
 342	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 343	spin_unlock(&EXT2_SB(sb)->s_lock);
 344	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 345}
 346
 347/*
 348 * ext2_xattr_set()
 349 *
 350 * Create, replace or remove an extended attribute for this inode.  Value
 351 * is NULL to remove an existing extended attribute, and non-NULL to
 352 * either replace an existing extended attribute, or create a new extended
 353 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 354 * specify that an extended attribute must exist and must not exist
 355 * previous to the call, respectively.
 356 *
 357 * Returns 0, or a negative error number on failure.
 358 */
 359int
 360ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 361	       const void *value, size_t value_len, int flags)
 362{
 363	struct super_block *sb = inode->i_sb;
 364	struct buffer_head *bh = NULL;
 365	struct ext2_xattr_header *header = NULL;
 366	struct ext2_xattr_entry *here, *last;
 367	size_t name_len, free, min_offs = sb->s_blocksize;
 368	int not_found = 1, error;
 369	char *end;
 370	
 371	/*
 372	 * header -- Points either into bh, or to a temporarily
 373	 *           allocated buffer.
 374	 * here -- The named entry found, or the place for inserting, within
 375	 *         the block pointed to by header.
 376	 * last -- Points right after the last named entry within the block
 377	 *         pointed to by header.
 378	 * min_offs -- The offset of the first value (values are aligned
 379	 *             towards the end of the block).
 380	 * end -- Points right after the block pointed to by header.
 381	 */
 382	
 383	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 384		  name_index, name, value, (long)value_len);
 385
 386	if (value == NULL)
 387		value_len = 0;
 388	if (name == NULL)
 389		return -EINVAL;
 390	name_len = strlen(name);
 391	if (name_len > 255 || value_len > sb->s_blocksize)
 392		return -ERANGE;
 
 
 
 393	down_write(&EXT2_I(inode)->xattr_sem);
 394	if (EXT2_I(inode)->i_file_acl) {
 395		/* The inode already has an extended attribute block. */
 396		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 397		error = -EIO;
 398		if (!bh)
 399			goto cleanup;
 400		ea_bdebug(bh, "b_count=%d, refcount=%d",
 401			atomic_read(&(bh->b_count)),
 402			le32_to_cpu(HDR(bh)->h_refcount));
 403		header = HDR(bh);
 404		end = bh->b_data + bh->b_size;
 405		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 406		    header->h_blocks != cpu_to_le32(1)) {
 407bad_block:		ext2_error(sb, "ext2_xattr_set",
 408				"inode %ld: bad block %d", inode->i_ino, 
 409				   EXT2_I(inode)->i_file_acl);
 410			error = -EIO;
 411			goto cleanup;
 412		}
 413		/* Find the named attribute. */
 414		here = FIRST_ENTRY(bh);
 415		while (!IS_LAST_ENTRY(here)) {
 416			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
 417			if ((char *)next >= end)
 418				goto bad_block;
 419			if (!here->e_value_block && here->e_value_size) {
 420				size_t offs = le16_to_cpu(here->e_value_offs);
 421				if (offs < min_offs)
 422					min_offs = offs;
 423			}
 424			not_found = name_index - here->e_name_index;
 425			if (!not_found)
 426				not_found = name_len - here->e_name_len;
 427			if (!not_found)
 428				not_found = memcmp(name, here->e_name,name_len);
 429			if (not_found <= 0)
 430				break;
 431			here = next;
 432		}
 433		last = here;
 434		/* We still need to compute min_offs and last. */
 435		while (!IS_LAST_ENTRY(last)) {
 436			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
 437			if ((char *)next >= end)
 438				goto bad_block;
 439			if (!last->e_value_block && last->e_value_size) {
 440				size_t offs = le16_to_cpu(last->e_value_offs);
 441				if (offs < min_offs)
 442					min_offs = offs;
 443			}
 444			last = next;
 
 
 
 
 
 
 
 445		}
 
 
 446
 447		/* Check whether we have enough space left. */
 448		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 449	} else {
 450		/* We will use a new extended attribute block. */
 451		free = sb->s_blocksize -
 452			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 453		here = last = NULL;  /* avoid gcc uninitialized warning. */
 454	}
 455
 456	if (not_found) {
 457		/* Request to remove a nonexistent attribute? */
 458		error = -ENODATA;
 459		if (flags & XATTR_REPLACE)
 460			goto cleanup;
 461		error = 0;
 462		if (value == NULL)
 463			goto cleanup;
 464	} else {
 465		/* Request to create an existing attribute? */
 466		error = -EEXIST;
 467		if (flags & XATTR_CREATE)
 468			goto cleanup;
 469		if (!here->e_value_block && here->e_value_size) {
 470			size_t size = le32_to_cpu(here->e_value_size);
 471
 472			if (le16_to_cpu(here->e_value_offs) + size > 
 473			    sb->s_blocksize || size > sb->s_blocksize)
 474				goto bad_block;
 475			free += EXT2_XATTR_SIZE(size);
 476		}
 477		free += EXT2_XATTR_LEN(name_len);
 478	}
 479	error = -ENOSPC;
 480	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 481		goto cleanup;
 482
 483	/* Here we know that we can set the new attribute. */
 484
 485	if (header) {
 486		/* assert(header == HDR(bh)); */
 
 487		lock_buffer(bh);
 488		if (header->h_refcount == cpu_to_le32(1)) {
 489			__u32 hash = le32_to_cpu(header->h_hash);
 
 490
 491			ea_bdebug(bh, "modifying in-place");
 
 
 
 
 
 492			/*
 493			 * This must happen under buffer lock for
 494			 * ext2_xattr_set2() to reliably detect modified block
 495			 */
 496			mb_cache_entry_delete_block(EXT2_SB(sb)->s_mb_cache,
 497						    hash, bh->b_blocknr);
 498
 499			/* keep the buffer locked while modifying it. */
 500		} else {
 501			int offset;
 
 
 
 502
 503			unlock_buffer(bh);
 504			ea_bdebug(bh, "cloning");
 505			header = kmalloc(bh->b_size, GFP_KERNEL);
 506			error = -ENOMEM;
 507			if (header == NULL)
 508				goto cleanup;
 509			memcpy(header, HDR(bh), bh->b_size);
 510			header->h_refcount = cpu_to_le32(1);
 511
 512			offset = (char *)here - bh->b_data;
 513			here = ENTRY((char *)header + offset);
 514			offset = (char *)last - bh->b_data;
 515			last = ENTRY((char *)header + offset);
 516		}
 517	} else {
 518		/* Allocate a buffer where we construct the new block. */
 519		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 520		error = -ENOMEM;
 521		if (header == NULL)
 522			goto cleanup;
 523		end = (char *)header + sb->s_blocksize;
 524		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 525		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 526		last = here = ENTRY(header+1);
 527	}
 528
 
 529	/* Iff we are modifying the block in-place, bh is locked here. */
 530
 531	if (not_found) {
 532		/* Insert the new name. */
 533		size_t size = EXT2_XATTR_LEN(name_len);
 534		size_t rest = (char *)last - (char *)here;
 535		memmove((char *)here + size, here, rest);
 536		memset(here, 0, size);
 537		here->e_name_index = name_index;
 538		here->e_name_len = name_len;
 539		memcpy(here->e_name, name, name_len);
 540	} else {
 541		if (!here->e_value_block && here->e_value_size) {
 542			char *first_val = (char *)header + min_offs;
 543			size_t offs = le16_to_cpu(here->e_value_offs);
 544			char *val = (char *)header + offs;
 545			size_t size = EXT2_XATTR_SIZE(
 546				le32_to_cpu(here->e_value_size));
 547
 548			if (size == EXT2_XATTR_SIZE(value_len)) {
 549				/* The old and the new value have the same
 550				   size. Just replace. */
 551				here->e_value_size = cpu_to_le32(value_len);
 552				memset(val + size - EXT2_XATTR_PAD, 0,
 553				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 554				memcpy(val, value, value_len);
 555				goto skip_replace;
 556			}
 557
 558			/* Remove the old value. */
 559			memmove(first_val + size, first_val, val - first_val);
 560			memset(first_val, 0, size);
 561			here->e_value_offs = 0;
 562			min_offs += size;
 563
 564			/* Adjust all value offsets. */
 565			last = ENTRY(header+1);
 566			while (!IS_LAST_ENTRY(last)) {
 567				size_t o = le16_to_cpu(last->e_value_offs);
 568				if (!last->e_value_block && o < offs)
 569					last->e_value_offs =
 570						cpu_to_le16(o + size);
 571				last = EXT2_XATTR_NEXT(last);
 572			}
 
 
 573		}
 574		if (value == NULL) {
 575			/* Remove the old name. */
 576			size_t size = EXT2_XATTR_LEN(name_len);
 577			last = ENTRY((char *)last - size);
 578			memmove(here, (char*)here + size,
 579				(char*)last - (char*)here);
 580			memset(last, 0, size);
 581		}
 582	}
 583
 584	if (value != NULL) {
 585		/* Insert the new value. */
 586		here->e_value_size = cpu_to_le32(value_len);
 587		if (value_len) {
 588			size_t size = EXT2_XATTR_SIZE(value_len);
 589			char *val = (char *)header + min_offs - size;
 590			here->e_value_offs =
 591				cpu_to_le16((char *)val - (char *)header);
 592			memset(val + size - EXT2_XATTR_PAD, 0,
 593			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 594			memcpy(val, value, value_len);
 595		}
 596	}
 597
 598skip_replace:
 599	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 600		/* This block is now empty. */
 601		if (bh && header == HDR(bh))
 602			unlock_buffer(bh);  /* we were modifying in-place. */
 603		error = ext2_xattr_set2(inode, bh, NULL);
 604	} else {
 605		ext2_xattr_rehash(header, here);
 606		if (bh && header == HDR(bh))
 607			unlock_buffer(bh);  /* we were modifying in-place. */
 608		error = ext2_xattr_set2(inode, bh, header);
 609	}
 610
 611cleanup:
 612	brelse(bh);
 613	if (!(bh && header == HDR(bh)))
 614		kfree(header);
 
 615	up_write(&EXT2_I(inode)->xattr_sem);
 616
 617	return error;
 618}
 619
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 620/*
 621 * Second half of ext2_xattr_set(): Update the file system.
 622 */
 623static int
 624ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 625		struct ext2_xattr_header *header)
 626{
 627	struct super_block *sb = inode->i_sb;
 628	struct buffer_head *new_bh = NULL;
 629	int error;
 630	struct mb_cache *ext2_mb_cache = EXT2_SB(sb)->s_mb_cache;
 631
 632	if (header) {
 633		new_bh = ext2_xattr_cache_find(inode, header);
 634		if (new_bh) {
 635			/* We found an identical block in the cache. */
 636			if (new_bh == old_bh) {
 637				ea_bdebug(new_bh, "keeping this block");
 638			} else {
 639				/* The old block is released after updating
 640				   the inode.  */
 641				ea_bdebug(new_bh, "reusing block");
 642
 643				error = dquot_alloc_block(inode, 1);
 644				if (error) {
 645					unlock_buffer(new_bh);
 646					goto cleanup;
 647				}
 648				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 649				ea_bdebug(new_bh, "refcount now=%d",
 650					le32_to_cpu(HDR(new_bh)->h_refcount));
 651			}
 652			unlock_buffer(new_bh);
 653		} else if (old_bh && header == HDR(old_bh)) {
 654			/* Keep this block. No need to lock the block as we
 655			   don't need to change the reference count. */
 656			new_bh = old_bh;
 657			get_bh(new_bh);
 658			ext2_xattr_cache_insert(ext2_mb_cache, new_bh);
 659		} else {
 660			/* We need to allocate a new block */
 661			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 662						EXT2_I(inode)->i_block_group);
 663			int block = ext2_new_block(inode, goal, &error);
 664			if (error)
 665				goto cleanup;
 666			ea_idebug(inode, "creating block %d", block);
 667
 668			new_bh = sb_getblk(sb, block);
 669			if (unlikely(!new_bh)) {
 670				ext2_free_blocks(inode, block, 1);
 671				mark_inode_dirty(inode);
 672				error = -ENOMEM;
 673				goto cleanup;
 674			}
 675			lock_buffer(new_bh);
 676			memcpy(new_bh->b_data, header, new_bh->b_size);
 677			set_buffer_uptodate(new_bh);
 678			unlock_buffer(new_bh);
 679			ext2_xattr_cache_insert(ext2_mb_cache, new_bh);
 680			
 681			ext2_xattr_update_super_block(sb);
 682		}
 683		mark_buffer_dirty(new_bh);
 684		if (IS_SYNC(inode)) {
 685			sync_dirty_buffer(new_bh);
 686			error = -EIO;
 687			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 688				goto cleanup;
 689		}
 690	}
 691
 692	/* Update the inode. */
 693	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 694	inode->i_ctime = CURRENT_TIME_SEC;
 695	if (IS_SYNC(inode)) {
 696		error = sync_inode_metadata(inode, 1);
 697		/* In case sync failed due to ENOSPC the inode was actually
 698		 * written (only some dirty data were not) so we just proceed
 699		 * as if nothing happened and cleanup the unused block */
 700		if (error && error != -ENOSPC) {
 701			if (new_bh && new_bh != old_bh) {
 702				dquot_free_block_nodirty(inode, 1);
 703				mark_inode_dirty(inode);
 704			}
 705			goto cleanup;
 706		}
 707	} else
 708		mark_inode_dirty(inode);
 709
 710	error = 0;
 711	if (old_bh && old_bh != new_bh) {
 712		/*
 713		 * If there was an old block and we are no longer using it,
 714		 * release the old block.
 715		 */
 716		lock_buffer(old_bh);
 717		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 718			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
 719
 720			/*
 721			 * This must happen under buffer lock for
 722			 * ext2_xattr_set2() to reliably detect freed block
 723			 */
 724			mb_cache_entry_delete_block(ext2_mb_cache,
 725						    hash, old_bh->b_blocknr);
 726			/* Free the old block. */
 727			ea_bdebug(old_bh, "freeing");
 728			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 729			mark_inode_dirty(inode);
 730			/* We let our caller release old_bh, so we
 731			 * need to duplicate the buffer before. */
 732			get_bh(old_bh);
 733			bforget(old_bh);
 734		} else {
 735			/* Decrement the refcount only. */
 736			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 737			dquot_free_block_nodirty(inode, 1);
 738			mark_inode_dirty(inode);
 739			mark_buffer_dirty(old_bh);
 740			ea_bdebug(old_bh, "refcount now=%d",
 741				le32_to_cpu(HDR(old_bh)->h_refcount));
 742		}
 743		unlock_buffer(old_bh);
 744	}
 745
 746cleanup:
 747	brelse(new_bh);
 748
 749	return error;
 750}
 751
 752/*
 753 * ext2_xattr_delete_inode()
 754 *
 755 * Free extended attribute resources associated with this inode. This
 756 * is called immediately before an inode is freed.
 757 */
 758void
 759ext2_xattr_delete_inode(struct inode *inode)
 760{
 761	struct buffer_head *bh = NULL;
 
 762
 763	down_write(&EXT2_I(inode)->xattr_sem);
 
 
 
 
 
 
 
 
 764	if (!EXT2_I(inode)->i_file_acl)
 765		goto cleanup;
 
 
 
 
 
 
 
 
 766	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 767	if (!bh) {
 768		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 769			"inode %ld: block %d read error", inode->i_ino,
 770			EXT2_I(inode)->i_file_acl);
 771		goto cleanup;
 772	}
 773	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 774	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 775	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 776		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 777			"inode %ld: bad block %d", inode->i_ino,
 778			EXT2_I(inode)->i_file_acl);
 779		goto cleanup;
 780	}
 781	lock_buffer(bh);
 782	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 783		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 784
 785		/*
 786		 * This must happen under buffer lock for ext2_xattr_set2() to
 787		 * reliably detect freed block
 788		 */
 789		mb_cache_entry_delete_block(EXT2_SB(inode->i_sb)->s_mb_cache,
 790					    hash, bh->b_blocknr);
 791		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 792		get_bh(bh);
 793		bforget(bh);
 794		unlock_buffer(bh);
 795	} else {
 796		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 797		ea_bdebug(bh, "refcount now=%d",
 798			le32_to_cpu(HDR(bh)->h_refcount));
 799		unlock_buffer(bh);
 800		mark_buffer_dirty(bh);
 801		if (IS_SYNC(inode))
 802			sync_dirty_buffer(bh);
 803		dquot_free_block_nodirty(inode, 1);
 804	}
 805	EXT2_I(inode)->i_file_acl = 0;
 806
 807cleanup:
 808	brelse(bh);
 809	up_write(&EXT2_I(inode)->xattr_sem);
 810}
 811
 812/*
 813 * ext2_xattr_cache_insert()
 814 *
 815 * Create a new entry in the extended attribute cache, and insert
 816 * it unless such an entry is already in the cache.
 817 *
 818 * Returns 0, or a negative error number on failure.
 819 */
 820static int
 821ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 822{
 823	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 824	int error;
 825
 826	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 1);
 
 827	if (error) {
 828		if (error == -EBUSY) {
 829			ea_bdebug(bh, "already in cache (%d cache entries)",
 830				atomic_read(&ext2_xattr_cache->c_entry_count));
 831			error = 0;
 832		}
 833	} else
 834		ea_bdebug(bh, "inserting [%x]", (int)hash);
 835	return error;
 836}
 837
 838/*
 839 * ext2_xattr_cmp()
 840 *
 841 * Compare two extended attribute blocks for equality.
 842 *
 843 * Returns 0 if the blocks are equal, 1 if they differ, and
 844 * a negative error number on errors.
 845 */
 846static int
 847ext2_xattr_cmp(struct ext2_xattr_header *header1,
 848	       struct ext2_xattr_header *header2)
 849{
 850	struct ext2_xattr_entry *entry1, *entry2;
 851
 852	entry1 = ENTRY(header1+1);
 853	entry2 = ENTRY(header2+1);
 854	while (!IS_LAST_ENTRY(entry1)) {
 855		if (IS_LAST_ENTRY(entry2))
 856			return 1;
 857		if (entry1->e_hash != entry2->e_hash ||
 858		    entry1->e_name_index != entry2->e_name_index ||
 859		    entry1->e_name_len != entry2->e_name_len ||
 860		    entry1->e_value_size != entry2->e_value_size ||
 861		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 862			return 1;
 863		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 864			return -EIO;
 865		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 866			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 867			   le32_to_cpu(entry1->e_value_size)))
 868			return 1;
 869
 870		entry1 = EXT2_XATTR_NEXT(entry1);
 871		entry2 = EXT2_XATTR_NEXT(entry2);
 872	}
 873	if (!IS_LAST_ENTRY(entry2))
 874		return 1;
 875	return 0;
 876}
 877
 878/*
 879 * ext2_xattr_cache_find()
 880 *
 881 * Find an identical extended attribute block.
 882 *
 883 * Returns a locked buffer head to the block found, or NULL if such
 884 * a block was not found or an error occurred.
 885 */
 886static struct buffer_head *
 887ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 888{
 889	__u32 hash = le32_to_cpu(header->h_hash);
 890	struct mb_cache_entry *ce;
 891	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
 892
 893	if (!header->h_hash)
 894		return NULL;  /* never share */
 895	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 896again:
 897	ce = mb_cache_entry_find_first(ext2_mb_cache, hash);
 898	while (ce) {
 899		struct buffer_head *bh;
 900
 901		bh = sb_bread(inode->i_sb, ce->e_block);
 902		if (!bh) {
 903			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 904				"inode %ld: block %ld read error",
 905				inode->i_ino, (unsigned long) ce->e_block);
 906		} else {
 907			lock_buffer(bh);
 908			/*
 909			 * We have to be careful about races with freeing or
 910			 * rehashing of xattr block. Once we hold buffer lock
 911			 * xattr block's state is stable so we can check
 912			 * whether the block got freed / rehashed or not.
 913			 * Since we unhash mbcache entry under buffer lock when
 914			 * freeing / rehashing xattr block, checking whether
 915			 * entry is still hashed is reliable.
 916			 */
 917			if (hlist_bl_unhashed(&ce->e_hash_list)) {
 918				mb_cache_entry_put(ext2_mb_cache, ce);
 919				unlock_buffer(bh);
 920				brelse(bh);
 921				goto again;
 922			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
 923				   EXT2_XATTR_REFCOUNT_MAX) {
 924				ea_idebug(inode, "block %ld refcount %d>%d",
 925					  (unsigned long) ce->e_block,
 926					  le32_to_cpu(HDR(bh)->h_refcount),
 927					  EXT2_XATTR_REFCOUNT_MAX);
 928			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 929				ea_bdebug(bh, "b_count=%d",
 930					  atomic_read(&(bh->b_count)));
 931				mb_cache_entry_touch(ext2_mb_cache, ce);
 932				mb_cache_entry_put(ext2_mb_cache, ce);
 933				return bh;
 934			}
 935			unlock_buffer(bh);
 936			brelse(bh);
 937		}
 938		ce = mb_cache_entry_find_next(ext2_mb_cache, ce);
 939	}
 940	return NULL;
 941}
 942
 943#define NAME_HASH_SHIFT 5
 944#define VALUE_HASH_SHIFT 16
 945
 946/*
 947 * ext2_xattr_hash_entry()
 948 *
 949 * Compute the hash of an extended attribute.
 950 */
 951static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 952					 struct ext2_xattr_entry *entry)
 953{
 954	__u32 hash = 0;
 955	char *name = entry->e_name;
 956	int n;
 957
 958	for (n=0; n < entry->e_name_len; n++) {
 959		hash = (hash << NAME_HASH_SHIFT) ^
 960		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 961		       *name++;
 962	}
 963
 964	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 965		__le32 *value = (__le32 *)((char *)header +
 966			le16_to_cpu(entry->e_value_offs));
 967		for (n = (le32_to_cpu(entry->e_value_size) +
 968		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 969			hash = (hash << VALUE_HASH_SHIFT) ^
 970			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 971			       le32_to_cpu(*value++);
 972		}
 973	}
 974	entry->e_hash = cpu_to_le32(hash);
 975}
 976
 977#undef NAME_HASH_SHIFT
 978#undef VALUE_HASH_SHIFT
 979
 980#define BLOCK_HASH_SHIFT 16
 981
 982/*
 983 * ext2_xattr_rehash()
 984 *
 985 * Re-compute the extended attribute hash value after an entry has changed.
 986 */
 987static void ext2_xattr_rehash(struct ext2_xattr_header *header,
 988			      struct ext2_xattr_entry *entry)
 989{
 990	struct ext2_xattr_entry *here;
 991	__u32 hash = 0;
 992	
 993	ext2_xattr_hash_entry(header, entry);
 994	here = ENTRY(header+1);
 995	while (!IS_LAST_ENTRY(here)) {
 996		if (!here->e_hash) {
 997			/* Block is not shared if an entry's hash value == 0 */
 998			hash = 0;
 999			break;
1000		}
1001		hash = (hash << BLOCK_HASH_SHIFT) ^
1002		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1003		       le32_to_cpu(here->e_hash);
1004		here = EXT2_XATTR_NEXT(here);
1005	}
1006	header->h_hash = cpu_to_le32(hash);
1007}
1008
1009#undef BLOCK_HASH_SHIFT
1010
1011#define HASH_BUCKET_BITS 10
1012
1013struct mb_cache *ext2_xattr_create_cache(void)
1014{
1015	return mb_cache_create(HASH_BUCKET_BITS);
1016}
1017
1018void ext2_xattr_destroy_cache(struct mb_cache *cache)
1019{
1020	if (cache)
1021		mb_cache_destroy(cache);
1022}