Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Feb 10-13, 2025
Register
Loading...
   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 * const 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]  = &nop_posix_acl_access,
 105	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
 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 * const ext2_xattr_handlers[] = {
 114	&ext2_xattr_user_handler,
 115	&ext2_xattr_trusted_handler,
 116#ifdef CONFIG_EXT2_FS_SECURITY
 117	&ext2_xattr_security_handler,
 118#endif
 119	NULL
 120};
 121
 122#define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
 123
 124static inline const char *ext2_xattr_prefix(int name_index,
 125					    struct dentry *dentry)
 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
 132	if (!xattr_handler_can_list(handler, dentry))
 133		return NULL;
 134
 135	return xattr_prefix(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 char *prefix;
 337
 338		prefix = ext2_xattr_prefix(entry->e_name_index, dentry);
 339		if (prefix) {
 340			size_t prefix_len = strlen(prefix);
 341			size_t size = prefix_len + entry->e_name_len + 1;
 342
 343			if (buffer) {
 344				if (size > rest) {
 345					error = -ERANGE;
 346					goto cleanup;
 347				}
 348				memcpy(buffer, prefix, prefix_len);
 349				buffer += prefix_len;
 350				memcpy(buffer, entry->e_name, entry->e_name_len);
 351				buffer += entry->e_name_len;
 352				*buffer++ = 0;
 353			}
 354			rest -= size;
 355		}
 356	}
 357	error = buffer_size - rest;  /* total size */
 358
 359cleanup:
 360	brelse(bh);
 361	up_read(&EXT2_I(inode)->xattr_sem);
 362
 363	return error;
 364}
 365
 366/*
 367 * Inode operation listxattr()
 368 *
 369 * d_inode(dentry)->i_mutex: don't care
 370 */
 371ssize_t
 372ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 373{
 374	return ext2_xattr_list(dentry, buffer, size);
 375}
 376
 377/*
 378 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 379 * not set, set it.
 380 */
 381static void ext2_xattr_update_super_block(struct super_block *sb)
 382{
 383	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 384		return;
 385
 386	spin_lock(&EXT2_SB(sb)->s_lock);
 387	ext2_update_dynamic_rev(sb);
 388	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 389	spin_unlock(&EXT2_SB(sb)->s_lock);
 390	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 391}
 392
 393/*
 394 * ext2_xattr_set()
 395 *
 396 * Create, replace or remove an extended attribute for this inode.  Value
 397 * is NULL to remove an existing extended attribute, and non-NULL to
 398 * either replace an existing extended attribute, or create a new extended
 399 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 400 * specify that an extended attribute must exist and must not exist
 401 * previous to the call, respectively.
 402 *
 403 * Returns 0, or a negative error number on failure.
 404 */
 405int
 406ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 407	       const void *value, size_t value_len, int flags)
 408{
 409	struct super_block *sb = inode->i_sb;
 410	struct buffer_head *bh = NULL;
 411	struct ext2_xattr_header *header = NULL;
 412	struct ext2_xattr_entry *here = NULL, *last = NULL;
 413	size_t name_len, free, min_offs = sb->s_blocksize;
 414	int not_found = 1, error;
 415	char *end;
 416	
 417	/*
 418	 * header -- Points either into bh, or to a temporarily
 419	 *           allocated buffer.
 420	 * here -- The named entry found, or the place for inserting, within
 421	 *         the block pointed to by header.
 422	 * last -- Points right after the last named entry within the block
 423	 *         pointed to by header.
 424	 * min_offs -- The offset of the first value (values are aligned
 425	 *             towards the end of the block).
 426	 * end -- Points right after the block pointed to by header.
 427	 */
 428	
 429	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 430		  name_index, name, value, (long)value_len);
 431
 432	if (value == NULL)
 433		value_len = 0;
 434	if (name == NULL)
 435		return -EINVAL;
 436	name_len = strlen(name);
 437	if (name_len > 255 || value_len > sb->s_blocksize)
 438		return -ERANGE;
 439	error = dquot_initialize(inode);
 440	if (error)
 441		return error;
 442	down_write(&EXT2_I(inode)->xattr_sem);
 443	if (EXT2_I(inode)->i_file_acl) {
 444		/* The inode already has an extended attribute block. */
 445		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 446		error = -EIO;
 447		if (!bh)
 448			goto cleanup;
 449		ea_bdebug(bh, "b_count=%d, refcount=%d",
 450			atomic_read(&(bh->b_count)),
 451			le32_to_cpu(HDR(bh)->h_refcount));
 452		header = HDR(bh);
 453		end = bh->b_data + bh->b_size;
 454		if (!ext2_xattr_header_valid(header)) {
 455bad_block:
 456			ext2_error(sb, "ext2_xattr_set",
 457				"inode %ld: bad block %d", inode->i_ino, 
 458				   EXT2_I(inode)->i_file_acl);
 459			error = -EIO;
 460			goto cleanup;
 461		}
 462		/*
 463		 * Find the named attribute. If not found, 'here' will point
 464		 * to entry where the new attribute should be inserted to
 465		 * maintain sorting.
 466		 */
 467		last = FIRST_ENTRY(bh);
 468		while (!IS_LAST_ENTRY(last)) {
 469			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
 470				goto bad_block;
 471			if (last->e_value_size) {
 472				size_t offs = le16_to_cpu(last->e_value_offs);
 473				if (offs < min_offs)
 474					min_offs = offs;
 475			}
 476			if (not_found > 0) {
 477				not_found = ext2_xattr_cmp_entry(name_index,
 478								 name_len,
 479								 name, last);
 480				if (not_found <= 0)
 481					here = last;
 482			}
 483			last = EXT2_XATTR_NEXT(last);
 484		}
 485		if (not_found > 0)
 486			here = last;
 487
 488		/* Check whether we have enough space left. */
 489		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 490	} else {
 491		/* We will use a new extended attribute block. */
 492		free = sb->s_blocksize -
 493			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 494	}
 495
 496	if (not_found) {
 497		/* Request to remove a nonexistent attribute? */
 498		error = -ENODATA;
 499		if (flags & XATTR_REPLACE)
 500			goto cleanup;
 501		error = 0;
 502		if (value == NULL)
 503			goto cleanup;
 504	} else {
 505		/* Request to create an existing attribute? */
 506		error = -EEXIST;
 507		if (flags & XATTR_CREATE)
 508			goto cleanup;
 509		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
 510		free += EXT2_XATTR_LEN(name_len);
 511	}
 512	error = -ENOSPC;
 513	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 514		goto cleanup;
 515
 516	/* Here we know that we can set the new attribute. */
 517
 518	if (header) {
 519		int offset;
 520
 521		lock_buffer(bh);
 522		if (header->h_refcount == cpu_to_le32(1)) {
 523			__u32 hash = le32_to_cpu(header->h_hash);
 524			struct mb_cache_entry *oe;
 525
 526			oe = mb_cache_entry_delete_or_get(EA_BLOCK_CACHE(inode),
 527					hash, bh->b_blocknr);
 528			if (!oe) {
 529				ea_bdebug(bh, "modifying in-place");
 530				goto update_block;
 531			}
 532			/*
 533			 * Someone is trying to reuse the block, leave it alone
 534			 */
 535			mb_cache_entry_put(EA_BLOCK_CACHE(inode), oe);
 536		}
 537		unlock_buffer(bh);
 538		ea_bdebug(bh, "cloning");
 539		header = kmemdup(HDR(bh), bh->b_size, GFP_KERNEL);
 540		error = -ENOMEM;
 541		if (header == NULL)
 542			goto cleanup;
 543		header->h_refcount = cpu_to_le32(1);
 544
 545		offset = (char *)here - bh->b_data;
 546		here = ENTRY((char *)header + offset);
 547		offset = (char *)last - bh->b_data;
 548		last = ENTRY((char *)header + offset);
 549	} else {
 550		/* Allocate a buffer where we construct the new block. */
 551		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 552		error = -ENOMEM;
 553		if (header == NULL)
 554			goto cleanup;
 555		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 556		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 557		last = here = ENTRY(header+1);
 558	}
 559
 560update_block:
 561	/* Iff we are modifying the block in-place, bh is locked here. */
 562
 563	if (not_found) {
 564		/* Insert the new name. */
 565		size_t size = EXT2_XATTR_LEN(name_len);
 566		size_t rest = (char *)last - (char *)here;
 567		memmove((char *)here + size, here, rest);
 568		memset(here, 0, size);
 569		here->e_name_index = name_index;
 570		here->e_name_len = name_len;
 571		memcpy(here->e_name, name, name_len);
 572	} else {
 573		if (here->e_value_size) {
 574			char *first_val = (char *)header + min_offs;
 575			size_t offs = le16_to_cpu(here->e_value_offs);
 576			char *val = (char *)header + offs;
 577			size_t size = EXT2_XATTR_SIZE(
 578				le32_to_cpu(here->e_value_size));
 579
 580			if (size == EXT2_XATTR_SIZE(value_len)) {
 581				/* The old and the new value have the same
 582				   size. Just replace. */
 583				here->e_value_size = cpu_to_le32(value_len);
 584				memset(val + size - EXT2_XATTR_PAD, 0,
 585				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 586				memcpy(val, value, value_len);
 587				goto skip_replace;
 588			}
 589
 590			/* Remove the old value. */
 591			memmove(first_val + size, first_val, val - first_val);
 592			memset(first_val, 0, size);
 593			min_offs += size;
 594
 595			/* Adjust all value offsets. */
 596			last = ENTRY(header+1);
 597			while (!IS_LAST_ENTRY(last)) {
 598				size_t o = le16_to_cpu(last->e_value_offs);
 599				if (o < offs)
 600					last->e_value_offs =
 601						cpu_to_le16(o + size);
 602				last = EXT2_XATTR_NEXT(last);
 603			}
 604
 605			here->e_value_offs = 0;
 606		}
 607		if (value == NULL) {
 608			/* Remove the old name. */
 609			size_t size = EXT2_XATTR_LEN(name_len);
 610			last = ENTRY((char *)last - size);
 611			memmove(here, (char*)here + size,
 612				(char*)last - (char*)here);
 613			memset(last, 0, size);
 614		}
 615	}
 616
 617	if (value != NULL) {
 618		/* Insert the new value. */
 619		here->e_value_size = cpu_to_le32(value_len);
 620		if (value_len) {
 621			size_t size = EXT2_XATTR_SIZE(value_len);
 622			char *val = (char *)header + min_offs - size;
 623			here->e_value_offs =
 624				cpu_to_le16((char *)val - (char *)header);
 625			memset(val + size - EXT2_XATTR_PAD, 0,
 626			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 627			memcpy(val, value, value_len);
 628		}
 629	}
 630
 631skip_replace:
 632	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 633		/* This block is now empty. */
 634		if (bh && header == HDR(bh))
 635			unlock_buffer(bh);  /* we were modifying in-place. */
 636		error = ext2_xattr_set2(inode, bh, NULL);
 637	} else {
 638		ext2_xattr_rehash(header, here);
 639		if (bh && header == HDR(bh))
 640			unlock_buffer(bh);  /* we were modifying in-place. */
 641		error = ext2_xattr_set2(inode, bh, header);
 642	}
 643
 644cleanup:
 645	if (!(bh && header == HDR(bh)))
 646		kfree(header);
 647	brelse(bh);
 648	up_write(&EXT2_I(inode)->xattr_sem);
 649
 650	return error;
 651}
 652
 653static void ext2_xattr_release_block(struct inode *inode,
 654				     struct buffer_head *bh)
 655{
 656	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 657
 658retry_ref:
 659	lock_buffer(bh);
 660	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 661		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 662		struct mb_cache_entry *oe;
 663
 664		/*
 665		 * This must happen under buffer lock to properly
 666		 * serialize with ext2_xattr_set() reusing the block.
 667		 */
 668		oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
 669						  bh->b_blocknr);
 670		if (oe) {
 671			/*
 672			 * Someone is trying to reuse the block. Wait
 673			 * and retry.
 674			 */
 675			unlock_buffer(bh);
 676			mb_cache_entry_wait_unused(oe);
 677			mb_cache_entry_put(ea_block_cache, oe);
 678			goto retry_ref;
 679		}
 680
 681		/* Free the old block. */
 682		ea_bdebug(bh, "freeing");
 683		ext2_free_blocks(inode, bh->b_blocknr, 1);
 684		/* We let our caller release bh, so we
 685		 * need to duplicate the buffer before. */
 686		get_bh(bh);
 687		bforget(bh);
 688		unlock_buffer(bh);
 689	} else {
 690		/* Decrement the refcount only. */
 691		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 692		dquot_free_block(inode, 1);
 693		mark_buffer_dirty(bh);
 694		unlock_buffer(bh);
 695		ea_bdebug(bh, "refcount now=%d",
 696			le32_to_cpu(HDR(bh)->h_refcount));
 697		if (IS_SYNC(inode))
 698			sync_dirty_buffer(bh);
 699	}
 700}
 701
 702/*
 703 * Second half of ext2_xattr_set(): Update the file system.
 704 */
 705static int
 706ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 707		struct ext2_xattr_header *header)
 708{
 709	struct super_block *sb = inode->i_sb;
 710	struct buffer_head *new_bh = NULL;
 711	int error;
 712	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 713
 714	if (header) {
 715		new_bh = ext2_xattr_cache_find(inode, header);
 716		if (new_bh) {
 717			/* We found an identical block in the cache. */
 718			if (new_bh == old_bh) {
 719				ea_bdebug(new_bh, "keeping this block");
 720			} else {
 721				/* The old block is released after updating
 722				   the inode.  */
 723				ea_bdebug(new_bh, "reusing block");
 724
 725				error = dquot_alloc_block(inode, 1);
 726				if (error) {
 727					unlock_buffer(new_bh);
 728					goto cleanup;
 729				}
 730				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 731				ea_bdebug(new_bh, "refcount now=%d",
 732					le32_to_cpu(HDR(new_bh)->h_refcount));
 733			}
 734			unlock_buffer(new_bh);
 735		} else if (old_bh && header == HDR(old_bh)) {
 736			/* Keep this block. No need to lock the block as we
 737			   don't need to change the reference count. */
 738			new_bh = old_bh;
 739			get_bh(new_bh);
 740			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 741		} else {
 742			/* We need to allocate a new block */
 743			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 744						EXT2_I(inode)->i_block_group);
 745			unsigned long count = 1;
 746			ext2_fsblk_t block = ext2_new_blocks(inode, goal,
 747						&count, &error,
 748						EXT2_ALLOC_NORESERVE);
 749			if (error)
 750				goto cleanup;
 751			ea_idebug(inode, "creating block %lu", block);
 752
 753			new_bh = sb_getblk(sb, block);
 754			if (unlikely(!new_bh)) {
 755				ext2_free_blocks(inode, block, 1);
 756				mark_inode_dirty(inode);
 757				error = -ENOMEM;
 758				goto cleanup;
 759			}
 760			lock_buffer(new_bh);
 761			memcpy(new_bh->b_data, header, new_bh->b_size);
 762			set_buffer_uptodate(new_bh);
 763			unlock_buffer(new_bh);
 764			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 765			
 766			ext2_xattr_update_super_block(sb);
 767		}
 768		mark_buffer_dirty(new_bh);
 769		if (IS_SYNC(inode)) {
 770			sync_dirty_buffer(new_bh);
 771			error = -EIO;
 772			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 773				goto cleanup;
 774		}
 775	}
 776
 777	/* Update the inode. */
 778	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 779	inode_set_ctime_current(inode);
 780	if (IS_SYNC(inode)) {
 781		error = sync_inode_metadata(inode, 1);
 782		/* In case sync failed due to ENOSPC the inode was actually
 783		 * written (only some dirty data were not) so we just proceed
 784		 * as if nothing happened and cleanup the unused block */
 785		if (error && error != -ENOSPC) {
 786			if (new_bh && new_bh != old_bh) {
 787				dquot_free_block_nodirty(inode, 1);
 788				mark_inode_dirty(inode);
 789			}
 790			goto cleanup;
 791		}
 792	} else
 793		mark_inode_dirty(inode);
 794
 795	error = 0;
 796	if (old_bh && old_bh != new_bh) {
 797		/*
 798		 * If there was an old block and we are no longer using it,
 799		 * release the old block.
 800		 */
 801		ext2_xattr_release_block(inode, old_bh);
 802	}
 803
 804cleanup:
 805	brelse(new_bh);
 806
 807	return error;
 808}
 809
 810/*
 811 * ext2_xattr_delete_inode()
 812 *
 813 * Free extended attribute resources associated with this inode. This
 814 * is called immediately before an inode is freed.
 815 */
 816void
 817ext2_xattr_delete_inode(struct inode *inode)
 818{
 819	struct buffer_head *bh = NULL;
 820	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 821
 822	/*
 823	 * We are the only ones holding inode reference. The xattr_sem should
 824	 * better be unlocked! We could as well just not acquire xattr_sem at
 825	 * all but this makes the code more futureproof. OTOH we need trylock
 826	 * here to avoid false-positive warning from lockdep about reclaim
 827	 * circular dependency.
 828	 */
 829	if (WARN_ON_ONCE(!down_write_trylock(&EXT2_I(inode)->xattr_sem)))
 830		return;
 831	if (!EXT2_I(inode)->i_file_acl)
 832		goto cleanup;
 833
 834	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
 835		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 836			"inode %ld: xattr block %d is out of data blocks range",
 837			inode->i_ino, EXT2_I(inode)->i_file_acl);
 838		goto cleanup;
 839	}
 840
 841	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 842	if (!bh) {
 843		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 844			"inode %ld: block %d read error", inode->i_ino,
 845			EXT2_I(inode)->i_file_acl);
 846		goto cleanup;
 847	}
 848	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 849	if (!ext2_xattr_header_valid(HDR(bh))) {
 850		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 851			"inode %ld: bad block %d", inode->i_ino,
 852			EXT2_I(inode)->i_file_acl);
 853		goto cleanup;
 854	}
 855	ext2_xattr_release_block(inode, bh);
 856	EXT2_I(inode)->i_file_acl = 0;
 857
 858cleanup:
 859	brelse(bh);
 860	up_write(&EXT2_I(inode)->xattr_sem);
 861}
 862
 863/*
 864 * ext2_xattr_cache_insert()
 865 *
 866 * Create a new entry in the extended attribute cache, and insert
 867 * it unless such an entry is already in the cache.
 868 *
 869 * Returns 0, or a negative error number on failure.
 870 */
 871static int
 872ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 873{
 874	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 875	int error;
 876
 877	error = mb_cache_entry_create(cache, GFP_KERNEL, hash, bh->b_blocknr,
 878				      true);
 879	if (error) {
 880		if (error == -EBUSY) {
 881			ea_bdebug(bh, "already in cache");
 882			error = 0;
 883		}
 884	} else
 885		ea_bdebug(bh, "inserting [%x]", (int)hash);
 886	return error;
 887}
 888
 889/*
 890 * ext2_xattr_cmp()
 891 *
 892 * Compare two extended attribute blocks for equality.
 893 *
 894 * Returns 0 if the blocks are equal, 1 if they differ, and
 895 * a negative error number on errors.
 896 */
 897static int
 898ext2_xattr_cmp(struct ext2_xattr_header *header1,
 899	       struct ext2_xattr_header *header2)
 900{
 901	struct ext2_xattr_entry *entry1, *entry2;
 902
 903	entry1 = ENTRY(header1+1);
 904	entry2 = ENTRY(header2+1);
 905	while (!IS_LAST_ENTRY(entry1)) {
 906		if (IS_LAST_ENTRY(entry2))
 907			return 1;
 908		if (entry1->e_hash != entry2->e_hash ||
 909		    entry1->e_name_index != entry2->e_name_index ||
 910		    entry1->e_name_len != entry2->e_name_len ||
 911		    entry1->e_value_size != entry2->e_value_size ||
 912		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 913			return 1;
 914		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 915			return -EIO;
 916		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 917			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 918			   le32_to_cpu(entry1->e_value_size)))
 919			return 1;
 920
 921		entry1 = EXT2_XATTR_NEXT(entry1);
 922		entry2 = EXT2_XATTR_NEXT(entry2);
 923	}
 924	if (!IS_LAST_ENTRY(entry2))
 925		return 1;
 926	return 0;
 927}
 928
 929/*
 930 * ext2_xattr_cache_find()
 931 *
 932 * Find an identical extended attribute block.
 933 *
 934 * Returns a locked buffer head to the block found, or NULL if such
 935 * a block was not found or an error occurred.
 936 */
 937static struct buffer_head *
 938ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 939{
 940	__u32 hash = le32_to_cpu(header->h_hash);
 941	struct mb_cache_entry *ce;
 942	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 943
 944	if (!header->h_hash)
 945		return NULL;  /* never share */
 946	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 947
 948	ce = mb_cache_entry_find_first(ea_block_cache, hash);
 949	while (ce) {
 950		struct buffer_head *bh;
 951
 952		bh = sb_bread(inode->i_sb, ce->e_value);
 953		if (!bh) {
 954			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 955				"inode %ld: block %ld read error",
 956				inode->i_ino, (unsigned long) ce->e_value);
 957		} else {
 958			lock_buffer(bh);
 959			if (le32_to_cpu(HDR(bh)->h_refcount) >
 960			    EXT2_XATTR_REFCOUNT_MAX) {
 961				ea_idebug(inode, "block %ld refcount %d>%d",
 962					  (unsigned long) ce->e_value,
 963					  le32_to_cpu(HDR(bh)->h_refcount),
 964					  EXT2_XATTR_REFCOUNT_MAX);
 965			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 966				ea_bdebug(bh, "b_count=%d",
 967					  atomic_read(&(bh->b_count)));
 968				mb_cache_entry_touch(ea_block_cache, ce);
 969				mb_cache_entry_put(ea_block_cache, ce);
 970				return bh;
 971			}
 972			unlock_buffer(bh);
 973			brelse(bh);
 974		}
 975		ce = mb_cache_entry_find_next(ea_block_cache, ce);
 976	}
 977	return NULL;
 978}
 979
 980#define NAME_HASH_SHIFT 5
 981#define VALUE_HASH_SHIFT 16
 982
 983/*
 984 * ext2_xattr_hash_entry()
 985 *
 986 * Compute the hash of an extended attribute.
 987 */
 988static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 989					 struct ext2_xattr_entry *entry)
 990{
 991	__u32 hash = 0;
 992	char *name = entry->e_name;
 993	int n;
 994
 995	for (n=0; n < entry->e_name_len; n++) {
 996		hash = (hash << NAME_HASH_SHIFT) ^
 997		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 998		       *name++;
 999	}
1000
1001	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1002		__le32 *value = (__le32 *)((char *)header +
1003			le16_to_cpu(entry->e_value_offs));
1004		for (n = (le32_to_cpu(entry->e_value_size) +
1005		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1006			hash = (hash << VALUE_HASH_SHIFT) ^
1007			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1008			       le32_to_cpu(*value++);
1009		}
1010	}
1011	entry->e_hash = cpu_to_le32(hash);
1012}
1013
1014#undef NAME_HASH_SHIFT
1015#undef VALUE_HASH_SHIFT
1016
1017#define BLOCK_HASH_SHIFT 16
1018
1019/*
1020 * ext2_xattr_rehash()
1021 *
1022 * Re-compute the extended attribute hash value after an entry has changed.
1023 */
1024static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1025			      struct ext2_xattr_entry *entry)
1026{
1027	struct ext2_xattr_entry *here;
1028	__u32 hash = 0;
1029	
1030	ext2_xattr_hash_entry(header, entry);
1031	here = ENTRY(header+1);
1032	while (!IS_LAST_ENTRY(here)) {
1033		if (!here->e_hash) {
1034			/* Block is not shared if an entry's hash value == 0 */
1035			hash = 0;
1036			break;
1037		}
1038		hash = (hash << BLOCK_HASH_SHIFT) ^
1039		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1040		       le32_to_cpu(here->e_hash);
1041		here = EXT2_XATTR_NEXT(here);
1042	}
1043	header->h_hash = cpu_to_le32(hash);
1044}
1045
1046#undef BLOCK_HASH_SHIFT
1047
1048#define HASH_BUCKET_BITS 10
1049
1050struct mb_cache *ext2_xattr_create_cache(void)
1051{
1052	return mb_cache_create(HASH_BUCKET_BITS);
1053}
1054
1055void ext2_xattr_destroy_cache(struct mb_cache *cache)
1056{
1057	if (cache)
1058		mb_cache_destroy(cache);
1059}
   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/module.h>
  58#include <linux/init.h>
  59#include <linux/slab.h>
  60#include <linux/mbcache.h>
  61#include <linux/quotaops.h>
  62#include <linux/rwsem.h>
  63#include <linux/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		char b[BDEVNAME_SIZE]; \
  82		printk(KERN_DEBUG "block %s:%lu: ", \
  83			bdevname(bh->b_bdev, b), \
  84			(unsigned long) bh->b_blocknr); \
  85		printk(f); \
  86		printk("\n"); \
  87	} while (0)
  88#else
  89# define ea_idebug(f...)
  90# define ea_bdebug(f...)
  91#endif
  92
  93static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  94			   struct ext2_xattr_header *);
  95
  96static int ext2_xattr_cache_insert(struct buffer_head *);
  97static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  98						 struct ext2_xattr_header *);
  99static void ext2_xattr_rehash(struct ext2_xattr_header *,
 100			      struct ext2_xattr_entry *);
 101
 102static struct mb_cache *ext2_xattr_cache;
 103
 104static const struct xattr_handler *ext2_xattr_handler_map[] = {
 105	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
 106#ifdef CONFIG_EXT2_FS_POSIX_ACL
 107	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext2_xattr_acl_access_handler,
 108	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
 109#endif
 110	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
 111#ifdef CONFIG_EXT2_FS_SECURITY
 112	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
 113#endif
 114};
 115
 116const struct xattr_handler *ext2_xattr_handlers[] = {
 117	&ext2_xattr_user_handler,
 118	&ext2_xattr_trusted_handler,
 119#ifdef CONFIG_EXT2_FS_POSIX_ACL
 120	&ext2_xattr_acl_access_handler,
 121	&ext2_xattr_acl_default_handler,
 122#endif
 123#ifdef CONFIG_EXT2_FS_SECURITY
 124	&ext2_xattr_security_handler,
 125#endif
 126	NULL
 127};
 128
 129static inline const struct xattr_handler *
 130ext2_xattr_handler(int name_index)
 131{
 132	const struct xattr_handler *handler = NULL;
 133
 134	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 135		handler = ext2_xattr_handler_map[name_index];
 136	return handler;
 137}
 138
 139/*
 140 * ext2_xattr_get()
 141 *
 142 * Copy an extended attribute into the buffer
 143 * provided, or compute the buffer size required.
 144 * Buffer is NULL to compute the size of the buffer required.
 145 *
 146 * Returns a negative error number on failure, or the number of bytes
 147 * used / required on success.
 148 */
 149int
 150ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 151	       void *buffer, size_t buffer_size)
 152{
 153	struct buffer_head *bh = NULL;
 154	struct ext2_xattr_entry *entry;
 155	size_t name_len, size;
 156	char *end;
 157	int error;
 158
 159	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 160		  name_index, name, buffer, (long)buffer_size);
 161
 162	if (name == NULL)
 163		return -EINVAL;
 164	name_len = strlen(name);
 165	if (name_len > 255)
 166		return -ERANGE;
 167
 168	down_read(&EXT2_I(inode)->xattr_sem);
 169	error = -ENODATA;
 170	if (!EXT2_I(inode)->i_file_acl)
 171		goto cleanup;
 172	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 173	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 174	error = -EIO;
 175	if (!bh)
 176		goto cleanup;
 177	ea_bdebug(bh, "b_count=%d, refcount=%d",
 178		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 179	end = bh->b_data + bh->b_size;
 180	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 181	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 182bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
 183			"inode %ld: bad block %d", inode->i_ino,
 184			EXT2_I(inode)->i_file_acl);
 185		error = -EIO;
 186		goto cleanup;
 187	}
 188
 189	/* find named attribute */
 190	entry = FIRST_ENTRY(bh);
 191	while (!IS_LAST_ENTRY(entry)) {
 192		struct ext2_xattr_entry *next =
 193			EXT2_XATTR_NEXT(entry);
 194		if ((char *)next >= end)
 195			goto bad_block;
 196		if (name_index == entry->e_name_index &&
 197		    name_len == entry->e_name_len &&
 198		    memcmp(name, entry->e_name, name_len) == 0)
 199			goto found;
 200		entry = next;
 201	}
 202	if (ext2_xattr_cache_insert(bh))
 203		ea_idebug(inode, "cache insert failed");
 204	error = -ENODATA;
 205	goto cleanup;
 206found:
 207	/* check the buffer size */
 208	if (entry->e_value_block != 0)
 209		goto bad_block;
 210	size = le32_to_cpu(entry->e_value_size);
 211	if (size > inode->i_sb->s_blocksize ||
 212	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
 213		goto bad_block;
 214
 215	if (ext2_xattr_cache_insert(bh))
 216		ea_idebug(inode, "cache insert failed");
 217	if (buffer) {
 218		error = -ERANGE;
 219		if (size > buffer_size)
 220			goto cleanup;
 221		/* return value of attribute */
 222		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 223			size);
 224	}
 225	error = size;
 226
 227cleanup:
 228	brelse(bh);
 229	up_read(&EXT2_I(inode)->xattr_sem);
 230
 231	return error;
 232}
 233
 234/*
 235 * ext2_xattr_list()
 236 *
 237 * Copy a list of attribute names into the buffer
 238 * provided, or compute the buffer size required.
 239 * Buffer is NULL to compute the size of the buffer required.
 240 *
 241 * Returns a negative error number on failure, or the number of bytes
 242 * used / required on success.
 243 */
 244static int
 245ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 246{
 247	struct inode *inode = dentry->d_inode;
 248	struct buffer_head *bh = NULL;
 249	struct ext2_xattr_entry *entry;
 250	char *end;
 251	size_t rest = buffer_size;
 252	int error;
 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)) {
 271bad_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(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) {
 297			size_t size = handler->list(dentry, buffer, rest,
 298						    entry->e_name,
 299						    entry->e_name_len,
 300						    handler->flags);
 301			if (buffer) {
 302				if (size > rest) {
 303					error = -ERANGE;
 304					goto cleanup;
 305				}
 306				buffer += size;
 307			}
 308			rest -= size;
 309		}
 310	}
 311	error = buffer_size - rest;  /* total size */
 312
 313cleanup:
 314	brelse(bh);
 315	up_read(&EXT2_I(inode)->xattr_sem);
 316
 317	return error;
 318}
 319
 320/*
 321 * Inode operation listxattr()
 322 *
 323 * dentry->d_inode->i_mutex: don't care
 324 */
 325ssize_t
 326ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 327{
 328	return ext2_xattr_list(dentry, buffer, size);
 329}
 330
 331/*
 332 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 333 * not set, set it.
 334 */
 335static void ext2_xattr_update_super_block(struct super_block *sb)
 336{
 337	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 338		return;
 339
 340	spin_lock(&EXT2_SB(sb)->s_lock);
 341	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 342	spin_unlock(&EXT2_SB(sb)->s_lock);
 343	sb->s_dirt = 1;
 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		struct mb_cache_entry *ce;
 487
 488		/* assert(header == HDR(bh)); */
 489		ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
 490					bh->b_blocknr);
 491		lock_buffer(bh);
 492		if (header->h_refcount == cpu_to_le32(1)) {
 493			ea_bdebug(bh, "modifying in-place");
 494			if (ce)
 495				mb_cache_entry_free(ce);
 496			/* keep the buffer locked while modifying it. */
 497		} else {
 498			int offset;
 499
 500			if (ce)
 501				mb_cache_entry_release(ce);
 502			unlock_buffer(bh);
 503			ea_bdebug(bh, "cloning");
 504			header = kmalloc(bh->b_size, GFP_KERNEL);
 505			error = -ENOMEM;
 506			if (header == NULL)
 507				goto cleanup;
 508			memcpy(header, HDR(bh), bh->b_size);
 509			header->h_refcount = cpu_to_le32(1);
 510
 511			offset = (char *)here - bh->b_data;
 512			here = ENTRY((char *)header + offset);
 513			offset = (char *)last - bh->b_data;
 514			last = ENTRY((char *)header + offset);
 515		}
 516	} else {
 517		/* Allocate a buffer where we construct the new block. */
 518		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 519		error = -ENOMEM;
 520		if (header == NULL)
 521			goto cleanup;
 522		end = (char *)header + sb->s_blocksize;
 523		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 524		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 525		last = here = ENTRY(header+1);
 526	}
 527
 528	/* Iff we are modifying the block in-place, bh is locked here. */
 529
 530	if (not_found) {
 531		/* Insert the new name. */
 532		size_t size = EXT2_XATTR_LEN(name_len);
 533		size_t rest = (char *)last - (char *)here;
 534		memmove((char *)here + size, here, rest);
 535		memset(here, 0, size);
 536		here->e_name_index = name_index;
 537		here->e_name_len = name_len;
 538		memcpy(here->e_name, name, name_len);
 539	} else {
 540		if (!here->e_value_block && here->e_value_size) {
 541			char *first_val = (char *)header + min_offs;
 542			size_t offs = le16_to_cpu(here->e_value_offs);
 543			char *val = (char *)header + offs;
 544			size_t size = EXT2_XATTR_SIZE(
 545				le32_to_cpu(here->e_value_size));
 546
 547			if (size == EXT2_XATTR_SIZE(value_len)) {
 548				/* The old and the new value have the same
 549				   size. Just replace. */
 550				here->e_value_size = cpu_to_le32(value_len);
 551				memset(val + size - EXT2_XATTR_PAD, 0,
 552				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 553				memcpy(val, value, value_len);
 554				goto skip_replace;
 555			}
 556
 557			/* Remove the old value. */
 558			memmove(first_val + size, first_val, val - first_val);
 559			memset(first_val, 0, size);
 560			here->e_value_offs = 0;
 561			min_offs += size;
 562
 563			/* Adjust all value offsets. */
 564			last = ENTRY(header+1);
 565			while (!IS_LAST_ENTRY(last)) {
 566				size_t o = le16_to_cpu(last->e_value_offs);
 567				if (!last->e_value_block && o < offs)
 568					last->e_value_offs =
 569						cpu_to_le16(o + size);
 570				last = EXT2_XATTR_NEXT(last);
 571			}
 572		}
 573		if (value == NULL) {
 574			/* Remove the old name. */
 575			size_t size = EXT2_XATTR_LEN(name_len);
 576			last = ENTRY((char *)last - size);
 577			memmove(here, (char*)here + size,
 578				(char*)last - (char*)here);
 579			memset(last, 0, size);
 580		}
 581	}
 582
 583	if (value != NULL) {
 584		/* Insert the new value. */
 585		here->e_value_size = cpu_to_le32(value_len);
 586		if (value_len) {
 587			size_t size = EXT2_XATTR_SIZE(value_len);
 588			char *val = (char *)header + min_offs - size;
 589			here->e_value_offs =
 590				cpu_to_le16((char *)val - (char *)header);
 591			memset(val + size - EXT2_XATTR_PAD, 0,
 592			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 593			memcpy(val, value, value_len);
 594		}
 595	}
 596
 597skip_replace:
 598	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 599		/* This block is now empty. */
 600		if (bh && header == HDR(bh))
 601			unlock_buffer(bh);  /* we were modifying in-place. */
 602		error = ext2_xattr_set2(inode, bh, NULL);
 603	} else {
 604		ext2_xattr_rehash(header, here);
 605		if (bh && header == HDR(bh))
 606			unlock_buffer(bh);  /* we were modifying in-place. */
 607		error = ext2_xattr_set2(inode, bh, header);
 608	}
 609
 610cleanup:
 611	brelse(bh);
 612	if (!(bh && header == HDR(bh)))
 613		kfree(header);
 614	up_write(&EXT2_I(inode)->xattr_sem);
 615
 616	return error;
 617}
 618
 619/*
 620 * Second half of ext2_xattr_set(): Update the file system.
 621 */
 622static int
 623ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 624		struct ext2_xattr_header *header)
 625{
 626	struct super_block *sb = inode->i_sb;
 627	struct buffer_head *new_bh = NULL;
 628	int error;
 629
 630	if (header) {
 631		new_bh = ext2_xattr_cache_find(inode, header);
 632		if (new_bh) {
 633			/* We found an identical block in the cache. */
 634			if (new_bh == old_bh) {
 635				ea_bdebug(new_bh, "keeping this block");
 636			} else {
 637				/* The old block is released after updating
 638				   the inode.  */
 639				ea_bdebug(new_bh, "reusing block");
 640
 641				error = dquot_alloc_block(inode, 1);
 642				if (error) {
 643					unlock_buffer(new_bh);
 644					goto cleanup;
 645				}
 646				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 647				ea_bdebug(new_bh, "refcount now=%d",
 648					le32_to_cpu(HDR(new_bh)->h_refcount));
 649			}
 650			unlock_buffer(new_bh);
 651		} else if (old_bh && header == HDR(old_bh)) {
 652			/* Keep this block. No need to lock the block as we
 653			   don't need to change the reference count. */
 654			new_bh = old_bh;
 655			get_bh(new_bh);
 656			ext2_xattr_cache_insert(new_bh);
 657		} else {
 658			/* We need to allocate a new block */
 659			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 660						EXT2_I(inode)->i_block_group);
 661			int block = ext2_new_block(inode, goal, &error);
 662			if (error)
 663				goto cleanup;
 664			ea_idebug(inode, "creating block %d", block);
 665
 666			new_bh = sb_getblk(sb, block);
 667			if (!new_bh) {
 668				ext2_free_blocks(inode, block, 1);
 669				mark_inode_dirty(inode);
 670				error = -EIO;
 671				goto cleanup;
 672			}
 673			lock_buffer(new_bh);
 674			memcpy(new_bh->b_data, header, new_bh->b_size);
 675			set_buffer_uptodate(new_bh);
 676			unlock_buffer(new_bh);
 677			ext2_xattr_cache_insert(new_bh);
 678			
 679			ext2_xattr_update_super_block(sb);
 680		}
 681		mark_buffer_dirty(new_bh);
 682		if (IS_SYNC(inode)) {
 683			sync_dirty_buffer(new_bh);
 684			error = -EIO;
 685			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 686				goto cleanup;
 687		}
 688	}
 689
 690	/* Update the inode. */
 691	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 692	inode->i_ctime = CURRENT_TIME_SEC;
 693	if (IS_SYNC(inode)) {
 694		error = sync_inode_metadata(inode, 1);
 695		/* In case sync failed due to ENOSPC the inode was actually
 696		 * written (only some dirty data were not) so we just proceed
 697		 * as if nothing happened and cleanup the unused block */
 698		if (error && error != -ENOSPC) {
 699			if (new_bh && new_bh != old_bh) {
 700				dquot_free_block_nodirty(inode, 1);
 701				mark_inode_dirty(inode);
 702			}
 703			goto cleanup;
 704		}
 705	} else
 706		mark_inode_dirty(inode);
 707
 708	error = 0;
 709	if (old_bh && old_bh != new_bh) {
 710		struct mb_cache_entry *ce;
 711
 712		/*
 713		 * If there was an old block and we are no longer using it,
 714		 * release the old block.
 715		 */
 716		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
 717					old_bh->b_blocknr);
 718		lock_buffer(old_bh);
 719		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 720			/* Free the old block. */
 721			if (ce)
 722				mb_cache_entry_free(ce);
 723			ea_bdebug(old_bh, "freeing");
 724			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 725			mark_inode_dirty(inode);
 726			/* We let our caller release old_bh, so we
 727			 * need to duplicate the buffer before. */
 728			get_bh(old_bh);
 729			bforget(old_bh);
 730		} else {
 731			/* Decrement the refcount only. */
 732			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 733			if (ce)
 734				mb_cache_entry_release(ce);
 735			dquot_free_block_nodirty(inode, 1);
 736			mark_inode_dirty(inode);
 737			mark_buffer_dirty(old_bh);
 738			ea_bdebug(old_bh, "refcount now=%d",
 739				le32_to_cpu(HDR(old_bh)->h_refcount));
 740		}
 741		unlock_buffer(old_bh);
 742	}
 743
 744cleanup:
 745	brelse(new_bh);
 746
 747	return error;
 748}
 749
 750/*
 751 * ext2_xattr_delete_inode()
 752 *
 753 * Free extended attribute resources associated with this inode. This
 754 * is called immediately before an inode is freed.
 755 */
 756void
 757ext2_xattr_delete_inode(struct inode *inode)
 758{
 759	struct buffer_head *bh = NULL;
 760	struct mb_cache_entry *ce;
 761
 762	down_write(&EXT2_I(inode)->xattr_sem);
 763	if (!EXT2_I(inode)->i_file_acl)
 764		goto cleanup;
 765	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 766	if (!bh) {
 767		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 768			"inode %ld: block %d read error", inode->i_ino,
 769			EXT2_I(inode)->i_file_acl);
 770		goto cleanup;
 771	}
 772	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 773	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 774	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 775		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 776			"inode %ld: bad block %d", inode->i_ino,
 777			EXT2_I(inode)->i_file_acl);
 778		goto cleanup;
 779	}
 780	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
 781	lock_buffer(bh);
 782	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 783		if (ce)
 784			mb_cache_entry_free(ce);
 785		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 786		get_bh(bh);
 787		bforget(bh);
 788		unlock_buffer(bh);
 789	} else {
 790		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 791		if (ce)
 792			mb_cache_entry_release(ce);
 793		ea_bdebug(bh, "refcount now=%d",
 794			le32_to_cpu(HDR(bh)->h_refcount));
 795		unlock_buffer(bh);
 796		mark_buffer_dirty(bh);
 797		if (IS_SYNC(inode))
 798			sync_dirty_buffer(bh);
 799		dquot_free_block_nodirty(inode, 1);
 800	}
 801	EXT2_I(inode)->i_file_acl = 0;
 802
 803cleanup:
 804	brelse(bh);
 805	up_write(&EXT2_I(inode)->xattr_sem);
 806}
 807
 808/*
 809 * ext2_xattr_put_super()
 810 *
 811 * This is called when a file system is unmounted.
 812 */
 813void
 814ext2_xattr_put_super(struct super_block *sb)
 815{
 816	mb_cache_shrink(sb->s_bdev);
 817}
 818
 819
 820/*
 821 * ext2_xattr_cache_insert()
 822 *
 823 * Create a new entry in the extended attribute cache, and insert
 824 * it unless such an entry is already in the cache.
 825 *
 826 * Returns 0, or a negative error number on failure.
 827 */
 828static int
 829ext2_xattr_cache_insert(struct buffer_head *bh)
 830{
 831	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 832	struct mb_cache_entry *ce;
 833	int error;
 834
 835	ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
 836	if (!ce)
 837		return -ENOMEM;
 838	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
 839	if (error) {
 840		mb_cache_entry_free(ce);
 841		if (error == -EBUSY) {
 842			ea_bdebug(bh, "already in cache (%d cache entries)",
 843				atomic_read(&ext2_xattr_cache->c_entry_count));
 844			error = 0;
 845		}
 846	} else {
 847		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
 848			  atomic_read(&ext2_xattr_cache->c_entry_count));
 849		mb_cache_entry_release(ce);
 850	}
 851	return error;
 852}
 853
 854/*
 855 * ext2_xattr_cmp()
 856 *
 857 * Compare two extended attribute blocks for equality.
 858 *
 859 * Returns 0 if the blocks are equal, 1 if they differ, and
 860 * a negative error number on errors.
 861 */
 862static int
 863ext2_xattr_cmp(struct ext2_xattr_header *header1,
 864	       struct ext2_xattr_header *header2)
 865{
 866	struct ext2_xattr_entry *entry1, *entry2;
 867
 868	entry1 = ENTRY(header1+1);
 869	entry2 = ENTRY(header2+1);
 870	while (!IS_LAST_ENTRY(entry1)) {
 871		if (IS_LAST_ENTRY(entry2))
 872			return 1;
 873		if (entry1->e_hash != entry2->e_hash ||
 874		    entry1->e_name_index != entry2->e_name_index ||
 875		    entry1->e_name_len != entry2->e_name_len ||
 876		    entry1->e_value_size != entry2->e_value_size ||
 877		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 878			return 1;
 879		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 880			return -EIO;
 881		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 882			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 883			   le32_to_cpu(entry1->e_value_size)))
 884			return 1;
 885
 886		entry1 = EXT2_XATTR_NEXT(entry1);
 887		entry2 = EXT2_XATTR_NEXT(entry2);
 888	}
 889	if (!IS_LAST_ENTRY(entry2))
 890		return 1;
 891	return 0;
 892}
 893
 894/*
 895 * ext2_xattr_cache_find()
 896 *
 897 * Find an identical extended attribute block.
 898 *
 899 * Returns a locked buffer head to the block found, or NULL if such
 900 * a block was not found or an error occurred.
 901 */
 902static struct buffer_head *
 903ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 904{
 905	__u32 hash = le32_to_cpu(header->h_hash);
 906	struct mb_cache_entry *ce;
 907
 908	if (!header->h_hash)
 909		return NULL;  /* never share */
 910	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 911again:
 912	ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev,
 913				       hash);
 914	while (ce) {
 915		struct buffer_head *bh;
 916
 917		if (IS_ERR(ce)) {
 918			if (PTR_ERR(ce) == -EAGAIN)
 919				goto again;
 920			break;
 921		}
 922
 923		bh = sb_bread(inode->i_sb, ce->e_block);
 924		if (!bh) {
 925			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 926				"inode %ld: block %ld read error",
 927				inode->i_ino, (unsigned long) ce->e_block);
 928		} else {
 929			lock_buffer(bh);
 930			if (le32_to_cpu(HDR(bh)->h_refcount) >
 931				   EXT2_XATTR_REFCOUNT_MAX) {
 932				ea_idebug(inode, "block %ld refcount %d>%d",
 933					  (unsigned long) ce->e_block,
 934					  le32_to_cpu(HDR(bh)->h_refcount),
 935					  EXT2_XATTR_REFCOUNT_MAX);
 936			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 937				ea_bdebug(bh, "b_count=%d",
 938					  atomic_read(&(bh->b_count)));
 939				mb_cache_entry_release(ce);
 940				return bh;
 941			}
 942			unlock_buffer(bh);
 943			brelse(bh);
 944		}
 945		ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
 946	}
 947	return NULL;
 948}
 949
 950#define NAME_HASH_SHIFT 5
 951#define VALUE_HASH_SHIFT 16
 952
 953/*
 954 * ext2_xattr_hash_entry()
 955 *
 956 * Compute the hash of an extended attribute.
 957 */
 958static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 959					 struct ext2_xattr_entry *entry)
 960{
 961	__u32 hash = 0;
 962	char *name = entry->e_name;
 963	int n;
 964
 965	for (n=0; n < entry->e_name_len; n++) {
 966		hash = (hash << NAME_HASH_SHIFT) ^
 967		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 968		       *name++;
 969	}
 970
 971	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 972		__le32 *value = (__le32 *)((char *)header +
 973			le16_to_cpu(entry->e_value_offs));
 974		for (n = (le32_to_cpu(entry->e_value_size) +
 975		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 976			hash = (hash << VALUE_HASH_SHIFT) ^
 977			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 978			       le32_to_cpu(*value++);
 979		}
 980	}
 981	entry->e_hash = cpu_to_le32(hash);
 982}
 983
 984#undef NAME_HASH_SHIFT
 985#undef VALUE_HASH_SHIFT
 986
 987#define BLOCK_HASH_SHIFT 16
 988
 989/*
 990 * ext2_xattr_rehash()
 991 *
 992 * Re-compute the extended attribute hash value after an entry has changed.
 993 */
 994static void ext2_xattr_rehash(struct ext2_xattr_header *header,
 995			      struct ext2_xattr_entry *entry)
 996{
 997	struct ext2_xattr_entry *here;
 998	__u32 hash = 0;
 999	
1000	ext2_xattr_hash_entry(header, entry);
1001	here = ENTRY(header+1);
1002	while (!IS_LAST_ENTRY(here)) {
1003		if (!here->e_hash) {
1004			/* Block is not shared if an entry's hash value == 0 */
1005			hash = 0;
1006			break;
1007		}
1008		hash = (hash << BLOCK_HASH_SHIFT) ^
1009		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1010		       le32_to_cpu(here->e_hash);
1011		here = EXT2_XATTR_NEXT(here);
1012	}
1013	header->h_hash = cpu_to_le32(hash);
1014}
1015
1016#undef BLOCK_HASH_SHIFT
1017
1018int __init
1019init_ext2_xattr(void)
1020{
1021	ext2_xattr_cache = mb_cache_create("ext2_xattr", 6);
1022	if (!ext2_xattr_cache)
1023		return -ENOMEM;
1024	return 0;
1025}
1026
1027void
1028exit_ext2_xattr(void)
1029{
1030	mb_cache_destroy(ext2_xattr_cache);
1031}