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}
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/ext2/xattr.c
   4 *
   5 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
   6 *
   7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8 * Extended attributes for symlinks and special files added per
   9 *  suggestion of Luka Renko <luka.renko@hermes.si>.
  10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  11 *  Red Hat Inc.
  12 *
  13 */
  14
  15/*
  16 * Extended attributes are stored on disk blocks allocated outside of
  17 * any inode. The i_file_acl field is then made to point to this allocated
  18 * block. If all extended attributes of an inode are identical, these
  19 * inodes may share the same extended attribute block. Such situations
  20 * are automatically detected by keeping a cache of recent attribute block
  21 * numbers and hashes over the block's contents in memory.
  22 *
  23 *
  24 * Extended attribute block layout:
  25 *
  26 *   +------------------+
  27 *   | header           |
  28 *   | entry 1          | |
  29 *   | entry 2          | | growing downwards
  30 *   | entry 3          | v
  31 *   | four null bytes  |
  32 *   | . . .            |
  33 *   | value 1          | ^
  34 *   | value 3          | | growing upwards
  35 *   | value 2          | |
  36 *   +------------------+
  37 *
  38 * The block header is followed by multiple entry descriptors. These entry
  39 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
  40 * byte boundaries. The entry descriptors are sorted by attribute name,
  41 * so that two extended attribute blocks can be compared efficiently.
  42 *
  43 * Attribute values are aligned to the end of the block, stored in
  44 * no specific order. They are also padded to EXT2_XATTR_PAD byte
  45 * boundaries. No additional gaps are left between them.
  46 *
  47 * Locking strategy
  48 * ----------------
  49 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
  50 * EA blocks are only changed if they are exclusive to an inode, so
  51 * holding xattr_sem also means that nothing but the EA block's reference
  52 * count will change. Multiple writers to an EA block are synchronized
  53 * by the bh lock. No more than a single bh lock is held at any time
  54 * to avoid deadlocks.
  55 */
  56
  57#include <linux/buffer_head.h>
  58#include <linux/init.h>
 
  59#include <linux/slab.h>
  60#include <linux/mbcache.h>
  61#include <linux/quotaops.h>
  62#include <linux/rwsem.h>
  63#include <linux/security.h>
  64#include "ext2.h"
  65#include "xattr.h"
  66#include "acl.h"
  67
  68#define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
  69#define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
  70#define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
  71#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
  72
  73#ifdef EXT2_XATTR_DEBUG
  74# define ea_idebug(inode, f...) do { \
  75		printk(KERN_DEBUG "inode %s:%ld: ", \
  76			inode->i_sb->s_id, inode->i_ino); \
  77		printk(f); \
  78		printk("\n"); \
  79	} while (0)
  80# define ea_bdebug(bh, f...) do { \
  81		printk(KERN_DEBUG "block %pg:%lu: ", \
  82			bh->b_bdev, (unsigned long) bh->b_blocknr); \
  83		printk(f); \
  84		printk("\n"); \
  85	} while (0)
  86#else
  87# define ea_idebug(f...)
  88# define ea_bdebug(f...)
  89#endif
  90
  91static int ext2_xattr_set2(struct inode *, struct buffer_head *,
  92			   struct ext2_xattr_header *);
  93
  94static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
  95static struct buffer_head *ext2_xattr_cache_find(struct inode *,
  96						 struct ext2_xattr_header *);
  97static void ext2_xattr_rehash(struct ext2_xattr_header *,
  98			      struct ext2_xattr_entry *);
  99
 100static const struct xattr_handler *ext2_xattr_handler_map[] = {
 101	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
 102#ifdef CONFIG_EXT2_FS_POSIX_ACL
 103	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
 104	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
 105#endif
 106	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
 107#ifdef CONFIG_EXT2_FS_SECURITY
 108	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
 109#endif
 110};
 111
 112const struct xattr_handler *ext2_xattr_handlers[] = {
 113	&ext2_xattr_user_handler,
 114	&ext2_xattr_trusted_handler,
 115#ifdef CONFIG_EXT2_FS_POSIX_ACL
 116	&posix_acl_access_xattr_handler,
 117	&posix_acl_default_xattr_handler,
 118#endif
 119#ifdef CONFIG_EXT2_FS_SECURITY
 120	&ext2_xattr_security_handler,
 121#endif
 122	NULL
 123};
 124
 125#define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
 126
 127static inline const struct xattr_handler *
 128ext2_xattr_handler(int name_index)
 129{
 130	const struct xattr_handler *handler = NULL;
 131
 132	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
 133		handler = ext2_xattr_handler_map[name_index];
 134	return handler;
 135}
 136
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 137/*
 138 * ext2_xattr_get()
 139 *
 140 * Copy an extended attribute into the buffer
 141 * provided, or compute the buffer size required.
 142 * Buffer is NULL to compute the size of the buffer required.
 143 *
 144 * Returns a negative error number on failure, or the number of bytes
 145 * used / required on success.
 146 */
 147int
 148ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 149	       void *buffer, size_t buffer_size)
 150{
 151	struct buffer_head *bh = NULL;
 152	struct ext2_xattr_entry *entry;
 153	size_t name_len, size;
 154	char *end;
 155	int error;
 156	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 157
 158	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 159		  name_index, name, buffer, (long)buffer_size);
 160
 161	if (name == NULL)
 162		return -EINVAL;
 163	name_len = strlen(name);
 164	if (name_len > 255)
 165		return -ERANGE;
 166
 167	down_read(&EXT2_I(inode)->xattr_sem);
 168	error = -ENODATA;
 169	if (!EXT2_I(inode)->i_file_acl)
 170		goto cleanup;
 171	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 172	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 173	error = -EIO;
 174	if (!bh)
 175		goto cleanup;
 176	ea_bdebug(bh, "b_count=%d, refcount=%d",
 177		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 178	end = bh->b_data + bh->b_size;
 179	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 180	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 181bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
 182			"inode %ld: bad block %d", inode->i_ino,
 183			EXT2_I(inode)->i_file_acl);
 184		error = -EIO;
 185		goto cleanup;
 186	}
 187
 188	/* find named attribute */
 189	entry = FIRST_ENTRY(bh);
 190	while (!IS_LAST_ENTRY(entry)) {
 191		struct ext2_xattr_entry *next =
 192			EXT2_XATTR_NEXT(entry);
 193		if ((char *)next >= end)
 194			goto bad_block;
 195		if (name_index == entry->e_name_index &&
 196		    name_len == entry->e_name_len &&
 197		    memcmp(name, entry->e_name, name_len) == 0)
 
 198			goto found;
 199		entry = next;
 
 
 
 200	}
 201	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 202		ea_idebug(inode, "cache insert failed");
 203	error = -ENODATA;
 204	goto cleanup;
 205found:
 206	/* check the buffer size */
 207	if (entry->e_value_block != 0)
 208		goto bad_block;
 209	size = le32_to_cpu(entry->e_value_size);
 210	if (size > inode->i_sb->s_blocksize ||
 211	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
 212		goto bad_block;
 213
 214	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 215		ea_idebug(inode, "cache insert failed");
 216	if (buffer) {
 217		error = -ERANGE;
 218		if (size > buffer_size)
 219			goto cleanup;
 220		/* return value of attribute */
 221		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 222			size);
 223	}
 224	error = size;
 225
 226cleanup:
 227	brelse(bh);
 228	up_read(&EXT2_I(inode)->xattr_sem);
 229
 230	return error;
 231}
 232
 233/*
 234 * ext2_xattr_list()
 235 *
 236 * Copy a list of attribute names into the buffer
 237 * provided, or compute the buffer size required.
 238 * Buffer is NULL to compute the size of the buffer required.
 239 *
 240 * Returns a negative error number on failure, or the number of bytes
 241 * used / required on success.
 242 */
 243static int
 244ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 245{
 246	struct inode *inode = d_inode(dentry);
 247	struct buffer_head *bh = NULL;
 248	struct ext2_xattr_entry *entry;
 249	char *end;
 250	size_t rest = buffer_size;
 251	int error;
 252	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 253
 254	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 255		  buffer, (long)buffer_size);
 256
 257	down_read(&EXT2_I(inode)->xattr_sem);
 258	error = 0;
 259	if (!EXT2_I(inode)->i_file_acl)
 260		goto cleanup;
 261	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 262	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 263	error = -EIO;
 264	if (!bh)
 265		goto cleanup;
 266	ea_bdebug(bh, "b_count=%d, refcount=%d",
 267		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 268	end = bh->b_data + bh->b_size;
 269	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 270	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 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(ea_block_cache, bh))
 288		ea_idebug(inode, "cache insert failed");
 289
 290	/* list the attribute names */
 291	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 292	     entry = EXT2_XATTR_NEXT(entry)) {
 293		const struct xattr_handler *handler =
 294			ext2_xattr_handler(entry->e_name_index);
 295
 296		if (handler && (!handler->list || handler->list(dentry))) {
 297			const char *prefix = handler->prefix ?: handler->name;
 298			size_t prefix_len = strlen(prefix);
 299			size_t size = prefix_len + entry->e_name_len + 1;
 300
 301			if (buffer) {
 302				if (size > rest) {
 303					error = -ERANGE;
 304					goto cleanup;
 305				}
 306				memcpy(buffer, prefix, prefix_len);
 307				buffer += prefix_len;
 308				memcpy(buffer, entry->e_name, entry->e_name_len);
 309				buffer += entry->e_name_len;
 310				*buffer++ = 0;
 311			}
 312			rest -= size;
 313		}
 314	}
 315	error = buffer_size - rest;  /* total size */
 316
 317cleanup:
 318	brelse(bh);
 319	up_read(&EXT2_I(inode)->xattr_sem);
 320
 321	return error;
 322}
 323
 324/*
 325 * Inode operation listxattr()
 326 *
 327 * d_inode(dentry)->i_mutex: don't care
 328 */
 329ssize_t
 330ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 331{
 332	return ext2_xattr_list(dentry, buffer, size);
 333}
 334
 335/*
 336 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 337 * not set, set it.
 338 */
 339static void ext2_xattr_update_super_block(struct super_block *sb)
 340{
 341	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 342		return;
 343
 344	spin_lock(&EXT2_SB(sb)->s_lock);
 
 345	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 346	spin_unlock(&EXT2_SB(sb)->s_lock);
 347	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 348}
 349
 350/*
 351 * ext2_xattr_set()
 352 *
 353 * Create, replace or remove an extended attribute for this inode.  Value
 354 * is NULL to remove an existing extended attribute, and non-NULL to
 355 * either replace an existing extended attribute, or create a new extended
 356 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 357 * specify that an extended attribute must exist and must not exist
 358 * previous to the call, respectively.
 359 *
 360 * Returns 0, or a negative error number on failure.
 361 */
 362int
 363ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 364	       const void *value, size_t value_len, int flags)
 365{
 366	struct super_block *sb = inode->i_sb;
 367	struct buffer_head *bh = NULL;
 368	struct ext2_xattr_header *header = NULL;
 369	struct ext2_xattr_entry *here, *last;
 370	size_t name_len, free, min_offs = sb->s_blocksize;
 371	int not_found = 1, error;
 372	char *end;
 373	
 374	/*
 375	 * header -- Points either into bh, or to a temporarily
 376	 *           allocated buffer.
 377	 * here -- The named entry found, or the place for inserting, within
 378	 *         the block pointed to by header.
 379	 * last -- Points right after the last named entry within the block
 380	 *         pointed to by header.
 381	 * min_offs -- The offset of the first value (values are aligned
 382	 *             towards the end of the block).
 383	 * end -- Points right after the block pointed to by header.
 384	 */
 385	
 386	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 387		  name_index, name, value, (long)value_len);
 388
 389	if (value == NULL)
 390		value_len = 0;
 391	if (name == NULL)
 392		return -EINVAL;
 393	name_len = strlen(name);
 394	if (name_len > 255 || value_len > sb->s_blocksize)
 395		return -ERANGE;
 
 
 
 396	down_write(&EXT2_I(inode)->xattr_sem);
 397	if (EXT2_I(inode)->i_file_acl) {
 398		/* The inode already has an extended attribute block. */
 399		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 400		error = -EIO;
 401		if (!bh)
 402			goto cleanup;
 403		ea_bdebug(bh, "b_count=%d, refcount=%d",
 404			atomic_read(&(bh->b_count)),
 405			le32_to_cpu(HDR(bh)->h_refcount));
 406		header = HDR(bh);
 407		end = bh->b_data + bh->b_size;
 408		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 409		    header->h_blocks != cpu_to_le32(1)) {
 410bad_block:		ext2_error(sb, "ext2_xattr_set",
 411				"inode %ld: bad block %d", inode->i_ino, 
 412				   EXT2_I(inode)->i_file_acl);
 413			error = -EIO;
 414			goto cleanup;
 415		}
 416		/* Find the named attribute. */
 417		here = FIRST_ENTRY(bh);
 418		while (!IS_LAST_ENTRY(here)) {
 419			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
 420			if ((char *)next >= end)
 421				goto bad_block;
 422			if (!here->e_value_block && here->e_value_size) {
 423				size_t offs = le16_to_cpu(here->e_value_offs);
 424				if (offs < min_offs)
 425					min_offs = offs;
 426			}
 427			not_found = name_index - here->e_name_index;
 428			if (!not_found)
 429				not_found = name_len - here->e_name_len;
 430			if (!not_found)
 431				not_found = memcmp(name, here->e_name,name_len);
 432			if (not_found <= 0)
 433				break;
 434			here = next;
 435		}
 436		last = here;
 437		/* We still need to compute min_offs and last. */
 438		while (!IS_LAST_ENTRY(last)) {
 439			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
 440			if ((char *)next >= end)
 441				goto bad_block;
 442			if (!last->e_value_block && last->e_value_size) {
 443				size_t offs = le16_to_cpu(last->e_value_offs);
 444				if (offs < min_offs)
 445					min_offs = offs;
 446			}
 447			last = next;
 
 
 
 
 
 
 
 448		}
 
 
 449
 450		/* Check whether we have enough space left. */
 451		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 452	} else {
 453		/* We will use a new extended attribute block. */
 454		free = sb->s_blocksize -
 455			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 456		here = last = NULL;  /* avoid gcc uninitialized warning. */
 457	}
 458
 459	if (not_found) {
 460		/* Request to remove a nonexistent attribute? */
 461		error = -ENODATA;
 462		if (flags & XATTR_REPLACE)
 463			goto cleanup;
 464		error = 0;
 465		if (value == NULL)
 466			goto cleanup;
 467	} else {
 468		/* Request to create an existing attribute? */
 469		error = -EEXIST;
 470		if (flags & XATTR_CREATE)
 471			goto cleanup;
 472		if (!here->e_value_block && here->e_value_size) {
 473			size_t size = le32_to_cpu(here->e_value_size);
 474
 475			if (le16_to_cpu(here->e_value_offs) + size > 
 476			    sb->s_blocksize || size > sb->s_blocksize)
 477				goto bad_block;
 478			free += EXT2_XATTR_SIZE(size);
 479		}
 480		free += EXT2_XATTR_LEN(name_len);
 481	}
 482	error = -ENOSPC;
 483	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 484		goto cleanup;
 485
 486	/* Here we know that we can set the new attribute. */
 487
 488	if (header) {
 489		/* assert(header == HDR(bh)); */
 
 490		lock_buffer(bh);
 491		if (header->h_refcount == cpu_to_le32(1)) {
 492			__u32 hash = le32_to_cpu(header->h_hash);
 
 493
 494			ea_bdebug(bh, "modifying in-place");
 
 
 
 
 
 495			/*
 496			 * This must happen under buffer lock for
 497			 * ext2_xattr_set2() to reliably detect modified block
 498			 */
 499			mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 500					      bh->b_blocknr);
 
 
 
 
 
 
 
 501
 502			/* keep the buffer locked while modifying it. */
 503		} else {
 504			int offset;
 505
 506			unlock_buffer(bh);
 507			ea_bdebug(bh, "cloning");
 508			header = kmalloc(bh->b_size, GFP_KERNEL);
 509			error = -ENOMEM;
 510			if (header == NULL)
 511				goto cleanup;
 512			memcpy(header, HDR(bh), bh->b_size);
 513			header->h_refcount = cpu_to_le32(1);
 514
 515			offset = (char *)here - bh->b_data;
 516			here = ENTRY((char *)header + offset);
 517			offset = (char *)last - bh->b_data;
 518			last = ENTRY((char *)header + offset);
 519		}
 520	} else {
 521		/* Allocate a buffer where we construct the new block. */
 522		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 523		error = -ENOMEM;
 524		if (header == NULL)
 525			goto cleanup;
 526		end = (char *)header + sb->s_blocksize;
 527		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 528		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 529		last = here = ENTRY(header+1);
 530	}
 531
 
 532	/* Iff we are modifying the block in-place, bh is locked here. */
 533
 534	if (not_found) {
 535		/* Insert the new name. */
 536		size_t size = EXT2_XATTR_LEN(name_len);
 537		size_t rest = (char *)last - (char *)here;
 538		memmove((char *)here + size, here, rest);
 539		memset(here, 0, size);
 540		here->e_name_index = name_index;
 541		here->e_name_len = name_len;
 542		memcpy(here->e_name, name, name_len);
 543	} else {
 544		if (!here->e_value_block && here->e_value_size) {
 545			char *first_val = (char *)header + min_offs;
 546			size_t offs = le16_to_cpu(here->e_value_offs);
 547			char *val = (char *)header + offs;
 548			size_t size = EXT2_XATTR_SIZE(
 549				le32_to_cpu(here->e_value_size));
 550
 551			if (size == EXT2_XATTR_SIZE(value_len)) {
 552				/* The old and the new value have the same
 553				   size. Just replace. */
 554				here->e_value_size = cpu_to_le32(value_len);
 555				memset(val + size - EXT2_XATTR_PAD, 0,
 556				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 557				memcpy(val, value, value_len);
 558				goto skip_replace;
 559			}
 560
 561			/* Remove the old value. */
 562			memmove(first_val + size, first_val, val - first_val);
 563			memset(first_val, 0, size);
 564			here->e_value_offs = 0;
 565			min_offs += size;
 566
 567			/* Adjust all value offsets. */
 568			last = ENTRY(header+1);
 569			while (!IS_LAST_ENTRY(last)) {
 570				size_t o = le16_to_cpu(last->e_value_offs);
 571				if (!last->e_value_block && o < offs)
 572					last->e_value_offs =
 573						cpu_to_le16(o + size);
 574				last = EXT2_XATTR_NEXT(last);
 575			}
 
 
 576		}
 577		if (value == NULL) {
 578			/* Remove the old name. */
 579			size_t size = EXT2_XATTR_LEN(name_len);
 580			last = ENTRY((char *)last - size);
 581			memmove(here, (char*)here + size,
 582				(char*)last - (char*)here);
 583			memset(last, 0, size);
 584		}
 585	}
 586
 587	if (value != NULL) {
 588		/* Insert the new value. */
 589		here->e_value_size = cpu_to_le32(value_len);
 590		if (value_len) {
 591			size_t size = EXT2_XATTR_SIZE(value_len);
 592			char *val = (char *)header + min_offs - size;
 593			here->e_value_offs =
 594				cpu_to_le16((char *)val - (char *)header);
 595			memset(val + size - EXT2_XATTR_PAD, 0,
 596			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 597			memcpy(val, value, value_len);
 598		}
 599	}
 600
 601skip_replace:
 602	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 603		/* This block is now empty. */
 604		if (bh && header == HDR(bh))
 605			unlock_buffer(bh);  /* we were modifying in-place. */
 606		error = ext2_xattr_set2(inode, bh, NULL);
 607	} else {
 608		ext2_xattr_rehash(header, here);
 609		if (bh && header == HDR(bh))
 610			unlock_buffer(bh);  /* we were modifying in-place. */
 611		error = ext2_xattr_set2(inode, bh, header);
 612	}
 613
 614cleanup:
 615	brelse(bh);
 616	if (!(bh && header == HDR(bh)))
 617		kfree(header);
 
 618	up_write(&EXT2_I(inode)->xattr_sem);
 619
 620	return error;
 621}
 622
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 623/*
 624 * Second half of ext2_xattr_set(): Update the file system.
 625 */
 626static int
 627ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 628		struct ext2_xattr_header *header)
 629{
 630	struct super_block *sb = inode->i_sb;
 631	struct buffer_head *new_bh = NULL;
 632	int error;
 633	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 634
 635	if (header) {
 636		new_bh = ext2_xattr_cache_find(inode, header);
 637		if (new_bh) {
 638			/* We found an identical block in the cache. */
 639			if (new_bh == old_bh) {
 640				ea_bdebug(new_bh, "keeping this block");
 641			} else {
 642				/* The old block is released after updating
 643				   the inode.  */
 644				ea_bdebug(new_bh, "reusing block");
 645
 646				error = dquot_alloc_block(inode, 1);
 647				if (error) {
 648					unlock_buffer(new_bh);
 649					goto cleanup;
 650				}
 651				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 652				ea_bdebug(new_bh, "refcount now=%d",
 653					le32_to_cpu(HDR(new_bh)->h_refcount));
 654			}
 655			unlock_buffer(new_bh);
 656		} else if (old_bh && header == HDR(old_bh)) {
 657			/* Keep this block. No need to lock the block as we
 658			   don't need to change the reference count. */
 659			new_bh = old_bh;
 660			get_bh(new_bh);
 661			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 662		} else {
 663			/* We need to allocate a new block */
 664			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 665						EXT2_I(inode)->i_block_group);
 666			int block = ext2_new_block(inode, goal, &error);
 667			if (error)
 668				goto cleanup;
 669			ea_idebug(inode, "creating block %d", block);
 670
 671			new_bh = sb_getblk(sb, block);
 672			if (unlikely(!new_bh)) {
 673				ext2_free_blocks(inode, block, 1);
 674				mark_inode_dirty(inode);
 675				error = -ENOMEM;
 676				goto cleanup;
 677			}
 678			lock_buffer(new_bh);
 679			memcpy(new_bh->b_data, header, new_bh->b_size);
 680			set_buffer_uptodate(new_bh);
 681			unlock_buffer(new_bh);
 682			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 683			
 684			ext2_xattr_update_super_block(sb);
 685		}
 686		mark_buffer_dirty(new_bh);
 687		if (IS_SYNC(inode)) {
 688			sync_dirty_buffer(new_bh);
 689			error = -EIO;
 690			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 691				goto cleanup;
 692		}
 693	}
 694
 695	/* Update the inode. */
 696	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 697	inode->i_ctime = current_time(inode);
 698	if (IS_SYNC(inode)) {
 699		error = sync_inode_metadata(inode, 1);
 700		/* In case sync failed due to ENOSPC the inode was actually
 701		 * written (only some dirty data were not) so we just proceed
 702		 * as if nothing happened and cleanup the unused block */
 703		if (error && error != -ENOSPC) {
 704			if (new_bh && new_bh != old_bh) {
 705				dquot_free_block_nodirty(inode, 1);
 706				mark_inode_dirty(inode);
 707			}
 708			goto cleanup;
 709		}
 710	} else
 711		mark_inode_dirty(inode);
 712
 713	error = 0;
 714	if (old_bh && old_bh != new_bh) {
 715		/*
 716		 * If there was an old block and we are no longer using it,
 717		 * release the old block.
 718		 */
 719		lock_buffer(old_bh);
 720		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 721			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
 722
 723			/*
 724			 * This must happen under buffer lock for
 725			 * ext2_xattr_set2() to reliably detect freed block
 726			 */
 727			mb_cache_entry_delete(ea_block_cache, hash,
 728					      old_bh->b_blocknr);
 729			/* Free the old block. */
 730			ea_bdebug(old_bh, "freeing");
 731			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 732			mark_inode_dirty(inode);
 733			/* We let our caller release old_bh, so we
 734			 * need to duplicate the buffer before. */
 735			get_bh(old_bh);
 736			bforget(old_bh);
 737		} else {
 738			/* Decrement the refcount only. */
 739			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 740			dquot_free_block_nodirty(inode, 1);
 741			mark_inode_dirty(inode);
 742			mark_buffer_dirty(old_bh);
 743			ea_bdebug(old_bh, "refcount now=%d",
 744				le32_to_cpu(HDR(old_bh)->h_refcount));
 745		}
 746		unlock_buffer(old_bh);
 747	}
 748
 749cleanup:
 750	brelse(new_bh);
 751
 752	return error;
 753}
 754
 755/*
 756 * ext2_xattr_delete_inode()
 757 *
 758 * Free extended attribute resources associated with this inode. This
 759 * is called immediately before an inode is freed.
 760 */
 761void
 762ext2_xattr_delete_inode(struct inode *inode)
 763{
 764	struct buffer_head *bh = NULL;
 765	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 766
 767	down_write(&EXT2_I(inode)->xattr_sem);
 
 
 
 
 
 
 
 
 768	if (!EXT2_I(inode)->i_file_acl)
 769		goto cleanup;
 770
 771	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
 772		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 773			"inode %ld: xattr block %d is out of data blocks range",
 774			inode->i_ino, EXT2_I(inode)->i_file_acl);
 775		goto cleanup;
 776	}
 777
 778	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 779	if (!bh) {
 780		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 781			"inode %ld: block %d read error", inode->i_ino,
 782			EXT2_I(inode)->i_file_acl);
 783		goto cleanup;
 784	}
 785	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 786	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 787	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 788		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 789			"inode %ld: bad block %d", inode->i_ino,
 790			EXT2_I(inode)->i_file_acl);
 791		goto cleanup;
 792	}
 793	lock_buffer(bh);
 794	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 795		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 796
 797		/*
 798		 * This must happen under buffer lock for ext2_xattr_set2() to
 799		 * reliably detect freed block
 800		 */
 801		mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 802				      bh->b_blocknr);
 803		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 804		get_bh(bh);
 805		bforget(bh);
 806		unlock_buffer(bh);
 807	} else {
 808		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 809		ea_bdebug(bh, "refcount now=%d",
 810			le32_to_cpu(HDR(bh)->h_refcount));
 811		unlock_buffer(bh);
 812		mark_buffer_dirty(bh);
 813		if (IS_SYNC(inode))
 814			sync_dirty_buffer(bh);
 815		dquot_free_block_nodirty(inode, 1);
 816	}
 817	EXT2_I(inode)->i_file_acl = 0;
 818
 819cleanup:
 820	brelse(bh);
 821	up_write(&EXT2_I(inode)->xattr_sem);
 822}
 823
 824/*
 825 * ext2_xattr_cache_insert()
 826 *
 827 * Create a new entry in the extended attribute cache, and insert
 828 * it unless such an entry is already in the cache.
 829 *
 830 * Returns 0, or a negative error number on failure.
 831 */
 832static int
 833ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 834{
 835	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 836	int error;
 837
 838	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 1);
 
 839	if (error) {
 840		if (error == -EBUSY) {
 841			ea_bdebug(bh, "already in cache (%d cache entries)",
 842				atomic_read(&ext2_xattr_cache->c_entry_count));
 843			error = 0;
 844		}
 845	} else
 846		ea_bdebug(bh, "inserting [%x]", (int)hash);
 847	return error;
 848}
 849
 850/*
 851 * ext2_xattr_cmp()
 852 *
 853 * Compare two extended attribute blocks for equality.
 854 *
 855 * Returns 0 if the blocks are equal, 1 if they differ, and
 856 * a negative error number on errors.
 857 */
 858static int
 859ext2_xattr_cmp(struct ext2_xattr_header *header1,
 860	       struct ext2_xattr_header *header2)
 861{
 862	struct ext2_xattr_entry *entry1, *entry2;
 863
 864	entry1 = ENTRY(header1+1);
 865	entry2 = ENTRY(header2+1);
 866	while (!IS_LAST_ENTRY(entry1)) {
 867		if (IS_LAST_ENTRY(entry2))
 868			return 1;
 869		if (entry1->e_hash != entry2->e_hash ||
 870		    entry1->e_name_index != entry2->e_name_index ||
 871		    entry1->e_name_len != entry2->e_name_len ||
 872		    entry1->e_value_size != entry2->e_value_size ||
 873		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 874			return 1;
 875		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 876			return -EIO;
 877		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 878			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 879			   le32_to_cpu(entry1->e_value_size)))
 880			return 1;
 881
 882		entry1 = EXT2_XATTR_NEXT(entry1);
 883		entry2 = EXT2_XATTR_NEXT(entry2);
 884	}
 885	if (!IS_LAST_ENTRY(entry2))
 886		return 1;
 887	return 0;
 888}
 889
 890/*
 891 * ext2_xattr_cache_find()
 892 *
 893 * Find an identical extended attribute block.
 894 *
 895 * Returns a locked buffer head to the block found, or NULL if such
 896 * a block was not found or an error occurred.
 897 */
 898static struct buffer_head *
 899ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 900{
 901	__u32 hash = le32_to_cpu(header->h_hash);
 902	struct mb_cache_entry *ce;
 903	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 904
 905	if (!header->h_hash)
 906		return NULL;  /* never share */
 907	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 908again:
 909	ce = mb_cache_entry_find_first(ea_block_cache, hash);
 910	while (ce) {
 911		struct buffer_head *bh;
 912
 913		bh = sb_bread(inode->i_sb, ce->e_value);
 914		if (!bh) {
 915			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 916				"inode %ld: block %ld read error",
 917				inode->i_ino, (unsigned long) ce->e_value);
 918		} else {
 919			lock_buffer(bh);
 920			/*
 921			 * We have to be careful about races with freeing or
 922			 * rehashing of xattr block. Once we hold buffer lock
 923			 * xattr block's state is stable so we can check
 924			 * whether the block got freed / rehashed or not.
 925			 * Since we unhash mbcache entry under buffer lock when
 926			 * freeing / rehashing xattr block, checking whether
 927			 * entry is still hashed is reliable.
 928			 */
 929			if (hlist_bl_unhashed(&ce->e_hash_list)) {
 930				mb_cache_entry_put(ea_block_cache, ce);
 931				unlock_buffer(bh);
 932				brelse(bh);
 933				goto again;
 934			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
 935				   EXT2_XATTR_REFCOUNT_MAX) {
 936				ea_idebug(inode, "block %ld refcount %d>%d",
 937					  (unsigned long) ce->e_value,
 938					  le32_to_cpu(HDR(bh)->h_refcount),
 939					  EXT2_XATTR_REFCOUNT_MAX);
 940			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 941				ea_bdebug(bh, "b_count=%d",
 942					  atomic_read(&(bh->b_count)));
 943				mb_cache_entry_touch(ea_block_cache, ce);
 944				mb_cache_entry_put(ea_block_cache, ce);
 945				return bh;
 946			}
 947			unlock_buffer(bh);
 948			brelse(bh);
 949		}
 950		ce = mb_cache_entry_find_next(ea_block_cache, ce);
 951	}
 952	return NULL;
 953}
 954
 955#define NAME_HASH_SHIFT 5
 956#define VALUE_HASH_SHIFT 16
 957
 958/*
 959 * ext2_xattr_hash_entry()
 960 *
 961 * Compute the hash of an extended attribute.
 962 */
 963static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 964					 struct ext2_xattr_entry *entry)
 965{
 966	__u32 hash = 0;
 967	char *name = entry->e_name;
 968	int n;
 969
 970	for (n=0; n < entry->e_name_len; n++) {
 971		hash = (hash << NAME_HASH_SHIFT) ^
 972		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 973		       *name++;
 974	}
 975
 976	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 977		__le32 *value = (__le32 *)((char *)header +
 978			le16_to_cpu(entry->e_value_offs));
 979		for (n = (le32_to_cpu(entry->e_value_size) +
 980		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 981			hash = (hash << VALUE_HASH_SHIFT) ^
 982			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 983			       le32_to_cpu(*value++);
 984		}
 985	}
 986	entry->e_hash = cpu_to_le32(hash);
 987}
 988
 989#undef NAME_HASH_SHIFT
 990#undef VALUE_HASH_SHIFT
 991
 992#define BLOCK_HASH_SHIFT 16
 993
 994/*
 995 * ext2_xattr_rehash()
 996 *
 997 * Re-compute the extended attribute hash value after an entry has changed.
 998 */
 999static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1000			      struct ext2_xattr_entry *entry)
1001{
1002	struct ext2_xattr_entry *here;
1003	__u32 hash = 0;
1004	
1005	ext2_xattr_hash_entry(header, entry);
1006	here = ENTRY(header+1);
1007	while (!IS_LAST_ENTRY(here)) {
1008		if (!here->e_hash) {
1009			/* Block is not shared if an entry's hash value == 0 */
1010			hash = 0;
1011			break;
1012		}
1013		hash = (hash << BLOCK_HASH_SHIFT) ^
1014		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1015		       le32_to_cpu(here->e_hash);
1016		here = EXT2_XATTR_NEXT(here);
1017	}
1018	header->h_hash = cpu_to_le32(hash);
1019}
1020
1021#undef BLOCK_HASH_SHIFT
1022
1023#define HASH_BUCKET_BITS 10
1024
1025struct mb_cache *ext2_xattr_create_cache(void)
1026{
1027	return mb_cache_create(HASH_BUCKET_BITS);
1028}
1029
1030void ext2_xattr_destroy_cache(struct mb_cache *cache)
1031{
1032	if (cache)
1033		mb_cache_destroy(cache);
1034}