Linux Audio

Check our new training course

Loading...
v5.4
   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
 137static bool
 138ext2_xattr_header_valid(struct ext2_xattr_header *header)
 139{
 140	if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 141	    header->h_blocks != cpu_to_le32(1))
 142		return false;
 143
 144	return true;
 145}
 146
 147static bool
 148ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
 149		       char *end, size_t end_offs)
 150{
 151	struct ext2_xattr_entry *next;
 152	size_t size;
 153
 154	next = EXT2_XATTR_NEXT(entry);
 155	if ((char *)next >= end)
 156		return false;
 157
 158	if (entry->e_value_block != 0)
 159		return false;
 160
 161	size = le32_to_cpu(entry->e_value_size);
 162	if (size > end_offs ||
 163	    le16_to_cpu(entry->e_value_offs) + size > end_offs)
 164		return false;
 165
 166	return true;
 167}
 168
 169static int
 170ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
 171		     struct ext2_xattr_entry *entry)
 172{
 173	int cmp;
 174
 175	cmp = name_index - entry->e_name_index;
 176	if (!cmp)
 177		cmp = name_len - entry->e_name_len;
 178	if (!cmp)
 179		cmp = memcmp(name, entry->e_name, name_len);
 180
 181	return cmp;
 182}
 183
 184/*
 185 * ext2_xattr_get()
 186 *
 187 * Copy an extended attribute into the buffer
 188 * provided, or compute the buffer size required.
 189 * Buffer is NULL to compute the size of the buffer required.
 190 *
 191 * Returns a negative error number on failure, or the number of bytes
 192 * used / required on success.
 193 */
 194int
 195ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 196	       void *buffer, size_t buffer_size)
 197{
 198	struct buffer_head *bh = NULL;
 199	struct ext2_xattr_entry *entry;
 200	size_t name_len, size;
 201	char *end;
 202	int error, not_found;
 203	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 204
 205	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 206		  name_index, name, buffer, (long)buffer_size);
 207
 208	if (name == NULL)
 209		return -EINVAL;
 210	name_len = strlen(name);
 211	if (name_len > 255)
 212		return -ERANGE;
 213
 214	down_read(&EXT2_I(inode)->xattr_sem);
 215	error = -ENODATA;
 216	if (!EXT2_I(inode)->i_file_acl)
 217		goto cleanup;
 218	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 219	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 220	error = -EIO;
 221	if (!bh)
 222		goto cleanup;
 223	ea_bdebug(bh, "b_count=%d, refcount=%d",
 224		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 225	end = bh->b_data + bh->b_size;
 226	if (!ext2_xattr_header_valid(HDR(bh))) {
 227bad_block:
 228		ext2_error(inode->i_sb, "ext2_xattr_get",
 229			"inode %ld: bad block %d", inode->i_ino,
 230			EXT2_I(inode)->i_file_acl);
 231		error = -EIO;
 232		goto cleanup;
 233	}
 234
 235	/* find named attribute */
 236	entry = FIRST_ENTRY(bh);
 237	while (!IS_LAST_ENTRY(entry)) {
 238		if (!ext2_xattr_entry_valid(entry, end,
 239		    inode->i_sb->s_blocksize))
 240			goto bad_block;
 241
 242		not_found = ext2_xattr_cmp_entry(name_index, name_len, name,
 243						 entry);
 244		if (!not_found)
 245			goto found;
 246		if (not_found < 0)
 247			break;
 248
 249		entry = EXT2_XATTR_NEXT(entry);
 250	}
 251	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 252		ea_idebug(inode, "cache insert failed");
 253	error = -ENODATA;
 254	goto cleanup;
 255found:
 256	size = le32_to_cpu(entry->e_value_size);
 257	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 258		ea_idebug(inode, "cache insert failed");
 259	if (buffer) {
 260		error = -ERANGE;
 261		if (size > buffer_size)
 262			goto cleanup;
 263		/* return value of attribute */
 264		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 265			size);
 266	}
 267	error = size;
 268
 269cleanup:
 270	brelse(bh);
 271	up_read(&EXT2_I(inode)->xattr_sem);
 272
 273	return error;
 274}
 275
 276/*
 277 * ext2_xattr_list()
 278 *
 279 * Copy a list of attribute names into the buffer
 280 * provided, or compute the buffer size required.
 281 * Buffer is NULL to compute the size of the buffer required.
 282 *
 283 * Returns a negative error number on failure, or the number of bytes
 284 * used / required on success.
 285 */
 286static int
 287ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 288{
 289	struct inode *inode = d_inode(dentry);
 290	struct buffer_head *bh = NULL;
 291	struct ext2_xattr_entry *entry;
 292	char *end;
 293	size_t rest = buffer_size;
 294	int error;
 295	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 296
 297	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 298		  buffer, (long)buffer_size);
 299
 300	down_read(&EXT2_I(inode)->xattr_sem);
 301	error = 0;
 302	if (!EXT2_I(inode)->i_file_acl)
 303		goto cleanup;
 304	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 305	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 306	error = -EIO;
 307	if (!bh)
 308		goto cleanup;
 309	ea_bdebug(bh, "b_count=%d, refcount=%d",
 310		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 311	end = bh->b_data + bh->b_size;
 312	if (!ext2_xattr_header_valid(HDR(bh))) {
 313bad_block:
 314		ext2_error(inode->i_sb, "ext2_xattr_list",
 315			"inode %ld: bad block %d", inode->i_ino,
 316			EXT2_I(inode)->i_file_acl);
 317		error = -EIO;
 318		goto cleanup;
 319	}
 320
 321	/* check the on-disk data structure */
 322	entry = FIRST_ENTRY(bh);
 323	while (!IS_LAST_ENTRY(entry)) {
 324		if (!ext2_xattr_entry_valid(entry, end,
 325		    inode->i_sb->s_blocksize))
 326			goto bad_block;
 327		entry = EXT2_XATTR_NEXT(entry);
 328	}
 329	if (ext2_xattr_cache_insert(ea_block_cache, bh))
 330		ea_idebug(inode, "cache insert failed");
 331
 332	/* list the attribute names */
 333	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 334	     entry = EXT2_XATTR_NEXT(entry)) {
 335		const struct xattr_handler *handler =
 336			ext2_xattr_handler(entry->e_name_index);
 337
 338		if (handler && (!handler->list || handler->list(dentry))) {
 339			const char *prefix = handler->prefix ?: handler->name;
 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	down_write(&EXT2_I(inode)->xattr_sem);
 440	if (EXT2_I(inode)->i_file_acl) {
 441		/* The inode already has an extended attribute block. */
 442		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 443		error = -EIO;
 444		if (!bh)
 445			goto cleanup;
 446		ea_bdebug(bh, "b_count=%d, refcount=%d",
 447			atomic_read(&(bh->b_count)),
 448			le32_to_cpu(HDR(bh)->h_refcount));
 449		header = HDR(bh);
 450		end = bh->b_data + bh->b_size;
 451		if (!ext2_xattr_header_valid(header)) {
 452bad_block:
 453			ext2_error(sb, "ext2_xattr_set",
 454				"inode %ld: bad block %d", inode->i_ino, 
 455				   EXT2_I(inode)->i_file_acl);
 456			error = -EIO;
 457			goto cleanup;
 458		}
 459		/*
 460		 * Find the named attribute. If not found, 'here' will point
 461		 * to entry where the new attribute should be inserted to
 462		 * maintain sorting.
 463		 */
 464		last = FIRST_ENTRY(bh);
 465		while (!IS_LAST_ENTRY(last)) {
 466			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
 467				goto bad_block;
 468			if (last->e_value_size) {
 469				size_t offs = le16_to_cpu(last->e_value_offs);
 470				if (offs < min_offs)
 471					min_offs = offs;
 472			}
 473			if (not_found > 0) {
 474				not_found = ext2_xattr_cmp_entry(name_index,
 475								 name_len,
 476								 name, last);
 477				if (not_found <= 0)
 478					here = last;
 479			}
 480			last = EXT2_XATTR_NEXT(last);
 481		}
 482		if (not_found > 0)
 483			here = last;
 484
 485		/* Check whether we have enough space left. */
 486		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 487	} else {
 488		/* We will use a new extended attribute block. */
 489		free = sb->s_blocksize -
 490			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 491	}
 492
 493	if (not_found) {
 494		/* Request to remove a nonexistent attribute? */
 495		error = -ENODATA;
 496		if (flags & XATTR_REPLACE)
 497			goto cleanup;
 498		error = 0;
 499		if (value == NULL)
 500			goto cleanup;
 501	} else {
 502		/* Request to create an existing attribute? */
 503		error = -EEXIST;
 504		if (flags & XATTR_CREATE)
 505			goto cleanup;
 506		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
 507		free += EXT2_XATTR_LEN(name_len);
 508	}
 509	error = -ENOSPC;
 510	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 511		goto cleanup;
 512
 513	/* Here we know that we can set the new attribute. */
 514
 515	if (header) {
 516		/* assert(header == HDR(bh)); */
 517		lock_buffer(bh);
 518		if (header->h_refcount == cpu_to_le32(1)) {
 519			__u32 hash = le32_to_cpu(header->h_hash);
 520
 521			ea_bdebug(bh, "modifying in-place");
 522			/*
 523			 * This must happen under buffer lock for
 524			 * ext2_xattr_set2() to reliably detect modified block
 525			 */
 526			mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 527					      bh->b_blocknr);
 528
 529			/* keep the buffer locked while modifying it. */
 530		} else {
 531			int offset;
 532
 533			unlock_buffer(bh);
 534			ea_bdebug(bh, "cloning");
 535			header = kmemdup(HDR(bh), bh->b_size, GFP_KERNEL);
 536			error = -ENOMEM;
 537			if (header == NULL)
 538				goto cleanup;
 539			header->h_refcount = cpu_to_le32(1);
 540
 541			offset = (char *)here - bh->b_data;
 542			here = ENTRY((char *)header + offset);
 543			offset = (char *)last - bh->b_data;
 544			last = ENTRY((char *)header + offset);
 545		}
 546	} else {
 547		/* Allocate a buffer where we construct the new block. */
 548		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 549		error = -ENOMEM;
 550		if (header == NULL)
 551			goto cleanup;
 552		end = (char *)header + sb->s_blocksize;
 553		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 554		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 555		last = here = ENTRY(header+1);
 556	}
 557
 558	/* Iff we are modifying the block in-place, bh is locked here. */
 559
 560	if (not_found) {
 561		/* Insert the new name. */
 562		size_t size = EXT2_XATTR_LEN(name_len);
 563		size_t rest = (char *)last - (char *)here;
 564		memmove((char *)here + size, here, rest);
 565		memset(here, 0, size);
 566		here->e_name_index = name_index;
 567		here->e_name_len = name_len;
 568		memcpy(here->e_name, name, name_len);
 569	} else {
 570		if (here->e_value_size) {
 571			char *first_val = (char *)header + min_offs;
 572			size_t offs = le16_to_cpu(here->e_value_offs);
 573			char *val = (char *)header + offs;
 574			size_t size = EXT2_XATTR_SIZE(
 575				le32_to_cpu(here->e_value_size));
 576
 577			if (size == EXT2_XATTR_SIZE(value_len)) {
 578				/* The old and the new value have the same
 579				   size. Just replace. */
 580				here->e_value_size = cpu_to_le32(value_len);
 581				memset(val + size - EXT2_XATTR_PAD, 0,
 582				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 583				memcpy(val, value, value_len);
 584				goto skip_replace;
 585			}
 586
 587			/* Remove the old value. */
 588			memmove(first_val + size, first_val, val - first_val);
 589			memset(first_val, 0, size);
 590			here->e_value_offs = 0;
 591			min_offs += size;
 592
 593			/* Adjust all value offsets. */
 594			last = ENTRY(header+1);
 595			while (!IS_LAST_ENTRY(last)) {
 596				size_t o = le16_to_cpu(last->e_value_offs);
 597				if (o < offs)
 598					last->e_value_offs =
 599						cpu_to_le16(o + size);
 600				last = EXT2_XATTR_NEXT(last);
 601			}
 
 
 602		}
 603		if (value == NULL) {
 604			/* Remove the old name. */
 605			size_t size = EXT2_XATTR_LEN(name_len);
 606			last = ENTRY((char *)last - size);
 607			memmove(here, (char*)here + size,
 608				(char*)last - (char*)here);
 609			memset(last, 0, size);
 610		}
 611	}
 612
 613	if (value != NULL) {
 614		/* Insert the new value. */
 615		here->e_value_size = cpu_to_le32(value_len);
 616		if (value_len) {
 617			size_t size = EXT2_XATTR_SIZE(value_len);
 618			char *val = (char *)header + min_offs - size;
 619			here->e_value_offs =
 620				cpu_to_le16((char *)val - (char *)header);
 621			memset(val + size - EXT2_XATTR_PAD, 0,
 622			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 623			memcpy(val, value, value_len);
 624		}
 625	}
 626
 627skip_replace:
 628	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 629		/* This block is now empty. */
 630		if (bh && header == HDR(bh))
 631			unlock_buffer(bh);  /* we were modifying in-place. */
 632		error = ext2_xattr_set2(inode, bh, NULL);
 633	} else {
 634		ext2_xattr_rehash(header, here);
 635		if (bh && header == HDR(bh))
 636			unlock_buffer(bh);  /* we were modifying in-place. */
 637		error = ext2_xattr_set2(inode, bh, header);
 638	}
 639
 640cleanup:
 641	if (!(bh && header == HDR(bh)))
 642		kfree(header);
 643	brelse(bh);
 644	up_write(&EXT2_I(inode)->xattr_sem);
 645
 646	return error;
 647}
 648
 649/*
 650 * Second half of ext2_xattr_set(): Update the file system.
 651 */
 652static int
 653ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 654		struct ext2_xattr_header *header)
 655{
 656	struct super_block *sb = inode->i_sb;
 657	struct buffer_head *new_bh = NULL;
 658	int error;
 659	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 660
 661	if (header) {
 662		new_bh = ext2_xattr_cache_find(inode, header);
 663		if (new_bh) {
 664			/* We found an identical block in the cache. */
 665			if (new_bh == old_bh) {
 666				ea_bdebug(new_bh, "keeping this block");
 667			} else {
 668				/* The old block is released after updating
 669				   the inode.  */
 670				ea_bdebug(new_bh, "reusing block");
 671
 672				error = dquot_alloc_block(inode, 1);
 673				if (error) {
 674					unlock_buffer(new_bh);
 675					goto cleanup;
 676				}
 677				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 678				ea_bdebug(new_bh, "refcount now=%d",
 679					le32_to_cpu(HDR(new_bh)->h_refcount));
 680			}
 681			unlock_buffer(new_bh);
 682		} else if (old_bh && header == HDR(old_bh)) {
 683			/* Keep this block. No need to lock the block as we
 684			   don't need to change the reference count. */
 685			new_bh = old_bh;
 686			get_bh(new_bh);
 687			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 688		} else {
 689			/* We need to allocate a new block */
 690			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 691						EXT2_I(inode)->i_block_group);
 692			int block = ext2_new_block(inode, goal, &error);
 693			if (error)
 694				goto cleanup;
 695			ea_idebug(inode, "creating block %d", block);
 696
 697			new_bh = sb_getblk(sb, block);
 698			if (unlikely(!new_bh)) {
 699				ext2_free_blocks(inode, block, 1);
 700				mark_inode_dirty(inode);
 701				error = -ENOMEM;
 702				goto cleanup;
 703			}
 704			lock_buffer(new_bh);
 705			memcpy(new_bh->b_data, header, new_bh->b_size);
 706			set_buffer_uptodate(new_bh);
 707			unlock_buffer(new_bh);
 708			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 709			
 710			ext2_xattr_update_super_block(sb);
 711		}
 712		mark_buffer_dirty(new_bh);
 713		if (IS_SYNC(inode)) {
 714			sync_dirty_buffer(new_bh);
 715			error = -EIO;
 716			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 717				goto cleanup;
 718		}
 719	}
 720
 721	/* Update the inode. */
 722	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 723	inode->i_ctime = current_time(inode);
 724	if (IS_SYNC(inode)) {
 725		error = sync_inode_metadata(inode, 1);
 726		/* In case sync failed due to ENOSPC the inode was actually
 727		 * written (only some dirty data were not) so we just proceed
 728		 * as if nothing happened and cleanup the unused block */
 729		if (error && error != -ENOSPC) {
 730			if (new_bh && new_bh != old_bh) {
 731				dquot_free_block_nodirty(inode, 1);
 732				mark_inode_dirty(inode);
 733			}
 734			goto cleanup;
 735		}
 736	} else
 737		mark_inode_dirty(inode);
 738
 739	error = 0;
 740	if (old_bh && old_bh != new_bh) {
 741		/*
 742		 * If there was an old block and we are no longer using it,
 743		 * release the old block.
 744		 */
 745		lock_buffer(old_bh);
 746		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 747			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
 748
 749			/*
 750			 * This must happen under buffer lock for
 751			 * ext2_xattr_set2() to reliably detect freed block
 752			 */
 753			mb_cache_entry_delete(ea_block_cache, hash,
 754					      old_bh->b_blocknr);
 755			/* Free the old block. */
 756			ea_bdebug(old_bh, "freeing");
 757			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 758			mark_inode_dirty(inode);
 759			/* We let our caller release old_bh, so we
 760			 * need to duplicate the buffer before. */
 761			get_bh(old_bh);
 762			bforget(old_bh);
 763		} else {
 764			/* Decrement the refcount only. */
 765			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 766			dquot_free_block_nodirty(inode, 1);
 767			mark_inode_dirty(inode);
 768			mark_buffer_dirty(old_bh);
 769			ea_bdebug(old_bh, "refcount now=%d",
 770				le32_to_cpu(HDR(old_bh)->h_refcount));
 771		}
 772		unlock_buffer(old_bh);
 773	}
 774
 775cleanup:
 776	brelse(new_bh);
 777
 778	return error;
 779}
 780
 781/*
 782 * ext2_xattr_delete_inode()
 783 *
 784 * Free extended attribute resources associated with this inode. This
 785 * is called immediately before an inode is freed.
 786 */
 787void
 788ext2_xattr_delete_inode(struct inode *inode)
 789{
 790	struct buffer_head *bh = NULL;
 791	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 792
 793	down_write(&EXT2_I(inode)->xattr_sem);
 
 
 
 
 
 
 
 
 794	if (!EXT2_I(inode)->i_file_acl)
 795		goto cleanup;
 796
 797	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
 798		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 799			"inode %ld: xattr block %d is out of data blocks range",
 800			inode->i_ino, EXT2_I(inode)->i_file_acl);
 801		goto cleanup;
 802	}
 803
 804	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 805	if (!bh) {
 806		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 807			"inode %ld: block %d read error", inode->i_ino,
 808			EXT2_I(inode)->i_file_acl);
 809		goto cleanup;
 810	}
 811	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 812	if (!ext2_xattr_header_valid(HDR(bh))) {
 813		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 814			"inode %ld: bad block %d", inode->i_ino,
 815			EXT2_I(inode)->i_file_acl);
 816		goto cleanup;
 817	}
 818	lock_buffer(bh);
 819	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 820		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 821
 822		/*
 823		 * This must happen under buffer lock for ext2_xattr_set2() to
 824		 * reliably detect freed block
 825		 */
 826		mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 827				      bh->b_blocknr);
 828		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 829		get_bh(bh);
 830		bforget(bh);
 831		unlock_buffer(bh);
 832	} else {
 833		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 834		ea_bdebug(bh, "refcount now=%d",
 835			le32_to_cpu(HDR(bh)->h_refcount));
 836		unlock_buffer(bh);
 837		mark_buffer_dirty(bh);
 838		if (IS_SYNC(inode))
 839			sync_dirty_buffer(bh);
 840		dquot_free_block_nodirty(inode, 1);
 841	}
 842	EXT2_I(inode)->i_file_acl = 0;
 843
 844cleanup:
 845	brelse(bh);
 846	up_write(&EXT2_I(inode)->xattr_sem);
 847}
 848
 849/*
 850 * ext2_xattr_cache_insert()
 851 *
 852 * Create a new entry in the extended attribute cache, and insert
 853 * it unless such an entry is already in the cache.
 854 *
 855 * Returns 0, or a negative error number on failure.
 856 */
 857static int
 858ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
 859{
 860	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 861	int error;
 862
 863	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
 864				      true);
 865	if (error) {
 866		if (error == -EBUSY) {
 867			ea_bdebug(bh, "already in cache (%d cache entries)",
 868				atomic_read(&ext2_xattr_cache->c_entry_count));
 869			error = 0;
 870		}
 871	} else
 872		ea_bdebug(bh, "inserting [%x]", (int)hash);
 873	return error;
 874}
 875
 876/*
 877 * ext2_xattr_cmp()
 878 *
 879 * Compare two extended attribute blocks for equality.
 880 *
 881 * Returns 0 if the blocks are equal, 1 if they differ, and
 882 * a negative error number on errors.
 883 */
 884static int
 885ext2_xattr_cmp(struct ext2_xattr_header *header1,
 886	       struct ext2_xattr_header *header2)
 887{
 888	struct ext2_xattr_entry *entry1, *entry2;
 889
 890	entry1 = ENTRY(header1+1);
 891	entry2 = ENTRY(header2+1);
 892	while (!IS_LAST_ENTRY(entry1)) {
 893		if (IS_LAST_ENTRY(entry2))
 894			return 1;
 895		if (entry1->e_hash != entry2->e_hash ||
 896		    entry1->e_name_index != entry2->e_name_index ||
 897		    entry1->e_name_len != entry2->e_name_len ||
 898		    entry1->e_value_size != entry2->e_value_size ||
 899		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 900			return 1;
 901		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 902			return -EIO;
 903		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 904			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 905			   le32_to_cpu(entry1->e_value_size)))
 906			return 1;
 907
 908		entry1 = EXT2_XATTR_NEXT(entry1);
 909		entry2 = EXT2_XATTR_NEXT(entry2);
 910	}
 911	if (!IS_LAST_ENTRY(entry2))
 912		return 1;
 913	return 0;
 914}
 915
 916/*
 917 * ext2_xattr_cache_find()
 918 *
 919 * Find an identical extended attribute block.
 920 *
 921 * Returns a locked buffer head to the block found, or NULL if such
 922 * a block was not found or an error occurred.
 923 */
 924static struct buffer_head *
 925ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 926{
 927	__u32 hash = le32_to_cpu(header->h_hash);
 928	struct mb_cache_entry *ce;
 929	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 930
 931	if (!header->h_hash)
 932		return NULL;  /* never share */
 933	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 934again:
 935	ce = mb_cache_entry_find_first(ea_block_cache, hash);
 936	while (ce) {
 937		struct buffer_head *bh;
 938
 939		bh = sb_bread(inode->i_sb, ce->e_value);
 940		if (!bh) {
 941			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 942				"inode %ld: block %ld read error",
 943				inode->i_ino, (unsigned long) ce->e_value);
 944		} else {
 945			lock_buffer(bh);
 946			/*
 947			 * We have to be careful about races with freeing or
 948			 * rehashing of xattr block. Once we hold buffer lock
 949			 * xattr block's state is stable so we can check
 950			 * whether the block got freed / rehashed or not.
 951			 * Since we unhash mbcache entry under buffer lock when
 952			 * freeing / rehashing xattr block, checking whether
 953			 * entry is still hashed is reliable.
 954			 */
 955			if (hlist_bl_unhashed(&ce->e_hash_list)) {
 956				mb_cache_entry_put(ea_block_cache, ce);
 957				unlock_buffer(bh);
 958				brelse(bh);
 959				goto again;
 960			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
 961				   EXT2_XATTR_REFCOUNT_MAX) {
 962				ea_idebug(inode, "block %ld refcount %d>%d",
 963					  (unsigned long) ce->e_value,
 964					  le32_to_cpu(HDR(bh)->h_refcount),
 965					  EXT2_XATTR_REFCOUNT_MAX);
 966			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 967				ea_bdebug(bh, "b_count=%d",
 968					  atomic_read(&(bh->b_count)));
 969				mb_cache_entry_touch(ea_block_cache, ce);
 970				mb_cache_entry_put(ea_block_cache, ce);
 971				return bh;
 972			}
 973			unlock_buffer(bh);
 974			brelse(bh);
 975		}
 976		ce = mb_cache_entry_find_next(ea_block_cache, ce);
 977	}
 978	return NULL;
 979}
 980
 981#define NAME_HASH_SHIFT 5
 982#define VALUE_HASH_SHIFT 16
 983
 984/*
 985 * ext2_xattr_hash_entry()
 986 *
 987 * Compute the hash of an extended attribute.
 988 */
 989static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 990					 struct ext2_xattr_entry *entry)
 991{
 992	__u32 hash = 0;
 993	char *name = entry->e_name;
 994	int n;
 995
 996	for (n=0; n < entry->e_name_len; n++) {
 997		hash = (hash << NAME_HASH_SHIFT) ^
 998		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 999		       *name++;
1000	}
1001
1002	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1003		__le32 *value = (__le32 *)((char *)header +
1004			le16_to_cpu(entry->e_value_offs));
1005		for (n = (le32_to_cpu(entry->e_value_size) +
1006		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1007			hash = (hash << VALUE_HASH_SHIFT) ^
1008			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1009			       le32_to_cpu(*value++);
1010		}
1011	}
1012	entry->e_hash = cpu_to_le32(hash);
1013}
1014
1015#undef NAME_HASH_SHIFT
1016#undef VALUE_HASH_SHIFT
1017
1018#define BLOCK_HASH_SHIFT 16
1019
1020/*
1021 * ext2_xattr_rehash()
1022 *
1023 * Re-compute the extended attribute hash value after an entry has changed.
1024 */
1025static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1026			      struct ext2_xattr_entry *entry)
1027{
1028	struct ext2_xattr_entry *here;
1029	__u32 hash = 0;
1030	
1031	ext2_xattr_hash_entry(header, entry);
1032	here = ENTRY(header+1);
1033	while (!IS_LAST_ENTRY(here)) {
1034		if (!here->e_hash) {
1035			/* Block is not shared if an entry's hash value == 0 */
1036			hash = 0;
1037			break;
1038		}
1039		hash = (hash << BLOCK_HASH_SHIFT) ^
1040		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1041		       le32_to_cpu(here->e_hash);
1042		here = EXT2_XATTR_NEXT(here);
1043	}
1044	header->h_hash = cpu_to_le32(hash);
1045}
1046
1047#undef BLOCK_HASH_SHIFT
1048
1049#define HASH_BUCKET_BITS 10
1050
1051struct mb_cache *ext2_xattr_create_cache(void)
1052{
1053	return mb_cache_create(HASH_BUCKET_BITS);
1054}
1055
1056void ext2_xattr_destroy_cache(struct mb_cache *cache)
1057{
1058	if (cache)
1059		mb_cache_destroy(cache);
1060}
v5.14.15
   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		/* assert(header == HDR(bh)); */
 521		lock_buffer(bh);
 522		if (header->h_refcount == cpu_to_le32(1)) {
 523			__u32 hash = le32_to_cpu(header->h_hash);
 524
 525			ea_bdebug(bh, "modifying in-place");
 526			/*
 527			 * This must happen under buffer lock for
 528			 * ext2_xattr_set2() to reliably detect modified block
 529			 */
 530			mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 531					      bh->b_blocknr);
 532
 533			/* keep the buffer locked while modifying it. */
 534		} else {
 535			int offset;
 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		}
 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
 562	/* Iff we are modifying the block in-place, bh is locked here. */
 563
 564	if (not_found) {
 565		/* Insert the new name. */
 566		size_t size = EXT2_XATTR_LEN(name_len);
 567		size_t rest = (char *)last - (char *)here;
 568		memmove((char *)here + size, here, rest);
 569		memset(here, 0, size);
 570		here->e_name_index = name_index;
 571		here->e_name_len = name_len;
 572		memcpy(here->e_name, name, name_len);
 573	} else {
 574		if (here->e_value_size) {
 575			char *first_val = (char *)header + min_offs;
 576			size_t offs = le16_to_cpu(here->e_value_offs);
 577			char *val = (char *)header + offs;
 578			size_t size = EXT2_XATTR_SIZE(
 579				le32_to_cpu(here->e_value_size));
 580
 581			if (size == EXT2_XATTR_SIZE(value_len)) {
 582				/* The old and the new value have the same
 583				   size. Just replace. */
 584				here->e_value_size = cpu_to_le32(value_len);
 585				memset(val + size - EXT2_XATTR_PAD, 0,
 586				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 587				memcpy(val, value, value_len);
 588				goto skip_replace;
 589			}
 590
 591			/* Remove the old value. */
 592			memmove(first_val + size, first_val, val - first_val);
 593			memset(first_val, 0, size);
 
 594			min_offs += size;
 595
 596			/* Adjust all value offsets. */
 597			last = ENTRY(header+1);
 598			while (!IS_LAST_ENTRY(last)) {
 599				size_t o = le16_to_cpu(last->e_value_offs);
 600				if (o < offs)
 601					last->e_value_offs =
 602						cpu_to_le16(o + size);
 603				last = EXT2_XATTR_NEXT(last);
 604			}
 605
 606			here->e_value_offs = 0;
 607		}
 608		if (value == NULL) {
 609			/* Remove the old name. */
 610			size_t size = EXT2_XATTR_LEN(name_len);
 611			last = ENTRY((char *)last - size);
 612			memmove(here, (char*)here + size,
 613				(char*)last - (char*)here);
 614			memset(last, 0, size);
 615		}
 616	}
 617
 618	if (value != NULL) {
 619		/* Insert the new value. */
 620		here->e_value_size = cpu_to_le32(value_len);
 621		if (value_len) {
 622			size_t size = EXT2_XATTR_SIZE(value_len);
 623			char *val = (char *)header + min_offs - size;
 624			here->e_value_offs =
 625				cpu_to_le16((char *)val - (char *)header);
 626			memset(val + size - EXT2_XATTR_PAD, 0,
 627			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 628			memcpy(val, value, value_len);
 629		}
 630	}
 631
 632skip_replace:
 633	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 634		/* This block is now empty. */
 635		if (bh && header == HDR(bh))
 636			unlock_buffer(bh);  /* we were modifying in-place. */
 637		error = ext2_xattr_set2(inode, bh, NULL);
 638	} else {
 639		ext2_xattr_rehash(header, here);
 640		if (bh && header == HDR(bh))
 641			unlock_buffer(bh);  /* we were modifying in-place. */
 642		error = ext2_xattr_set2(inode, bh, header);
 643	}
 644
 645cleanup:
 646	if (!(bh && header == HDR(bh)))
 647		kfree(header);
 648	brelse(bh);
 649	up_write(&EXT2_I(inode)->xattr_sem);
 650
 651	return error;
 652}
 653
 654/*
 655 * Second half of ext2_xattr_set(): Update the file system.
 656 */
 657static int
 658ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 659		struct ext2_xattr_header *header)
 660{
 661	struct super_block *sb = inode->i_sb;
 662	struct buffer_head *new_bh = NULL;
 663	int error;
 664	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 665
 666	if (header) {
 667		new_bh = ext2_xattr_cache_find(inode, header);
 668		if (new_bh) {
 669			/* We found an identical block in the cache. */
 670			if (new_bh == old_bh) {
 671				ea_bdebug(new_bh, "keeping this block");
 672			} else {
 673				/* The old block is released after updating
 674				   the inode.  */
 675				ea_bdebug(new_bh, "reusing block");
 676
 677				error = dquot_alloc_block(inode, 1);
 678				if (error) {
 679					unlock_buffer(new_bh);
 680					goto cleanup;
 681				}
 682				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 683				ea_bdebug(new_bh, "refcount now=%d",
 684					le32_to_cpu(HDR(new_bh)->h_refcount));
 685			}
 686			unlock_buffer(new_bh);
 687		} else if (old_bh && header == HDR(old_bh)) {
 688			/* Keep this block. No need to lock the block as we
 689			   don't need to change the reference count. */
 690			new_bh = old_bh;
 691			get_bh(new_bh);
 692			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 693		} else {
 694			/* We need to allocate a new block */
 695			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 696						EXT2_I(inode)->i_block_group);
 697			int block = ext2_new_block(inode, goal, &error);
 698			if (error)
 699				goto cleanup;
 700			ea_idebug(inode, "creating block %d", block);
 701
 702			new_bh = sb_getblk(sb, block);
 703			if (unlikely(!new_bh)) {
 704				ext2_free_blocks(inode, block, 1);
 705				mark_inode_dirty(inode);
 706				error = -ENOMEM;
 707				goto cleanup;
 708			}
 709			lock_buffer(new_bh);
 710			memcpy(new_bh->b_data, header, new_bh->b_size);
 711			set_buffer_uptodate(new_bh);
 712			unlock_buffer(new_bh);
 713			ext2_xattr_cache_insert(ea_block_cache, new_bh);
 714			
 715			ext2_xattr_update_super_block(sb);
 716		}
 717		mark_buffer_dirty(new_bh);
 718		if (IS_SYNC(inode)) {
 719			sync_dirty_buffer(new_bh);
 720			error = -EIO;
 721			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 722				goto cleanup;
 723		}
 724	}
 725
 726	/* Update the inode. */
 727	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 728	inode->i_ctime = current_time(inode);
 729	if (IS_SYNC(inode)) {
 730		error = sync_inode_metadata(inode, 1);
 731		/* In case sync failed due to ENOSPC the inode was actually
 732		 * written (only some dirty data were not) so we just proceed
 733		 * as if nothing happened and cleanup the unused block */
 734		if (error && error != -ENOSPC) {
 735			if (new_bh && new_bh != old_bh) {
 736				dquot_free_block_nodirty(inode, 1);
 737				mark_inode_dirty(inode);
 738			}
 739			goto cleanup;
 740		}
 741	} else
 742		mark_inode_dirty(inode);
 743
 744	error = 0;
 745	if (old_bh && old_bh != new_bh) {
 746		/*
 747		 * If there was an old block and we are no longer using it,
 748		 * release the old block.
 749		 */
 750		lock_buffer(old_bh);
 751		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 752			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
 753
 754			/*
 755			 * This must happen under buffer lock for
 756			 * ext2_xattr_set2() to reliably detect freed block
 757			 */
 758			mb_cache_entry_delete(ea_block_cache, hash,
 759					      old_bh->b_blocknr);
 760			/* Free the old block. */
 761			ea_bdebug(old_bh, "freeing");
 762			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 763			mark_inode_dirty(inode);
 764			/* We let our caller release old_bh, so we
 765			 * need to duplicate the buffer before. */
 766			get_bh(old_bh);
 767			bforget(old_bh);
 768		} else {
 769			/* Decrement the refcount only. */
 770			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 771			dquot_free_block_nodirty(inode, 1);
 772			mark_inode_dirty(inode);
 773			mark_buffer_dirty(old_bh);
 774			ea_bdebug(old_bh, "refcount now=%d",
 775				le32_to_cpu(HDR(old_bh)->h_refcount));
 776		}
 777		unlock_buffer(old_bh);
 778	}
 779
 780cleanup:
 781	brelse(new_bh);
 782
 783	return error;
 784}
 785
 786/*
 787 * ext2_xattr_delete_inode()
 788 *
 789 * Free extended attribute resources associated with this inode. This
 790 * is called immediately before an inode is freed.
 791 */
 792void
 793ext2_xattr_delete_inode(struct inode *inode)
 794{
 795	struct buffer_head *bh = NULL;
 796	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
 797
 798	/*
 799	 * We are the only ones holding inode reference. The xattr_sem should
 800	 * better be unlocked! We could as well just not acquire xattr_sem at
 801	 * all but this makes the code more futureproof. OTOH we need trylock
 802	 * here to avoid false-positive warning from lockdep about reclaim
 803	 * circular dependency.
 804	 */
 805	if (WARN_ON_ONCE(!down_write_trylock(&EXT2_I(inode)->xattr_sem)))
 806		return;
 807	if (!EXT2_I(inode)->i_file_acl)
 808		goto cleanup;
 809
 810	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
 811		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 812			"inode %ld: xattr block %d is out of data blocks range",
 813			inode->i_ino, EXT2_I(inode)->i_file_acl);
 814		goto cleanup;
 815	}
 816
 817	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 818	if (!bh) {
 819		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 820			"inode %ld: block %d read error", inode->i_ino,
 821			EXT2_I(inode)->i_file_acl);
 822		goto cleanup;
 823	}
 824	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 825	if (!ext2_xattr_header_valid(HDR(bh))) {
 826		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 827			"inode %ld: bad block %d", inode->i_ino,
 828			EXT2_I(inode)->i_file_acl);
 829		goto cleanup;
 830	}
 831	lock_buffer(bh);
 832	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 833		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 834
 835		/*
 836		 * This must happen under buffer lock for ext2_xattr_set2() to
 837		 * reliably detect freed block
 838		 */
 839		mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
 840				      bh->b_blocknr);
 841		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 842		get_bh(bh);
 843		bforget(bh);
 844		unlock_buffer(bh);
 845	} else {
 846		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 847		ea_bdebug(bh, "refcount now=%d",
 848			le32_to_cpu(HDR(bh)->h_refcount));
 849		unlock_buffer(bh);
 850		mark_buffer_dirty(bh);
 851		if (IS_SYNC(inode))
 852			sync_dirty_buffer(bh);
 853		dquot_free_block_nodirty(inode, 1);
 854	}
 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);
 946again:
 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			/*
 959			 * We have to be careful about races with freeing or
 960			 * rehashing of xattr block. Once we hold buffer lock
 961			 * xattr block's state is stable so we can check
 962			 * whether the block got freed / rehashed or not.
 963			 * Since we unhash mbcache entry under buffer lock when
 964			 * freeing / rehashing xattr block, checking whether
 965			 * entry is still hashed is reliable.
 966			 */
 967			if (hlist_bl_unhashed(&ce->e_hash_list)) {
 968				mb_cache_entry_put(ea_block_cache, ce);
 969				unlock_buffer(bh);
 970				brelse(bh);
 971				goto again;
 972			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
 973				   EXT2_XATTR_REFCOUNT_MAX) {
 974				ea_idebug(inode, "block %ld refcount %d>%d",
 975					  (unsigned long) ce->e_value,
 976					  le32_to_cpu(HDR(bh)->h_refcount),
 977					  EXT2_XATTR_REFCOUNT_MAX);
 978			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 979				ea_bdebug(bh, "b_count=%d",
 980					  atomic_read(&(bh->b_count)));
 981				mb_cache_entry_touch(ea_block_cache, ce);
 982				mb_cache_entry_put(ea_block_cache, ce);
 983				return bh;
 984			}
 985			unlock_buffer(bh);
 986			brelse(bh);
 987		}
 988		ce = mb_cache_entry_find_next(ea_block_cache, ce);
 989	}
 990	return NULL;
 991}
 992
 993#define NAME_HASH_SHIFT 5
 994#define VALUE_HASH_SHIFT 16
 995
 996/*
 997 * ext2_xattr_hash_entry()
 998 *
 999 * Compute the hash of an extended attribute.
1000 */
1001static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
1002					 struct ext2_xattr_entry *entry)
1003{
1004	__u32 hash = 0;
1005	char *name = entry->e_name;
1006	int n;
1007
1008	for (n=0; n < entry->e_name_len; n++) {
1009		hash = (hash << NAME_HASH_SHIFT) ^
1010		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1011		       *name++;
1012	}
1013
1014	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1015		__le32 *value = (__le32 *)((char *)header +
1016			le16_to_cpu(entry->e_value_offs));
1017		for (n = (le32_to_cpu(entry->e_value_size) +
1018		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1019			hash = (hash << VALUE_HASH_SHIFT) ^
1020			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1021			       le32_to_cpu(*value++);
1022		}
1023	}
1024	entry->e_hash = cpu_to_le32(hash);
1025}
1026
1027#undef NAME_HASH_SHIFT
1028#undef VALUE_HASH_SHIFT
1029
1030#define BLOCK_HASH_SHIFT 16
1031
1032/*
1033 * ext2_xattr_rehash()
1034 *
1035 * Re-compute the extended attribute hash value after an entry has changed.
1036 */
1037static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1038			      struct ext2_xattr_entry *entry)
1039{
1040	struct ext2_xattr_entry *here;
1041	__u32 hash = 0;
1042	
1043	ext2_xattr_hash_entry(header, entry);
1044	here = ENTRY(header+1);
1045	while (!IS_LAST_ENTRY(here)) {
1046		if (!here->e_hash) {
1047			/* Block is not shared if an entry's hash value == 0 */
1048			hash = 0;
1049			break;
1050		}
1051		hash = (hash << BLOCK_HASH_SHIFT) ^
1052		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1053		       le32_to_cpu(here->e_hash);
1054		here = EXT2_XATTR_NEXT(here);
1055	}
1056	header->h_hash = cpu_to_le32(hash);
1057}
1058
1059#undef BLOCK_HASH_SHIFT
1060
1061#define HASH_BUCKET_BITS 10
1062
1063struct mb_cache *ext2_xattr_create_cache(void)
1064{
1065	return mb_cache_create(HASH_BUCKET_BITS);
1066}
1067
1068void ext2_xattr_destroy_cache(struct mb_cache *cache)
1069{
1070	if (cache)
1071		mb_cache_destroy(cache);
1072}