Linux Audio

Check our new training course

Loading...
v5.9
   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}
v3.5.6
 
   1/*
   2 * linux/fs/ext2/xattr.c
   3 *
   4 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
   5 *
   6 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   7 * Extended attributes for symlinks and special files added per
   8 *  suggestion of Luka Renko <luka.renko@hermes.si>.
   9 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  10 *  Red Hat Inc.
  11 *
  12 */
  13
  14/*
  15 * Extended attributes are stored on disk blocks allocated outside of
  16 * any inode. The i_file_acl field is then made to point to this allocated
  17 * block. If all extended attributes of an inode are identical, these
  18 * inodes may share the same extended attribute block. Such situations
  19 * are automatically detected by keeping a cache of recent attribute block
  20 * numbers and hashes over the block's contents in memory.
  21 *
  22 *
  23 * Extended attribute block layout:
  24 *
  25 *   +------------------+
  26 *   | header           |
  27 *   | entry 1          | |
  28 *   | entry 2          | | growing downwards
  29 *   | entry 3          | v
  30 *   | four null bytes  |
  31 *   | . . .            |
  32 *   | value 1          | ^
  33 *   | value 3          | | growing upwards
  34 *   | value 2          | |
  35 *   +------------------+
  36 *
  37 * The block header is followed by multiple entry descriptors. These entry
  38 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
  39 * byte boundaries. The entry descriptors are sorted by attribute name,
  40 * so that two extended attribute blocks can be compared efficiently.
  41 *
  42 * Attribute values are aligned to the end of the block, stored in
  43 * no specific order. They are also padded to EXT2_XATTR_PAD byte
  44 * boundaries. No additional gaps are left between them.
  45 *
  46 * Locking strategy
  47 * ----------------
  48 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
  49 * EA blocks are only changed if they are exclusive to an inode, so
  50 * holding xattr_sem also means that nothing but the EA block's reference
  51 * count will change. Multiple writers to an EA block are synchronized
  52 * by the bh lock. No more than a single bh lock is held at any time
  53 * to avoid deadlocks.
  54 */
  55
  56#include <linux/buffer_head.h>
  57#include <linux/init.h>
 
  58#include <linux/slab.h>
  59#include <linux/mbcache.h>
  60#include <linux/quotaops.h>
  61#include <linux/rwsem.h>
  62#include <linux/security.h>
  63#include "ext2.h"
  64#include "xattr.h"
  65#include "acl.h"
  66
  67#define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
  68#define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
  69#define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
  70#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
  71
  72#ifdef EXT2_XATTR_DEBUG
  73# define ea_idebug(inode, f...) do { \
  74		printk(KERN_DEBUG "inode %s:%ld: ", \
  75			inode->i_sb->s_id, inode->i_ino); \
  76		printk(f); \
  77		printk("\n"); \
  78	} while (0)
  79# define ea_bdebug(bh, f...) do { \
  80		char b[BDEVNAME_SIZE]; \
  81		printk(KERN_DEBUG "block %s:%lu: ", \
  82			bdevname(bh->b_bdev, b), \
  83			(unsigned long) bh->b_blocknr); \
  84		printk(f); \
  85		printk("\n"); \
  86	} while (0)
  87#else
  88# define ea_idebug(f...)
  89# define ea_bdebug(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 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 struct mb_cache *ext2_xattr_cache;
 102
 103static const struct xattr_handler *ext2_xattr_handler_map[] = {
 104	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
 105#ifdef CONFIG_EXT2_FS_POSIX_ACL
 106	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext2_xattr_acl_access_handler,
 107	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
 108#endif
 109	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
 110#ifdef CONFIG_EXT2_FS_SECURITY
 111	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
 112#endif
 113};
 114
 115const struct xattr_handler *ext2_xattr_handlers[] = {
 116	&ext2_xattr_user_handler,
 117	&ext2_xattr_trusted_handler,
 118#ifdef CONFIG_EXT2_FS_POSIX_ACL
 119	&ext2_xattr_acl_access_handler,
 120	&ext2_xattr_acl_default_handler,
 121#endif
 122#ifdef CONFIG_EXT2_FS_SECURITY
 123	&ext2_xattr_security_handler,
 124#endif
 125	NULL
 126};
 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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 138/*
 139 * ext2_xattr_get()
 140 *
 141 * Copy an extended attribute into the buffer
 142 * provided, or compute the buffer size required.
 143 * Buffer is NULL to compute the size of the buffer required.
 144 *
 145 * Returns a negative error number on failure, or the number of bytes
 146 * used / required on success.
 147 */
 148int
 149ext2_xattr_get(struct inode *inode, int name_index, const char *name,
 150	       void *buffer, size_t buffer_size)
 151{
 152	struct buffer_head *bh = NULL;
 153	struct ext2_xattr_entry *entry;
 154	size_t name_len, size;
 155	char *end;
 156	int error;
 
 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(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(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 = dentry->d_inode;
 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
 253	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 254		  buffer, (long)buffer_size);
 255
 256	down_read(&EXT2_I(inode)->xattr_sem);
 257	error = 0;
 258	if (!EXT2_I(inode)->i_file_acl)
 259		goto cleanup;
 260	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
 261	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 262	error = -EIO;
 263	if (!bh)
 264		goto cleanup;
 265	ea_bdebug(bh, "b_count=%d, refcount=%d",
 266		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
 267	end = bh->b_data + bh->b_size;
 268	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 269	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 270bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
 271			"inode %ld: bad block %d", inode->i_ino,
 272			EXT2_I(inode)->i_file_acl);
 273		error = -EIO;
 274		goto cleanup;
 275	}
 276
 277	/* check the on-disk data structure */
 278	entry = FIRST_ENTRY(bh);
 279	while (!IS_LAST_ENTRY(entry)) {
 280		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
 281
 282		if ((char *)next >= end)
 283			goto bad_block;
 284		entry = next;
 285	}
 286	if (ext2_xattr_cache_insert(bh))
 287		ea_idebug(inode, "cache insert failed");
 288
 289	/* list the attribute names */
 290	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
 291	     entry = EXT2_XATTR_NEXT(entry)) {
 292		const struct xattr_handler *handler =
 293			ext2_xattr_handler(entry->e_name_index);
 294
 295		if (handler) {
 296			size_t size = handler->list(dentry, buffer, rest,
 297						    entry->e_name,
 298						    entry->e_name_len,
 299						    handler->flags);
 300			if (buffer) {
 301				if (size > rest) {
 302					error = -ERANGE;
 303					goto cleanup;
 304				}
 305				buffer += size;
 
 
 
 
 306			}
 307			rest -= size;
 308		}
 309	}
 310	error = buffer_size - rest;  /* total size */
 311
 312cleanup:
 313	brelse(bh);
 314	up_read(&EXT2_I(inode)->xattr_sem);
 315
 316	return error;
 317}
 318
 319/*
 320 * Inode operation listxattr()
 321 *
 322 * dentry->d_inode->i_mutex: don't care
 323 */
 324ssize_t
 325ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
 326{
 327	return ext2_xattr_list(dentry, buffer, size);
 328}
 329
 330/*
 331 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 332 * not set, set it.
 333 */
 334static void ext2_xattr_update_super_block(struct super_block *sb)
 335{
 336	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
 337		return;
 338
 339	spin_lock(&EXT2_SB(sb)->s_lock);
 
 340	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
 341	spin_unlock(&EXT2_SB(sb)->s_lock);
 342	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
 343}
 344
 345/*
 346 * ext2_xattr_set()
 347 *
 348 * Create, replace or remove an extended attribute for this inode.  Value
 349 * is NULL to remove an existing extended attribute, and non-NULL to
 350 * either replace an existing extended attribute, or create a new extended
 351 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
 352 * specify that an extended attribute must exist and must not exist
 353 * previous to the call, respectively.
 354 *
 355 * Returns 0, or a negative error number on failure.
 356 */
 357int
 358ext2_xattr_set(struct inode *inode, int name_index, const char *name,
 359	       const void *value, size_t value_len, int flags)
 360{
 361	struct super_block *sb = inode->i_sb;
 362	struct buffer_head *bh = NULL;
 363	struct ext2_xattr_header *header = NULL;
 364	struct ext2_xattr_entry *here, *last;
 365	size_t name_len, free, min_offs = sb->s_blocksize;
 366	int not_found = 1, error;
 367	char *end;
 368	
 369	/*
 370	 * header -- Points either into bh, or to a temporarily
 371	 *           allocated buffer.
 372	 * here -- The named entry found, or the place for inserting, within
 373	 *         the block pointed to by header.
 374	 * last -- Points right after the last named entry within the block
 375	 *         pointed to by header.
 376	 * min_offs -- The offset of the first value (values are aligned
 377	 *             towards the end of the block).
 378	 * end -- Points right after the block pointed to by header.
 379	 */
 380	
 381	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
 382		  name_index, name, value, (long)value_len);
 383
 384	if (value == NULL)
 385		value_len = 0;
 386	if (name == NULL)
 387		return -EINVAL;
 388	name_len = strlen(name);
 389	if (name_len > 255 || value_len > sb->s_blocksize)
 390		return -ERANGE;
 
 
 
 391	down_write(&EXT2_I(inode)->xattr_sem);
 392	if (EXT2_I(inode)->i_file_acl) {
 393		/* The inode already has an extended attribute block. */
 394		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
 395		error = -EIO;
 396		if (!bh)
 397			goto cleanup;
 398		ea_bdebug(bh, "b_count=%d, refcount=%d",
 399			atomic_read(&(bh->b_count)),
 400			le32_to_cpu(HDR(bh)->h_refcount));
 401		header = HDR(bh);
 402		end = bh->b_data + bh->b_size;
 403		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 404		    header->h_blocks != cpu_to_le32(1)) {
 405bad_block:		ext2_error(sb, "ext2_xattr_set",
 406				"inode %ld: bad block %d", inode->i_ino, 
 407				   EXT2_I(inode)->i_file_acl);
 408			error = -EIO;
 409			goto cleanup;
 410		}
 411		/* Find the named attribute. */
 412		here = FIRST_ENTRY(bh);
 413		while (!IS_LAST_ENTRY(here)) {
 414			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
 415			if ((char *)next >= end)
 416				goto bad_block;
 417			if (!here->e_value_block && here->e_value_size) {
 418				size_t offs = le16_to_cpu(here->e_value_offs);
 419				if (offs < min_offs)
 420					min_offs = offs;
 421			}
 422			not_found = name_index - here->e_name_index;
 423			if (!not_found)
 424				not_found = name_len - here->e_name_len;
 425			if (!not_found)
 426				not_found = memcmp(name, here->e_name,name_len);
 427			if (not_found <= 0)
 428				break;
 429			here = next;
 430		}
 431		last = here;
 432		/* We still need to compute min_offs and last. */
 433		while (!IS_LAST_ENTRY(last)) {
 434			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
 435			if ((char *)next >= end)
 436				goto bad_block;
 437			if (!last->e_value_block && last->e_value_size) {
 438				size_t offs = le16_to_cpu(last->e_value_offs);
 439				if (offs < min_offs)
 440					min_offs = offs;
 441			}
 442			last = next;
 
 
 
 
 
 
 
 443		}
 
 
 444
 445		/* Check whether we have enough space left. */
 446		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
 447	} else {
 448		/* We will use a new extended attribute block. */
 449		free = sb->s_blocksize -
 450			sizeof(struct ext2_xattr_header) - sizeof(__u32);
 451		here = last = NULL;  /* avoid gcc uninitialized warning. */
 452	}
 453
 454	if (not_found) {
 455		/* Request to remove a nonexistent attribute? */
 456		error = -ENODATA;
 457		if (flags & XATTR_REPLACE)
 458			goto cleanup;
 459		error = 0;
 460		if (value == NULL)
 461			goto cleanup;
 462	} else {
 463		/* Request to create an existing attribute? */
 464		error = -EEXIST;
 465		if (flags & XATTR_CREATE)
 466			goto cleanup;
 467		if (!here->e_value_block && here->e_value_size) {
 468			size_t size = le32_to_cpu(here->e_value_size);
 469
 470			if (le16_to_cpu(here->e_value_offs) + size > 
 471			    sb->s_blocksize || size > sb->s_blocksize)
 472				goto bad_block;
 473			free += EXT2_XATTR_SIZE(size);
 474		}
 475		free += EXT2_XATTR_LEN(name_len);
 476	}
 477	error = -ENOSPC;
 478	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
 479		goto cleanup;
 480
 481	/* Here we know that we can set the new attribute. */
 482
 483	if (header) {
 484		struct mb_cache_entry *ce;
 485
 486		/* assert(header == HDR(bh)); */
 487		ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
 488					bh->b_blocknr);
 489		lock_buffer(bh);
 490		if (header->h_refcount == cpu_to_le32(1)) {
 
 
 491			ea_bdebug(bh, "modifying in-place");
 492			if (ce)
 493				mb_cache_entry_free(ce);
 
 
 
 
 
 494			/* keep the buffer locked while modifying it. */
 495		} else {
 496			int offset;
 497
 498			if (ce)
 499				mb_cache_entry_release(ce);
 500			unlock_buffer(bh);
 501			ea_bdebug(bh, "cloning");
 502			header = kmalloc(bh->b_size, GFP_KERNEL);
 503			error = -ENOMEM;
 504			if (header == NULL)
 505				goto cleanup;
 506			memcpy(header, HDR(bh), bh->b_size);
 507			header->h_refcount = cpu_to_le32(1);
 508
 509			offset = (char *)here - bh->b_data;
 510			here = ENTRY((char *)header + offset);
 511			offset = (char *)last - bh->b_data;
 512			last = ENTRY((char *)header + offset);
 513		}
 514	} else {
 515		/* Allocate a buffer where we construct the new block. */
 516		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
 517		error = -ENOMEM;
 518		if (header == NULL)
 519			goto cleanup;
 520		end = (char *)header + sb->s_blocksize;
 521		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
 522		header->h_blocks = header->h_refcount = cpu_to_le32(1);
 523		last = here = ENTRY(header+1);
 524	}
 525
 526	/* Iff we are modifying the block in-place, bh is locked here. */
 527
 528	if (not_found) {
 529		/* Insert the new name. */
 530		size_t size = EXT2_XATTR_LEN(name_len);
 531		size_t rest = (char *)last - (char *)here;
 532		memmove((char *)here + size, here, rest);
 533		memset(here, 0, size);
 534		here->e_name_index = name_index;
 535		here->e_name_len = name_len;
 536		memcpy(here->e_name, name, name_len);
 537	} else {
 538		if (!here->e_value_block && here->e_value_size) {
 539			char *first_val = (char *)header + min_offs;
 540			size_t offs = le16_to_cpu(here->e_value_offs);
 541			char *val = (char *)header + offs;
 542			size_t size = EXT2_XATTR_SIZE(
 543				le32_to_cpu(here->e_value_size));
 544
 545			if (size == EXT2_XATTR_SIZE(value_len)) {
 546				/* The old and the new value have the same
 547				   size. Just replace. */
 548				here->e_value_size = cpu_to_le32(value_len);
 549				memset(val + size - EXT2_XATTR_PAD, 0,
 550				       EXT2_XATTR_PAD); /* Clear pad bytes. */
 551				memcpy(val, value, value_len);
 552				goto skip_replace;
 553			}
 554
 555			/* Remove the old value. */
 556			memmove(first_val + size, first_val, val - first_val);
 557			memset(first_val, 0, size);
 558			here->e_value_offs = 0;
 559			min_offs += size;
 560
 561			/* Adjust all value offsets. */
 562			last = ENTRY(header+1);
 563			while (!IS_LAST_ENTRY(last)) {
 564				size_t o = le16_to_cpu(last->e_value_offs);
 565				if (!last->e_value_block && o < offs)
 566					last->e_value_offs =
 567						cpu_to_le16(o + size);
 568				last = EXT2_XATTR_NEXT(last);
 569			}
 
 
 570		}
 571		if (value == NULL) {
 572			/* Remove the old name. */
 573			size_t size = EXT2_XATTR_LEN(name_len);
 574			last = ENTRY((char *)last - size);
 575			memmove(here, (char*)here + size,
 576				(char*)last - (char*)here);
 577			memset(last, 0, size);
 578		}
 579	}
 580
 581	if (value != NULL) {
 582		/* Insert the new value. */
 583		here->e_value_size = cpu_to_le32(value_len);
 584		if (value_len) {
 585			size_t size = EXT2_XATTR_SIZE(value_len);
 586			char *val = (char *)header + min_offs - size;
 587			here->e_value_offs =
 588				cpu_to_le16((char *)val - (char *)header);
 589			memset(val + size - EXT2_XATTR_PAD, 0,
 590			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
 591			memcpy(val, value, value_len);
 592		}
 593	}
 594
 595skip_replace:
 596	if (IS_LAST_ENTRY(ENTRY(header+1))) {
 597		/* This block is now empty. */
 598		if (bh && header == HDR(bh))
 599			unlock_buffer(bh);  /* we were modifying in-place. */
 600		error = ext2_xattr_set2(inode, bh, NULL);
 601	} else {
 602		ext2_xattr_rehash(header, here);
 603		if (bh && header == HDR(bh))
 604			unlock_buffer(bh);  /* we were modifying in-place. */
 605		error = ext2_xattr_set2(inode, bh, header);
 606	}
 607
 608cleanup:
 609	brelse(bh);
 610	if (!(bh && header == HDR(bh)))
 611		kfree(header);
 
 612	up_write(&EXT2_I(inode)->xattr_sem);
 613
 614	return error;
 615}
 616
 617/*
 618 * Second half of ext2_xattr_set(): Update the file system.
 619 */
 620static int
 621ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
 622		struct ext2_xattr_header *header)
 623{
 624	struct super_block *sb = inode->i_sb;
 625	struct buffer_head *new_bh = NULL;
 626	int error;
 
 627
 628	if (header) {
 629		new_bh = ext2_xattr_cache_find(inode, header);
 630		if (new_bh) {
 631			/* We found an identical block in the cache. */
 632			if (new_bh == old_bh) {
 633				ea_bdebug(new_bh, "keeping this block");
 634			} else {
 635				/* The old block is released after updating
 636				   the inode.  */
 637				ea_bdebug(new_bh, "reusing block");
 638
 639				error = dquot_alloc_block(inode, 1);
 640				if (error) {
 641					unlock_buffer(new_bh);
 642					goto cleanup;
 643				}
 644				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
 645				ea_bdebug(new_bh, "refcount now=%d",
 646					le32_to_cpu(HDR(new_bh)->h_refcount));
 647			}
 648			unlock_buffer(new_bh);
 649		} else if (old_bh && header == HDR(old_bh)) {
 650			/* Keep this block. No need to lock the block as we
 651			   don't need to change the reference count. */
 652			new_bh = old_bh;
 653			get_bh(new_bh);
 654			ext2_xattr_cache_insert(new_bh);
 655		} else {
 656			/* We need to allocate a new block */
 657			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
 658						EXT2_I(inode)->i_block_group);
 659			int block = ext2_new_block(inode, goal, &error);
 660			if (error)
 661				goto cleanup;
 662			ea_idebug(inode, "creating block %d", block);
 663
 664			new_bh = sb_getblk(sb, block);
 665			if (!new_bh) {
 666				ext2_free_blocks(inode, block, 1);
 667				mark_inode_dirty(inode);
 668				error = -EIO;
 669				goto cleanup;
 670			}
 671			lock_buffer(new_bh);
 672			memcpy(new_bh->b_data, header, new_bh->b_size);
 673			set_buffer_uptodate(new_bh);
 674			unlock_buffer(new_bh);
 675			ext2_xattr_cache_insert(new_bh);
 676			
 677			ext2_xattr_update_super_block(sb);
 678		}
 679		mark_buffer_dirty(new_bh);
 680		if (IS_SYNC(inode)) {
 681			sync_dirty_buffer(new_bh);
 682			error = -EIO;
 683			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
 684				goto cleanup;
 685		}
 686	}
 687
 688	/* Update the inode. */
 689	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 690	inode->i_ctime = CURRENT_TIME_SEC;
 691	if (IS_SYNC(inode)) {
 692		error = sync_inode_metadata(inode, 1);
 693		/* In case sync failed due to ENOSPC the inode was actually
 694		 * written (only some dirty data were not) so we just proceed
 695		 * as if nothing happened and cleanup the unused block */
 696		if (error && error != -ENOSPC) {
 697			if (new_bh && new_bh != old_bh) {
 698				dquot_free_block_nodirty(inode, 1);
 699				mark_inode_dirty(inode);
 700			}
 701			goto cleanup;
 702		}
 703	} else
 704		mark_inode_dirty(inode);
 705
 706	error = 0;
 707	if (old_bh && old_bh != new_bh) {
 708		struct mb_cache_entry *ce;
 709
 710		/*
 711		 * If there was an old block and we are no longer using it,
 712		 * release the old block.
 713		 */
 714		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
 715					old_bh->b_blocknr);
 716		lock_buffer(old_bh);
 717		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
 
 
 
 
 
 
 
 
 718			/* Free the old block. */
 719			if (ce)
 720				mb_cache_entry_free(ce);
 721			ea_bdebug(old_bh, "freeing");
 722			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
 723			mark_inode_dirty(inode);
 724			/* We let our caller release old_bh, so we
 725			 * need to duplicate the buffer before. */
 726			get_bh(old_bh);
 727			bforget(old_bh);
 728		} else {
 729			/* Decrement the refcount only. */
 730			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
 731			if (ce)
 732				mb_cache_entry_release(ce);
 733			dquot_free_block_nodirty(inode, 1);
 734			mark_inode_dirty(inode);
 735			mark_buffer_dirty(old_bh);
 736			ea_bdebug(old_bh, "refcount now=%d",
 737				le32_to_cpu(HDR(old_bh)->h_refcount));
 738		}
 739		unlock_buffer(old_bh);
 740	}
 741
 742cleanup:
 743	brelse(new_bh);
 744
 745	return error;
 746}
 747
 748/*
 749 * ext2_xattr_delete_inode()
 750 *
 751 * Free extended attribute resources associated with this inode. This
 752 * is called immediately before an inode is freed.
 753 */
 754void
 755ext2_xattr_delete_inode(struct inode *inode)
 756{
 757	struct buffer_head *bh = NULL;
 758	struct mb_cache_entry *ce;
 759
 760	down_write(&EXT2_I(inode)->xattr_sem);
 
 
 
 
 
 
 
 
 761	if (!EXT2_I(inode)->i_file_acl)
 762		goto cleanup;
 
 
 
 
 
 
 
 
 763	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
 764	if (!bh) {
 765		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 766			"inode %ld: block %d read error", inode->i_ino,
 767			EXT2_I(inode)->i_file_acl);
 768		goto cleanup;
 769	}
 770	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
 771	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
 772	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
 773		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
 774			"inode %ld: bad block %d", inode->i_ino,
 775			EXT2_I(inode)->i_file_acl);
 776		goto cleanup;
 777	}
 778	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
 779	lock_buffer(bh);
 780	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
 781		if (ce)
 782			mb_cache_entry_free(ce);
 
 
 
 
 
 
 783		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
 784		get_bh(bh);
 785		bforget(bh);
 786		unlock_buffer(bh);
 787	} else {
 788		le32_add_cpu(&HDR(bh)->h_refcount, -1);
 789		if (ce)
 790			mb_cache_entry_release(ce);
 791		ea_bdebug(bh, "refcount now=%d",
 792			le32_to_cpu(HDR(bh)->h_refcount));
 793		unlock_buffer(bh);
 794		mark_buffer_dirty(bh);
 795		if (IS_SYNC(inode))
 796			sync_dirty_buffer(bh);
 797		dquot_free_block_nodirty(inode, 1);
 798	}
 799	EXT2_I(inode)->i_file_acl = 0;
 800
 801cleanup:
 802	brelse(bh);
 803	up_write(&EXT2_I(inode)->xattr_sem);
 804}
 805
 806/*
 807 * ext2_xattr_put_super()
 808 *
 809 * This is called when a file system is unmounted.
 810 */
 811void
 812ext2_xattr_put_super(struct super_block *sb)
 813{
 814	mb_cache_shrink(sb->s_bdev);
 815}
 816
 817
 818/*
 819 * ext2_xattr_cache_insert()
 820 *
 821 * Create a new entry in the extended attribute cache, and insert
 822 * it unless such an entry is already in the cache.
 823 *
 824 * Returns 0, or a negative error number on failure.
 825 */
 826static int
 827ext2_xattr_cache_insert(struct buffer_head *bh)
 828{
 829	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
 830	struct mb_cache_entry *ce;
 831	int error;
 832
 833	ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
 834	if (!ce)
 835		return -ENOMEM;
 836	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
 837	if (error) {
 838		mb_cache_entry_free(ce);
 839		if (error == -EBUSY) {
 840			ea_bdebug(bh, "already in cache (%d cache entries)",
 841				atomic_read(&ext2_xattr_cache->c_entry_count));
 842			error = 0;
 843		}
 844	} else {
 845		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
 846			  atomic_read(&ext2_xattr_cache->c_entry_count));
 847		mb_cache_entry_release(ce);
 848	}
 849	return error;
 850}
 851
 852/*
 853 * ext2_xattr_cmp()
 854 *
 855 * Compare two extended attribute blocks for equality.
 856 *
 857 * Returns 0 if the blocks are equal, 1 if they differ, and
 858 * a negative error number on errors.
 859 */
 860static int
 861ext2_xattr_cmp(struct ext2_xattr_header *header1,
 862	       struct ext2_xattr_header *header2)
 863{
 864	struct ext2_xattr_entry *entry1, *entry2;
 865
 866	entry1 = ENTRY(header1+1);
 867	entry2 = ENTRY(header2+1);
 868	while (!IS_LAST_ENTRY(entry1)) {
 869		if (IS_LAST_ENTRY(entry2))
 870			return 1;
 871		if (entry1->e_hash != entry2->e_hash ||
 872		    entry1->e_name_index != entry2->e_name_index ||
 873		    entry1->e_name_len != entry2->e_name_len ||
 874		    entry1->e_value_size != entry2->e_value_size ||
 875		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 876			return 1;
 877		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 878			return -EIO;
 879		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 880			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 881			   le32_to_cpu(entry1->e_value_size)))
 882			return 1;
 883
 884		entry1 = EXT2_XATTR_NEXT(entry1);
 885		entry2 = EXT2_XATTR_NEXT(entry2);
 886	}
 887	if (!IS_LAST_ENTRY(entry2))
 888		return 1;
 889	return 0;
 890}
 891
 892/*
 893 * ext2_xattr_cache_find()
 894 *
 895 * Find an identical extended attribute block.
 896 *
 897 * Returns a locked buffer head to the block found, or NULL if such
 898 * a block was not found or an error occurred.
 899 */
 900static struct buffer_head *
 901ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
 902{
 903	__u32 hash = le32_to_cpu(header->h_hash);
 904	struct mb_cache_entry *ce;
 
 905
 906	if (!header->h_hash)
 907		return NULL;  /* never share */
 908	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
 909again:
 910	ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev,
 911				       hash);
 912	while (ce) {
 913		struct buffer_head *bh;
 914
 915		if (IS_ERR(ce)) {
 916			if (PTR_ERR(ce) == -EAGAIN)
 917				goto again;
 918			break;
 919		}
 920
 921		bh = sb_bread(inode->i_sb, ce->e_block);
 922		if (!bh) {
 923			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
 924				"inode %ld: block %ld read error",
 925				inode->i_ino, (unsigned long) ce->e_block);
 926		} else {
 927			lock_buffer(bh);
 928			if (le32_to_cpu(HDR(bh)->h_refcount) >
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 929				   EXT2_XATTR_REFCOUNT_MAX) {
 930				ea_idebug(inode, "block %ld refcount %d>%d",
 931					  (unsigned long) ce->e_block,
 932					  le32_to_cpu(HDR(bh)->h_refcount),
 933					  EXT2_XATTR_REFCOUNT_MAX);
 934			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
 935				ea_bdebug(bh, "b_count=%d",
 936					  atomic_read(&(bh->b_count)));
 937				mb_cache_entry_release(ce);
 
 938				return bh;
 939			}
 940			unlock_buffer(bh);
 941			brelse(bh);
 942		}
 943		ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
 944	}
 945	return NULL;
 946}
 947
 948#define NAME_HASH_SHIFT 5
 949#define VALUE_HASH_SHIFT 16
 950
 951/*
 952 * ext2_xattr_hash_entry()
 953 *
 954 * Compute the hash of an extended attribute.
 955 */
 956static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
 957					 struct ext2_xattr_entry *entry)
 958{
 959	__u32 hash = 0;
 960	char *name = entry->e_name;
 961	int n;
 962
 963	for (n=0; n < entry->e_name_len; n++) {
 964		hash = (hash << NAME_HASH_SHIFT) ^
 965		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
 966		       *name++;
 967	}
 968
 969	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 970		__le32 *value = (__le32 *)((char *)header +
 971			le16_to_cpu(entry->e_value_offs));
 972		for (n = (le32_to_cpu(entry->e_value_size) +
 973		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
 974			hash = (hash << VALUE_HASH_SHIFT) ^
 975			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
 976			       le32_to_cpu(*value++);
 977		}
 978	}
 979	entry->e_hash = cpu_to_le32(hash);
 980}
 981
 982#undef NAME_HASH_SHIFT
 983#undef VALUE_HASH_SHIFT
 984
 985#define BLOCK_HASH_SHIFT 16
 986
 987/*
 988 * ext2_xattr_rehash()
 989 *
 990 * Re-compute the extended attribute hash value after an entry has changed.
 991 */
 992static void ext2_xattr_rehash(struct ext2_xattr_header *header,
 993			      struct ext2_xattr_entry *entry)
 994{
 995	struct ext2_xattr_entry *here;
 996	__u32 hash = 0;
 997	
 998	ext2_xattr_hash_entry(header, entry);
 999	here = ENTRY(header+1);
1000	while (!IS_LAST_ENTRY(here)) {
1001		if (!here->e_hash) {
1002			/* Block is not shared if an entry's hash value == 0 */
1003			hash = 0;
1004			break;
1005		}
1006		hash = (hash << BLOCK_HASH_SHIFT) ^
1007		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1008		       le32_to_cpu(here->e_hash);
1009		here = EXT2_XATTR_NEXT(here);
1010	}
1011	header->h_hash = cpu_to_le32(hash);
1012}
1013
1014#undef BLOCK_HASH_SHIFT
1015
1016int __init
1017init_ext2_xattr(void)
 
1018{
1019	ext2_xattr_cache = mb_cache_create("ext2_xattr", 6);
1020	if (!ext2_xattr_cache)
1021		return -ENOMEM;
1022	return 0;
1023}
1024
1025void
1026exit_ext2_xattr(void)
1027{
1028	mb_cache_destroy(ext2_xattr_cache);
 
1029}