Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/ext4/xattr.c
   4 *
   5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6 *
   7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9 * Extended attributes for symlinks and special files added per
  10 *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12 *  Red Hat Inc.
  13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14 *  and Andreas Gruenbacher <agruen@suse.de>.
  15 */
  16
  17/*
  18 * Extended attributes are stored directly in inodes (on file systems with
  19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20 * field contains the block number if an inode uses an additional block. All
  21 * attributes must fit in the inode and one additional block. Blocks that
  22 * contain the identical set of attributes may be shared among several inodes.
  23 * Identical blocks are detected by keeping a cache of blocks that have
  24 * recently been accessed.
  25 *
  26 * The attributes in inodes and on blocks have a different header; the entries
  27 * are stored in the same format:
  28 *
  29 *   +------------------+
  30 *   | header           |
  31 *   | entry 1          | |
  32 *   | entry 2          | | growing downwards
  33 *   | entry 3          | v
  34 *   | four null bytes  |
  35 *   | . . .            |
  36 *   | value 1          | ^
  37 *   | value 3          | | growing upwards
  38 *   | value 2          | |
  39 *   +------------------+
  40 *
  41 * The header is followed by multiple entry descriptors. In disk blocks, the
  42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43 * attribute values are aligned to the end of the block in no specific order.
  44 *
  45 * Locking strategy
  46 * ----------------
  47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48 * EA blocks are only changed if they are exclusive to an inode, so
  49 * holding xattr_sem also means that nothing but the EA block's reference
  50 * count can change. Multiple writers to the same block are synchronized
  51 * by the buffer lock.
  52 */
  53
  54#include <linux/init.h>
  55#include <linux/fs.h>
  56#include <linux/slab.h>
  57#include <linux/mbcache.h>
  58#include <linux/quotaops.h>
  59#include <linux/iversion.h>
  60#include "ext4_jbd2.h"
  61#include "ext4.h"
  62#include "xattr.h"
  63#include "acl.h"
  64
  65#ifdef EXT4_XATTR_DEBUG
  66# define ea_idebug(inode, fmt, ...)					\
  67	printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",			\
  68	       inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69# define ea_bdebug(bh, fmt, ...)					\
  70	printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",			\
  71	       bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72#else
  73# define ea_idebug(inode, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
  74# define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
  75#endif
  76
  77static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78					  struct buffer_head *);
  79static struct buffer_head *
  80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81			    struct mb_cache_entry **);
  82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83				    size_t value_count);
  84static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
  85				    size_t value_count);
  86static void ext4_xattr_rehash(struct ext4_xattr_header *);
  87
  88static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  89	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
  90#ifdef CONFIG_EXT4_FS_POSIX_ACL
  91	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  92	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  93#endif
  94	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
  95#ifdef CONFIG_EXT4_FS_SECURITY
  96	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
  97#endif
  98	[EXT4_XATTR_INDEX_HURD]		     = &ext4_xattr_hurd_handler,
  99};
 100
 101const struct xattr_handler *ext4_xattr_handlers[] = {
 102	&ext4_xattr_user_handler,
 103	&ext4_xattr_trusted_handler,
 104#ifdef CONFIG_EXT4_FS_POSIX_ACL
 105	&posix_acl_access_xattr_handler,
 106	&posix_acl_default_xattr_handler,
 107#endif
 108#ifdef CONFIG_EXT4_FS_SECURITY
 109	&ext4_xattr_security_handler,
 110#endif
 111	&ext4_xattr_hurd_handler,
 112	NULL
 113};
 114
 115#define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
 116				inode->i_sb->s_fs_info)->s_ea_block_cache)
 117
 118#define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
 119				inode->i_sb->s_fs_info)->s_ea_inode_cache)
 120
 121static int
 122ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 123			struct inode *inode);
 124
 125#ifdef CONFIG_LOCKDEP
 126void ext4_xattr_inode_set_class(struct inode *ea_inode)
 127{
 128	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 129}
 130#endif
 131
 132static __le32 ext4_xattr_block_csum(struct inode *inode,
 133				    sector_t block_nr,
 134				    struct ext4_xattr_header *hdr)
 135{
 136	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 137	__u32 csum;
 138	__le64 dsk_block_nr = cpu_to_le64(block_nr);
 139	__u32 dummy_csum = 0;
 140	int offset = offsetof(struct ext4_xattr_header, h_checksum);
 141
 142	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 143			   sizeof(dsk_block_nr));
 144	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 145	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 146	offset += sizeof(dummy_csum);
 147	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 148			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 149
 150	return cpu_to_le32(csum);
 151}
 152
 153static int ext4_xattr_block_csum_verify(struct inode *inode,
 154					struct buffer_head *bh)
 155{
 156	struct ext4_xattr_header *hdr = BHDR(bh);
 157	int ret = 1;
 158
 159	if (ext4_has_metadata_csum(inode->i_sb)) {
 160		lock_buffer(bh);
 161		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 162							bh->b_blocknr, hdr));
 163		unlock_buffer(bh);
 164	}
 165	return ret;
 166}
 167
 168static void ext4_xattr_block_csum_set(struct inode *inode,
 169				      struct buffer_head *bh)
 170{
 171	if (ext4_has_metadata_csum(inode->i_sb))
 172		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 173						bh->b_blocknr, BHDR(bh));
 174}
 175
 176static inline const struct xattr_handler *
 177ext4_xattr_handler(int name_index)
 178{
 179	const struct xattr_handler *handler = NULL;
 180
 181	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 182		handler = ext4_xattr_handler_map[name_index];
 183	return handler;
 184}
 185
 186static int
 187ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 188			 void *value_start)
 189{
 190	struct ext4_xattr_entry *e = entry;
 191
 192	/* Find the end of the names list */
 193	while (!IS_LAST_ENTRY(e)) {
 194		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 195		if ((void *)next >= end)
 196			return -EFSCORRUPTED;
 197		if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 198			return -EFSCORRUPTED;
 199		e = next;
 200	}
 201
 202	/* Check the values */
 203	while (!IS_LAST_ENTRY(entry)) {
 204		u32 size = le32_to_cpu(entry->e_value_size);
 205
 206		if (size > EXT4_XATTR_SIZE_MAX)
 207			return -EFSCORRUPTED;
 208
 209		if (size != 0 && entry->e_value_inum == 0) {
 210			u16 offs = le16_to_cpu(entry->e_value_offs);
 211			void *value;
 212
 213			/*
 214			 * The value cannot overlap the names, and the value
 215			 * with padding cannot extend beyond 'end'.  Check both
 216			 * the padded and unpadded sizes, since the size may
 217			 * overflow to 0 when adding padding.
 218			 */
 219			if (offs > end - value_start)
 220				return -EFSCORRUPTED;
 221			value = value_start + offs;
 222			if (value < (void *)e + sizeof(u32) ||
 223			    size > end - value ||
 224			    EXT4_XATTR_SIZE(size) > end - value)
 225				return -EFSCORRUPTED;
 226		}
 227		entry = EXT4_XATTR_NEXT(entry);
 228	}
 229
 230	return 0;
 231}
 232
 233static inline int
 234__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 235			 const char *function, unsigned int line)
 236{
 237	int error = -EFSCORRUPTED;
 238
 239	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 240	    BHDR(bh)->h_blocks != cpu_to_le32(1))
 241		goto errout;
 242	if (buffer_verified(bh))
 243		return 0;
 244
 245	error = -EFSBADCRC;
 246	if (!ext4_xattr_block_csum_verify(inode, bh))
 247		goto errout;
 248	error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 249					 bh->b_data);
 250errout:
 251	if (error)
 252		__ext4_error_inode(inode, function, line, 0, -error,
 253				   "corrupted xattr block %llu",
 254				   (unsigned long long) bh->b_blocknr);
 255	else
 256		set_buffer_verified(bh);
 257	return error;
 258}
 259
 260#define ext4_xattr_check_block(inode, bh) \
 261	__ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 262
 263
 264static int
 265__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 266			 void *end, const char *function, unsigned int line)
 267{
 268	int error = -EFSCORRUPTED;
 269
 270	if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 271	    (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 272		goto errout;
 273	error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 274errout:
 275	if (error)
 276		__ext4_error_inode(inode, function, line, 0, -error,
 277				   "corrupted in-inode xattr");
 278	return error;
 279}
 280
 281#define xattr_check_inode(inode, header, end) \
 282	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 283
 284static int
 285xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 286		 void *end, int name_index, const char *name, int sorted)
 287{
 288	struct ext4_xattr_entry *entry, *next;
 289	size_t name_len;
 290	int cmp = 1;
 291
 292	if (name == NULL)
 293		return -EINVAL;
 294	name_len = strlen(name);
 295	for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 296		next = EXT4_XATTR_NEXT(entry);
 297		if ((void *) next >= end) {
 298			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 299			return -EFSCORRUPTED;
 300		}
 301		cmp = name_index - entry->e_name_index;
 302		if (!cmp)
 303			cmp = name_len - entry->e_name_len;
 304		if (!cmp)
 305			cmp = memcmp(name, entry->e_name, name_len);
 306		if (cmp <= 0 && (sorted || cmp == 0))
 307			break;
 308	}
 309	*pentry = entry;
 310	return cmp ? -ENODATA : 0;
 311}
 312
 313static u32
 314ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 315{
 316	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 317}
 318
 319static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 320{
 321	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 322		(u32) inode_peek_iversion_raw(ea_inode);
 323}
 324
 325static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 326{
 327	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 328	inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 329}
 330
 331static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 332{
 333	return (u32)ea_inode->i_atime.tv_sec;
 334}
 335
 336static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 337{
 338	ea_inode->i_atime.tv_sec = hash;
 339}
 340
 341/*
 342 * Read the EA value from an inode.
 343 */
 344static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 345{
 346	int blocksize = 1 << ea_inode->i_blkbits;
 347	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 348	int tail_size = (size % blocksize) ?: blocksize;
 349	struct buffer_head *bhs_inline[8];
 350	struct buffer_head **bhs = bhs_inline;
 351	int i, ret;
 352
 353	if (bh_count > ARRAY_SIZE(bhs_inline)) {
 354		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 355		if (!bhs)
 356			return -ENOMEM;
 357	}
 358
 359	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 360			       true /* wait */, bhs);
 361	if (ret)
 362		goto free_bhs;
 363
 364	for (i = 0; i < bh_count; i++) {
 365		/* There shouldn't be any holes in ea_inode. */
 366		if (!bhs[i]) {
 367			ret = -EFSCORRUPTED;
 368			goto put_bhs;
 369		}
 370		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 371		       i < bh_count - 1 ? blocksize : tail_size);
 372	}
 373	ret = 0;
 374put_bhs:
 375	for (i = 0; i < bh_count; i++)
 376		brelse(bhs[i]);
 377free_bhs:
 378	if (bhs != bhs_inline)
 379		kfree(bhs);
 380	return ret;
 381}
 382
 383#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 384
 385static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 386				 u32 ea_inode_hash, struct inode **ea_inode)
 387{
 388	struct inode *inode;
 389	int err;
 390
 391	inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 392	if (IS_ERR(inode)) {
 393		err = PTR_ERR(inode);
 394		ext4_error(parent->i_sb,
 395			   "error while reading EA inode %lu err=%d", ea_ino,
 396			   err);
 397		return err;
 398	}
 399
 400	if (is_bad_inode(inode)) {
 401		ext4_error(parent->i_sb,
 402			   "error while reading EA inode %lu is_bad_inode",
 403			   ea_ino);
 404		err = -EIO;
 405		goto error;
 406	}
 407
 408	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 409		ext4_error(parent->i_sb,
 410			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 411			    ea_ino);
 412		err = -EINVAL;
 413		goto error;
 414	}
 415
 416	ext4_xattr_inode_set_class(inode);
 417
 418	/*
 419	 * Check whether this is an old Lustre-style xattr inode. Lustre
 420	 * implementation does not have hash validation, rather it has a
 421	 * backpointer from ea_inode to the parent inode.
 422	 */
 423	if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 424	    EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 425	    inode->i_generation == parent->i_generation) {
 426		ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 427		ext4_xattr_inode_set_ref(inode, 1);
 428	} else {
 429		inode_lock(inode);
 430		inode->i_flags |= S_NOQUOTA;
 431		inode_unlock(inode);
 432	}
 433
 434	*ea_inode = inode;
 435	return 0;
 436error:
 437	iput(inode);
 438	return err;
 439}
 440
 441/* Remove entry from mbcache when EA inode is getting evicted */
 442void ext4_evict_ea_inode(struct inode *inode)
 443{
 444	struct mb_cache_entry *oe;
 445
 446	if (!EA_INODE_CACHE(inode))
 447		return;
 448	/* Wait for entry to get unused so that we can remove it */
 449	while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
 450			ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
 451		mb_cache_entry_wait_unused(oe);
 452		mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
 453	}
 454}
 455
 456static int
 457ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 458			       struct ext4_xattr_entry *entry, void *buffer,
 459			       size_t size)
 460{
 461	u32 hash;
 462
 463	/* Verify stored hash matches calculated hash. */
 464	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 465	if (hash != ext4_xattr_inode_get_hash(ea_inode))
 466		return -EFSCORRUPTED;
 467
 468	if (entry) {
 469		__le32 e_hash, tmp_data;
 470
 471		/* Verify entry hash. */
 472		tmp_data = cpu_to_le32(hash);
 473		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 474					       &tmp_data, 1);
 475		/* All good? */
 476		if (e_hash == entry->e_hash)
 477			return 0;
 478
 479		/*
 480		 * Not good. Maybe the entry hash was calculated
 481		 * using the buggy signed char version?
 482		 */
 483		e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
 484							&tmp_data, 1);
 485		/* Still no match - bad */
 486		if (e_hash != entry->e_hash)
 487			return -EFSCORRUPTED;
 488
 489		/* Let people know about old hash */
 490		pr_warn_once("ext4: filesystem with signed xattr name hash");
 491	}
 492	return 0;
 493}
 494
 495/*
 496 * Read xattr value from the EA inode.
 497 */
 498static int
 499ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 500		     void *buffer, size_t size)
 501{
 502	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 503	struct inode *ea_inode;
 504	int err;
 505
 506	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 507				    le32_to_cpu(entry->e_hash), &ea_inode);
 508	if (err) {
 509		ea_inode = NULL;
 510		goto out;
 511	}
 512
 513	if (i_size_read(ea_inode) != size) {
 514		ext4_warning_inode(ea_inode,
 515				   "ea_inode file size=%llu entry size=%zu",
 516				   i_size_read(ea_inode), size);
 517		err = -EFSCORRUPTED;
 518		goto out;
 519	}
 520
 521	err = ext4_xattr_inode_read(ea_inode, buffer, size);
 522	if (err)
 523		goto out;
 524
 525	if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 526		err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 527						     size);
 528		if (err) {
 529			ext4_warning_inode(ea_inode,
 530					   "EA inode hash validation failed");
 531			goto out;
 532		}
 533
 534		if (ea_inode_cache)
 535			mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 536					ext4_xattr_inode_get_hash(ea_inode),
 537					ea_inode->i_ino, true /* reusable */);
 538	}
 539out:
 540	iput(ea_inode);
 541	return err;
 542}
 543
 544static int
 545ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 546		     void *buffer, size_t buffer_size)
 547{
 548	struct buffer_head *bh = NULL;
 549	struct ext4_xattr_entry *entry;
 550	size_t size;
 551	void *end;
 552	int error;
 553	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 554
 555	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 556		  name_index, name, buffer, (long)buffer_size);
 557
 558	if (!EXT4_I(inode)->i_file_acl)
 559		return -ENODATA;
 560	ea_idebug(inode, "reading block %llu",
 561		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 562	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 563	if (IS_ERR(bh))
 564		return PTR_ERR(bh);
 565	ea_bdebug(bh, "b_count=%d, refcount=%d",
 566		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 567	error = ext4_xattr_check_block(inode, bh);
 568	if (error)
 569		goto cleanup;
 570	ext4_xattr_block_cache_insert(ea_block_cache, bh);
 571	entry = BFIRST(bh);
 572	end = bh->b_data + bh->b_size;
 573	error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 574	if (error)
 575		goto cleanup;
 576	size = le32_to_cpu(entry->e_value_size);
 577	error = -ERANGE;
 578	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 579		goto cleanup;
 580	if (buffer) {
 581		if (size > buffer_size)
 582			goto cleanup;
 583		if (entry->e_value_inum) {
 584			error = ext4_xattr_inode_get(inode, entry, buffer,
 585						     size);
 586			if (error)
 587				goto cleanup;
 588		} else {
 589			u16 offset = le16_to_cpu(entry->e_value_offs);
 590			void *p = bh->b_data + offset;
 591
 592			if (unlikely(p + size > end))
 593				goto cleanup;
 594			memcpy(buffer, p, size);
 595		}
 596	}
 597	error = size;
 598
 599cleanup:
 600	brelse(bh);
 601	return error;
 602}
 603
 604int
 605ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 606		     void *buffer, size_t buffer_size)
 607{
 608	struct ext4_xattr_ibody_header *header;
 609	struct ext4_xattr_entry *entry;
 610	struct ext4_inode *raw_inode;
 611	struct ext4_iloc iloc;
 612	size_t size;
 613	void *end;
 614	int error;
 615
 616	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 617		return -ENODATA;
 618	error = ext4_get_inode_loc(inode, &iloc);
 619	if (error)
 620		return error;
 621	raw_inode = ext4_raw_inode(&iloc);
 622	header = IHDR(inode, raw_inode);
 623	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 624	error = xattr_check_inode(inode, header, end);
 625	if (error)
 626		goto cleanup;
 627	entry = IFIRST(header);
 628	error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 629	if (error)
 630		goto cleanup;
 631	size = le32_to_cpu(entry->e_value_size);
 632	error = -ERANGE;
 633	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 634		goto cleanup;
 635	if (buffer) {
 636		if (size > buffer_size)
 637			goto cleanup;
 638		if (entry->e_value_inum) {
 639			error = ext4_xattr_inode_get(inode, entry, buffer,
 640						     size);
 641			if (error)
 642				goto cleanup;
 643		} else {
 644			u16 offset = le16_to_cpu(entry->e_value_offs);
 645			void *p = (void *)IFIRST(header) + offset;
 646
 647			if (unlikely(p + size > end))
 648				goto cleanup;
 649			memcpy(buffer, p, size);
 650		}
 651	}
 652	error = size;
 653
 654cleanup:
 655	brelse(iloc.bh);
 656	return error;
 657}
 658
 659/*
 660 * ext4_xattr_get()
 661 *
 662 * Copy an extended attribute into the buffer
 663 * provided, or compute the buffer size required.
 664 * Buffer is NULL to compute the size of the buffer required.
 665 *
 666 * Returns a negative error number on failure, or the number of bytes
 667 * used / required on success.
 668 */
 669int
 670ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 671	       void *buffer, size_t buffer_size)
 672{
 673	int error;
 674
 675	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 676		return -EIO;
 677
 678	if (strlen(name) > 255)
 679		return -ERANGE;
 680
 681	down_read(&EXT4_I(inode)->xattr_sem);
 682	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 683				     buffer_size);
 684	if (error == -ENODATA)
 685		error = ext4_xattr_block_get(inode, name_index, name, buffer,
 686					     buffer_size);
 687	up_read(&EXT4_I(inode)->xattr_sem);
 688	return error;
 689}
 690
 691static int
 692ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 693			char *buffer, size_t buffer_size)
 694{
 695	size_t rest = buffer_size;
 696
 697	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 698		const struct xattr_handler *handler =
 699			ext4_xattr_handler(entry->e_name_index);
 700
 701		if (handler && (!handler->list || handler->list(dentry))) {
 702			const char *prefix = handler->prefix ?: handler->name;
 703			size_t prefix_len = strlen(prefix);
 704			size_t size = prefix_len + entry->e_name_len + 1;
 705
 706			if (buffer) {
 707				if (size > rest)
 708					return -ERANGE;
 709				memcpy(buffer, prefix, prefix_len);
 710				buffer += prefix_len;
 711				memcpy(buffer, entry->e_name, entry->e_name_len);
 712				buffer += entry->e_name_len;
 713				*buffer++ = 0;
 714			}
 715			rest -= size;
 716		}
 717	}
 718	return buffer_size - rest;  /* total size */
 719}
 720
 721static int
 722ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 723{
 724	struct inode *inode = d_inode(dentry);
 725	struct buffer_head *bh = NULL;
 726	int error;
 727
 728	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 729		  buffer, (long)buffer_size);
 730
 731	if (!EXT4_I(inode)->i_file_acl)
 732		return 0;
 733	ea_idebug(inode, "reading block %llu",
 734		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 735	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 736	if (IS_ERR(bh))
 737		return PTR_ERR(bh);
 738	ea_bdebug(bh, "b_count=%d, refcount=%d",
 739		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 740	error = ext4_xattr_check_block(inode, bh);
 741	if (error)
 742		goto cleanup;
 743	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 744	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 745					buffer_size);
 746cleanup:
 747	brelse(bh);
 748	return error;
 749}
 750
 751static int
 752ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 753{
 754	struct inode *inode = d_inode(dentry);
 755	struct ext4_xattr_ibody_header *header;
 756	struct ext4_inode *raw_inode;
 757	struct ext4_iloc iloc;
 758	void *end;
 759	int error;
 760
 761	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 762		return 0;
 763	error = ext4_get_inode_loc(inode, &iloc);
 764	if (error)
 765		return error;
 766	raw_inode = ext4_raw_inode(&iloc);
 767	header = IHDR(inode, raw_inode);
 768	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 769	error = xattr_check_inode(inode, header, end);
 770	if (error)
 771		goto cleanup;
 772	error = ext4_xattr_list_entries(dentry, IFIRST(header),
 773					buffer, buffer_size);
 774
 775cleanup:
 776	brelse(iloc.bh);
 777	return error;
 778}
 779
 780/*
 781 * Inode operation listxattr()
 782 *
 783 * d_inode(dentry)->i_rwsem: don't care
 784 *
 785 * Copy a list of attribute names into the buffer
 786 * provided, or compute the buffer size required.
 787 * Buffer is NULL to compute the size of the buffer required.
 788 *
 789 * Returns a negative error number on failure, or the number of bytes
 790 * used / required on success.
 791 */
 792ssize_t
 793ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 794{
 795	int ret, ret2;
 796
 797	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 798	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 799	if (ret < 0)
 800		goto errout;
 801	if (buffer) {
 802		buffer += ret;
 803		buffer_size -= ret;
 804	}
 805	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 806	if (ret < 0)
 807		goto errout;
 808	ret += ret2;
 809errout:
 810	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 811	return ret;
 812}
 813
 814/*
 815 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 816 * not set, set it.
 817 */
 818static void ext4_xattr_update_super_block(handle_t *handle,
 819					  struct super_block *sb)
 820{
 821	if (ext4_has_feature_xattr(sb))
 822		return;
 823
 824	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 825	if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 826					  EXT4_JTR_NONE) == 0) {
 827		lock_buffer(EXT4_SB(sb)->s_sbh);
 828		ext4_set_feature_xattr(sb);
 829		ext4_superblock_csum_set(sb);
 830		unlock_buffer(EXT4_SB(sb)->s_sbh);
 831		ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 832	}
 833}
 834
 835int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 836{
 837	struct ext4_iloc iloc = { .bh = NULL };
 838	struct buffer_head *bh = NULL;
 839	struct ext4_inode *raw_inode;
 840	struct ext4_xattr_ibody_header *header;
 841	struct ext4_xattr_entry *entry;
 842	qsize_t ea_inode_refs = 0;
 843	void *end;
 844	int ret;
 845
 846	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 847
 848	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 849		ret = ext4_get_inode_loc(inode, &iloc);
 850		if (ret)
 851			goto out;
 852		raw_inode = ext4_raw_inode(&iloc);
 853		header = IHDR(inode, raw_inode);
 854		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 855		ret = xattr_check_inode(inode, header, end);
 856		if (ret)
 857			goto out;
 858
 859		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 860		     entry = EXT4_XATTR_NEXT(entry))
 861			if (entry->e_value_inum)
 862				ea_inode_refs++;
 863	}
 864
 865	if (EXT4_I(inode)->i_file_acl) {
 866		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 867		if (IS_ERR(bh)) {
 868			ret = PTR_ERR(bh);
 869			bh = NULL;
 870			goto out;
 871		}
 872
 873		ret = ext4_xattr_check_block(inode, bh);
 874		if (ret)
 875			goto out;
 876
 877		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 878		     entry = EXT4_XATTR_NEXT(entry))
 879			if (entry->e_value_inum)
 880				ea_inode_refs++;
 881	}
 882	*usage = ea_inode_refs + 1;
 883	ret = 0;
 884out:
 885	brelse(iloc.bh);
 886	brelse(bh);
 887	return ret;
 888}
 889
 890static inline size_t round_up_cluster(struct inode *inode, size_t length)
 891{
 892	struct super_block *sb = inode->i_sb;
 893	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 894				    inode->i_blkbits);
 895	size_t mask = ~(cluster_size - 1);
 896
 897	return (length + cluster_size - 1) & mask;
 898}
 899
 900static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 901{
 902	int err;
 903
 904	err = dquot_alloc_inode(inode);
 905	if (err)
 906		return err;
 907	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 908	if (err)
 909		dquot_free_inode(inode);
 910	return err;
 911}
 912
 913static void ext4_xattr_inode_free_quota(struct inode *parent,
 914					struct inode *ea_inode,
 915					size_t len)
 916{
 917	if (ea_inode &&
 918	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 919		return;
 920	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 921	dquot_free_inode(parent);
 922}
 923
 924int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 925			     struct buffer_head *block_bh, size_t value_len,
 926			     bool is_create)
 927{
 928	int credits;
 929	int blocks;
 930
 931	/*
 932	 * 1) Owner inode update
 933	 * 2) Ref count update on old xattr block
 934	 * 3) new xattr block
 935	 * 4) block bitmap update for new xattr block
 936	 * 5) group descriptor for new xattr block
 937	 * 6) block bitmap update for old xattr block
 938	 * 7) group descriptor for old block
 939	 *
 940	 * 6 & 7 can happen if we have two racing threads T_a and T_b
 941	 * which are each trying to set an xattr on inodes I_a and I_b
 942	 * which were both initially sharing an xattr block.
 943	 */
 944	credits = 7;
 945
 946	/* Quota updates. */
 947	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 948
 949	/*
 950	 * In case of inline data, we may push out the data to a block,
 951	 * so we need to reserve credits for this eventuality
 952	 */
 953	if (inode && ext4_has_inline_data(inode))
 954		credits += ext4_writepage_trans_blocks(inode) + 1;
 955
 956	/* We are done if ea_inode feature is not enabled. */
 957	if (!ext4_has_feature_ea_inode(sb))
 958		return credits;
 959
 960	/* New ea_inode, inode map, block bitmap, group descriptor. */
 961	credits += 4;
 962
 963	/* Data blocks. */
 964	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 965
 966	/* Indirection block or one level of extent tree. */
 967	blocks += 1;
 968
 969	/* Block bitmap and group descriptor updates for each block. */
 970	credits += blocks * 2;
 971
 972	/* Blocks themselves. */
 973	credits += blocks;
 974
 975	if (!is_create) {
 976		/* Dereference ea_inode holding old xattr value.
 977		 * Old ea_inode, inode map, block bitmap, group descriptor.
 978		 */
 979		credits += 4;
 980
 981		/* Data blocks for old ea_inode. */
 982		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 983
 984		/* Indirection block or one level of extent tree for old
 985		 * ea_inode.
 986		 */
 987		blocks += 1;
 988
 989		/* Block bitmap and group descriptor updates for each block. */
 990		credits += blocks * 2;
 991	}
 992
 993	/* We may need to clone the existing xattr block in which case we need
 994	 * to increment ref counts for existing ea_inodes referenced by it.
 995	 */
 996	if (block_bh) {
 997		struct ext4_xattr_entry *entry = BFIRST(block_bh);
 998
 999		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1000			if (entry->e_value_inum)
1001				/* Ref count update on ea_inode. */
1002				credits += 1;
1003	}
1004	return credits;
1005}
1006
1007static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1008				       int ref_change)
1009{
 
1010	struct ext4_iloc iloc;
1011	s64 ref_count;
 
1012	int ret;
1013
1014	inode_lock(ea_inode);
1015
1016	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1017	if (ret)
1018		goto out;
1019
1020	ref_count = ext4_xattr_inode_get_ref(ea_inode);
1021	ref_count += ref_change;
1022	ext4_xattr_inode_set_ref(ea_inode, ref_count);
1023
1024	if (ref_change > 0) {
1025		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1026			  ea_inode->i_ino, ref_count);
1027
1028		if (ref_count == 1) {
1029			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1030				  ea_inode->i_ino, ea_inode->i_nlink);
1031
1032			set_nlink(ea_inode, 1);
1033			ext4_orphan_del(handle, ea_inode);
 
 
 
 
 
 
 
 
1034		}
1035	} else {
1036		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1037			  ea_inode->i_ino, ref_count);
1038
1039		if (ref_count == 0) {
1040			WARN_ONCE(ea_inode->i_nlink != 1,
1041				  "EA inode %lu i_nlink=%u",
1042				  ea_inode->i_ino, ea_inode->i_nlink);
1043
1044			clear_nlink(ea_inode);
1045			ext4_orphan_add(handle, ea_inode);
 
 
 
 
 
 
1046		}
1047	}
1048
1049	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1050	if (ret)
1051		ext4_warning_inode(ea_inode,
1052				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1053out:
1054	inode_unlock(ea_inode);
1055	return ret;
1056}
1057
1058static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1059{
1060	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1061}
1062
1063static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1064{
1065	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1066}
1067
1068static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1069					struct ext4_xattr_entry *first)
1070{
1071	struct inode *ea_inode;
1072	struct ext4_xattr_entry *entry;
1073	struct ext4_xattr_entry *failed_entry;
1074	unsigned int ea_ino;
1075	int err, saved_err;
1076
1077	for (entry = first; !IS_LAST_ENTRY(entry);
1078	     entry = EXT4_XATTR_NEXT(entry)) {
1079		if (!entry->e_value_inum)
1080			continue;
1081		ea_ino = le32_to_cpu(entry->e_value_inum);
1082		err = ext4_xattr_inode_iget(parent, ea_ino,
1083					    le32_to_cpu(entry->e_hash),
1084					    &ea_inode);
1085		if (err)
1086			goto cleanup;
1087		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1088		if (err) {
1089			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1090			iput(ea_inode);
1091			goto cleanup;
1092		}
1093		iput(ea_inode);
1094	}
1095	return 0;
1096
1097cleanup:
1098	saved_err = err;
1099	failed_entry = entry;
1100
1101	for (entry = first; entry != failed_entry;
1102	     entry = EXT4_XATTR_NEXT(entry)) {
1103		if (!entry->e_value_inum)
1104			continue;
1105		ea_ino = le32_to_cpu(entry->e_value_inum);
1106		err = ext4_xattr_inode_iget(parent, ea_ino,
1107					    le32_to_cpu(entry->e_hash),
1108					    &ea_inode);
1109		if (err) {
1110			ext4_warning(parent->i_sb,
1111				     "cleanup ea_ino %u iget error %d", ea_ino,
1112				     err);
1113			continue;
1114		}
1115		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1116		if (err)
1117			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1118					   err);
1119		iput(ea_inode);
1120	}
1121	return saved_err;
1122}
1123
1124static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1125			struct buffer_head *bh, bool block_csum, bool dirty)
1126{
1127	int error;
1128
1129	if (bh && dirty) {
1130		if (block_csum)
1131			ext4_xattr_block_csum_set(inode, bh);
1132		error = ext4_handle_dirty_metadata(handle, NULL, bh);
1133		if (error) {
1134			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1135				     error);
1136			return error;
1137		}
1138	}
1139	return 0;
1140}
1141
1142static void
1143ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1144			     struct buffer_head *bh,
1145			     struct ext4_xattr_entry *first, bool block_csum,
1146			     struct ext4_xattr_inode_array **ea_inode_array,
1147			     int extra_credits, bool skip_quota)
1148{
1149	struct inode *ea_inode;
1150	struct ext4_xattr_entry *entry;
1151	bool dirty = false;
1152	unsigned int ea_ino;
1153	int err;
1154	int credits;
1155
1156	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1157	credits = 2 + extra_credits;
1158
1159	for (entry = first; !IS_LAST_ENTRY(entry);
1160	     entry = EXT4_XATTR_NEXT(entry)) {
1161		if (!entry->e_value_inum)
1162			continue;
1163		ea_ino = le32_to_cpu(entry->e_value_inum);
1164		err = ext4_xattr_inode_iget(parent, ea_ino,
1165					    le32_to_cpu(entry->e_hash),
1166					    &ea_inode);
1167		if (err)
1168			continue;
1169
1170		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1171		if (err) {
1172			ext4_warning_inode(ea_inode,
1173					   "Expand inode array err=%d", err);
1174			iput(ea_inode);
1175			continue;
1176		}
1177
1178		err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1179			ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1180			ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1181					      dirty));
1182		if (err < 0) {
1183			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1184					   err);
1185			continue;
1186		}
1187		if (err > 0) {
1188			err = ext4_journal_get_write_access(handle,
1189					parent->i_sb, bh, EXT4_JTR_NONE);
1190			if (err) {
1191				ext4_warning_inode(ea_inode,
1192						"Re-get write access err=%d",
1193						err);
1194				continue;
1195			}
1196		}
1197
1198		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1199		if (err) {
1200			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1201					   err);
1202			continue;
1203		}
1204
1205		if (!skip_quota)
1206			ext4_xattr_inode_free_quota(parent, ea_inode,
1207					      le32_to_cpu(entry->e_value_size));
1208
1209		/*
1210		 * Forget about ea_inode within the same transaction that
1211		 * decrements the ref count. This avoids duplicate decrements in
1212		 * case the rest of the work spills over to subsequent
1213		 * transactions.
1214		 */
1215		entry->e_value_inum = 0;
1216		entry->e_value_size = 0;
1217
1218		dirty = true;
1219	}
1220
1221	if (dirty) {
1222		/*
1223		 * Note that we are deliberately skipping csum calculation for
1224		 * the final update because we do not expect any journal
1225		 * restarts until xattr block is freed.
1226		 */
1227
1228		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1229		if (err)
1230			ext4_warning_inode(parent,
1231					   "handle dirty metadata err=%d", err);
1232	}
1233}
1234
1235/*
1236 * Release the xattr block BH: If the reference count is > 1, decrement it;
1237 * otherwise free the block.
1238 */
1239static void
1240ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1241			 struct buffer_head *bh,
1242			 struct ext4_xattr_inode_array **ea_inode_array,
1243			 int extra_credits)
1244{
1245	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1246	u32 hash, ref;
1247	int error = 0;
1248
1249	BUFFER_TRACE(bh, "get_write_access");
1250	error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1251					      EXT4_JTR_NONE);
1252	if (error)
1253		goto out;
1254
1255retry_ref:
1256	lock_buffer(bh);
1257	hash = le32_to_cpu(BHDR(bh)->h_hash);
1258	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1259	if (ref == 1) {
1260		ea_bdebug(bh, "refcount now=0; freeing");
1261		/*
1262		 * This must happen under buffer lock for
1263		 * ext4_xattr_block_set() to reliably detect freed block
1264		 */
1265		if (ea_block_cache) {
1266			struct mb_cache_entry *oe;
1267
1268			oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1269							  bh->b_blocknr);
1270			if (oe) {
1271				unlock_buffer(bh);
1272				mb_cache_entry_wait_unused(oe);
1273				mb_cache_entry_put(ea_block_cache, oe);
1274				goto retry_ref;
1275			}
1276		}
1277		get_bh(bh);
1278		unlock_buffer(bh);
1279
1280		if (ext4_has_feature_ea_inode(inode->i_sb))
1281			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1282						     BFIRST(bh),
1283						     true /* block_csum */,
1284						     ea_inode_array,
1285						     extra_credits,
1286						     true /* skip_quota */);
1287		ext4_free_blocks(handle, inode, bh, 0, 1,
1288				 EXT4_FREE_BLOCKS_METADATA |
1289				 EXT4_FREE_BLOCKS_FORGET);
1290	} else {
1291		ref--;
1292		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1293		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1294			struct mb_cache_entry *ce;
1295
1296			if (ea_block_cache) {
1297				ce = mb_cache_entry_get(ea_block_cache, hash,
1298							bh->b_blocknr);
1299				if (ce) {
1300					set_bit(MBE_REUSABLE_B, &ce->e_flags);
1301					mb_cache_entry_put(ea_block_cache, ce);
1302				}
1303			}
1304		}
1305
1306		ext4_xattr_block_csum_set(inode, bh);
1307		/*
1308		 * Beware of this ugliness: Releasing of xattr block references
1309		 * from different inodes can race and so we have to protect
1310		 * from a race where someone else frees the block (and releases
1311		 * its journal_head) before we are done dirtying the buffer. In
1312		 * nojournal mode this race is harmless and we actually cannot
1313		 * call ext4_handle_dirty_metadata() with locked buffer as
1314		 * that function can call sync_dirty_buffer() so for that case
1315		 * we handle the dirtying after unlocking the buffer.
1316		 */
1317		if (ext4_handle_valid(handle))
1318			error = ext4_handle_dirty_metadata(handle, inode, bh);
1319		unlock_buffer(bh);
1320		if (!ext4_handle_valid(handle))
1321			error = ext4_handle_dirty_metadata(handle, inode, bh);
1322		if (IS_SYNC(inode))
1323			ext4_handle_sync(handle);
1324		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1325		ea_bdebug(bh, "refcount now=%d; releasing",
1326			  le32_to_cpu(BHDR(bh)->h_refcount));
1327	}
1328out:
1329	ext4_std_error(inode->i_sb, error);
1330	return;
1331}
1332
1333/*
1334 * Find the available free space for EAs. This also returns the total number of
1335 * bytes used by EA entries.
1336 */
1337static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1338				    size_t *min_offs, void *base, int *total)
1339{
1340	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1341		if (!last->e_value_inum && last->e_value_size) {
1342			size_t offs = le16_to_cpu(last->e_value_offs);
1343			if (offs < *min_offs)
1344				*min_offs = offs;
1345		}
1346		if (total)
1347			*total += EXT4_XATTR_LEN(last->e_name_len);
1348	}
1349	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1350}
1351
1352/*
1353 * Write the value of the EA in an inode.
1354 */
1355static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1356				  const void *buf, int bufsize)
1357{
1358	struct buffer_head *bh = NULL;
1359	unsigned long block = 0;
1360	int blocksize = ea_inode->i_sb->s_blocksize;
1361	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1362	int csize, wsize = 0;
1363	int ret = 0, ret2 = 0;
1364	int retries = 0;
1365
1366retry:
1367	while (ret >= 0 && ret < max_blocks) {
1368		struct ext4_map_blocks map;
1369		map.m_lblk = block += ret;
1370		map.m_len = max_blocks -= ret;
1371
1372		ret = ext4_map_blocks(handle, ea_inode, &map,
1373				      EXT4_GET_BLOCKS_CREATE);
1374		if (ret <= 0) {
1375			ext4_mark_inode_dirty(handle, ea_inode);
1376			if (ret == -ENOSPC &&
1377			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1378				ret = 0;
1379				goto retry;
1380			}
1381			break;
1382		}
1383	}
1384
1385	if (ret < 0)
1386		return ret;
1387
1388	block = 0;
1389	while (wsize < bufsize) {
1390		brelse(bh);
1391		csize = (bufsize - wsize) > blocksize ? blocksize :
1392								bufsize - wsize;
1393		bh = ext4_getblk(handle, ea_inode, block, 0);
1394		if (IS_ERR(bh))
1395			return PTR_ERR(bh);
1396		if (!bh) {
1397			WARN_ON_ONCE(1);
1398			EXT4_ERROR_INODE(ea_inode,
1399					 "ext4_getblk() return bh = NULL");
1400			return -EFSCORRUPTED;
1401		}
1402		ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1403						   EXT4_JTR_NONE);
1404		if (ret)
1405			goto out;
1406
1407		memcpy(bh->b_data, buf, csize);
1408		set_buffer_uptodate(bh);
1409		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1410
1411		buf += csize;
1412		wsize += csize;
1413		block += 1;
1414	}
1415
1416	inode_lock(ea_inode);
1417	i_size_write(ea_inode, wsize);
1418	ext4_update_i_disksize(ea_inode, wsize);
1419	inode_unlock(ea_inode);
1420
1421	ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1422	if (unlikely(ret2 && !ret))
1423		ret = ret2;
1424
1425out:
1426	brelse(bh);
1427
1428	return ret;
1429}
1430
1431/*
1432 * Create an inode to store the value of a large EA.
1433 */
1434static struct inode *ext4_xattr_inode_create(handle_t *handle,
1435					     struct inode *inode, u32 hash)
1436{
1437	struct inode *ea_inode = NULL;
1438	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1439	int err;
1440
1441	/*
1442	 * Let the next inode be the goal, so we try and allocate the EA inode
1443	 * in the same group, or nearby one.
1444	 */
1445	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1446				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1447				  EXT4_EA_INODE_FL);
1448	if (!IS_ERR(ea_inode)) {
1449		ea_inode->i_op = &ext4_file_inode_operations;
1450		ea_inode->i_fop = &ext4_file_operations;
1451		ext4_set_aops(ea_inode);
1452		ext4_xattr_inode_set_class(ea_inode);
1453		unlock_new_inode(ea_inode);
1454		ext4_xattr_inode_set_ref(ea_inode, 1);
1455		ext4_xattr_inode_set_hash(ea_inode, hash);
1456		err = ext4_mark_inode_dirty(handle, ea_inode);
1457		if (!err)
1458			err = ext4_inode_attach_jinode(ea_inode);
1459		if (err) {
1460			if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1461				ext4_warning_inode(ea_inode,
1462					"cleanup dec ref error %d", err);
1463			iput(ea_inode);
1464			return ERR_PTR(err);
1465		}
1466
1467		/*
1468		 * Xattr inodes are shared therefore quota charging is performed
1469		 * at a higher level.
1470		 */
1471		dquot_free_inode(ea_inode);
1472		dquot_drop(ea_inode);
1473		inode_lock(ea_inode);
1474		ea_inode->i_flags |= S_NOQUOTA;
1475		inode_unlock(ea_inode);
1476	}
1477
1478	return ea_inode;
1479}
1480
1481static struct inode *
1482ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1483			    size_t value_len, u32 hash)
1484{
1485	struct inode *ea_inode;
1486	struct mb_cache_entry *ce;
1487	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1488	void *ea_data;
1489
1490	if (!ea_inode_cache)
1491		return NULL;
1492
1493	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1494	if (!ce)
1495		return NULL;
1496
1497	WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1498		     !(current->flags & PF_MEMALLOC_NOFS));
1499
1500	ea_data = kvmalloc(value_len, GFP_KERNEL);
1501	if (!ea_data) {
1502		mb_cache_entry_put(ea_inode_cache, ce);
1503		return NULL;
1504	}
1505
1506	while (ce) {
1507		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1508				     EXT4_IGET_NORMAL);
1509		if (!IS_ERR(ea_inode) &&
1510		    !is_bad_inode(ea_inode) &&
1511		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1512		    i_size_read(ea_inode) == value_len &&
1513		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1514		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1515						    value_len) &&
1516		    !memcmp(value, ea_data, value_len)) {
1517			mb_cache_entry_touch(ea_inode_cache, ce);
1518			mb_cache_entry_put(ea_inode_cache, ce);
1519			kvfree(ea_data);
1520			return ea_inode;
1521		}
1522
1523		if (!IS_ERR(ea_inode))
1524			iput(ea_inode);
1525		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1526	}
1527	kvfree(ea_data);
1528	return NULL;
1529}
1530
1531/*
1532 * Add value of the EA in an inode.
1533 */
1534static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1535					  const void *value, size_t value_len,
1536					  struct inode **ret_inode)
1537{
1538	struct inode *ea_inode;
1539	u32 hash;
1540	int err;
1541
1542	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1543	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1544	if (ea_inode) {
1545		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1546		if (err) {
1547			iput(ea_inode);
1548			return err;
1549		}
1550
1551		*ret_inode = ea_inode;
1552		return 0;
1553	}
1554
1555	/* Create an inode for the EA value */
1556	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1557	if (IS_ERR(ea_inode))
1558		return PTR_ERR(ea_inode);
1559
1560	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1561	if (err) {
1562		if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1563			ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1564		iput(ea_inode);
1565		return err;
1566	}
1567
1568	if (EA_INODE_CACHE(inode))
1569		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1570				      ea_inode->i_ino, true /* reusable */);
1571
1572	*ret_inode = ea_inode;
1573	return 0;
1574}
1575
1576/*
1577 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1578 * feature is enabled.
1579 */
1580#define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1581
1582static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1583				struct ext4_xattr_search *s,
1584				handle_t *handle, struct inode *inode,
1585				bool is_block)
1586{
1587	struct ext4_xattr_entry *last, *next;
1588	struct ext4_xattr_entry *here = s->here;
1589	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1590	int in_inode = i->in_inode;
1591	struct inode *old_ea_inode = NULL;
1592	struct inode *new_ea_inode = NULL;
1593	size_t old_size, new_size;
1594	int ret;
1595
1596	/* Space used by old and new values. */
1597	old_size = (!s->not_found && !here->e_value_inum) ?
1598			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1599	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1600
1601	/*
1602	 * Optimization for the simple case when old and new values have the
1603	 * same padded sizes. Not applicable if external inodes are involved.
1604	 */
1605	if (new_size && new_size == old_size) {
1606		size_t offs = le16_to_cpu(here->e_value_offs);
1607		void *val = s->base + offs;
1608
1609		here->e_value_size = cpu_to_le32(i->value_len);
1610		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1611			memset(val, 0, new_size);
1612		} else {
1613			memcpy(val, i->value, i->value_len);
1614			/* Clear padding bytes. */
1615			memset(val + i->value_len, 0, new_size - i->value_len);
1616		}
1617		goto update_hash;
1618	}
1619
1620	/* Compute min_offs and last. */
1621	last = s->first;
1622	for (; !IS_LAST_ENTRY(last); last = next) {
1623		next = EXT4_XATTR_NEXT(last);
1624		if ((void *)next >= s->end) {
1625			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1626			ret = -EFSCORRUPTED;
1627			goto out;
1628		}
1629		if (!last->e_value_inum && last->e_value_size) {
1630			size_t offs = le16_to_cpu(last->e_value_offs);
1631			if (offs < min_offs)
1632				min_offs = offs;
1633		}
1634	}
1635
1636	/* Check whether we have enough space. */
1637	if (i->value) {
1638		size_t free;
1639
1640		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1641		if (!s->not_found)
1642			free += EXT4_XATTR_LEN(name_len) + old_size;
1643
1644		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1645			ret = -ENOSPC;
1646			goto out;
1647		}
1648
1649		/*
1650		 * If storing the value in an external inode is an option,
1651		 * reserve space for xattr entries/names in the external
1652		 * attribute block so that a long value does not occupy the
1653		 * whole space and prevent further entries being added.
1654		 */
1655		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1656		    new_size && is_block &&
1657		    (min_offs + old_size - new_size) <
1658					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1659			ret = -ENOSPC;
1660			goto out;
1661		}
1662	}
1663
1664	/*
1665	 * Getting access to old and new ea inodes is subject to failures.
1666	 * Finish that work before doing any modifications to the xattr data.
1667	 */
1668	if (!s->not_found && here->e_value_inum) {
1669		ret = ext4_xattr_inode_iget(inode,
1670					    le32_to_cpu(here->e_value_inum),
1671					    le32_to_cpu(here->e_hash),
1672					    &old_ea_inode);
1673		if (ret) {
1674			old_ea_inode = NULL;
1675			goto out;
1676		}
1677	}
1678	if (i->value && in_inode) {
1679		WARN_ON_ONCE(!i->value_len);
1680
1681		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1682		if (ret)
1683			goto out;
1684
1685		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1686						     i->value_len,
1687						     &new_ea_inode);
1688		if (ret) {
1689			new_ea_inode = NULL;
1690			ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1691			goto out;
1692		}
1693	}
1694
1695	if (old_ea_inode) {
1696		/* We are ready to release ref count on the old_ea_inode. */
1697		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1698		if (ret) {
1699			/* Release newly required ref count on new_ea_inode. */
1700			if (new_ea_inode) {
1701				int err;
1702
1703				err = ext4_xattr_inode_dec_ref(handle,
1704							       new_ea_inode);
1705				if (err)
1706					ext4_warning_inode(new_ea_inode,
1707						  "dec ref new_ea_inode err=%d",
1708						  err);
1709				ext4_xattr_inode_free_quota(inode, new_ea_inode,
1710							    i->value_len);
1711			}
1712			goto out;
1713		}
1714
1715		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1716					    le32_to_cpu(here->e_value_size));
1717	}
1718
1719	/* No failures allowed past this point. */
1720
1721	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1722		/* Remove the old value. */
1723		void *first_val = s->base + min_offs;
1724		size_t offs = le16_to_cpu(here->e_value_offs);
1725		void *val = s->base + offs;
1726
1727		memmove(first_val + old_size, first_val, val - first_val);
1728		memset(first_val, 0, old_size);
1729		min_offs += old_size;
1730
1731		/* Adjust all value offsets. */
1732		last = s->first;
1733		while (!IS_LAST_ENTRY(last)) {
1734			size_t o = le16_to_cpu(last->e_value_offs);
1735
1736			if (!last->e_value_inum &&
1737			    last->e_value_size && o < offs)
1738				last->e_value_offs = cpu_to_le16(o + old_size);
1739			last = EXT4_XATTR_NEXT(last);
1740		}
1741	}
1742
1743	if (!i->value) {
1744		/* Remove old name. */
1745		size_t size = EXT4_XATTR_LEN(name_len);
1746
1747		last = ENTRY((void *)last - size);
1748		memmove(here, (void *)here + size,
1749			(void *)last - (void *)here + sizeof(__u32));
1750		memset(last, 0, size);
1751	} else if (s->not_found) {
1752		/* Insert new name. */
1753		size_t size = EXT4_XATTR_LEN(name_len);
1754		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1755
1756		memmove((void *)here + size, here, rest);
1757		memset(here, 0, size);
1758		here->e_name_index = i->name_index;
1759		here->e_name_len = name_len;
1760		memcpy(here->e_name, i->name, name_len);
1761	} else {
1762		/* This is an update, reset value info. */
1763		here->e_value_inum = 0;
1764		here->e_value_offs = 0;
1765		here->e_value_size = 0;
1766	}
1767
1768	if (i->value) {
1769		/* Insert new value. */
1770		if (in_inode) {
1771			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1772		} else if (i->value_len) {
1773			void *val = s->base + min_offs - new_size;
1774
1775			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1776			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1777				memset(val, 0, new_size);
1778			} else {
1779				memcpy(val, i->value, i->value_len);
1780				/* Clear padding bytes. */
1781				memset(val + i->value_len, 0,
1782				       new_size - i->value_len);
1783			}
1784		}
1785		here->e_value_size = cpu_to_le32(i->value_len);
1786	}
1787
1788update_hash:
1789	if (i->value) {
1790		__le32 hash = 0;
1791
1792		/* Entry hash calculation. */
1793		if (in_inode) {
1794			__le32 crc32c_hash;
1795
1796			/*
1797			 * Feed crc32c hash instead of the raw value for entry
1798			 * hash calculation. This is to avoid walking
1799			 * potentially long value buffer again.
1800			 */
1801			crc32c_hash = cpu_to_le32(
1802				       ext4_xattr_inode_get_hash(new_ea_inode));
1803			hash = ext4_xattr_hash_entry(here->e_name,
1804						     here->e_name_len,
1805						     &crc32c_hash, 1);
1806		} else if (is_block) {
1807			__le32 *value = s->base + le16_to_cpu(
1808							here->e_value_offs);
1809
1810			hash = ext4_xattr_hash_entry(here->e_name,
1811						     here->e_name_len, value,
1812						     new_size >> 2);
1813		}
1814		here->e_hash = hash;
1815	}
1816
1817	if (is_block)
1818		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1819
1820	ret = 0;
1821out:
1822	iput(old_ea_inode);
1823	iput(new_ea_inode);
1824	return ret;
1825}
1826
1827struct ext4_xattr_block_find {
1828	struct ext4_xattr_search s;
1829	struct buffer_head *bh;
1830};
1831
1832static int
1833ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1834		      struct ext4_xattr_block_find *bs)
1835{
1836	struct super_block *sb = inode->i_sb;
1837	int error;
1838
1839	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1840		  i->name_index, i->name, i->value, (long)i->value_len);
1841
1842	if (EXT4_I(inode)->i_file_acl) {
1843		/* The inode already has an extended attribute block. */
1844		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1845		if (IS_ERR(bs->bh)) {
1846			error = PTR_ERR(bs->bh);
1847			bs->bh = NULL;
1848			return error;
1849		}
1850		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1851			atomic_read(&(bs->bh->b_count)),
1852			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1853		error = ext4_xattr_check_block(inode, bs->bh);
1854		if (error)
1855			return error;
1856		/* Find the named attribute. */
1857		bs->s.base = BHDR(bs->bh);
1858		bs->s.first = BFIRST(bs->bh);
1859		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1860		bs->s.here = bs->s.first;
1861		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1862					 i->name_index, i->name, 1);
1863		if (error && error != -ENODATA)
1864			return error;
1865		bs->s.not_found = error;
1866	}
1867	return 0;
1868}
1869
1870static int
1871ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1872		     struct ext4_xattr_info *i,
1873		     struct ext4_xattr_block_find *bs)
1874{
1875	struct super_block *sb = inode->i_sb;
1876	struct buffer_head *new_bh = NULL;
1877	struct ext4_xattr_search s_copy = bs->s;
1878	struct ext4_xattr_search *s = &s_copy;
1879	struct mb_cache_entry *ce = NULL;
1880	int error = 0;
1881	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1882	struct inode *ea_inode = NULL, *tmp_inode;
1883	size_t old_ea_inode_quota = 0;
1884	unsigned int ea_ino;
1885
1886
1887#define header(x) ((struct ext4_xattr_header *)(x))
1888
1889	if (s->base) {
1890		int offset = (char *)s->here - bs->bh->b_data;
1891
1892		BUFFER_TRACE(bs->bh, "get_write_access");
1893		error = ext4_journal_get_write_access(handle, sb, bs->bh,
1894						      EXT4_JTR_NONE);
1895		if (error)
1896			goto cleanup;
1897		lock_buffer(bs->bh);
1898
1899		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1900			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1901
1902			/*
1903			 * This must happen under buffer lock for
1904			 * ext4_xattr_block_set() to reliably detect modified
1905			 * block
1906			 */
1907			if (ea_block_cache) {
1908				struct mb_cache_entry *oe;
1909
1910				oe = mb_cache_entry_delete_or_get(ea_block_cache,
1911					hash, bs->bh->b_blocknr);
1912				if (oe) {
1913					/*
1914					 * Xattr block is getting reused. Leave
1915					 * it alone.
1916					 */
1917					mb_cache_entry_put(ea_block_cache, oe);
1918					goto clone_block;
1919				}
1920			}
1921			ea_bdebug(bs->bh, "modifying in-place");
1922			error = ext4_xattr_set_entry(i, s, handle, inode,
1923						     true /* is_block */);
1924			ext4_xattr_block_csum_set(inode, bs->bh);
1925			unlock_buffer(bs->bh);
1926			if (error == -EFSCORRUPTED)
1927				goto bad_block;
1928			if (!error)
1929				error = ext4_handle_dirty_metadata(handle,
1930								   inode,
1931								   bs->bh);
1932			if (error)
1933				goto cleanup;
1934			goto inserted;
1935		}
1936clone_block:
1937		unlock_buffer(bs->bh);
1938		ea_bdebug(bs->bh, "cloning");
1939		s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1940		error = -ENOMEM;
1941		if (s->base == NULL)
1942			goto cleanup;
1943		s->first = ENTRY(header(s->base)+1);
1944		header(s->base)->h_refcount = cpu_to_le32(1);
1945		s->here = ENTRY(s->base + offset);
1946		s->end = s->base + bs->bh->b_size;
1947
1948		/*
1949		 * If existing entry points to an xattr inode, we need
1950		 * to prevent ext4_xattr_set_entry() from decrementing
1951		 * ref count on it because the reference belongs to the
1952		 * original block. In this case, make the entry look
1953		 * like it has an empty value.
1954		 */
1955		if (!s->not_found && s->here->e_value_inum) {
1956			ea_ino = le32_to_cpu(s->here->e_value_inum);
1957			error = ext4_xattr_inode_iget(inode, ea_ino,
1958				      le32_to_cpu(s->here->e_hash),
1959				      &tmp_inode);
1960			if (error)
1961				goto cleanup;
 
 
 
 
 
1962
1963			if (!ext4_test_inode_state(tmp_inode,
1964					EXT4_STATE_LUSTRE_EA_INODE)) {
1965				/*
1966				 * Defer quota free call for previous
1967				 * inode until success is guaranteed.
1968				 */
1969				old_ea_inode_quota = le32_to_cpu(
1970						s->here->e_value_size);
1971			}
1972			iput(tmp_inode);
 
 
 
 
1973
1974			s->here->e_value_inum = 0;
1975			s->here->e_value_size = 0;
 
 
 
 
 
 
 
 
 
 
 
 
1976		}
1977	} else {
1978		/* Allocate a buffer where we construct the new block. */
1979		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
 
1980		error = -ENOMEM;
1981		if (s->base == NULL)
1982			goto cleanup;
1983		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1984		header(s->base)->h_blocks = cpu_to_le32(1);
1985		header(s->base)->h_refcount = cpu_to_le32(1);
1986		s->first = ENTRY(header(s->base)+1);
1987		s->here = ENTRY(header(s->base)+1);
1988		s->end = s->base + sb->s_blocksize;
1989	}
1990
1991	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1992	if (error == -EFSCORRUPTED)
1993		goto bad_block;
1994	if (error)
1995		goto cleanup;
1996
1997	if (i->value && s->here->e_value_inum) {
1998		/*
1999		 * A ref count on ea_inode has been taken as part of the call to
2000		 * ext4_xattr_set_entry() above. We would like to drop this
2001		 * extra ref but we have to wait until the xattr block is
2002		 * initialized and has its own ref count on the ea_inode.
2003		 */
2004		ea_ino = le32_to_cpu(s->here->e_value_inum);
2005		error = ext4_xattr_inode_iget(inode, ea_ino,
2006					      le32_to_cpu(s->here->e_hash),
2007					      &ea_inode);
2008		if (error) {
2009			ea_inode = NULL;
2010			goto cleanup;
2011		}
2012	}
2013
2014inserted:
2015	if (!IS_LAST_ENTRY(s->first)) {
2016		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2017						     &ce);
2018		if (new_bh) {
2019			/* We found an identical block in the cache. */
2020			if (new_bh == bs->bh)
2021				ea_bdebug(new_bh, "keeping");
2022			else {
2023				u32 ref;
2024
2025				WARN_ON_ONCE(dquot_initialize_needed(inode));
2026
2027				/* The old block is released after updating
2028				   the inode. */
2029				error = dquot_alloc_block(inode,
2030						EXT4_C2B(EXT4_SB(sb), 1));
2031				if (error)
2032					goto cleanup;
2033				BUFFER_TRACE(new_bh, "get_write_access");
2034				error = ext4_journal_get_write_access(
2035						handle, sb, new_bh,
2036						EXT4_JTR_NONE);
2037				if (error)
2038					goto cleanup_dquot;
2039				lock_buffer(new_bh);
2040				/*
2041				 * We have to be careful about races with
2042				 * adding references to xattr block. Once we
2043				 * hold buffer lock xattr block's state is
2044				 * stable so we can check the additional
2045				 * reference fits.
 
 
 
 
 
2046				 */
2047				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2048				if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2049					/*
2050					 * Undo everything and check mbcache
2051					 * again.
2052					 */
2053					unlock_buffer(new_bh);
2054					dquot_free_block(inode,
2055							 EXT4_C2B(EXT4_SB(sb),
2056								  1));
2057					brelse(new_bh);
2058					mb_cache_entry_put(ea_block_cache, ce);
2059					ce = NULL;
2060					new_bh = NULL;
2061					goto inserted;
2062				}
 
2063				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2064				if (ref == EXT4_XATTR_REFCOUNT_MAX)
2065					clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2066				ea_bdebug(new_bh, "reusing; refcount now=%d",
2067					  ref);
2068				ext4_xattr_block_csum_set(inode, new_bh);
2069				unlock_buffer(new_bh);
2070				error = ext4_handle_dirty_metadata(handle,
2071								   inode,
2072								   new_bh);
2073				if (error)
2074					goto cleanup_dquot;
2075			}
2076			mb_cache_entry_touch(ea_block_cache, ce);
2077			mb_cache_entry_put(ea_block_cache, ce);
2078			ce = NULL;
2079		} else if (bs->bh && s->base == bs->bh->b_data) {
2080			/* We were modifying this block in-place. */
2081			ea_bdebug(bs->bh, "keeping this block");
2082			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2083			new_bh = bs->bh;
2084			get_bh(new_bh);
2085		} else {
2086			/* We need to allocate a new block */
2087			ext4_fsblk_t goal, block;
2088
2089			WARN_ON_ONCE(dquot_initialize_needed(inode));
2090
2091			goal = ext4_group_first_block_no(sb,
2092						EXT4_I(inode)->i_block_group);
 
 
 
 
 
2093			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2094						     NULL, &error);
2095			if (error)
2096				goto cleanup;
2097
 
 
 
2098			ea_idebug(inode, "creating block %llu",
2099				  (unsigned long long)block);
2100
2101			new_bh = sb_getblk(sb, block);
2102			if (unlikely(!new_bh)) {
2103				error = -ENOMEM;
2104getblk_failed:
2105				ext4_free_blocks(handle, inode, NULL, block, 1,
2106						 EXT4_FREE_BLOCKS_METADATA);
2107				goto cleanup;
2108			}
2109			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2110						      ENTRY(header(s->base)+1));
2111			if (error)
2112				goto getblk_failed;
2113			if (ea_inode) {
2114				/* Drop the extra ref on ea_inode. */
2115				error = ext4_xattr_inode_dec_ref(handle,
2116								 ea_inode);
2117				if (error)
2118					ext4_warning_inode(ea_inode,
2119							   "dec ref error=%d",
2120							   error);
2121				iput(ea_inode);
2122				ea_inode = NULL;
2123			}
2124
2125			lock_buffer(new_bh);
2126			error = ext4_journal_get_create_access(handle, sb,
2127							new_bh, EXT4_JTR_NONE);
2128			if (error) {
2129				unlock_buffer(new_bh);
2130				error = -EIO;
2131				goto getblk_failed;
2132			}
2133			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2134			ext4_xattr_block_csum_set(inode, new_bh);
2135			set_buffer_uptodate(new_bh);
2136			unlock_buffer(new_bh);
2137			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2138			error = ext4_handle_dirty_metadata(handle, inode,
2139							   new_bh);
2140			if (error)
2141				goto cleanup;
2142		}
2143	}
2144
2145	if (old_ea_inode_quota)
2146		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2147
2148	/* Update the inode. */
2149	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2150
2151	/* Drop the previous xattr block. */
2152	if (bs->bh && bs->bh != new_bh) {
2153		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2154
2155		ext4_xattr_release_block(handle, inode, bs->bh,
2156					 &ea_inode_array,
2157					 0 /* extra_credits */);
2158		ext4_xattr_inode_array_free(ea_inode_array);
2159	}
2160	error = 0;
2161
2162cleanup:
2163	if (ea_inode) {
2164		int error2;
2165
2166		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2167		if (error2)
2168			ext4_warning_inode(ea_inode, "dec ref error=%d",
2169					   error2);
2170
2171		/* If there was an error, revert the quota charge. */
2172		if (error)
2173			ext4_xattr_inode_free_quota(inode, ea_inode,
2174						    i_size_read(ea_inode));
2175		iput(ea_inode);
2176	}
2177	if (ce)
2178		mb_cache_entry_put(ea_block_cache, ce);
2179	brelse(new_bh);
2180	if (!(bs->bh && s->base == bs->bh->b_data))
2181		kfree(s->base);
2182
2183	return error;
2184
2185cleanup_dquot:
2186	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2187	goto cleanup;
2188
2189bad_block:
2190	EXT4_ERROR_INODE(inode, "bad block %llu",
2191			 EXT4_I(inode)->i_file_acl);
2192	goto cleanup;
2193
2194#undef header
2195}
2196
2197int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2198			  struct ext4_xattr_ibody_find *is)
2199{
2200	struct ext4_xattr_ibody_header *header;
2201	struct ext4_inode *raw_inode;
2202	int error;
2203
2204	if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2205		return 0;
2206
2207	raw_inode = ext4_raw_inode(&is->iloc);
2208	header = IHDR(inode, raw_inode);
2209	is->s.base = is->s.first = IFIRST(header);
2210	is->s.here = is->s.first;
2211	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2212	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2213		error = xattr_check_inode(inode, header, is->s.end);
2214		if (error)
2215			return error;
2216		/* Find the named attribute. */
2217		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2218					 i->name_index, i->name, 0);
2219		if (error && error != -ENODATA)
2220			return error;
2221		is->s.not_found = error;
2222	}
2223	return 0;
2224}
2225
2226int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2227				struct ext4_xattr_info *i,
2228				struct ext4_xattr_ibody_find *is)
2229{
2230	struct ext4_xattr_ibody_header *header;
2231	struct ext4_xattr_search *s = &is->s;
2232	int error;
2233
2234	if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2235		return -ENOSPC;
 
 
 
 
 
 
 
 
 
 
 
 
 
2236
 
 
 
 
 
 
 
 
 
 
2237	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2238	if (error)
2239		return error;
2240	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2241	if (!IS_LAST_ENTRY(s->first)) {
2242		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2243		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2244	} else {
2245		header->h_magic = cpu_to_le32(0);
2246		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2247	}
2248	return 0;
2249}
2250
2251static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2252				 struct ext4_xattr_info *i)
2253{
2254	void *value;
2255
2256	/* When e_value_inum is set the value is stored externally. */
2257	if (s->here->e_value_inum)
2258		return 0;
2259	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2260		return 0;
2261	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2262	return !memcmp(value, i->value, i->value_len);
2263}
2264
2265static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2266{
2267	struct buffer_head *bh;
2268	int error;
2269
2270	if (!EXT4_I(inode)->i_file_acl)
2271		return NULL;
2272	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2273	if (IS_ERR(bh))
2274		return bh;
2275	error = ext4_xattr_check_block(inode, bh);
2276	if (error) {
2277		brelse(bh);
2278		return ERR_PTR(error);
2279	}
2280	return bh;
2281}
2282
2283/*
2284 * ext4_xattr_set_handle()
2285 *
2286 * Create, replace or remove an extended attribute for this inode.  Value
2287 * is NULL to remove an existing extended attribute, and non-NULL to
2288 * either replace an existing extended attribute, or create a new extended
2289 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2290 * specify that an extended attribute must exist and must not exist
2291 * previous to the call, respectively.
2292 *
2293 * Returns 0, or a negative error number on failure.
2294 */
2295int
2296ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2297		      const char *name, const void *value, size_t value_len,
2298		      int flags)
2299{
2300	struct ext4_xattr_info i = {
2301		.name_index = name_index,
2302		.name = name,
2303		.value = value,
2304		.value_len = value_len,
2305		.in_inode = 0,
2306	};
2307	struct ext4_xattr_ibody_find is = {
2308		.s = { .not_found = -ENODATA, },
2309	};
2310	struct ext4_xattr_block_find bs = {
2311		.s = { .not_found = -ENODATA, },
2312	};
2313	int no_expand;
2314	int error;
2315
2316	if (!name)
2317		return -EINVAL;
2318	if (strlen(name) > 255)
2319		return -ERANGE;
2320
2321	ext4_write_lock_xattr(inode, &no_expand);
2322
2323	/* Check journal credits under write lock. */
2324	if (ext4_handle_valid(handle)) {
2325		struct buffer_head *bh;
2326		int credits;
2327
2328		bh = ext4_xattr_get_block(inode);
2329		if (IS_ERR(bh)) {
2330			error = PTR_ERR(bh);
2331			goto cleanup;
2332		}
2333
2334		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2335						   value_len,
2336						   flags & XATTR_CREATE);
2337		brelse(bh);
2338
2339		if (jbd2_handle_buffer_credits(handle) < credits) {
2340			error = -ENOSPC;
2341			goto cleanup;
2342		}
2343		WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2344	}
2345
2346	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2347	if (error)
2348		goto cleanup;
2349
2350	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2351		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2352		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2353		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2354	}
2355
2356	error = ext4_xattr_ibody_find(inode, &i, &is);
2357	if (error)
2358		goto cleanup;
2359	if (is.s.not_found)
2360		error = ext4_xattr_block_find(inode, &i, &bs);
2361	if (error)
2362		goto cleanup;
2363	if (is.s.not_found && bs.s.not_found) {
2364		error = -ENODATA;
2365		if (flags & XATTR_REPLACE)
2366			goto cleanup;
2367		error = 0;
2368		if (!value)
2369			goto cleanup;
2370	} else {
2371		error = -EEXIST;
2372		if (flags & XATTR_CREATE)
2373			goto cleanup;
2374	}
2375
2376	if (!value) {
2377		if (!is.s.not_found)
2378			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379		else if (!bs.s.not_found)
2380			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2381	} else {
2382		error = 0;
2383		/* Xattr value did not change? Save us some work and bail out */
2384		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2385			goto cleanup;
2386		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2387			goto cleanup;
2388
2389		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2390		    (EXT4_XATTR_SIZE(i.value_len) >
2391			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2392			i.in_inode = 1;
2393retry_inode:
2394		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2395		if (!error && !bs.s.not_found) {
2396			i.value = NULL;
2397			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2398		} else if (error == -ENOSPC) {
2399			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2400				brelse(bs.bh);
2401				bs.bh = NULL;
2402				error = ext4_xattr_block_find(inode, &i, &bs);
2403				if (error)
2404					goto cleanup;
2405			}
2406			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2407			if (!error && !is.s.not_found) {
2408				i.value = NULL;
2409				error = ext4_xattr_ibody_set(handle, inode, &i,
2410							     &is);
2411			} else if (error == -ENOSPC) {
2412				/*
2413				 * Xattr does not fit in the block, store at
2414				 * external inode if possible.
2415				 */
2416				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2417				    i.value_len && !i.in_inode) {
2418					i.in_inode = 1;
2419					goto retry_inode;
2420				}
2421			}
2422		}
2423	}
2424	if (!error) {
2425		ext4_xattr_update_super_block(handle, inode->i_sb);
2426		inode->i_ctime = current_time(inode);
2427		inode_inc_iversion(inode);
2428		if (!value)
2429			no_expand = 0;
2430		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2431		/*
2432		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2433		 * error != 0.
2434		 */
2435		is.iloc.bh = NULL;
2436		if (IS_SYNC(inode))
2437			ext4_handle_sync(handle);
2438	}
2439	ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2440
2441cleanup:
2442	brelse(is.iloc.bh);
2443	brelse(bs.bh);
2444	ext4_write_unlock_xattr(inode, &no_expand);
2445	return error;
2446}
2447
2448int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2449			   bool is_create, int *credits)
2450{
2451	struct buffer_head *bh;
2452	int err;
2453
2454	*credits = 0;
2455
2456	if (!EXT4_SB(inode->i_sb)->s_journal)
2457		return 0;
2458
2459	down_read(&EXT4_I(inode)->xattr_sem);
2460
2461	bh = ext4_xattr_get_block(inode);
2462	if (IS_ERR(bh)) {
2463		err = PTR_ERR(bh);
2464	} else {
2465		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2466						    value_len, is_create);
2467		brelse(bh);
2468		err = 0;
2469	}
2470
2471	up_read(&EXT4_I(inode)->xattr_sem);
2472	return err;
2473}
2474
2475/*
2476 * ext4_xattr_set()
2477 *
2478 * Like ext4_xattr_set_handle, but start from an inode. This extended
2479 * attribute modification is a filesystem transaction by itself.
2480 *
2481 * Returns 0, or a negative error number on failure.
2482 */
2483int
2484ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2485	       const void *value, size_t value_len, int flags)
2486{
2487	handle_t *handle;
2488	struct super_block *sb = inode->i_sb;
2489	int error, retries = 0;
2490	int credits;
2491
2492	error = dquot_initialize(inode);
2493	if (error)
2494		return error;
2495
2496retry:
2497	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2498				       &credits);
2499	if (error)
2500		return error;
2501
2502	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2503	if (IS_ERR(handle)) {
2504		error = PTR_ERR(handle);
2505	} else {
2506		int error2;
2507
2508		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2509					      value, value_len, flags);
2510		error2 = ext4_journal_stop(handle);
2511		if (error == -ENOSPC &&
2512		    ext4_should_retry_alloc(sb, &retries))
2513			goto retry;
2514		if (error == 0)
2515			error = error2;
2516	}
2517	ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2518
2519	return error;
2520}
2521
2522/*
2523 * Shift the EA entries in the inode to create space for the increased
2524 * i_extra_isize.
2525 */
2526static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2527				     int value_offs_shift, void *to,
2528				     void *from, size_t n)
2529{
2530	struct ext4_xattr_entry *last = entry;
2531	int new_offs;
2532
2533	/* We always shift xattr headers further thus offsets get lower */
2534	BUG_ON(value_offs_shift > 0);
2535
2536	/* Adjust the value offsets of the entries */
2537	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2538		if (!last->e_value_inum && last->e_value_size) {
2539			new_offs = le16_to_cpu(last->e_value_offs) +
2540							value_offs_shift;
2541			last->e_value_offs = cpu_to_le16(new_offs);
2542		}
2543	}
2544	/* Shift the entries by n bytes */
2545	memmove(to, from, n);
2546}
2547
2548/*
2549 * Move xattr pointed to by 'entry' from inode into external xattr block
2550 */
2551static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2552				    struct ext4_inode *raw_inode,
2553				    struct ext4_xattr_entry *entry)
2554{
2555	struct ext4_xattr_ibody_find *is = NULL;
2556	struct ext4_xattr_block_find *bs = NULL;
2557	char *buffer = NULL, *b_entry_name = NULL;
2558	size_t value_size = le32_to_cpu(entry->e_value_size);
2559	struct ext4_xattr_info i = {
2560		.value = NULL,
2561		.value_len = 0,
2562		.name_index = entry->e_name_index,
2563		.in_inode = !!entry->e_value_inum,
2564	};
2565	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2566	int error;
2567
2568	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2569	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2570	buffer = kvmalloc(value_size, GFP_NOFS);
2571	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2572	if (!is || !bs || !buffer || !b_entry_name) {
2573		error = -ENOMEM;
2574		goto out;
2575	}
2576
2577	is->s.not_found = -ENODATA;
2578	bs->s.not_found = -ENODATA;
2579	is->iloc.bh = NULL;
2580	bs->bh = NULL;
2581
2582	/* Save the entry name and the entry value */
2583	if (entry->e_value_inum) {
2584		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2585		if (error)
2586			goto out;
2587	} else {
2588		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2589		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2590	}
2591
2592	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2593	b_entry_name[entry->e_name_len] = '\0';
2594	i.name = b_entry_name;
2595
2596	error = ext4_get_inode_loc(inode, &is->iloc);
2597	if (error)
2598		goto out;
2599
2600	error = ext4_xattr_ibody_find(inode, &i, is);
2601	if (error)
2602		goto out;
2603
2604	/* Remove the chosen entry from the inode */
2605	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2606	if (error)
2607		goto out;
2608
2609	i.value = buffer;
2610	i.value_len = value_size;
2611	error = ext4_xattr_block_find(inode, &i, bs);
2612	if (error)
2613		goto out;
2614
2615	/* Add entry which was removed from the inode into the block */
2616	error = ext4_xattr_block_set(handle, inode, &i, bs);
2617	if (error)
2618		goto out;
2619	error = 0;
2620out:
2621	kfree(b_entry_name);
2622	kvfree(buffer);
2623	if (is)
2624		brelse(is->iloc.bh);
2625	if (bs)
2626		brelse(bs->bh);
2627	kfree(is);
2628	kfree(bs);
2629
2630	return error;
2631}
2632
2633static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2634				       struct ext4_inode *raw_inode,
2635				       int isize_diff, size_t ifree,
2636				       size_t bfree, int *total_ino)
2637{
2638	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2639	struct ext4_xattr_entry *small_entry;
2640	struct ext4_xattr_entry *entry;
2641	struct ext4_xattr_entry *last;
2642	unsigned int entry_size;	/* EA entry size */
2643	unsigned int total_size;	/* EA entry size + value size */
2644	unsigned int min_total_size;
2645	int error;
2646
2647	while (isize_diff > ifree) {
2648		entry = NULL;
2649		small_entry = NULL;
2650		min_total_size = ~0U;
2651		last = IFIRST(header);
2652		/* Find the entry best suited to be pushed into EA block */
2653		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2654			/* never move system.data out of the inode */
2655			if ((last->e_name_len == 4) &&
2656			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2657			    !memcmp(last->e_name, "data", 4))
2658				continue;
2659			total_size = EXT4_XATTR_LEN(last->e_name_len);
2660			if (!last->e_value_inum)
2661				total_size += EXT4_XATTR_SIZE(
2662					       le32_to_cpu(last->e_value_size));
2663			if (total_size <= bfree &&
2664			    total_size < min_total_size) {
2665				if (total_size + ifree < isize_diff) {
2666					small_entry = last;
2667				} else {
2668					entry = last;
2669					min_total_size = total_size;
2670				}
2671			}
2672		}
2673
2674		if (entry == NULL) {
2675			if (small_entry == NULL)
2676				return -ENOSPC;
2677			entry = small_entry;
2678		}
2679
2680		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2681		total_size = entry_size;
2682		if (!entry->e_value_inum)
2683			total_size += EXT4_XATTR_SIZE(
2684					      le32_to_cpu(entry->e_value_size));
2685		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2686						 entry);
2687		if (error)
2688			return error;
2689
2690		*total_ino -= entry_size;
2691		ifree += total_size;
2692		bfree -= total_size;
2693	}
2694
2695	return 0;
2696}
2697
2698/*
2699 * Expand an inode by new_extra_isize bytes when EAs are present.
2700 * Returns 0 on success or negative error number on failure.
2701 */
2702int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2703			       struct ext4_inode *raw_inode, handle_t *handle)
2704{
2705	struct ext4_xattr_ibody_header *header;
2706	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2707	static unsigned int mnt_count;
2708	size_t min_offs;
2709	size_t ifree, bfree;
2710	int total_ino;
2711	void *base, *end;
2712	int error = 0, tried_min_extra_isize = 0;
2713	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2714	int isize_diff;	/* How much do we need to grow i_extra_isize */
2715
2716retry:
2717	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2718	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2719		return 0;
2720
2721	header = IHDR(inode, raw_inode);
2722
2723	/*
2724	 * Check if enough free space is available in the inode to shift the
2725	 * entries ahead by new_extra_isize.
2726	 */
2727
2728	base = IFIRST(header);
2729	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2730	min_offs = end - base;
2731	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2732
2733	error = xattr_check_inode(inode, header, end);
2734	if (error)
2735		goto cleanup;
2736
2737	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2738	if (ifree >= isize_diff)
2739		goto shift;
2740
2741	/*
2742	 * Enough free space isn't available in the inode, check if
2743	 * EA block can hold new_extra_isize bytes.
2744	 */
2745	if (EXT4_I(inode)->i_file_acl) {
2746		struct buffer_head *bh;
2747
2748		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2749		if (IS_ERR(bh)) {
2750			error = PTR_ERR(bh);
2751			goto cleanup;
2752		}
2753		error = ext4_xattr_check_block(inode, bh);
2754		if (error) {
2755			brelse(bh);
2756			goto cleanup;
2757		}
2758		base = BHDR(bh);
2759		end = bh->b_data + bh->b_size;
2760		min_offs = end - base;
2761		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2762					      NULL);
2763		brelse(bh);
2764		if (bfree + ifree < isize_diff) {
2765			if (!tried_min_extra_isize && s_min_extra_isize) {
2766				tried_min_extra_isize++;
2767				new_extra_isize = s_min_extra_isize;
2768				goto retry;
2769			}
2770			error = -ENOSPC;
2771			goto cleanup;
2772		}
2773	} else {
2774		bfree = inode->i_sb->s_blocksize;
2775	}
2776
2777	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2778					    isize_diff, ifree, bfree,
2779					    &total_ino);
2780	if (error) {
2781		if (error == -ENOSPC && !tried_min_extra_isize &&
2782		    s_min_extra_isize) {
2783			tried_min_extra_isize++;
2784			new_extra_isize = s_min_extra_isize;
2785			goto retry;
2786		}
2787		goto cleanup;
2788	}
2789shift:
2790	/* Adjust the offsets and shift the remaining entries ahead */
2791	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2792			- new_extra_isize, (void *)raw_inode +
2793			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2794			(void *)header, total_ino);
2795	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2796
2797cleanup:
2798	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2799		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2800			     inode->i_ino);
2801		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2802	}
2803	return error;
2804}
2805
2806#define EIA_INCR 16 /* must be 2^n */
2807#define EIA_MASK (EIA_INCR - 1)
2808
2809/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2810 * If @ea_inode_array is new or full it will be grown and the old
2811 * contents copied over.
2812 */
2813static int
2814ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2815			struct inode *inode)
2816{
2817	if (*ea_inode_array == NULL) {
2818		/*
2819		 * Start with 15 inodes, so it fits into a power-of-two size.
2820		 * If *ea_inode_array is NULL, this is essentially offsetof()
2821		 */
2822		(*ea_inode_array) =
2823			kmalloc(offsetof(struct ext4_xattr_inode_array,
2824					 inodes[EIA_MASK]),
2825				GFP_NOFS);
2826		if (*ea_inode_array == NULL)
2827			return -ENOMEM;
2828		(*ea_inode_array)->count = 0;
2829	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2830		/* expand the array once all 15 + n * 16 slots are full */
2831		struct ext4_xattr_inode_array *new_array = NULL;
2832		int count = (*ea_inode_array)->count;
2833
2834		/* if new_array is NULL, this is essentially offsetof() */
2835		new_array = kmalloc(
2836				offsetof(struct ext4_xattr_inode_array,
2837					 inodes[count + EIA_INCR]),
2838				GFP_NOFS);
2839		if (new_array == NULL)
2840			return -ENOMEM;
2841		memcpy(new_array, *ea_inode_array,
2842		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2843		kfree(*ea_inode_array);
2844		*ea_inode_array = new_array;
2845	}
2846	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2847	return 0;
2848}
2849
2850/*
2851 * ext4_xattr_delete_inode()
2852 *
2853 * Free extended attribute resources associated with this inode. Traverse
2854 * all entries and decrement reference on any xattr inodes associated with this
2855 * inode. This is called immediately before an inode is freed. We have exclusive
2856 * access to the inode. If an orphan inode is deleted it will also release its
2857 * references on xattr block and xattr inodes.
2858 */
2859int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2860			    struct ext4_xattr_inode_array **ea_inode_array,
2861			    int extra_credits)
2862{
2863	struct buffer_head *bh = NULL;
2864	struct ext4_xattr_ibody_header *header;
2865	struct ext4_iloc iloc = { .bh = NULL };
2866	struct ext4_xattr_entry *entry;
2867	struct inode *ea_inode;
2868	int error;
2869
2870	error = ext4_journal_ensure_credits(handle, extra_credits,
2871			ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2872	if (error < 0) {
2873		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2874		goto cleanup;
2875	}
2876
2877	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2878	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2879
2880		error = ext4_get_inode_loc(inode, &iloc);
2881		if (error) {
2882			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2883			goto cleanup;
2884		}
2885
2886		error = ext4_journal_get_write_access(handle, inode->i_sb,
2887						iloc.bh, EXT4_JTR_NONE);
2888		if (error) {
2889			EXT4_ERROR_INODE(inode, "write access (error %d)",
2890					 error);
2891			goto cleanup;
2892		}
2893
2894		header = IHDR(inode, ext4_raw_inode(&iloc));
2895		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2896			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2897						     IFIRST(header),
2898						     false /* block_csum */,
2899						     ea_inode_array,
2900						     extra_credits,
2901						     false /* skip_quota */);
2902	}
2903
2904	if (EXT4_I(inode)->i_file_acl) {
2905		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2906		if (IS_ERR(bh)) {
2907			error = PTR_ERR(bh);
2908			if (error == -EIO) {
2909				EXT4_ERROR_INODE_ERR(inode, EIO,
2910						     "block %llu read error",
2911						     EXT4_I(inode)->i_file_acl);
2912			}
2913			bh = NULL;
2914			goto cleanup;
2915		}
2916		error = ext4_xattr_check_block(inode, bh);
2917		if (error)
2918			goto cleanup;
2919
2920		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2921			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2922			     entry = EXT4_XATTR_NEXT(entry)) {
2923				if (!entry->e_value_inum)
2924					continue;
2925				error = ext4_xattr_inode_iget(inode,
2926					      le32_to_cpu(entry->e_value_inum),
2927					      le32_to_cpu(entry->e_hash),
2928					      &ea_inode);
2929				if (error)
2930					continue;
2931				ext4_xattr_inode_free_quota(inode, ea_inode,
2932					      le32_to_cpu(entry->e_value_size));
2933				iput(ea_inode);
2934			}
2935
2936		}
2937
2938		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2939					 extra_credits);
2940		/*
2941		 * Update i_file_acl value in the same transaction that releases
2942		 * block.
2943		 */
2944		EXT4_I(inode)->i_file_acl = 0;
2945		error = ext4_mark_inode_dirty(handle, inode);
2946		if (error) {
2947			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2948					 error);
2949			goto cleanup;
2950		}
2951		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2952	}
2953	error = 0;
2954cleanup:
2955	brelse(iloc.bh);
2956	brelse(bh);
2957	return error;
2958}
2959
2960void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2961{
2962	int idx;
2963
2964	if (ea_inode_array == NULL)
2965		return;
2966
2967	for (idx = 0; idx < ea_inode_array->count; ++idx)
2968		iput(ea_inode_array->inodes[idx]);
2969	kfree(ea_inode_array);
2970}
2971
2972/*
2973 * ext4_xattr_block_cache_insert()
2974 *
2975 * Create a new entry in the extended attribute block cache, and insert
2976 * it unless such an entry is already in the cache.
2977 *
2978 * Returns 0, or a negative error number on failure.
2979 */
2980static void
2981ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2982			      struct buffer_head *bh)
2983{
2984	struct ext4_xattr_header *header = BHDR(bh);
2985	__u32 hash = le32_to_cpu(header->h_hash);
2986	int reusable = le32_to_cpu(header->h_refcount) <
2987		       EXT4_XATTR_REFCOUNT_MAX;
2988	int error;
2989
2990	if (!ea_block_cache)
2991		return;
2992	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2993				      bh->b_blocknr, reusable);
2994	if (error) {
2995		if (error == -EBUSY)
2996			ea_bdebug(bh, "already in cache");
2997	} else
2998		ea_bdebug(bh, "inserting [%x]", (int)hash);
2999}
3000
3001/*
3002 * ext4_xattr_cmp()
3003 *
3004 * Compare two extended attribute blocks for equality.
3005 *
3006 * Returns 0 if the blocks are equal, 1 if they differ, and
3007 * a negative error number on errors.
3008 */
3009static int
3010ext4_xattr_cmp(struct ext4_xattr_header *header1,
3011	       struct ext4_xattr_header *header2)
3012{
3013	struct ext4_xattr_entry *entry1, *entry2;
3014
3015	entry1 = ENTRY(header1+1);
3016	entry2 = ENTRY(header2+1);
3017	while (!IS_LAST_ENTRY(entry1)) {
3018		if (IS_LAST_ENTRY(entry2))
3019			return 1;
3020		if (entry1->e_hash != entry2->e_hash ||
3021		    entry1->e_name_index != entry2->e_name_index ||
3022		    entry1->e_name_len != entry2->e_name_len ||
3023		    entry1->e_value_size != entry2->e_value_size ||
3024		    entry1->e_value_inum != entry2->e_value_inum ||
3025		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3026			return 1;
3027		if (!entry1->e_value_inum &&
3028		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3029			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3030			   le32_to_cpu(entry1->e_value_size)))
3031			return 1;
3032
3033		entry1 = EXT4_XATTR_NEXT(entry1);
3034		entry2 = EXT4_XATTR_NEXT(entry2);
3035	}
3036	if (!IS_LAST_ENTRY(entry2))
3037		return 1;
3038	return 0;
3039}
3040
3041/*
3042 * ext4_xattr_block_cache_find()
3043 *
3044 * Find an identical extended attribute block.
3045 *
3046 * Returns a pointer to the block found, or NULL if such a block was
3047 * not found or an error occurred.
3048 */
3049static struct buffer_head *
3050ext4_xattr_block_cache_find(struct inode *inode,
3051			    struct ext4_xattr_header *header,
3052			    struct mb_cache_entry **pce)
3053{
3054	__u32 hash = le32_to_cpu(header->h_hash);
3055	struct mb_cache_entry *ce;
3056	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3057
3058	if (!ea_block_cache)
3059		return NULL;
3060	if (!header->h_hash)
3061		return NULL;  /* never share */
3062	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3063	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3064	while (ce) {
3065		struct buffer_head *bh;
3066
3067		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3068		if (IS_ERR(bh)) {
3069			if (PTR_ERR(bh) == -ENOMEM)
3070				return NULL;
3071			bh = NULL;
3072			EXT4_ERROR_INODE(inode, "block %lu read error",
3073					 (unsigned long)ce->e_value);
3074		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3075			*pce = ce;
3076			return bh;
3077		}
3078		brelse(bh);
3079		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3080	}
3081	return NULL;
3082}
3083
3084#define NAME_HASH_SHIFT 5
3085#define VALUE_HASH_SHIFT 16
3086
3087/*
3088 * ext4_xattr_hash_entry()
3089 *
3090 * Compute the hash of an extended attribute.
3091 */
3092static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3093				    size_t value_count)
3094{
3095	__u32 hash = 0;
3096
3097	while (name_len--) {
3098		hash = (hash << NAME_HASH_SHIFT) ^
3099		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3100		       (unsigned char)*name++;
3101	}
3102	while (value_count--) {
3103		hash = (hash << VALUE_HASH_SHIFT) ^
3104		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3105		       le32_to_cpu(*value++);
3106	}
3107	return cpu_to_le32(hash);
3108}
3109
3110/*
3111 * ext4_xattr_hash_entry_signed()
3112 *
3113 * Compute the hash of an extended attribute incorrectly.
3114 */
3115static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3116{
3117	__u32 hash = 0;
3118
3119	while (name_len--) {
3120		hash = (hash << NAME_HASH_SHIFT) ^
3121		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3122		       (signed char)*name++;
3123	}
3124	while (value_count--) {
3125		hash = (hash << VALUE_HASH_SHIFT) ^
3126		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3127		       le32_to_cpu(*value++);
3128	}
3129	return cpu_to_le32(hash);
3130}
3131
3132#undef NAME_HASH_SHIFT
3133#undef VALUE_HASH_SHIFT
3134
3135#define BLOCK_HASH_SHIFT 16
3136
3137/*
3138 * ext4_xattr_rehash()
3139 *
3140 * Re-compute the extended attribute hash value after an entry has changed.
3141 */
3142static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3143{
3144	struct ext4_xattr_entry *here;
3145	__u32 hash = 0;
3146
3147	here = ENTRY(header+1);
3148	while (!IS_LAST_ENTRY(here)) {
3149		if (!here->e_hash) {
3150			/* Block is not shared if an entry's hash value == 0 */
3151			hash = 0;
3152			break;
3153		}
3154		hash = (hash << BLOCK_HASH_SHIFT) ^
3155		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3156		       le32_to_cpu(here->e_hash);
3157		here = EXT4_XATTR_NEXT(here);
3158	}
3159	header->h_hash = cpu_to_le32(hash);
3160}
3161
3162#undef BLOCK_HASH_SHIFT
3163
3164#define	HASH_BUCKET_BITS	10
3165
3166struct mb_cache *
3167ext4_xattr_create_cache(void)
3168{
3169	return mb_cache_create(HASH_BUCKET_BITS);
3170}
3171
3172void ext4_xattr_destroy_cache(struct mb_cache *cache)
3173{
3174	if (cache)
3175		mb_cache_destroy(cache);
3176}
3177
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/ext4/xattr.c
   4 *
   5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6 *
   7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9 * Extended attributes for symlinks and special files added per
  10 *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12 *  Red Hat Inc.
  13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14 *  and Andreas Gruenbacher <agruen@suse.de>.
  15 */
  16
  17/*
  18 * Extended attributes are stored directly in inodes (on file systems with
  19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20 * field contains the block number if an inode uses an additional block. All
  21 * attributes must fit in the inode and one additional block. Blocks that
  22 * contain the identical set of attributes may be shared among several inodes.
  23 * Identical blocks are detected by keeping a cache of blocks that have
  24 * recently been accessed.
  25 *
  26 * The attributes in inodes and on blocks have a different header; the entries
  27 * are stored in the same format:
  28 *
  29 *   +------------------+
  30 *   | header           |
  31 *   | entry 1          | |
  32 *   | entry 2          | | growing downwards
  33 *   | entry 3          | v
  34 *   | four null bytes  |
  35 *   | . . .            |
  36 *   | value 1          | ^
  37 *   | value 3          | | growing upwards
  38 *   | value 2          | |
  39 *   +------------------+
  40 *
  41 * The header is followed by multiple entry descriptors. In disk blocks, the
  42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43 * attribute values are aligned to the end of the block in no specific order.
  44 *
  45 * Locking strategy
  46 * ----------------
  47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48 * EA blocks are only changed if they are exclusive to an inode, so
  49 * holding xattr_sem also means that nothing but the EA block's reference
  50 * count can change. Multiple writers to the same block are synchronized
  51 * by the buffer lock.
  52 */
  53
  54#include <linux/init.h>
  55#include <linux/fs.h>
  56#include <linux/slab.h>
  57#include <linux/mbcache.h>
  58#include <linux/quotaops.h>
  59#include <linux/iversion.h>
  60#include "ext4_jbd2.h"
  61#include "ext4.h"
  62#include "xattr.h"
  63#include "acl.h"
  64
  65#ifdef EXT4_XATTR_DEBUG
  66# define ea_idebug(inode, fmt, ...)					\
  67	printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",			\
  68	       inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69# define ea_bdebug(bh, fmt, ...)					\
  70	printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",			\
  71	       bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72#else
  73# define ea_idebug(inode, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
  74# define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
  75#endif
  76
  77static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78					  struct buffer_head *);
  79static struct buffer_head *
  80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81			    struct mb_cache_entry **);
  82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83				    size_t value_count);
 
 
  84static void ext4_xattr_rehash(struct ext4_xattr_header *);
  85
  86static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  87	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
  88#ifdef CONFIG_EXT4_FS_POSIX_ACL
  89	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  90	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  91#endif
  92	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
  93#ifdef CONFIG_EXT4_FS_SECURITY
  94	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
  95#endif
  96	[EXT4_XATTR_INDEX_HURD]		     = &ext4_xattr_hurd_handler,
  97};
  98
  99const struct xattr_handler *ext4_xattr_handlers[] = {
 100	&ext4_xattr_user_handler,
 101	&ext4_xattr_trusted_handler,
 102#ifdef CONFIG_EXT4_FS_POSIX_ACL
 103	&posix_acl_access_xattr_handler,
 104	&posix_acl_default_xattr_handler,
 105#endif
 106#ifdef CONFIG_EXT4_FS_SECURITY
 107	&ext4_xattr_security_handler,
 108#endif
 109	&ext4_xattr_hurd_handler,
 110	NULL
 111};
 112
 113#define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
 114				inode->i_sb->s_fs_info)->s_ea_block_cache)
 115
 116#define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
 117				inode->i_sb->s_fs_info)->s_ea_inode_cache)
 118
 119static int
 120ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 121			struct inode *inode);
 122
 123#ifdef CONFIG_LOCKDEP
 124void ext4_xattr_inode_set_class(struct inode *ea_inode)
 125{
 126	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 127}
 128#endif
 129
 130static __le32 ext4_xattr_block_csum(struct inode *inode,
 131				    sector_t block_nr,
 132				    struct ext4_xattr_header *hdr)
 133{
 134	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 135	__u32 csum;
 136	__le64 dsk_block_nr = cpu_to_le64(block_nr);
 137	__u32 dummy_csum = 0;
 138	int offset = offsetof(struct ext4_xattr_header, h_checksum);
 139
 140	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 141			   sizeof(dsk_block_nr));
 142	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 143	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 144	offset += sizeof(dummy_csum);
 145	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 146			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 147
 148	return cpu_to_le32(csum);
 149}
 150
 151static int ext4_xattr_block_csum_verify(struct inode *inode,
 152					struct buffer_head *bh)
 153{
 154	struct ext4_xattr_header *hdr = BHDR(bh);
 155	int ret = 1;
 156
 157	if (ext4_has_metadata_csum(inode->i_sb)) {
 158		lock_buffer(bh);
 159		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 160							bh->b_blocknr, hdr));
 161		unlock_buffer(bh);
 162	}
 163	return ret;
 164}
 165
 166static void ext4_xattr_block_csum_set(struct inode *inode,
 167				      struct buffer_head *bh)
 168{
 169	if (ext4_has_metadata_csum(inode->i_sb))
 170		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 171						bh->b_blocknr, BHDR(bh));
 172}
 173
 174static inline const struct xattr_handler *
 175ext4_xattr_handler(int name_index)
 176{
 177	const struct xattr_handler *handler = NULL;
 178
 179	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 180		handler = ext4_xattr_handler_map[name_index];
 181	return handler;
 182}
 183
 184static int
 185ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 186			 void *value_start)
 187{
 188	struct ext4_xattr_entry *e = entry;
 189
 190	/* Find the end of the names list */
 191	while (!IS_LAST_ENTRY(e)) {
 192		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 193		if ((void *)next >= end)
 194			return -EFSCORRUPTED;
 195		if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 196			return -EFSCORRUPTED;
 197		e = next;
 198	}
 199
 200	/* Check the values */
 201	while (!IS_LAST_ENTRY(entry)) {
 202		u32 size = le32_to_cpu(entry->e_value_size);
 203
 204		if (size > EXT4_XATTR_SIZE_MAX)
 205			return -EFSCORRUPTED;
 206
 207		if (size != 0 && entry->e_value_inum == 0) {
 208			u16 offs = le16_to_cpu(entry->e_value_offs);
 209			void *value;
 210
 211			/*
 212			 * The value cannot overlap the names, and the value
 213			 * with padding cannot extend beyond 'end'.  Check both
 214			 * the padded and unpadded sizes, since the size may
 215			 * overflow to 0 when adding padding.
 216			 */
 217			if (offs > end - value_start)
 218				return -EFSCORRUPTED;
 219			value = value_start + offs;
 220			if (value < (void *)e + sizeof(u32) ||
 221			    size > end - value ||
 222			    EXT4_XATTR_SIZE(size) > end - value)
 223				return -EFSCORRUPTED;
 224		}
 225		entry = EXT4_XATTR_NEXT(entry);
 226	}
 227
 228	return 0;
 229}
 230
 231static inline int
 232__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 233			 const char *function, unsigned int line)
 234{
 235	int error = -EFSCORRUPTED;
 236
 237	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 238	    BHDR(bh)->h_blocks != cpu_to_le32(1))
 239		goto errout;
 240	if (buffer_verified(bh))
 241		return 0;
 242
 243	error = -EFSBADCRC;
 244	if (!ext4_xattr_block_csum_verify(inode, bh))
 245		goto errout;
 246	error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 247					 bh->b_data);
 248errout:
 249	if (error)
 250		__ext4_error_inode(inode, function, line, 0, -error,
 251				   "corrupted xattr block %llu",
 252				   (unsigned long long) bh->b_blocknr);
 253	else
 254		set_buffer_verified(bh);
 255	return error;
 256}
 257
 258#define ext4_xattr_check_block(inode, bh) \
 259	__ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 260
 261
 262static int
 263__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 264			 void *end, const char *function, unsigned int line)
 265{
 266	int error = -EFSCORRUPTED;
 267
 268	if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 269	    (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 270		goto errout;
 271	error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 272errout:
 273	if (error)
 274		__ext4_error_inode(inode, function, line, 0, -error,
 275				   "corrupted in-inode xattr");
 276	return error;
 277}
 278
 279#define xattr_check_inode(inode, header, end) \
 280	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 281
 282static int
 283xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 284		 void *end, int name_index, const char *name, int sorted)
 285{
 286	struct ext4_xattr_entry *entry, *next;
 287	size_t name_len;
 288	int cmp = 1;
 289
 290	if (name == NULL)
 291		return -EINVAL;
 292	name_len = strlen(name);
 293	for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 294		next = EXT4_XATTR_NEXT(entry);
 295		if ((void *) next >= end) {
 296			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 297			return -EFSCORRUPTED;
 298		}
 299		cmp = name_index - entry->e_name_index;
 300		if (!cmp)
 301			cmp = name_len - entry->e_name_len;
 302		if (!cmp)
 303			cmp = memcmp(name, entry->e_name, name_len);
 304		if (cmp <= 0 && (sorted || cmp == 0))
 305			break;
 306	}
 307	*pentry = entry;
 308	return cmp ? -ENODATA : 0;
 309}
 310
 311static u32
 312ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 313{
 314	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 315}
 316
 317static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 318{
 319	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 320		(u32) inode_peek_iversion_raw(ea_inode);
 321}
 322
 323static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 324{
 325	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 326	inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 327}
 328
 329static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 330{
 331	return (u32)ea_inode->i_atime.tv_sec;
 332}
 333
 334static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 335{
 336	ea_inode->i_atime.tv_sec = hash;
 337}
 338
 339/*
 340 * Read the EA value from an inode.
 341 */
 342static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 343{
 344	int blocksize = 1 << ea_inode->i_blkbits;
 345	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 346	int tail_size = (size % blocksize) ?: blocksize;
 347	struct buffer_head *bhs_inline[8];
 348	struct buffer_head **bhs = bhs_inline;
 349	int i, ret;
 350
 351	if (bh_count > ARRAY_SIZE(bhs_inline)) {
 352		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 353		if (!bhs)
 354			return -ENOMEM;
 355	}
 356
 357	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 358			       true /* wait */, bhs);
 359	if (ret)
 360		goto free_bhs;
 361
 362	for (i = 0; i < bh_count; i++) {
 363		/* There shouldn't be any holes in ea_inode. */
 364		if (!bhs[i]) {
 365			ret = -EFSCORRUPTED;
 366			goto put_bhs;
 367		}
 368		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 369		       i < bh_count - 1 ? blocksize : tail_size);
 370	}
 371	ret = 0;
 372put_bhs:
 373	for (i = 0; i < bh_count; i++)
 374		brelse(bhs[i]);
 375free_bhs:
 376	if (bhs != bhs_inline)
 377		kfree(bhs);
 378	return ret;
 379}
 380
 381#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 382
 383static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 384				 u32 ea_inode_hash, struct inode **ea_inode)
 385{
 386	struct inode *inode;
 387	int err;
 388
 389	inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 390	if (IS_ERR(inode)) {
 391		err = PTR_ERR(inode);
 392		ext4_error(parent->i_sb,
 393			   "error while reading EA inode %lu err=%d", ea_ino,
 394			   err);
 395		return err;
 396	}
 397
 398	if (is_bad_inode(inode)) {
 399		ext4_error(parent->i_sb,
 400			   "error while reading EA inode %lu is_bad_inode",
 401			   ea_ino);
 402		err = -EIO;
 403		goto error;
 404	}
 405
 406	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 407		ext4_error(parent->i_sb,
 408			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 409			    ea_ino);
 410		err = -EINVAL;
 411		goto error;
 412	}
 413
 414	ext4_xattr_inode_set_class(inode);
 415
 416	/*
 417	 * Check whether this is an old Lustre-style xattr inode. Lustre
 418	 * implementation does not have hash validation, rather it has a
 419	 * backpointer from ea_inode to the parent inode.
 420	 */
 421	if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 422	    EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 423	    inode->i_generation == parent->i_generation) {
 424		ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 425		ext4_xattr_inode_set_ref(inode, 1);
 426	} else {
 427		inode_lock(inode);
 428		inode->i_flags |= S_NOQUOTA;
 429		inode_unlock(inode);
 430	}
 431
 432	*ea_inode = inode;
 433	return 0;
 434error:
 435	iput(inode);
 436	return err;
 437}
 438
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 439static int
 440ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 441			       struct ext4_xattr_entry *entry, void *buffer,
 442			       size_t size)
 443{
 444	u32 hash;
 445
 446	/* Verify stored hash matches calculated hash. */
 447	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 448	if (hash != ext4_xattr_inode_get_hash(ea_inode))
 449		return -EFSCORRUPTED;
 450
 451	if (entry) {
 452		__le32 e_hash, tmp_data;
 453
 454		/* Verify entry hash. */
 455		tmp_data = cpu_to_le32(hash);
 456		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 457					       &tmp_data, 1);
 
 
 
 
 
 
 
 
 
 
 
 458		if (e_hash != entry->e_hash)
 459			return -EFSCORRUPTED;
 
 
 
 460	}
 461	return 0;
 462}
 463
 464/*
 465 * Read xattr value from the EA inode.
 466 */
 467static int
 468ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 469		     void *buffer, size_t size)
 470{
 471	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 472	struct inode *ea_inode;
 473	int err;
 474
 475	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 476				    le32_to_cpu(entry->e_hash), &ea_inode);
 477	if (err) {
 478		ea_inode = NULL;
 479		goto out;
 480	}
 481
 482	if (i_size_read(ea_inode) != size) {
 483		ext4_warning_inode(ea_inode,
 484				   "ea_inode file size=%llu entry size=%zu",
 485				   i_size_read(ea_inode), size);
 486		err = -EFSCORRUPTED;
 487		goto out;
 488	}
 489
 490	err = ext4_xattr_inode_read(ea_inode, buffer, size);
 491	if (err)
 492		goto out;
 493
 494	if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 495		err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 496						     size);
 497		if (err) {
 498			ext4_warning_inode(ea_inode,
 499					   "EA inode hash validation failed");
 500			goto out;
 501		}
 502
 503		if (ea_inode_cache)
 504			mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 505					ext4_xattr_inode_get_hash(ea_inode),
 506					ea_inode->i_ino, true /* reusable */);
 507	}
 508out:
 509	iput(ea_inode);
 510	return err;
 511}
 512
 513static int
 514ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 515		     void *buffer, size_t buffer_size)
 516{
 517	struct buffer_head *bh = NULL;
 518	struct ext4_xattr_entry *entry;
 519	size_t size;
 520	void *end;
 521	int error;
 522	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 523
 524	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 525		  name_index, name, buffer, (long)buffer_size);
 526
 527	if (!EXT4_I(inode)->i_file_acl)
 528		return -ENODATA;
 529	ea_idebug(inode, "reading block %llu",
 530		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 531	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 532	if (IS_ERR(bh))
 533		return PTR_ERR(bh);
 534	ea_bdebug(bh, "b_count=%d, refcount=%d",
 535		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 536	error = ext4_xattr_check_block(inode, bh);
 537	if (error)
 538		goto cleanup;
 539	ext4_xattr_block_cache_insert(ea_block_cache, bh);
 540	entry = BFIRST(bh);
 541	end = bh->b_data + bh->b_size;
 542	error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 543	if (error)
 544		goto cleanup;
 545	size = le32_to_cpu(entry->e_value_size);
 546	error = -ERANGE;
 547	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 548		goto cleanup;
 549	if (buffer) {
 550		if (size > buffer_size)
 551			goto cleanup;
 552		if (entry->e_value_inum) {
 553			error = ext4_xattr_inode_get(inode, entry, buffer,
 554						     size);
 555			if (error)
 556				goto cleanup;
 557		} else {
 558			u16 offset = le16_to_cpu(entry->e_value_offs);
 559			void *p = bh->b_data + offset;
 560
 561			if (unlikely(p + size > end))
 562				goto cleanup;
 563			memcpy(buffer, p, size);
 564		}
 565	}
 566	error = size;
 567
 568cleanup:
 569	brelse(bh);
 570	return error;
 571}
 572
 573int
 574ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 575		     void *buffer, size_t buffer_size)
 576{
 577	struct ext4_xattr_ibody_header *header;
 578	struct ext4_xattr_entry *entry;
 579	struct ext4_inode *raw_inode;
 580	struct ext4_iloc iloc;
 581	size_t size;
 582	void *end;
 583	int error;
 584
 585	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 586		return -ENODATA;
 587	error = ext4_get_inode_loc(inode, &iloc);
 588	if (error)
 589		return error;
 590	raw_inode = ext4_raw_inode(&iloc);
 591	header = IHDR(inode, raw_inode);
 592	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 593	error = xattr_check_inode(inode, header, end);
 594	if (error)
 595		goto cleanup;
 596	entry = IFIRST(header);
 597	error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 598	if (error)
 599		goto cleanup;
 600	size = le32_to_cpu(entry->e_value_size);
 601	error = -ERANGE;
 602	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 603		goto cleanup;
 604	if (buffer) {
 605		if (size > buffer_size)
 606			goto cleanup;
 607		if (entry->e_value_inum) {
 608			error = ext4_xattr_inode_get(inode, entry, buffer,
 609						     size);
 610			if (error)
 611				goto cleanup;
 612		} else {
 613			u16 offset = le16_to_cpu(entry->e_value_offs);
 614			void *p = (void *)IFIRST(header) + offset;
 615
 616			if (unlikely(p + size > end))
 617				goto cleanup;
 618			memcpy(buffer, p, size);
 619		}
 620	}
 621	error = size;
 622
 623cleanup:
 624	brelse(iloc.bh);
 625	return error;
 626}
 627
 628/*
 629 * ext4_xattr_get()
 630 *
 631 * Copy an extended attribute into the buffer
 632 * provided, or compute the buffer size required.
 633 * Buffer is NULL to compute the size of the buffer required.
 634 *
 635 * Returns a negative error number on failure, or the number of bytes
 636 * used / required on success.
 637 */
 638int
 639ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 640	       void *buffer, size_t buffer_size)
 641{
 642	int error;
 643
 644	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 645		return -EIO;
 646
 647	if (strlen(name) > 255)
 648		return -ERANGE;
 649
 650	down_read(&EXT4_I(inode)->xattr_sem);
 651	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 652				     buffer_size);
 653	if (error == -ENODATA)
 654		error = ext4_xattr_block_get(inode, name_index, name, buffer,
 655					     buffer_size);
 656	up_read(&EXT4_I(inode)->xattr_sem);
 657	return error;
 658}
 659
 660static int
 661ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 662			char *buffer, size_t buffer_size)
 663{
 664	size_t rest = buffer_size;
 665
 666	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 667		const struct xattr_handler *handler =
 668			ext4_xattr_handler(entry->e_name_index);
 669
 670		if (handler && (!handler->list || handler->list(dentry))) {
 671			const char *prefix = handler->prefix ?: handler->name;
 672			size_t prefix_len = strlen(prefix);
 673			size_t size = prefix_len + entry->e_name_len + 1;
 674
 675			if (buffer) {
 676				if (size > rest)
 677					return -ERANGE;
 678				memcpy(buffer, prefix, prefix_len);
 679				buffer += prefix_len;
 680				memcpy(buffer, entry->e_name, entry->e_name_len);
 681				buffer += entry->e_name_len;
 682				*buffer++ = 0;
 683			}
 684			rest -= size;
 685		}
 686	}
 687	return buffer_size - rest;  /* total size */
 688}
 689
 690static int
 691ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 692{
 693	struct inode *inode = d_inode(dentry);
 694	struct buffer_head *bh = NULL;
 695	int error;
 696
 697	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 698		  buffer, (long)buffer_size);
 699
 700	if (!EXT4_I(inode)->i_file_acl)
 701		return 0;
 702	ea_idebug(inode, "reading block %llu",
 703		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 704	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 705	if (IS_ERR(bh))
 706		return PTR_ERR(bh);
 707	ea_bdebug(bh, "b_count=%d, refcount=%d",
 708		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 709	error = ext4_xattr_check_block(inode, bh);
 710	if (error)
 711		goto cleanup;
 712	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 713	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 714					buffer_size);
 715cleanup:
 716	brelse(bh);
 717	return error;
 718}
 719
 720static int
 721ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 722{
 723	struct inode *inode = d_inode(dentry);
 724	struct ext4_xattr_ibody_header *header;
 725	struct ext4_inode *raw_inode;
 726	struct ext4_iloc iloc;
 727	void *end;
 728	int error;
 729
 730	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 731		return 0;
 732	error = ext4_get_inode_loc(inode, &iloc);
 733	if (error)
 734		return error;
 735	raw_inode = ext4_raw_inode(&iloc);
 736	header = IHDR(inode, raw_inode);
 737	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 738	error = xattr_check_inode(inode, header, end);
 739	if (error)
 740		goto cleanup;
 741	error = ext4_xattr_list_entries(dentry, IFIRST(header),
 742					buffer, buffer_size);
 743
 744cleanup:
 745	brelse(iloc.bh);
 746	return error;
 747}
 748
 749/*
 750 * Inode operation listxattr()
 751 *
 752 * d_inode(dentry)->i_rwsem: don't care
 753 *
 754 * Copy a list of attribute names into the buffer
 755 * provided, or compute the buffer size required.
 756 * Buffer is NULL to compute the size of the buffer required.
 757 *
 758 * Returns a negative error number on failure, or the number of bytes
 759 * used / required on success.
 760 */
 761ssize_t
 762ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 763{
 764	int ret, ret2;
 765
 766	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 767	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 768	if (ret < 0)
 769		goto errout;
 770	if (buffer) {
 771		buffer += ret;
 772		buffer_size -= ret;
 773	}
 774	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 775	if (ret < 0)
 776		goto errout;
 777	ret += ret2;
 778errout:
 779	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 780	return ret;
 781}
 782
 783/*
 784 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 785 * not set, set it.
 786 */
 787static void ext4_xattr_update_super_block(handle_t *handle,
 788					  struct super_block *sb)
 789{
 790	if (ext4_has_feature_xattr(sb))
 791		return;
 792
 793	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 794	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 
 
 795		ext4_set_feature_xattr(sb);
 796		ext4_handle_dirty_super(handle, sb);
 
 
 797	}
 798}
 799
 800int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 801{
 802	struct ext4_iloc iloc = { .bh = NULL };
 803	struct buffer_head *bh = NULL;
 804	struct ext4_inode *raw_inode;
 805	struct ext4_xattr_ibody_header *header;
 806	struct ext4_xattr_entry *entry;
 807	qsize_t ea_inode_refs = 0;
 808	void *end;
 809	int ret;
 810
 811	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 812
 813	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 814		ret = ext4_get_inode_loc(inode, &iloc);
 815		if (ret)
 816			goto out;
 817		raw_inode = ext4_raw_inode(&iloc);
 818		header = IHDR(inode, raw_inode);
 819		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 820		ret = xattr_check_inode(inode, header, end);
 821		if (ret)
 822			goto out;
 823
 824		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 825		     entry = EXT4_XATTR_NEXT(entry))
 826			if (entry->e_value_inum)
 827				ea_inode_refs++;
 828	}
 829
 830	if (EXT4_I(inode)->i_file_acl) {
 831		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 832		if (IS_ERR(bh)) {
 833			ret = PTR_ERR(bh);
 834			bh = NULL;
 835			goto out;
 836		}
 837
 838		ret = ext4_xattr_check_block(inode, bh);
 839		if (ret)
 840			goto out;
 841
 842		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 843		     entry = EXT4_XATTR_NEXT(entry))
 844			if (entry->e_value_inum)
 845				ea_inode_refs++;
 846	}
 847	*usage = ea_inode_refs + 1;
 848	ret = 0;
 849out:
 850	brelse(iloc.bh);
 851	brelse(bh);
 852	return ret;
 853}
 854
 855static inline size_t round_up_cluster(struct inode *inode, size_t length)
 856{
 857	struct super_block *sb = inode->i_sb;
 858	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 859				    inode->i_blkbits);
 860	size_t mask = ~(cluster_size - 1);
 861
 862	return (length + cluster_size - 1) & mask;
 863}
 864
 865static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 866{
 867	int err;
 868
 869	err = dquot_alloc_inode(inode);
 870	if (err)
 871		return err;
 872	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 873	if (err)
 874		dquot_free_inode(inode);
 875	return err;
 876}
 877
 878static void ext4_xattr_inode_free_quota(struct inode *parent,
 879					struct inode *ea_inode,
 880					size_t len)
 881{
 882	if (ea_inode &&
 883	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 884		return;
 885	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 886	dquot_free_inode(parent);
 887}
 888
 889int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 890			     struct buffer_head *block_bh, size_t value_len,
 891			     bool is_create)
 892{
 893	int credits;
 894	int blocks;
 895
 896	/*
 897	 * 1) Owner inode update
 898	 * 2) Ref count update on old xattr block
 899	 * 3) new xattr block
 900	 * 4) block bitmap update for new xattr block
 901	 * 5) group descriptor for new xattr block
 902	 * 6) block bitmap update for old xattr block
 903	 * 7) group descriptor for old block
 904	 *
 905	 * 6 & 7 can happen if we have two racing threads T_a and T_b
 906	 * which are each trying to set an xattr on inodes I_a and I_b
 907	 * which were both initially sharing an xattr block.
 908	 */
 909	credits = 7;
 910
 911	/* Quota updates. */
 912	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 913
 914	/*
 915	 * In case of inline data, we may push out the data to a block,
 916	 * so we need to reserve credits for this eventuality
 917	 */
 918	if (inode && ext4_has_inline_data(inode))
 919		credits += ext4_writepage_trans_blocks(inode) + 1;
 920
 921	/* We are done if ea_inode feature is not enabled. */
 922	if (!ext4_has_feature_ea_inode(sb))
 923		return credits;
 924
 925	/* New ea_inode, inode map, block bitmap, group descriptor. */
 926	credits += 4;
 927
 928	/* Data blocks. */
 929	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 930
 931	/* Indirection block or one level of extent tree. */
 932	blocks += 1;
 933
 934	/* Block bitmap and group descriptor updates for each block. */
 935	credits += blocks * 2;
 936
 937	/* Blocks themselves. */
 938	credits += blocks;
 939
 940	if (!is_create) {
 941		/* Dereference ea_inode holding old xattr value.
 942		 * Old ea_inode, inode map, block bitmap, group descriptor.
 943		 */
 944		credits += 4;
 945
 946		/* Data blocks for old ea_inode. */
 947		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 948
 949		/* Indirection block or one level of extent tree for old
 950		 * ea_inode.
 951		 */
 952		blocks += 1;
 953
 954		/* Block bitmap and group descriptor updates for each block. */
 955		credits += blocks * 2;
 956	}
 957
 958	/* We may need to clone the existing xattr block in which case we need
 959	 * to increment ref counts for existing ea_inodes referenced by it.
 960	 */
 961	if (block_bh) {
 962		struct ext4_xattr_entry *entry = BFIRST(block_bh);
 963
 964		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 965			if (entry->e_value_inum)
 966				/* Ref count update on ea_inode. */
 967				credits += 1;
 968	}
 969	return credits;
 970}
 971
 972static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
 973				       int ref_change)
 974{
 975	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
 976	struct ext4_iloc iloc;
 977	s64 ref_count;
 978	u32 hash;
 979	int ret;
 980
 981	inode_lock(ea_inode);
 982
 983	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
 984	if (ret)
 985		goto out;
 986
 987	ref_count = ext4_xattr_inode_get_ref(ea_inode);
 988	ref_count += ref_change;
 989	ext4_xattr_inode_set_ref(ea_inode, ref_count);
 990
 991	if (ref_change > 0) {
 992		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
 993			  ea_inode->i_ino, ref_count);
 994
 995		if (ref_count == 1) {
 996			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
 997				  ea_inode->i_ino, ea_inode->i_nlink);
 998
 999			set_nlink(ea_inode, 1);
1000			ext4_orphan_del(handle, ea_inode);
1001
1002			if (ea_inode_cache) {
1003				hash = ext4_xattr_inode_get_hash(ea_inode);
1004				mb_cache_entry_create(ea_inode_cache,
1005						      GFP_NOFS, hash,
1006						      ea_inode->i_ino,
1007						      true /* reusable */);
1008			}
1009		}
1010	} else {
1011		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1012			  ea_inode->i_ino, ref_count);
1013
1014		if (ref_count == 0) {
1015			WARN_ONCE(ea_inode->i_nlink != 1,
1016				  "EA inode %lu i_nlink=%u",
1017				  ea_inode->i_ino, ea_inode->i_nlink);
1018
1019			clear_nlink(ea_inode);
1020			ext4_orphan_add(handle, ea_inode);
1021
1022			if (ea_inode_cache) {
1023				hash = ext4_xattr_inode_get_hash(ea_inode);
1024				mb_cache_entry_delete(ea_inode_cache, hash,
1025						      ea_inode->i_ino);
1026			}
1027		}
1028	}
1029
1030	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1031	if (ret)
1032		ext4_warning_inode(ea_inode,
1033				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1034out:
1035	inode_unlock(ea_inode);
1036	return ret;
1037}
1038
1039static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1040{
1041	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1042}
1043
1044static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1045{
1046	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1047}
1048
1049static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1050					struct ext4_xattr_entry *first)
1051{
1052	struct inode *ea_inode;
1053	struct ext4_xattr_entry *entry;
1054	struct ext4_xattr_entry *failed_entry;
1055	unsigned int ea_ino;
1056	int err, saved_err;
1057
1058	for (entry = first; !IS_LAST_ENTRY(entry);
1059	     entry = EXT4_XATTR_NEXT(entry)) {
1060		if (!entry->e_value_inum)
1061			continue;
1062		ea_ino = le32_to_cpu(entry->e_value_inum);
1063		err = ext4_xattr_inode_iget(parent, ea_ino,
1064					    le32_to_cpu(entry->e_hash),
1065					    &ea_inode);
1066		if (err)
1067			goto cleanup;
1068		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1069		if (err) {
1070			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1071			iput(ea_inode);
1072			goto cleanup;
1073		}
1074		iput(ea_inode);
1075	}
1076	return 0;
1077
1078cleanup:
1079	saved_err = err;
1080	failed_entry = entry;
1081
1082	for (entry = first; entry != failed_entry;
1083	     entry = EXT4_XATTR_NEXT(entry)) {
1084		if (!entry->e_value_inum)
1085			continue;
1086		ea_ino = le32_to_cpu(entry->e_value_inum);
1087		err = ext4_xattr_inode_iget(parent, ea_ino,
1088					    le32_to_cpu(entry->e_hash),
1089					    &ea_inode);
1090		if (err) {
1091			ext4_warning(parent->i_sb,
1092				     "cleanup ea_ino %u iget error %d", ea_ino,
1093				     err);
1094			continue;
1095		}
1096		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1097		if (err)
1098			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1099					   err);
1100		iput(ea_inode);
1101	}
1102	return saved_err;
1103}
1104
1105static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1106			struct buffer_head *bh, bool block_csum, bool dirty)
1107{
1108	int error;
1109
1110	if (bh && dirty) {
1111		if (block_csum)
1112			ext4_xattr_block_csum_set(inode, bh);
1113		error = ext4_handle_dirty_metadata(handle, NULL, bh);
1114		if (error) {
1115			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1116				     error);
1117			return error;
1118		}
1119	}
1120	return 0;
1121}
1122
1123static void
1124ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1125			     struct buffer_head *bh,
1126			     struct ext4_xattr_entry *first, bool block_csum,
1127			     struct ext4_xattr_inode_array **ea_inode_array,
1128			     int extra_credits, bool skip_quota)
1129{
1130	struct inode *ea_inode;
1131	struct ext4_xattr_entry *entry;
1132	bool dirty = false;
1133	unsigned int ea_ino;
1134	int err;
1135	int credits;
1136
1137	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1138	credits = 2 + extra_credits;
1139
1140	for (entry = first; !IS_LAST_ENTRY(entry);
1141	     entry = EXT4_XATTR_NEXT(entry)) {
1142		if (!entry->e_value_inum)
1143			continue;
1144		ea_ino = le32_to_cpu(entry->e_value_inum);
1145		err = ext4_xattr_inode_iget(parent, ea_ino,
1146					    le32_to_cpu(entry->e_hash),
1147					    &ea_inode);
1148		if (err)
1149			continue;
1150
1151		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1152		if (err) {
1153			ext4_warning_inode(ea_inode,
1154					   "Expand inode array err=%d", err);
1155			iput(ea_inode);
1156			continue;
1157		}
1158
1159		err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1160			ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1161			ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1162					      dirty));
1163		if (err < 0) {
1164			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1165					   err);
1166			continue;
1167		}
1168		if (err > 0) {
1169			err = ext4_journal_get_write_access(handle, bh);
 
1170			if (err) {
1171				ext4_warning_inode(ea_inode,
1172						"Re-get write access err=%d",
1173						err);
1174				continue;
1175			}
1176		}
1177
1178		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1179		if (err) {
1180			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1181					   err);
1182			continue;
1183		}
1184
1185		if (!skip_quota)
1186			ext4_xattr_inode_free_quota(parent, ea_inode,
1187					      le32_to_cpu(entry->e_value_size));
1188
1189		/*
1190		 * Forget about ea_inode within the same transaction that
1191		 * decrements the ref count. This avoids duplicate decrements in
1192		 * case the rest of the work spills over to subsequent
1193		 * transactions.
1194		 */
1195		entry->e_value_inum = 0;
1196		entry->e_value_size = 0;
1197
1198		dirty = true;
1199	}
1200
1201	if (dirty) {
1202		/*
1203		 * Note that we are deliberately skipping csum calculation for
1204		 * the final update because we do not expect any journal
1205		 * restarts until xattr block is freed.
1206		 */
1207
1208		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1209		if (err)
1210			ext4_warning_inode(parent,
1211					   "handle dirty metadata err=%d", err);
1212	}
1213}
1214
1215/*
1216 * Release the xattr block BH: If the reference count is > 1, decrement it;
1217 * otherwise free the block.
1218 */
1219static void
1220ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1221			 struct buffer_head *bh,
1222			 struct ext4_xattr_inode_array **ea_inode_array,
1223			 int extra_credits)
1224{
1225	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1226	u32 hash, ref;
1227	int error = 0;
1228
1229	BUFFER_TRACE(bh, "get_write_access");
1230	error = ext4_journal_get_write_access(handle, bh);
 
1231	if (error)
1232		goto out;
1233
 
1234	lock_buffer(bh);
1235	hash = le32_to_cpu(BHDR(bh)->h_hash);
1236	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1237	if (ref == 1) {
1238		ea_bdebug(bh, "refcount now=0; freeing");
1239		/*
1240		 * This must happen under buffer lock for
1241		 * ext4_xattr_block_set() to reliably detect freed block
1242		 */
1243		if (ea_block_cache)
1244			mb_cache_entry_delete(ea_block_cache, hash,
1245					      bh->b_blocknr);
 
 
 
 
 
 
 
 
 
1246		get_bh(bh);
1247		unlock_buffer(bh);
1248
1249		if (ext4_has_feature_ea_inode(inode->i_sb))
1250			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1251						     BFIRST(bh),
1252						     true /* block_csum */,
1253						     ea_inode_array,
1254						     extra_credits,
1255						     true /* skip_quota */);
1256		ext4_free_blocks(handle, inode, bh, 0, 1,
1257				 EXT4_FREE_BLOCKS_METADATA |
1258				 EXT4_FREE_BLOCKS_FORGET);
1259	} else {
1260		ref--;
1261		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1262		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1263			struct mb_cache_entry *ce;
1264
1265			if (ea_block_cache) {
1266				ce = mb_cache_entry_get(ea_block_cache, hash,
1267							bh->b_blocknr);
1268				if (ce) {
1269					ce->e_reusable = 1;
1270					mb_cache_entry_put(ea_block_cache, ce);
1271				}
1272			}
1273		}
1274
1275		ext4_xattr_block_csum_set(inode, bh);
1276		/*
1277		 * Beware of this ugliness: Releasing of xattr block references
1278		 * from different inodes can race and so we have to protect
1279		 * from a race where someone else frees the block (and releases
1280		 * its journal_head) before we are done dirtying the buffer. In
1281		 * nojournal mode this race is harmless and we actually cannot
1282		 * call ext4_handle_dirty_metadata() with locked buffer as
1283		 * that function can call sync_dirty_buffer() so for that case
1284		 * we handle the dirtying after unlocking the buffer.
1285		 */
1286		if (ext4_handle_valid(handle))
1287			error = ext4_handle_dirty_metadata(handle, inode, bh);
1288		unlock_buffer(bh);
1289		if (!ext4_handle_valid(handle))
1290			error = ext4_handle_dirty_metadata(handle, inode, bh);
1291		if (IS_SYNC(inode))
1292			ext4_handle_sync(handle);
1293		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1294		ea_bdebug(bh, "refcount now=%d; releasing",
1295			  le32_to_cpu(BHDR(bh)->h_refcount));
1296	}
1297out:
1298	ext4_std_error(inode->i_sb, error);
1299	return;
1300}
1301
1302/*
1303 * Find the available free space for EAs. This also returns the total number of
1304 * bytes used by EA entries.
1305 */
1306static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1307				    size_t *min_offs, void *base, int *total)
1308{
1309	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1310		if (!last->e_value_inum && last->e_value_size) {
1311			size_t offs = le16_to_cpu(last->e_value_offs);
1312			if (offs < *min_offs)
1313				*min_offs = offs;
1314		}
1315		if (total)
1316			*total += EXT4_XATTR_LEN(last->e_name_len);
1317	}
1318	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1319}
1320
1321/*
1322 * Write the value of the EA in an inode.
1323 */
1324static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1325				  const void *buf, int bufsize)
1326{
1327	struct buffer_head *bh = NULL;
1328	unsigned long block = 0;
1329	int blocksize = ea_inode->i_sb->s_blocksize;
1330	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1331	int csize, wsize = 0;
1332	int ret = 0, ret2 = 0;
1333	int retries = 0;
1334
1335retry:
1336	while (ret >= 0 && ret < max_blocks) {
1337		struct ext4_map_blocks map;
1338		map.m_lblk = block += ret;
1339		map.m_len = max_blocks -= ret;
1340
1341		ret = ext4_map_blocks(handle, ea_inode, &map,
1342				      EXT4_GET_BLOCKS_CREATE);
1343		if (ret <= 0) {
1344			ext4_mark_inode_dirty(handle, ea_inode);
1345			if (ret == -ENOSPC &&
1346			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1347				ret = 0;
1348				goto retry;
1349			}
1350			break;
1351		}
1352	}
1353
1354	if (ret < 0)
1355		return ret;
1356
1357	block = 0;
1358	while (wsize < bufsize) {
1359		brelse(bh);
1360		csize = (bufsize - wsize) > blocksize ? blocksize :
1361								bufsize - wsize;
1362		bh = ext4_getblk(handle, ea_inode, block, 0);
1363		if (IS_ERR(bh))
1364			return PTR_ERR(bh);
1365		if (!bh) {
1366			WARN_ON_ONCE(1);
1367			EXT4_ERROR_INODE(ea_inode,
1368					 "ext4_getblk() return bh = NULL");
1369			return -EFSCORRUPTED;
1370		}
1371		ret = ext4_journal_get_write_access(handle, bh);
 
1372		if (ret)
1373			goto out;
1374
1375		memcpy(bh->b_data, buf, csize);
1376		set_buffer_uptodate(bh);
1377		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1378
1379		buf += csize;
1380		wsize += csize;
1381		block += 1;
1382	}
1383
1384	inode_lock(ea_inode);
1385	i_size_write(ea_inode, wsize);
1386	ext4_update_i_disksize(ea_inode, wsize);
1387	inode_unlock(ea_inode);
1388
1389	ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1390	if (unlikely(ret2 && !ret))
1391		ret = ret2;
1392
1393out:
1394	brelse(bh);
1395
1396	return ret;
1397}
1398
1399/*
1400 * Create an inode to store the value of a large EA.
1401 */
1402static struct inode *ext4_xattr_inode_create(handle_t *handle,
1403					     struct inode *inode, u32 hash)
1404{
1405	struct inode *ea_inode = NULL;
1406	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1407	int err;
1408
1409	/*
1410	 * Let the next inode be the goal, so we try and allocate the EA inode
1411	 * in the same group, or nearby one.
1412	 */
1413	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1414				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1415				  EXT4_EA_INODE_FL);
1416	if (!IS_ERR(ea_inode)) {
1417		ea_inode->i_op = &ext4_file_inode_operations;
1418		ea_inode->i_fop = &ext4_file_operations;
1419		ext4_set_aops(ea_inode);
1420		ext4_xattr_inode_set_class(ea_inode);
1421		unlock_new_inode(ea_inode);
1422		ext4_xattr_inode_set_ref(ea_inode, 1);
1423		ext4_xattr_inode_set_hash(ea_inode, hash);
1424		err = ext4_mark_inode_dirty(handle, ea_inode);
1425		if (!err)
1426			err = ext4_inode_attach_jinode(ea_inode);
1427		if (err) {
 
 
 
1428			iput(ea_inode);
1429			return ERR_PTR(err);
1430		}
1431
1432		/*
1433		 * Xattr inodes are shared therefore quota charging is performed
1434		 * at a higher level.
1435		 */
1436		dquot_free_inode(ea_inode);
1437		dquot_drop(ea_inode);
1438		inode_lock(ea_inode);
1439		ea_inode->i_flags |= S_NOQUOTA;
1440		inode_unlock(ea_inode);
1441	}
1442
1443	return ea_inode;
1444}
1445
1446static struct inode *
1447ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1448			    size_t value_len, u32 hash)
1449{
1450	struct inode *ea_inode;
1451	struct mb_cache_entry *ce;
1452	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1453	void *ea_data;
1454
1455	if (!ea_inode_cache)
1456		return NULL;
1457
1458	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1459	if (!ce)
1460		return NULL;
1461
 
 
 
1462	ea_data = kvmalloc(value_len, GFP_KERNEL);
1463	if (!ea_data) {
1464		mb_cache_entry_put(ea_inode_cache, ce);
1465		return NULL;
1466	}
1467
1468	while (ce) {
1469		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1470				     EXT4_IGET_NORMAL);
1471		if (!IS_ERR(ea_inode) &&
1472		    !is_bad_inode(ea_inode) &&
1473		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1474		    i_size_read(ea_inode) == value_len &&
1475		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1476		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1477						    value_len) &&
1478		    !memcmp(value, ea_data, value_len)) {
1479			mb_cache_entry_touch(ea_inode_cache, ce);
1480			mb_cache_entry_put(ea_inode_cache, ce);
1481			kvfree(ea_data);
1482			return ea_inode;
1483		}
1484
1485		if (!IS_ERR(ea_inode))
1486			iput(ea_inode);
1487		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1488	}
1489	kvfree(ea_data);
1490	return NULL;
1491}
1492
1493/*
1494 * Add value of the EA in an inode.
1495 */
1496static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1497					  const void *value, size_t value_len,
1498					  struct inode **ret_inode)
1499{
1500	struct inode *ea_inode;
1501	u32 hash;
1502	int err;
1503
1504	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1505	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1506	if (ea_inode) {
1507		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1508		if (err) {
1509			iput(ea_inode);
1510			return err;
1511		}
1512
1513		*ret_inode = ea_inode;
1514		return 0;
1515	}
1516
1517	/* Create an inode for the EA value */
1518	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1519	if (IS_ERR(ea_inode))
1520		return PTR_ERR(ea_inode);
1521
1522	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1523	if (err) {
1524		ext4_xattr_inode_dec_ref(handle, ea_inode);
 
1525		iput(ea_inode);
1526		return err;
1527	}
1528
1529	if (EA_INODE_CACHE(inode))
1530		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1531				      ea_inode->i_ino, true /* reusable */);
1532
1533	*ret_inode = ea_inode;
1534	return 0;
1535}
1536
1537/*
1538 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1539 * feature is enabled.
1540 */
1541#define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1542
1543static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1544				struct ext4_xattr_search *s,
1545				handle_t *handle, struct inode *inode,
1546				bool is_block)
1547{
1548	struct ext4_xattr_entry *last, *next;
1549	struct ext4_xattr_entry *here = s->here;
1550	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1551	int in_inode = i->in_inode;
1552	struct inode *old_ea_inode = NULL;
1553	struct inode *new_ea_inode = NULL;
1554	size_t old_size, new_size;
1555	int ret;
1556
1557	/* Space used by old and new values. */
1558	old_size = (!s->not_found && !here->e_value_inum) ?
1559			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1560	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1561
1562	/*
1563	 * Optimization for the simple case when old and new values have the
1564	 * same padded sizes. Not applicable if external inodes are involved.
1565	 */
1566	if (new_size && new_size == old_size) {
1567		size_t offs = le16_to_cpu(here->e_value_offs);
1568		void *val = s->base + offs;
1569
1570		here->e_value_size = cpu_to_le32(i->value_len);
1571		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1572			memset(val, 0, new_size);
1573		} else {
1574			memcpy(val, i->value, i->value_len);
1575			/* Clear padding bytes. */
1576			memset(val + i->value_len, 0, new_size - i->value_len);
1577		}
1578		goto update_hash;
1579	}
1580
1581	/* Compute min_offs and last. */
1582	last = s->first;
1583	for (; !IS_LAST_ENTRY(last); last = next) {
1584		next = EXT4_XATTR_NEXT(last);
1585		if ((void *)next >= s->end) {
1586			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1587			ret = -EFSCORRUPTED;
1588			goto out;
1589		}
1590		if (!last->e_value_inum && last->e_value_size) {
1591			size_t offs = le16_to_cpu(last->e_value_offs);
1592			if (offs < min_offs)
1593				min_offs = offs;
1594		}
1595	}
1596
1597	/* Check whether we have enough space. */
1598	if (i->value) {
1599		size_t free;
1600
1601		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1602		if (!s->not_found)
1603			free += EXT4_XATTR_LEN(name_len) + old_size;
1604
1605		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1606			ret = -ENOSPC;
1607			goto out;
1608		}
1609
1610		/*
1611		 * If storing the value in an external inode is an option,
1612		 * reserve space for xattr entries/names in the external
1613		 * attribute block so that a long value does not occupy the
1614		 * whole space and prevent futher entries being added.
1615		 */
1616		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1617		    new_size && is_block &&
1618		    (min_offs + old_size - new_size) <
1619					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1620			ret = -ENOSPC;
1621			goto out;
1622		}
1623	}
1624
1625	/*
1626	 * Getting access to old and new ea inodes is subject to failures.
1627	 * Finish that work before doing any modifications to the xattr data.
1628	 */
1629	if (!s->not_found && here->e_value_inum) {
1630		ret = ext4_xattr_inode_iget(inode,
1631					    le32_to_cpu(here->e_value_inum),
1632					    le32_to_cpu(here->e_hash),
1633					    &old_ea_inode);
1634		if (ret) {
1635			old_ea_inode = NULL;
1636			goto out;
1637		}
1638	}
1639	if (i->value && in_inode) {
1640		WARN_ON_ONCE(!i->value_len);
1641
1642		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1643		if (ret)
1644			goto out;
1645
1646		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1647						     i->value_len,
1648						     &new_ea_inode);
1649		if (ret) {
1650			new_ea_inode = NULL;
1651			ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1652			goto out;
1653		}
1654	}
1655
1656	if (old_ea_inode) {
1657		/* We are ready to release ref count on the old_ea_inode. */
1658		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1659		if (ret) {
1660			/* Release newly required ref count on new_ea_inode. */
1661			if (new_ea_inode) {
1662				int err;
1663
1664				err = ext4_xattr_inode_dec_ref(handle,
1665							       new_ea_inode);
1666				if (err)
1667					ext4_warning_inode(new_ea_inode,
1668						  "dec ref new_ea_inode err=%d",
1669						  err);
1670				ext4_xattr_inode_free_quota(inode, new_ea_inode,
1671							    i->value_len);
1672			}
1673			goto out;
1674		}
1675
1676		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1677					    le32_to_cpu(here->e_value_size));
1678	}
1679
1680	/* No failures allowed past this point. */
1681
1682	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1683		/* Remove the old value. */
1684		void *first_val = s->base + min_offs;
1685		size_t offs = le16_to_cpu(here->e_value_offs);
1686		void *val = s->base + offs;
1687
1688		memmove(first_val + old_size, first_val, val - first_val);
1689		memset(first_val, 0, old_size);
1690		min_offs += old_size;
1691
1692		/* Adjust all value offsets. */
1693		last = s->first;
1694		while (!IS_LAST_ENTRY(last)) {
1695			size_t o = le16_to_cpu(last->e_value_offs);
1696
1697			if (!last->e_value_inum &&
1698			    last->e_value_size && o < offs)
1699				last->e_value_offs = cpu_to_le16(o + old_size);
1700			last = EXT4_XATTR_NEXT(last);
1701		}
1702	}
1703
1704	if (!i->value) {
1705		/* Remove old name. */
1706		size_t size = EXT4_XATTR_LEN(name_len);
1707
1708		last = ENTRY((void *)last - size);
1709		memmove(here, (void *)here + size,
1710			(void *)last - (void *)here + sizeof(__u32));
1711		memset(last, 0, size);
1712	} else if (s->not_found) {
1713		/* Insert new name. */
1714		size_t size = EXT4_XATTR_LEN(name_len);
1715		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1716
1717		memmove((void *)here + size, here, rest);
1718		memset(here, 0, size);
1719		here->e_name_index = i->name_index;
1720		here->e_name_len = name_len;
1721		memcpy(here->e_name, i->name, name_len);
1722	} else {
1723		/* This is an update, reset value info. */
1724		here->e_value_inum = 0;
1725		here->e_value_offs = 0;
1726		here->e_value_size = 0;
1727	}
1728
1729	if (i->value) {
1730		/* Insert new value. */
1731		if (in_inode) {
1732			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1733		} else if (i->value_len) {
1734			void *val = s->base + min_offs - new_size;
1735
1736			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1737			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1738				memset(val, 0, new_size);
1739			} else {
1740				memcpy(val, i->value, i->value_len);
1741				/* Clear padding bytes. */
1742				memset(val + i->value_len, 0,
1743				       new_size - i->value_len);
1744			}
1745		}
1746		here->e_value_size = cpu_to_le32(i->value_len);
1747	}
1748
1749update_hash:
1750	if (i->value) {
1751		__le32 hash = 0;
1752
1753		/* Entry hash calculation. */
1754		if (in_inode) {
1755			__le32 crc32c_hash;
1756
1757			/*
1758			 * Feed crc32c hash instead of the raw value for entry
1759			 * hash calculation. This is to avoid walking
1760			 * potentially long value buffer again.
1761			 */
1762			crc32c_hash = cpu_to_le32(
1763				       ext4_xattr_inode_get_hash(new_ea_inode));
1764			hash = ext4_xattr_hash_entry(here->e_name,
1765						     here->e_name_len,
1766						     &crc32c_hash, 1);
1767		} else if (is_block) {
1768			__le32 *value = s->base + le16_to_cpu(
1769							here->e_value_offs);
1770
1771			hash = ext4_xattr_hash_entry(here->e_name,
1772						     here->e_name_len, value,
1773						     new_size >> 2);
1774		}
1775		here->e_hash = hash;
1776	}
1777
1778	if (is_block)
1779		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1780
1781	ret = 0;
1782out:
1783	iput(old_ea_inode);
1784	iput(new_ea_inode);
1785	return ret;
1786}
1787
1788struct ext4_xattr_block_find {
1789	struct ext4_xattr_search s;
1790	struct buffer_head *bh;
1791};
1792
1793static int
1794ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1795		      struct ext4_xattr_block_find *bs)
1796{
1797	struct super_block *sb = inode->i_sb;
1798	int error;
1799
1800	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1801		  i->name_index, i->name, i->value, (long)i->value_len);
1802
1803	if (EXT4_I(inode)->i_file_acl) {
1804		/* The inode already has an extended attribute block. */
1805		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1806		if (IS_ERR(bs->bh)) {
1807			error = PTR_ERR(bs->bh);
1808			bs->bh = NULL;
1809			return error;
1810		}
1811		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1812			atomic_read(&(bs->bh->b_count)),
1813			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1814		error = ext4_xattr_check_block(inode, bs->bh);
1815		if (error)
1816			return error;
1817		/* Find the named attribute. */
1818		bs->s.base = BHDR(bs->bh);
1819		bs->s.first = BFIRST(bs->bh);
1820		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1821		bs->s.here = bs->s.first;
1822		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1823					 i->name_index, i->name, 1);
1824		if (error && error != -ENODATA)
1825			return error;
1826		bs->s.not_found = error;
1827	}
1828	return 0;
1829}
1830
1831static int
1832ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1833		     struct ext4_xattr_info *i,
1834		     struct ext4_xattr_block_find *bs)
1835{
1836	struct super_block *sb = inode->i_sb;
1837	struct buffer_head *new_bh = NULL;
1838	struct ext4_xattr_search s_copy = bs->s;
1839	struct ext4_xattr_search *s = &s_copy;
1840	struct mb_cache_entry *ce = NULL;
1841	int error = 0;
1842	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1843	struct inode *ea_inode = NULL, *tmp_inode;
1844	size_t old_ea_inode_quota = 0;
1845	unsigned int ea_ino;
1846
1847
1848#define header(x) ((struct ext4_xattr_header *)(x))
1849
1850	if (s->base) {
 
 
1851		BUFFER_TRACE(bs->bh, "get_write_access");
1852		error = ext4_journal_get_write_access(handle, bs->bh);
 
1853		if (error)
1854			goto cleanup;
1855		lock_buffer(bs->bh);
1856
1857		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1858			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1859
1860			/*
1861			 * This must happen under buffer lock for
1862			 * ext4_xattr_block_set() to reliably detect modified
1863			 * block
1864			 */
1865			if (ea_block_cache)
1866				mb_cache_entry_delete(ea_block_cache, hash,
1867						      bs->bh->b_blocknr);
 
 
 
 
 
 
 
 
 
 
 
1868			ea_bdebug(bs->bh, "modifying in-place");
1869			error = ext4_xattr_set_entry(i, s, handle, inode,
1870						     true /* is_block */);
1871			ext4_xattr_block_csum_set(inode, bs->bh);
1872			unlock_buffer(bs->bh);
1873			if (error == -EFSCORRUPTED)
1874				goto bad_block;
1875			if (!error)
1876				error = ext4_handle_dirty_metadata(handle,
1877								   inode,
1878								   bs->bh);
1879			if (error)
1880				goto cleanup;
1881			goto inserted;
1882		} else {
1883			int offset = (char *)s->here - bs->bh->b_data;
 
 
 
 
 
 
 
 
 
 
1884
1885			unlock_buffer(bs->bh);
1886			ea_bdebug(bs->bh, "cloning");
1887			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1888			error = -ENOMEM;
1889			if (s->base == NULL)
 
 
 
 
 
 
 
 
1890				goto cleanup;
1891			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1892			s->first = ENTRY(header(s->base)+1);
1893			header(s->base)->h_refcount = cpu_to_le32(1);
1894			s->here = ENTRY(s->base + offset);
1895			s->end = s->base + bs->bh->b_size;
1896
1897			/*
1898			 * If existing entry points to an xattr inode, we need
1899			 * to prevent ext4_xattr_set_entry() from decrementing
1900			 * ref count on it because the reference belongs to the
1901			 * original block. In this case, make the entry look
1902			 * like it has an empty value.
1903			 */
1904			if (!s->not_found && s->here->e_value_inum) {
1905				ea_ino = le32_to_cpu(s->here->e_value_inum);
1906				error = ext4_xattr_inode_iget(inode, ea_ino,
1907					      le32_to_cpu(s->here->e_hash),
1908					      &tmp_inode);
1909				if (error)
1910					goto cleanup;
1911
1912				if (!ext4_test_inode_state(tmp_inode,
1913						EXT4_STATE_LUSTRE_EA_INODE)) {
1914					/*
1915					 * Defer quota free call for previous
1916					 * inode until success is guaranteed.
1917					 */
1918					old_ea_inode_quota = le32_to_cpu(
1919							s->here->e_value_size);
1920				}
1921				iput(tmp_inode);
1922
1923				s->here->e_value_inum = 0;
1924				s->here->e_value_size = 0;
1925			}
1926		}
1927	} else {
1928		/* Allocate a buffer where we construct the new block. */
1929		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1930		/* assert(header == s->base) */
1931		error = -ENOMEM;
1932		if (s->base == NULL)
1933			goto cleanup;
1934		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1935		header(s->base)->h_blocks = cpu_to_le32(1);
1936		header(s->base)->h_refcount = cpu_to_le32(1);
1937		s->first = ENTRY(header(s->base)+1);
1938		s->here = ENTRY(header(s->base)+1);
1939		s->end = s->base + sb->s_blocksize;
1940	}
1941
1942	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1943	if (error == -EFSCORRUPTED)
1944		goto bad_block;
1945	if (error)
1946		goto cleanup;
1947
1948	if (i->value && s->here->e_value_inum) {
1949		/*
1950		 * A ref count on ea_inode has been taken as part of the call to
1951		 * ext4_xattr_set_entry() above. We would like to drop this
1952		 * extra ref but we have to wait until the xattr block is
1953		 * initialized and has its own ref count on the ea_inode.
1954		 */
1955		ea_ino = le32_to_cpu(s->here->e_value_inum);
1956		error = ext4_xattr_inode_iget(inode, ea_ino,
1957					      le32_to_cpu(s->here->e_hash),
1958					      &ea_inode);
1959		if (error) {
1960			ea_inode = NULL;
1961			goto cleanup;
1962		}
1963	}
1964
1965inserted:
1966	if (!IS_LAST_ENTRY(s->first)) {
1967		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1968						     &ce);
1969		if (new_bh) {
1970			/* We found an identical block in the cache. */
1971			if (new_bh == bs->bh)
1972				ea_bdebug(new_bh, "keeping");
1973			else {
1974				u32 ref;
1975
1976				WARN_ON_ONCE(dquot_initialize_needed(inode));
1977
1978				/* The old block is released after updating
1979				   the inode. */
1980				error = dquot_alloc_block(inode,
1981						EXT4_C2B(EXT4_SB(sb), 1));
1982				if (error)
1983					goto cleanup;
1984				BUFFER_TRACE(new_bh, "get_write_access");
1985				error = ext4_journal_get_write_access(handle,
1986								      new_bh);
 
1987				if (error)
1988					goto cleanup_dquot;
1989				lock_buffer(new_bh);
1990				/*
1991				 * We have to be careful about races with
1992				 * freeing, rehashing or adding references to
1993				 * xattr block. Once we hold buffer lock xattr
1994				 * block's state is stable so we can check
1995				 * whether the block got freed / rehashed or
1996				 * not.  Since we unhash mbcache entry under
1997				 * buffer lock when freeing / rehashing xattr
1998				 * block, checking whether entry is still
1999				 * hashed is reliable. Same rules hold for
2000				 * e_reusable handling.
2001				 */
2002				if (hlist_bl_unhashed(&ce->e_hash_list) ||
2003				    !ce->e_reusable) {
2004					/*
2005					 * Undo everything and check mbcache
2006					 * again.
2007					 */
2008					unlock_buffer(new_bh);
2009					dquot_free_block(inode,
2010							 EXT4_C2B(EXT4_SB(sb),
2011								  1));
2012					brelse(new_bh);
2013					mb_cache_entry_put(ea_block_cache, ce);
2014					ce = NULL;
2015					new_bh = NULL;
2016					goto inserted;
2017				}
2018				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2019				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2020				if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2021					ce->e_reusable = 0;
2022				ea_bdebug(new_bh, "reusing; refcount now=%d",
2023					  ref);
2024				ext4_xattr_block_csum_set(inode, new_bh);
2025				unlock_buffer(new_bh);
2026				error = ext4_handle_dirty_metadata(handle,
2027								   inode,
2028								   new_bh);
2029				if (error)
2030					goto cleanup_dquot;
2031			}
2032			mb_cache_entry_touch(ea_block_cache, ce);
2033			mb_cache_entry_put(ea_block_cache, ce);
2034			ce = NULL;
2035		} else if (bs->bh && s->base == bs->bh->b_data) {
2036			/* We were modifying this block in-place. */
2037			ea_bdebug(bs->bh, "keeping this block");
2038			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2039			new_bh = bs->bh;
2040			get_bh(new_bh);
2041		} else {
2042			/* We need to allocate a new block */
2043			ext4_fsblk_t goal, block;
2044
2045			WARN_ON_ONCE(dquot_initialize_needed(inode));
2046
2047			goal = ext4_group_first_block_no(sb,
2048						EXT4_I(inode)->i_block_group);
2049
2050			/* non-extent files can't have physical blocks past 2^32 */
2051			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2052				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2053
2054			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2055						     NULL, &error);
2056			if (error)
2057				goto cleanup;
2058
2059			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2060				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2061
2062			ea_idebug(inode, "creating block %llu",
2063				  (unsigned long long)block);
2064
2065			new_bh = sb_getblk(sb, block);
2066			if (unlikely(!new_bh)) {
2067				error = -ENOMEM;
2068getblk_failed:
2069				ext4_free_blocks(handle, inode, NULL, block, 1,
2070						 EXT4_FREE_BLOCKS_METADATA);
2071				goto cleanup;
2072			}
2073			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2074						      ENTRY(header(s->base)+1));
2075			if (error)
2076				goto getblk_failed;
2077			if (ea_inode) {
2078				/* Drop the extra ref on ea_inode. */
2079				error = ext4_xattr_inode_dec_ref(handle,
2080								 ea_inode);
2081				if (error)
2082					ext4_warning_inode(ea_inode,
2083							   "dec ref error=%d",
2084							   error);
2085				iput(ea_inode);
2086				ea_inode = NULL;
2087			}
2088
2089			lock_buffer(new_bh);
2090			error = ext4_journal_get_create_access(handle, new_bh);
 
2091			if (error) {
2092				unlock_buffer(new_bh);
2093				error = -EIO;
2094				goto getblk_failed;
2095			}
2096			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2097			ext4_xattr_block_csum_set(inode, new_bh);
2098			set_buffer_uptodate(new_bh);
2099			unlock_buffer(new_bh);
2100			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2101			error = ext4_handle_dirty_metadata(handle, inode,
2102							   new_bh);
2103			if (error)
2104				goto cleanup;
2105		}
2106	}
2107
2108	if (old_ea_inode_quota)
2109		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2110
2111	/* Update the inode. */
2112	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2113
2114	/* Drop the previous xattr block. */
2115	if (bs->bh && bs->bh != new_bh) {
2116		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2117
2118		ext4_xattr_release_block(handle, inode, bs->bh,
2119					 &ea_inode_array,
2120					 0 /* extra_credits */);
2121		ext4_xattr_inode_array_free(ea_inode_array);
2122	}
2123	error = 0;
2124
2125cleanup:
2126	if (ea_inode) {
2127		int error2;
2128
2129		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2130		if (error2)
2131			ext4_warning_inode(ea_inode, "dec ref error=%d",
2132					   error2);
2133
2134		/* If there was an error, revert the quota charge. */
2135		if (error)
2136			ext4_xattr_inode_free_quota(inode, ea_inode,
2137						    i_size_read(ea_inode));
2138		iput(ea_inode);
2139	}
2140	if (ce)
2141		mb_cache_entry_put(ea_block_cache, ce);
2142	brelse(new_bh);
2143	if (!(bs->bh && s->base == bs->bh->b_data))
2144		kfree(s->base);
2145
2146	return error;
2147
2148cleanup_dquot:
2149	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2150	goto cleanup;
2151
2152bad_block:
2153	EXT4_ERROR_INODE(inode, "bad block %llu",
2154			 EXT4_I(inode)->i_file_acl);
2155	goto cleanup;
2156
2157#undef header
2158}
2159
2160int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2161			  struct ext4_xattr_ibody_find *is)
2162{
2163	struct ext4_xattr_ibody_header *header;
2164	struct ext4_inode *raw_inode;
2165	int error;
2166
2167	if (EXT4_I(inode)->i_extra_isize == 0)
2168		return 0;
 
2169	raw_inode = ext4_raw_inode(&is->iloc);
2170	header = IHDR(inode, raw_inode);
2171	is->s.base = is->s.first = IFIRST(header);
2172	is->s.here = is->s.first;
2173	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2174	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2175		error = xattr_check_inode(inode, header, is->s.end);
2176		if (error)
2177			return error;
2178		/* Find the named attribute. */
2179		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2180					 i->name_index, i->name, 0);
2181		if (error && error != -ENODATA)
2182			return error;
2183		is->s.not_found = error;
2184	}
2185	return 0;
2186}
2187
2188int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2189				struct ext4_xattr_info *i,
2190				struct ext4_xattr_ibody_find *is)
2191{
2192	struct ext4_xattr_ibody_header *header;
2193	struct ext4_xattr_search *s = &is->s;
2194	int error;
2195
2196	if (EXT4_I(inode)->i_extra_isize == 0)
2197		return -ENOSPC;
2198	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2199	if (error)
2200		return error;
2201	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2202	if (!IS_LAST_ENTRY(s->first)) {
2203		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2204		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2205	} else {
2206		header->h_magic = cpu_to_le32(0);
2207		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2208	}
2209	return 0;
2210}
2211
2212static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2213				struct ext4_xattr_info *i,
2214				struct ext4_xattr_ibody_find *is)
2215{
2216	struct ext4_xattr_ibody_header *header;
2217	struct ext4_xattr_search *s = &is->s;
2218	int error;
2219
2220	if (EXT4_I(inode)->i_extra_isize == 0)
2221		return -ENOSPC;
2222	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2223	if (error)
2224		return error;
2225	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2226	if (!IS_LAST_ENTRY(s->first)) {
2227		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2228		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2229	} else {
2230		header->h_magic = cpu_to_le32(0);
2231		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2232	}
2233	return 0;
2234}
2235
2236static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2237				 struct ext4_xattr_info *i)
2238{
2239	void *value;
2240
2241	/* When e_value_inum is set the value is stored externally. */
2242	if (s->here->e_value_inum)
2243		return 0;
2244	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2245		return 0;
2246	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2247	return !memcmp(value, i->value, i->value_len);
2248}
2249
2250static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2251{
2252	struct buffer_head *bh;
2253	int error;
2254
2255	if (!EXT4_I(inode)->i_file_acl)
2256		return NULL;
2257	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2258	if (IS_ERR(bh))
2259		return bh;
2260	error = ext4_xattr_check_block(inode, bh);
2261	if (error) {
2262		brelse(bh);
2263		return ERR_PTR(error);
2264	}
2265	return bh;
2266}
2267
2268/*
2269 * ext4_xattr_set_handle()
2270 *
2271 * Create, replace or remove an extended attribute for this inode.  Value
2272 * is NULL to remove an existing extended attribute, and non-NULL to
2273 * either replace an existing extended attribute, or create a new extended
2274 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2275 * specify that an extended attribute must exist and must not exist
2276 * previous to the call, respectively.
2277 *
2278 * Returns 0, or a negative error number on failure.
2279 */
2280int
2281ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2282		      const char *name, const void *value, size_t value_len,
2283		      int flags)
2284{
2285	struct ext4_xattr_info i = {
2286		.name_index = name_index,
2287		.name = name,
2288		.value = value,
2289		.value_len = value_len,
2290		.in_inode = 0,
2291	};
2292	struct ext4_xattr_ibody_find is = {
2293		.s = { .not_found = -ENODATA, },
2294	};
2295	struct ext4_xattr_block_find bs = {
2296		.s = { .not_found = -ENODATA, },
2297	};
2298	int no_expand;
2299	int error;
2300
2301	if (!name)
2302		return -EINVAL;
2303	if (strlen(name) > 255)
2304		return -ERANGE;
2305
2306	ext4_write_lock_xattr(inode, &no_expand);
2307
2308	/* Check journal credits under write lock. */
2309	if (ext4_handle_valid(handle)) {
2310		struct buffer_head *bh;
2311		int credits;
2312
2313		bh = ext4_xattr_get_block(inode);
2314		if (IS_ERR(bh)) {
2315			error = PTR_ERR(bh);
2316			goto cleanup;
2317		}
2318
2319		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2320						   value_len,
2321						   flags & XATTR_CREATE);
2322		brelse(bh);
2323
2324		if (jbd2_handle_buffer_credits(handle) < credits) {
2325			error = -ENOSPC;
2326			goto cleanup;
2327		}
 
2328	}
2329
2330	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2331	if (error)
2332		goto cleanup;
2333
2334	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2335		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2336		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2337		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2338	}
2339
2340	error = ext4_xattr_ibody_find(inode, &i, &is);
2341	if (error)
2342		goto cleanup;
2343	if (is.s.not_found)
2344		error = ext4_xattr_block_find(inode, &i, &bs);
2345	if (error)
2346		goto cleanup;
2347	if (is.s.not_found && bs.s.not_found) {
2348		error = -ENODATA;
2349		if (flags & XATTR_REPLACE)
2350			goto cleanup;
2351		error = 0;
2352		if (!value)
2353			goto cleanup;
2354	} else {
2355		error = -EEXIST;
2356		if (flags & XATTR_CREATE)
2357			goto cleanup;
2358	}
2359
2360	if (!value) {
2361		if (!is.s.not_found)
2362			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2363		else if (!bs.s.not_found)
2364			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2365	} else {
2366		error = 0;
2367		/* Xattr value did not change? Save us some work and bail out */
2368		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2369			goto cleanup;
2370		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2371			goto cleanup;
2372
2373		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2374		    (EXT4_XATTR_SIZE(i.value_len) >
2375			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2376			i.in_inode = 1;
2377retry_inode:
2378		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379		if (!error && !bs.s.not_found) {
2380			i.value = NULL;
2381			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2382		} else if (error == -ENOSPC) {
2383			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2384				brelse(bs.bh);
2385				bs.bh = NULL;
2386				error = ext4_xattr_block_find(inode, &i, &bs);
2387				if (error)
2388					goto cleanup;
2389			}
2390			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2391			if (!error && !is.s.not_found) {
2392				i.value = NULL;
2393				error = ext4_xattr_ibody_set(handle, inode, &i,
2394							     &is);
2395			} else if (error == -ENOSPC) {
2396				/*
2397				 * Xattr does not fit in the block, store at
2398				 * external inode if possible.
2399				 */
2400				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2401				    !i.in_inode) {
2402					i.in_inode = 1;
2403					goto retry_inode;
2404				}
2405			}
2406		}
2407	}
2408	if (!error) {
2409		ext4_xattr_update_super_block(handle, inode->i_sb);
2410		inode->i_ctime = current_time(inode);
 
2411		if (!value)
2412			no_expand = 0;
2413		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2414		/*
2415		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2416		 * error != 0.
2417		 */
2418		is.iloc.bh = NULL;
2419		if (IS_SYNC(inode))
2420			ext4_handle_sync(handle);
2421	}
 
2422
2423cleanup:
2424	brelse(is.iloc.bh);
2425	brelse(bs.bh);
2426	ext4_write_unlock_xattr(inode, &no_expand);
2427	return error;
2428}
2429
2430int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2431			   bool is_create, int *credits)
2432{
2433	struct buffer_head *bh;
2434	int err;
2435
2436	*credits = 0;
2437
2438	if (!EXT4_SB(inode->i_sb)->s_journal)
2439		return 0;
2440
2441	down_read(&EXT4_I(inode)->xattr_sem);
2442
2443	bh = ext4_xattr_get_block(inode);
2444	if (IS_ERR(bh)) {
2445		err = PTR_ERR(bh);
2446	} else {
2447		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2448						    value_len, is_create);
2449		brelse(bh);
2450		err = 0;
2451	}
2452
2453	up_read(&EXT4_I(inode)->xattr_sem);
2454	return err;
2455}
2456
2457/*
2458 * ext4_xattr_set()
2459 *
2460 * Like ext4_xattr_set_handle, but start from an inode. This extended
2461 * attribute modification is a filesystem transaction by itself.
2462 *
2463 * Returns 0, or a negative error number on failure.
2464 */
2465int
2466ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2467	       const void *value, size_t value_len, int flags)
2468{
2469	handle_t *handle;
2470	struct super_block *sb = inode->i_sb;
2471	int error, retries = 0;
2472	int credits;
2473
2474	error = dquot_initialize(inode);
2475	if (error)
2476		return error;
2477
2478retry:
2479	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2480				       &credits);
2481	if (error)
2482		return error;
2483
2484	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2485	if (IS_ERR(handle)) {
2486		error = PTR_ERR(handle);
2487	} else {
2488		int error2;
2489
2490		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2491					      value, value_len, flags);
2492		error2 = ext4_journal_stop(handle);
2493		if (error == -ENOSPC &&
2494		    ext4_should_retry_alloc(sb, &retries))
2495			goto retry;
2496		if (error == 0)
2497			error = error2;
2498	}
 
2499
2500	return error;
2501}
2502
2503/*
2504 * Shift the EA entries in the inode to create space for the increased
2505 * i_extra_isize.
2506 */
2507static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2508				     int value_offs_shift, void *to,
2509				     void *from, size_t n)
2510{
2511	struct ext4_xattr_entry *last = entry;
2512	int new_offs;
2513
2514	/* We always shift xattr headers further thus offsets get lower */
2515	BUG_ON(value_offs_shift > 0);
2516
2517	/* Adjust the value offsets of the entries */
2518	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2519		if (!last->e_value_inum && last->e_value_size) {
2520			new_offs = le16_to_cpu(last->e_value_offs) +
2521							value_offs_shift;
2522			last->e_value_offs = cpu_to_le16(new_offs);
2523		}
2524	}
2525	/* Shift the entries by n bytes */
2526	memmove(to, from, n);
2527}
2528
2529/*
2530 * Move xattr pointed to by 'entry' from inode into external xattr block
2531 */
2532static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2533				    struct ext4_inode *raw_inode,
2534				    struct ext4_xattr_entry *entry)
2535{
2536	struct ext4_xattr_ibody_find *is = NULL;
2537	struct ext4_xattr_block_find *bs = NULL;
2538	char *buffer = NULL, *b_entry_name = NULL;
2539	size_t value_size = le32_to_cpu(entry->e_value_size);
2540	struct ext4_xattr_info i = {
2541		.value = NULL,
2542		.value_len = 0,
2543		.name_index = entry->e_name_index,
2544		.in_inode = !!entry->e_value_inum,
2545	};
2546	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2547	int error;
2548
2549	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2550	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2551	buffer = kmalloc(value_size, GFP_NOFS);
2552	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2553	if (!is || !bs || !buffer || !b_entry_name) {
2554		error = -ENOMEM;
2555		goto out;
2556	}
2557
2558	is->s.not_found = -ENODATA;
2559	bs->s.not_found = -ENODATA;
2560	is->iloc.bh = NULL;
2561	bs->bh = NULL;
2562
2563	/* Save the entry name and the entry value */
2564	if (entry->e_value_inum) {
2565		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2566		if (error)
2567			goto out;
2568	} else {
2569		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2570		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2571	}
2572
2573	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2574	b_entry_name[entry->e_name_len] = '\0';
2575	i.name = b_entry_name;
2576
2577	error = ext4_get_inode_loc(inode, &is->iloc);
2578	if (error)
2579		goto out;
2580
2581	error = ext4_xattr_ibody_find(inode, &i, is);
2582	if (error)
2583		goto out;
2584
2585	/* Remove the chosen entry from the inode */
2586	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2587	if (error)
2588		goto out;
2589
2590	i.value = buffer;
2591	i.value_len = value_size;
2592	error = ext4_xattr_block_find(inode, &i, bs);
2593	if (error)
2594		goto out;
2595
2596	/* Add entry which was removed from the inode into the block */
2597	error = ext4_xattr_block_set(handle, inode, &i, bs);
2598	if (error)
2599		goto out;
2600	error = 0;
2601out:
2602	kfree(b_entry_name);
2603	kfree(buffer);
2604	if (is)
2605		brelse(is->iloc.bh);
2606	if (bs)
2607		brelse(bs->bh);
2608	kfree(is);
2609	kfree(bs);
2610
2611	return error;
2612}
2613
2614static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2615				       struct ext4_inode *raw_inode,
2616				       int isize_diff, size_t ifree,
2617				       size_t bfree, int *total_ino)
2618{
2619	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2620	struct ext4_xattr_entry *small_entry;
2621	struct ext4_xattr_entry *entry;
2622	struct ext4_xattr_entry *last;
2623	unsigned int entry_size;	/* EA entry size */
2624	unsigned int total_size;	/* EA entry size + value size */
2625	unsigned int min_total_size;
2626	int error;
2627
2628	while (isize_diff > ifree) {
2629		entry = NULL;
2630		small_entry = NULL;
2631		min_total_size = ~0U;
2632		last = IFIRST(header);
2633		/* Find the entry best suited to be pushed into EA block */
2634		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2635			/* never move system.data out of the inode */
2636			if ((last->e_name_len == 4) &&
2637			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2638			    !memcmp(last->e_name, "data", 4))
2639				continue;
2640			total_size = EXT4_XATTR_LEN(last->e_name_len);
2641			if (!last->e_value_inum)
2642				total_size += EXT4_XATTR_SIZE(
2643					       le32_to_cpu(last->e_value_size));
2644			if (total_size <= bfree &&
2645			    total_size < min_total_size) {
2646				if (total_size + ifree < isize_diff) {
2647					small_entry = last;
2648				} else {
2649					entry = last;
2650					min_total_size = total_size;
2651				}
2652			}
2653		}
2654
2655		if (entry == NULL) {
2656			if (small_entry == NULL)
2657				return -ENOSPC;
2658			entry = small_entry;
2659		}
2660
2661		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2662		total_size = entry_size;
2663		if (!entry->e_value_inum)
2664			total_size += EXT4_XATTR_SIZE(
2665					      le32_to_cpu(entry->e_value_size));
2666		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2667						 entry);
2668		if (error)
2669			return error;
2670
2671		*total_ino -= entry_size;
2672		ifree += total_size;
2673		bfree -= total_size;
2674	}
2675
2676	return 0;
2677}
2678
2679/*
2680 * Expand an inode by new_extra_isize bytes when EAs are present.
2681 * Returns 0 on success or negative error number on failure.
2682 */
2683int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2684			       struct ext4_inode *raw_inode, handle_t *handle)
2685{
2686	struct ext4_xattr_ibody_header *header;
2687	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2688	static unsigned int mnt_count;
2689	size_t min_offs;
2690	size_t ifree, bfree;
2691	int total_ino;
2692	void *base, *end;
2693	int error = 0, tried_min_extra_isize = 0;
2694	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2695	int isize_diff;	/* How much do we need to grow i_extra_isize */
2696
2697retry:
2698	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2699	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2700		return 0;
2701
2702	header = IHDR(inode, raw_inode);
2703
2704	/*
2705	 * Check if enough free space is available in the inode to shift the
2706	 * entries ahead by new_extra_isize.
2707	 */
2708
2709	base = IFIRST(header);
2710	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2711	min_offs = end - base;
2712	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2713
2714	error = xattr_check_inode(inode, header, end);
2715	if (error)
2716		goto cleanup;
2717
2718	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2719	if (ifree >= isize_diff)
2720		goto shift;
2721
2722	/*
2723	 * Enough free space isn't available in the inode, check if
2724	 * EA block can hold new_extra_isize bytes.
2725	 */
2726	if (EXT4_I(inode)->i_file_acl) {
2727		struct buffer_head *bh;
2728
2729		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2730		if (IS_ERR(bh)) {
2731			error = PTR_ERR(bh);
2732			goto cleanup;
2733		}
2734		error = ext4_xattr_check_block(inode, bh);
2735		if (error) {
2736			brelse(bh);
2737			goto cleanup;
2738		}
2739		base = BHDR(bh);
2740		end = bh->b_data + bh->b_size;
2741		min_offs = end - base;
2742		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2743					      NULL);
2744		brelse(bh);
2745		if (bfree + ifree < isize_diff) {
2746			if (!tried_min_extra_isize && s_min_extra_isize) {
2747				tried_min_extra_isize++;
2748				new_extra_isize = s_min_extra_isize;
2749				goto retry;
2750			}
2751			error = -ENOSPC;
2752			goto cleanup;
2753		}
2754	} else {
2755		bfree = inode->i_sb->s_blocksize;
2756	}
2757
2758	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2759					    isize_diff, ifree, bfree,
2760					    &total_ino);
2761	if (error) {
2762		if (error == -ENOSPC && !tried_min_extra_isize &&
2763		    s_min_extra_isize) {
2764			tried_min_extra_isize++;
2765			new_extra_isize = s_min_extra_isize;
2766			goto retry;
2767		}
2768		goto cleanup;
2769	}
2770shift:
2771	/* Adjust the offsets and shift the remaining entries ahead */
2772	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2773			- new_extra_isize, (void *)raw_inode +
2774			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2775			(void *)header, total_ino);
2776	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2777
2778cleanup:
2779	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2780		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2781			     inode->i_ino);
2782		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2783	}
2784	return error;
2785}
2786
2787#define EIA_INCR 16 /* must be 2^n */
2788#define EIA_MASK (EIA_INCR - 1)
2789
2790/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2791 * If @ea_inode_array is new or full it will be grown and the old
2792 * contents copied over.
2793 */
2794static int
2795ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2796			struct inode *inode)
2797{
2798	if (*ea_inode_array == NULL) {
2799		/*
2800		 * Start with 15 inodes, so it fits into a power-of-two size.
2801		 * If *ea_inode_array is NULL, this is essentially offsetof()
2802		 */
2803		(*ea_inode_array) =
2804			kmalloc(offsetof(struct ext4_xattr_inode_array,
2805					 inodes[EIA_MASK]),
2806				GFP_NOFS);
2807		if (*ea_inode_array == NULL)
2808			return -ENOMEM;
2809		(*ea_inode_array)->count = 0;
2810	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2811		/* expand the array once all 15 + n * 16 slots are full */
2812		struct ext4_xattr_inode_array *new_array = NULL;
2813		int count = (*ea_inode_array)->count;
2814
2815		/* if new_array is NULL, this is essentially offsetof() */
2816		new_array = kmalloc(
2817				offsetof(struct ext4_xattr_inode_array,
2818					 inodes[count + EIA_INCR]),
2819				GFP_NOFS);
2820		if (new_array == NULL)
2821			return -ENOMEM;
2822		memcpy(new_array, *ea_inode_array,
2823		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2824		kfree(*ea_inode_array);
2825		*ea_inode_array = new_array;
2826	}
2827	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2828	return 0;
2829}
2830
2831/*
2832 * ext4_xattr_delete_inode()
2833 *
2834 * Free extended attribute resources associated with this inode. Traverse
2835 * all entries and decrement reference on any xattr inodes associated with this
2836 * inode. This is called immediately before an inode is freed. We have exclusive
2837 * access to the inode. If an orphan inode is deleted it will also release its
2838 * references on xattr block and xattr inodes.
2839 */
2840int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2841			    struct ext4_xattr_inode_array **ea_inode_array,
2842			    int extra_credits)
2843{
2844	struct buffer_head *bh = NULL;
2845	struct ext4_xattr_ibody_header *header;
2846	struct ext4_iloc iloc = { .bh = NULL };
2847	struct ext4_xattr_entry *entry;
2848	struct inode *ea_inode;
2849	int error;
2850
2851	error = ext4_journal_ensure_credits(handle, extra_credits,
2852			ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2853	if (error < 0) {
2854		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2855		goto cleanup;
2856	}
2857
2858	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2859	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2860
2861		error = ext4_get_inode_loc(inode, &iloc);
2862		if (error) {
2863			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2864			goto cleanup;
2865		}
2866
2867		error = ext4_journal_get_write_access(handle, iloc.bh);
 
2868		if (error) {
2869			EXT4_ERROR_INODE(inode, "write access (error %d)",
2870					 error);
2871			goto cleanup;
2872		}
2873
2874		header = IHDR(inode, ext4_raw_inode(&iloc));
2875		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2876			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2877						     IFIRST(header),
2878						     false /* block_csum */,
2879						     ea_inode_array,
2880						     extra_credits,
2881						     false /* skip_quota */);
2882	}
2883
2884	if (EXT4_I(inode)->i_file_acl) {
2885		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2886		if (IS_ERR(bh)) {
2887			error = PTR_ERR(bh);
2888			if (error == -EIO) {
2889				EXT4_ERROR_INODE_ERR(inode, EIO,
2890						     "block %llu read error",
2891						     EXT4_I(inode)->i_file_acl);
2892			}
2893			bh = NULL;
2894			goto cleanup;
2895		}
2896		error = ext4_xattr_check_block(inode, bh);
2897		if (error)
2898			goto cleanup;
2899
2900		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2901			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2902			     entry = EXT4_XATTR_NEXT(entry)) {
2903				if (!entry->e_value_inum)
2904					continue;
2905				error = ext4_xattr_inode_iget(inode,
2906					      le32_to_cpu(entry->e_value_inum),
2907					      le32_to_cpu(entry->e_hash),
2908					      &ea_inode);
2909				if (error)
2910					continue;
2911				ext4_xattr_inode_free_quota(inode, ea_inode,
2912					      le32_to_cpu(entry->e_value_size));
2913				iput(ea_inode);
2914			}
2915
2916		}
2917
2918		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2919					 extra_credits);
2920		/*
2921		 * Update i_file_acl value in the same transaction that releases
2922		 * block.
2923		 */
2924		EXT4_I(inode)->i_file_acl = 0;
2925		error = ext4_mark_inode_dirty(handle, inode);
2926		if (error) {
2927			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2928					 error);
2929			goto cleanup;
2930		}
 
2931	}
2932	error = 0;
2933cleanup:
2934	brelse(iloc.bh);
2935	brelse(bh);
2936	return error;
2937}
2938
2939void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2940{
2941	int idx;
2942
2943	if (ea_inode_array == NULL)
2944		return;
2945
2946	for (idx = 0; idx < ea_inode_array->count; ++idx)
2947		iput(ea_inode_array->inodes[idx]);
2948	kfree(ea_inode_array);
2949}
2950
2951/*
2952 * ext4_xattr_block_cache_insert()
2953 *
2954 * Create a new entry in the extended attribute block cache, and insert
2955 * it unless such an entry is already in the cache.
2956 *
2957 * Returns 0, or a negative error number on failure.
2958 */
2959static void
2960ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2961			      struct buffer_head *bh)
2962{
2963	struct ext4_xattr_header *header = BHDR(bh);
2964	__u32 hash = le32_to_cpu(header->h_hash);
2965	int reusable = le32_to_cpu(header->h_refcount) <
2966		       EXT4_XATTR_REFCOUNT_MAX;
2967	int error;
2968
2969	if (!ea_block_cache)
2970		return;
2971	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2972				      bh->b_blocknr, reusable);
2973	if (error) {
2974		if (error == -EBUSY)
2975			ea_bdebug(bh, "already in cache");
2976	} else
2977		ea_bdebug(bh, "inserting [%x]", (int)hash);
2978}
2979
2980/*
2981 * ext4_xattr_cmp()
2982 *
2983 * Compare two extended attribute blocks for equality.
2984 *
2985 * Returns 0 if the blocks are equal, 1 if they differ, and
2986 * a negative error number on errors.
2987 */
2988static int
2989ext4_xattr_cmp(struct ext4_xattr_header *header1,
2990	       struct ext4_xattr_header *header2)
2991{
2992	struct ext4_xattr_entry *entry1, *entry2;
2993
2994	entry1 = ENTRY(header1+1);
2995	entry2 = ENTRY(header2+1);
2996	while (!IS_LAST_ENTRY(entry1)) {
2997		if (IS_LAST_ENTRY(entry2))
2998			return 1;
2999		if (entry1->e_hash != entry2->e_hash ||
3000		    entry1->e_name_index != entry2->e_name_index ||
3001		    entry1->e_name_len != entry2->e_name_len ||
3002		    entry1->e_value_size != entry2->e_value_size ||
3003		    entry1->e_value_inum != entry2->e_value_inum ||
3004		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3005			return 1;
3006		if (!entry1->e_value_inum &&
3007		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3008			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3009			   le32_to_cpu(entry1->e_value_size)))
3010			return 1;
3011
3012		entry1 = EXT4_XATTR_NEXT(entry1);
3013		entry2 = EXT4_XATTR_NEXT(entry2);
3014	}
3015	if (!IS_LAST_ENTRY(entry2))
3016		return 1;
3017	return 0;
3018}
3019
3020/*
3021 * ext4_xattr_block_cache_find()
3022 *
3023 * Find an identical extended attribute block.
3024 *
3025 * Returns a pointer to the block found, or NULL if such a block was
3026 * not found or an error occurred.
3027 */
3028static struct buffer_head *
3029ext4_xattr_block_cache_find(struct inode *inode,
3030			    struct ext4_xattr_header *header,
3031			    struct mb_cache_entry **pce)
3032{
3033	__u32 hash = le32_to_cpu(header->h_hash);
3034	struct mb_cache_entry *ce;
3035	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3036
3037	if (!ea_block_cache)
3038		return NULL;
3039	if (!header->h_hash)
3040		return NULL;  /* never share */
3041	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3042	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3043	while (ce) {
3044		struct buffer_head *bh;
3045
3046		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3047		if (IS_ERR(bh)) {
3048			if (PTR_ERR(bh) == -ENOMEM)
3049				return NULL;
3050			bh = NULL;
3051			EXT4_ERROR_INODE(inode, "block %lu read error",
3052					 (unsigned long)ce->e_value);
3053		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3054			*pce = ce;
3055			return bh;
3056		}
3057		brelse(bh);
3058		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3059	}
3060	return NULL;
3061}
3062
3063#define NAME_HASH_SHIFT 5
3064#define VALUE_HASH_SHIFT 16
3065
3066/*
3067 * ext4_xattr_hash_entry()
3068 *
3069 * Compute the hash of an extended attribute.
3070 */
3071static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3072				    size_t value_count)
3073{
3074	__u32 hash = 0;
3075
3076	while (name_len--) {
3077		hash = (hash << NAME_HASH_SHIFT) ^
3078		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3079		       *name++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3080	}
3081	while (value_count--) {
3082		hash = (hash << VALUE_HASH_SHIFT) ^
3083		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3084		       le32_to_cpu(*value++);
3085	}
3086	return cpu_to_le32(hash);
3087}
3088
3089#undef NAME_HASH_SHIFT
3090#undef VALUE_HASH_SHIFT
3091
3092#define BLOCK_HASH_SHIFT 16
3093
3094/*
3095 * ext4_xattr_rehash()
3096 *
3097 * Re-compute the extended attribute hash value after an entry has changed.
3098 */
3099static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3100{
3101	struct ext4_xattr_entry *here;
3102	__u32 hash = 0;
3103
3104	here = ENTRY(header+1);
3105	while (!IS_LAST_ENTRY(here)) {
3106		if (!here->e_hash) {
3107			/* Block is not shared if an entry's hash value == 0 */
3108			hash = 0;
3109			break;
3110		}
3111		hash = (hash << BLOCK_HASH_SHIFT) ^
3112		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3113		       le32_to_cpu(here->e_hash);
3114		here = EXT4_XATTR_NEXT(here);
3115	}
3116	header->h_hash = cpu_to_le32(hash);
3117}
3118
3119#undef BLOCK_HASH_SHIFT
3120
3121#define	HASH_BUCKET_BITS	10
3122
3123struct mb_cache *
3124ext4_xattr_create_cache(void)
3125{
3126	return mb_cache_create(HASH_BUCKET_BITS);
3127}
3128
3129void ext4_xattr_destroy_cache(struct mb_cache *cache)
3130{
3131	if (cache)
3132		mb_cache_destroy(cache);
3133}
3134