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
v5.4
   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};
  97
  98const struct xattr_handler *ext4_xattr_handlers[] = {
  99	&ext4_xattr_user_handler,
 100	&ext4_xattr_trusted_handler,
 101#ifdef CONFIG_EXT4_FS_POSIX_ACL
 102	&posix_acl_access_xattr_handler,
 103	&posix_acl_default_xattr_handler,
 104#endif
 105#ifdef CONFIG_EXT4_FS_SECURITY
 106	&ext4_xattr_security_handler,
 107#endif
 
 108	NULL
 109};
 110
 111#define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
 112				inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114#define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
 115				inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117static int
 118ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119			struct inode *inode);
 120
 121#ifdef CONFIG_LOCKDEP
 122void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123{
 124	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 125}
 126#endif
 127
 128static __le32 ext4_xattr_block_csum(struct inode *inode,
 129				    sector_t block_nr,
 130				    struct ext4_xattr_header *hdr)
 131{
 132	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133	__u32 csum;
 134	__le64 dsk_block_nr = cpu_to_le64(block_nr);
 135	__u32 dummy_csum = 0;
 136	int offset = offsetof(struct ext4_xattr_header, h_checksum);
 137
 138	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 139			   sizeof(dsk_block_nr));
 140	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 141	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 142	offset += sizeof(dummy_csum);
 143	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 144			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 145
 146	return cpu_to_le32(csum);
 147}
 148
 149static int ext4_xattr_block_csum_verify(struct inode *inode,
 150					struct buffer_head *bh)
 151{
 152	struct ext4_xattr_header *hdr = BHDR(bh);
 153	int ret = 1;
 154
 155	if (ext4_has_metadata_csum(inode->i_sb)) {
 156		lock_buffer(bh);
 157		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 158							bh->b_blocknr, hdr));
 159		unlock_buffer(bh);
 160	}
 161	return ret;
 162}
 163
 164static void ext4_xattr_block_csum_set(struct inode *inode,
 165				      struct buffer_head *bh)
 166{
 167	if (ext4_has_metadata_csum(inode->i_sb))
 168		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 169						bh->b_blocknr, BHDR(bh));
 170}
 171
 172static inline const struct xattr_handler *
 173ext4_xattr_handler(int name_index)
 174{
 175	const struct xattr_handler *handler = NULL;
 176
 177	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 178		handler = ext4_xattr_handler_map[name_index];
 179	return handler;
 180}
 181
 182static int
 183ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 184			 void *value_start)
 185{
 186	struct ext4_xattr_entry *e = entry;
 187
 188	/* Find the end of the names list */
 189	while (!IS_LAST_ENTRY(e)) {
 190		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 191		if ((void *)next >= end)
 192			return -EFSCORRUPTED;
 193		if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 194			return -EFSCORRUPTED;
 195		e = next;
 196	}
 197
 198	/* Check the values */
 199	while (!IS_LAST_ENTRY(entry)) {
 200		u32 size = le32_to_cpu(entry->e_value_size);
 201
 202		if (size > EXT4_XATTR_SIZE_MAX)
 203			return -EFSCORRUPTED;
 204
 205		if (size != 0 && entry->e_value_inum == 0) {
 206			u16 offs = le16_to_cpu(entry->e_value_offs);
 207			void *value;
 208
 209			/*
 210			 * The value cannot overlap the names, and the value
 211			 * with padding cannot extend beyond 'end'.  Check both
 212			 * the padded and unpadded sizes, since the size may
 213			 * overflow to 0 when adding padding.
 214			 */
 215			if (offs > end - value_start)
 216				return -EFSCORRUPTED;
 217			value = value_start + offs;
 218			if (value < (void *)e + sizeof(u32) ||
 219			    size > end - value ||
 220			    EXT4_XATTR_SIZE(size) > end - value)
 221				return -EFSCORRUPTED;
 222		}
 223		entry = EXT4_XATTR_NEXT(entry);
 224	}
 225
 226	return 0;
 227}
 228
 229static inline int
 230__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 231			 const char *function, unsigned int line)
 232{
 233	int error = -EFSCORRUPTED;
 234
 235	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 236	    BHDR(bh)->h_blocks != cpu_to_le32(1))
 237		goto errout;
 238	if (buffer_verified(bh))
 239		return 0;
 240
 241	error = -EFSBADCRC;
 242	if (!ext4_xattr_block_csum_verify(inode, bh))
 243		goto errout;
 244	error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 245					 bh->b_data);
 246errout:
 247	if (error)
 248		__ext4_error_inode(inode, function, line, 0,
 249				   "corrupted xattr block %llu",
 250				   (unsigned long long) bh->b_blocknr);
 251	else
 252		set_buffer_verified(bh);
 253	return error;
 254}
 255
 256#define ext4_xattr_check_block(inode, bh) \
 257	__ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 258
 259
 260static int
 261__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 262			 void *end, const char *function, unsigned int line)
 263{
 264	int error = -EFSCORRUPTED;
 265
 266	if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 267	    (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 268		goto errout;
 269	error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 270errout:
 271	if (error)
 272		__ext4_error_inode(inode, function, line, 0,
 273				   "corrupted in-inode xattr");
 274	return error;
 275}
 276
 277#define xattr_check_inode(inode, header, end) \
 278	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 279
 280static int
 281xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 282		 void *end, int name_index, const char *name, int sorted)
 283{
 284	struct ext4_xattr_entry *entry, *next;
 285	size_t name_len;
 286	int cmp = 1;
 287
 288	if (name == NULL)
 289		return -EINVAL;
 290	name_len = strlen(name);
 291	for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 292		next = EXT4_XATTR_NEXT(entry);
 293		if ((void *) next >= end) {
 294			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 295			return -EFSCORRUPTED;
 296		}
 297		cmp = name_index - entry->e_name_index;
 298		if (!cmp)
 299			cmp = name_len - entry->e_name_len;
 300		if (!cmp)
 301			cmp = memcmp(name, entry->e_name, name_len);
 302		if (cmp <= 0 && (sorted || cmp == 0))
 303			break;
 304	}
 305	*pentry = entry;
 306	return cmp ? -ENODATA : 0;
 307}
 308
 309static u32
 310ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 311{
 312	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 313}
 314
 315static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 316{
 317	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 318		(u32) inode_peek_iversion_raw(ea_inode);
 319}
 320
 321static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 322{
 323	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 324	inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 325}
 326
 327static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 328{
 329	return (u32)ea_inode->i_atime.tv_sec;
 330}
 331
 332static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 333{
 334	ea_inode->i_atime.tv_sec = hash;
 335}
 336
 337/*
 338 * Read the EA value from an inode.
 339 */
 340static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 341{
 342	int blocksize = 1 << ea_inode->i_blkbits;
 343	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 344	int tail_size = (size % blocksize) ?: blocksize;
 345	struct buffer_head *bhs_inline[8];
 346	struct buffer_head **bhs = bhs_inline;
 347	int i, ret;
 348
 349	if (bh_count > ARRAY_SIZE(bhs_inline)) {
 350		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 351		if (!bhs)
 352			return -ENOMEM;
 353	}
 354
 355	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 356			       true /* wait */, bhs);
 357	if (ret)
 358		goto free_bhs;
 359
 360	for (i = 0; i < bh_count; i++) {
 361		/* There shouldn't be any holes in ea_inode. */
 362		if (!bhs[i]) {
 363			ret = -EFSCORRUPTED;
 364			goto put_bhs;
 365		}
 366		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 367		       i < bh_count - 1 ? blocksize : tail_size);
 368	}
 369	ret = 0;
 370put_bhs:
 371	for (i = 0; i < bh_count; i++)
 372		brelse(bhs[i]);
 373free_bhs:
 374	if (bhs != bhs_inline)
 375		kfree(bhs);
 376	return ret;
 377}
 378
 379#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 380
 381static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 382				 u32 ea_inode_hash, struct inode **ea_inode)
 383{
 384	struct inode *inode;
 385	int err;
 386
 387	inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 388	if (IS_ERR(inode)) {
 389		err = PTR_ERR(inode);
 390		ext4_error(parent->i_sb,
 391			   "error while reading EA inode %lu err=%d", ea_ino,
 392			   err);
 393		return err;
 394	}
 395
 396	if (is_bad_inode(inode)) {
 397		ext4_error(parent->i_sb,
 398			   "error while reading EA inode %lu is_bad_inode",
 399			   ea_ino);
 400		err = -EIO;
 401		goto error;
 402	}
 403
 404	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 405		ext4_error(parent->i_sb,
 406			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 407			    ea_ino);
 408		err = -EINVAL;
 409		goto error;
 410	}
 411
 412	ext4_xattr_inode_set_class(inode);
 413
 414	/*
 415	 * Check whether this is an old Lustre-style xattr inode. Lustre
 416	 * implementation does not have hash validation, rather it has a
 417	 * backpointer from ea_inode to the parent inode.
 418	 */
 419	if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 420	    EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 421	    inode->i_generation == parent->i_generation) {
 422		ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 423		ext4_xattr_inode_set_ref(inode, 1);
 424	} else {
 425		inode_lock(inode);
 426		inode->i_flags |= S_NOQUOTA;
 427		inode_unlock(inode);
 428	}
 429
 430	*ea_inode = inode;
 431	return 0;
 432error:
 433	iput(inode);
 434	return err;
 435}
 436
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 437static int
 438ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 439			       struct ext4_xattr_entry *entry, void *buffer,
 440			       size_t size)
 441{
 442	u32 hash;
 443
 444	/* Verify stored hash matches calculated hash. */
 445	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 446	if (hash != ext4_xattr_inode_get_hash(ea_inode))
 447		return -EFSCORRUPTED;
 448
 449	if (entry) {
 450		__le32 e_hash, tmp_data;
 451
 452		/* Verify entry hash. */
 453		tmp_data = cpu_to_le32(hash);
 454		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 455					       &tmp_data, 1);
 
 
 
 
 
 
 
 
 
 
 
 456		if (e_hash != entry->e_hash)
 457			return -EFSCORRUPTED;
 
 
 
 458	}
 459	return 0;
 460}
 461
 462/*
 463 * Read xattr value from the EA inode.
 464 */
 465static int
 466ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 467		     void *buffer, size_t size)
 468{
 469	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 470	struct inode *ea_inode;
 471	int err;
 472
 473	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 474				    le32_to_cpu(entry->e_hash), &ea_inode);
 475	if (err) {
 476		ea_inode = NULL;
 477		goto out;
 478	}
 479
 480	if (i_size_read(ea_inode) != size) {
 481		ext4_warning_inode(ea_inode,
 482				   "ea_inode file size=%llu entry size=%zu",
 483				   i_size_read(ea_inode), size);
 484		err = -EFSCORRUPTED;
 485		goto out;
 486	}
 487
 488	err = ext4_xattr_inode_read(ea_inode, buffer, size);
 489	if (err)
 490		goto out;
 491
 492	if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 493		err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 494						     size);
 495		if (err) {
 496			ext4_warning_inode(ea_inode,
 497					   "EA inode hash validation failed");
 498			goto out;
 499		}
 500
 501		if (ea_inode_cache)
 502			mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 503					ext4_xattr_inode_get_hash(ea_inode),
 504					ea_inode->i_ino, true /* reusable */);
 505	}
 506out:
 507	iput(ea_inode);
 508	return err;
 509}
 510
 511static int
 512ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 513		     void *buffer, size_t buffer_size)
 514{
 515	struct buffer_head *bh = NULL;
 516	struct ext4_xattr_entry *entry;
 517	size_t size;
 518	void *end;
 519	int error;
 520	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 521
 522	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 523		  name_index, name, buffer, (long)buffer_size);
 524
 525	if (!EXT4_I(inode)->i_file_acl)
 526		return -ENODATA;
 527	ea_idebug(inode, "reading block %llu",
 528		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 529	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 530	if (IS_ERR(bh))
 531		return PTR_ERR(bh);
 532	ea_bdebug(bh, "b_count=%d, refcount=%d",
 533		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 534	error = ext4_xattr_check_block(inode, bh);
 535	if (error)
 536		goto cleanup;
 537	ext4_xattr_block_cache_insert(ea_block_cache, bh);
 538	entry = BFIRST(bh);
 539	end = bh->b_data + bh->b_size;
 540	error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 541	if (error)
 542		goto cleanup;
 543	size = le32_to_cpu(entry->e_value_size);
 544	error = -ERANGE;
 545	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 546		goto cleanup;
 547	if (buffer) {
 548		if (size > buffer_size)
 549			goto cleanup;
 550		if (entry->e_value_inum) {
 551			error = ext4_xattr_inode_get(inode, entry, buffer,
 552						     size);
 553			if (error)
 554				goto cleanup;
 555		} else {
 556			u16 offset = le16_to_cpu(entry->e_value_offs);
 557			void *p = bh->b_data + offset;
 558
 559			if (unlikely(p + size > end))
 560				goto cleanup;
 561			memcpy(buffer, p, size);
 562		}
 563	}
 564	error = size;
 565
 566cleanup:
 567	brelse(bh);
 568	return error;
 569}
 570
 571int
 572ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 573		     void *buffer, size_t buffer_size)
 574{
 575	struct ext4_xattr_ibody_header *header;
 576	struct ext4_xattr_entry *entry;
 577	struct ext4_inode *raw_inode;
 578	struct ext4_iloc iloc;
 579	size_t size;
 580	void *end;
 581	int error;
 582
 583	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 584		return -ENODATA;
 585	error = ext4_get_inode_loc(inode, &iloc);
 586	if (error)
 587		return error;
 588	raw_inode = ext4_raw_inode(&iloc);
 589	header = IHDR(inode, raw_inode);
 590	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 591	error = xattr_check_inode(inode, header, end);
 592	if (error)
 593		goto cleanup;
 594	entry = IFIRST(header);
 595	error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 596	if (error)
 597		goto cleanup;
 598	size = le32_to_cpu(entry->e_value_size);
 599	error = -ERANGE;
 600	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 601		goto cleanup;
 602	if (buffer) {
 603		if (size > buffer_size)
 604			goto cleanup;
 605		if (entry->e_value_inum) {
 606			error = ext4_xattr_inode_get(inode, entry, buffer,
 607						     size);
 608			if (error)
 609				goto cleanup;
 610		} else {
 611			u16 offset = le16_to_cpu(entry->e_value_offs);
 612			void *p = (void *)IFIRST(header) + offset;
 613
 614			if (unlikely(p + size > end))
 615				goto cleanup;
 616			memcpy(buffer, p, size);
 617		}
 618	}
 619	error = size;
 620
 621cleanup:
 622	brelse(iloc.bh);
 623	return error;
 624}
 625
 626/*
 627 * ext4_xattr_get()
 628 *
 629 * Copy an extended attribute into the buffer
 630 * provided, or compute the buffer size required.
 631 * Buffer is NULL to compute the size of the buffer required.
 632 *
 633 * Returns a negative error number on failure, or the number of bytes
 634 * used / required on success.
 635 */
 636int
 637ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 638	       void *buffer, size_t buffer_size)
 639{
 640	int error;
 641
 642	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 643		return -EIO;
 644
 645	if (strlen(name) > 255)
 646		return -ERANGE;
 647
 648	down_read(&EXT4_I(inode)->xattr_sem);
 649	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 650				     buffer_size);
 651	if (error == -ENODATA)
 652		error = ext4_xattr_block_get(inode, name_index, name, buffer,
 653					     buffer_size);
 654	up_read(&EXT4_I(inode)->xattr_sem);
 655	return error;
 656}
 657
 658static int
 659ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 660			char *buffer, size_t buffer_size)
 661{
 662	size_t rest = buffer_size;
 663
 664	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 665		const struct xattr_handler *handler =
 666			ext4_xattr_handler(entry->e_name_index);
 667
 668		if (handler && (!handler->list || handler->list(dentry))) {
 669			const char *prefix = handler->prefix ?: handler->name;
 670			size_t prefix_len = strlen(prefix);
 671			size_t size = prefix_len + entry->e_name_len + 1;
 672
 673			if (buffer) {
 674				if (size > rest)
 675					return -ERANGE;
 676				memcpy(buffer, prefix, prefix_len);
 677				buffer += prefix_len;
 678				memcpy(buffer, entry->e_name, entry->e_name_len);
 679				buffer += entry->e_name_len;
 680				*buffer++ = 0;
 681			}
 682			rest -= size;
 683		}
 684	}
 685	return buffer_size - rest;  /* total size */
 686}
 687
 688static int
 689ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 690{
 691	struct inode *inode = d_inode(dentry);
 692	struct buffer_head *bh = NULL;
 693	int error;
 694
 695	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 696		  buffer, (long)buffer_size);
 697
 698	if (!EXT4_I(inode)->i_file_acl)
 699		return 0;
 700	ea_idebug(inode, "reading block %llu",
 701		  (unsigned long long)EXT4_I(inode)->i_file_acl);
 702	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 703	if (IS_ERR(bh))
 704		return PTR_ERR(bh);
 705	ea_bdebug(bh, "b_count=%d, refcount=%d",
 706		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 707	error = ext4_xattr_check_block(inode, bh);
 708	if (error)
 709		goto cleanup;
 710	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 711	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 712					buffer_size);
 713cleanup:
 714	brelse(bh);
 715	return error;
 716}
 717
 718static int
 719ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 720{
 721	struct inode *inode = d_inode(dentry);
 722	struct ext4_xattr_ibody_header *header;
 723	struct ext4_inode *raw_inode;
 724	struct ext4_iloc iloc;
 725	void *end;
 726	int error;
 727
 728	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 729		return 0;
 730	error = ext4_get_inode_loc(inode, &iloc);
 731	if (error)
 732		return error;
 733	raw_inode = ext4_raw_inode(&iloc);
 734	header = IHDR(inode, raw_inode);
 735	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 736	error = xattr_check_inode(inode, header, end);
 737	if (error)
 738		goto cleanup;
 739	error = ext4_xattr_list_entries(dentry, IFIRST(header),
 740					buffer, buffer_size);
 741
 742cleanup:
 743	brelse(iloc.bh);
 744	return error;
 745}
 746
 747/*
 748 * Inode operation listxattr()
 749 *
 750 * d_inode(dentry)->i_rwsem: don't care
 751 *
 752 * Copy a list of attribute names into the buffer
 753 * provided, or compute the buffer size required.
 754 * Buffer is NULL to compute the size of the buffer required.
 755 *
 756 * Returns a negative error number on failure, or the number of bytes
 757 * used / required on success.
 758 */
 759ssize_t
 760ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 761{
 762	int ret, ret2;
 763
 764	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 765	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 766	if (ret < 0)
 767		goto errout;
 768	if (buffer) {
 769		buffer += ret;
 770		buffer_size -= ret;
 771	}
 772	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 773	if (ret < 0)
 774		goto errout;
 775	ret += ret2;
 776errout:
 777	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 778	return ret;
 779}
 780
 781/*
 782 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 783 * not set, set it.
 784 */
 785static void ext4_xattr_update_super_block(handle_t *handle,
 786					  struct super_block *sb)
 787{
 788	if (ext4_has_feature_xattr(sb))
 789		return;
 790
 791	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 792	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 
 
 793		ext4_set_feature_xattr(sb);
 794		ext4_handle_dirty_super(handle, sb);
 
 
 795	}
 796}
 797
 798int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 799{
 800	struct ext4_iloc iloc = { .bh = NULL };
 801	struct buffer_head *bh = NULL;
 802	struct ext4_inode *raw_inode;
 803	struct ext4_xattr_ibody_header *header;
 804	struct ext4_xattr_entry *entry;
 805	qsize_t ea_inode_refs = 0;
 806	void *end;
 807	int ret;
 808
 809	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 810
 811	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 812		ret = ext4_get_inode_loc(inode, &iloc);
 813		if (ret)
 814			goto out;
 815		raw_inode = ext4_raw_inode(&iloc);
 816		header = IHDR(inode, raw_inode);
 817		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 818		ret = xattr_check_inode(inode, header, end);
 819		if (ret)
 820			goto out;
 821
 822		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 823		     entry = EXT4_XATTR_NEXT(entry))
 824			if (entry->e_value_inum)
 825				ea_inode_refs++;
 826	}
 827
 828	if (EXT4_I(inode)->i_file_acl) {
 829		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 830		if (IS_ERR(bh)) {
 831			ret = PTR_ERR(bh);
 832			bh = NULL;
 833			goto out;
 834		}
 835
 836		ret = ext4_xattr_check_block(inode, bh);
 837		if (ret)
 838			goto out;
 839
 840		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 841		     entry = EXT4_XATTR_NEXT(entry))
 842			if (entry->e_value_inum)
 843				ea_inode_refs++;
 844	}
 845	*usage = ea_inode_refs + 1;
 846	ret = 0;
 847out:
 848	brelse(iloc.bh);
 849	brelse(bh);
 850	return ret;
 851}
 852
 853static inline size_t round_up_cluster(struct inode *inode, size_t length)
 854{
 855	struct super_block *sb = inode->i_sb;
 856	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 857				    inode->i_blkbits);
 858	size_t mask = ~(cluster_size - 1);
 859
 860	return (length + cluster_size - 1) & mask;
 861}
 862
 863static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 864{
 865	int err;
 866
 867	err = dquot_alloc_inode(inode);
 868	if (err)
 869		return err;
 870	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 871	if (err)
 872		dquot_free_inode(inode);
 873	return err;
 874}
 875
 876static void ext4_xattr_inode_free_quota(struct inode *parent,
 877					struct inode *ea_inode,
 878					size_t len)
 879{
 880	if (ea_inode &&
 881	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 882		return;
 883	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 884	dquot_free_inode(parent);
 885}
 886
 887int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 888			     struct buffer_head *block_bh, size_t value_len,
 889			     bool is_create)
 890{
 891	int credits;
 892	int blocks;
 893
 894	/*
 895	 * 1) Owner inode update
 896	 * 2) Ref count update on old xattr block
 897	 * 3) new xattr block
 898	 * 4) block bitmap update for new xattr block
 899	 * 5) group descriptor for new xattr block
 900	 * 6) block bitmap update for old xattr block
 901	 * 7) group descriptor for old block
 902	 *
 903	 * 6 & 7 can happen if we have two racing threads T_a and T_b
 904	 * which are each trying to set an xattr on inodes I_a and I_b
 905	 * which were both initially sharing an xattr block.
 906	 */
 907	credits = 7;
 908
 909	/* Quota updates. */
 910	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 911
 912	/*
 913	 * In case of inline data, we may push out the data to a block,
 914	 * so we need to reserve credits for this eventuality
 915	 */
 916	if (inode && ext4_has_inline_data(inode))
 917		credits += ext4_writepage_trans_blocks(inode) + 1;
 918
 919	/* We are done if ea_inode feature is not enabled. */
 920	if (!ext4_has_feature_ea_inode(sb))
 921		return credits;
 922
 923	/* New ea_inode, inode map, block bitmap, group descriptor. */
 924	credits += 4;
 925
 926	/* Data blocks. */
 927	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 928
 929	/* Indirection block or one level of extent tree. */
 930	blocks += 1;
 931
 932	/* Block bitmap and group descriptor updates for each block. */
 933	credits += blocks * 2;
 934
 935	/* Blocks themselves. */
 936	credits += blocks;
 937
 938	if (!is_create) {
 939		/* Dereference ea_inode holding old xattr value.
 940		 * Old ea_inode, inode map, block bitmap, group descriptor.
 941		 */
 942		credits += 4;
 943
 944		/* Data blocks for old ea_inode. */
 945		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 946
 947		/* Indirection block or one level of extent tree for old
 948		 * ea_inode.
 949		 */
 950		blocks += 1;
 951
 952		/* Block bitmap and group descriptor updates for each block. */
 953		credits += blocks * 2;
 954	}
 955
 956	/* We may need to clone the existing xattr block in which case we need
 957	 * to increment ref counts for existing ea_inodes referenced by it.
 958	 */
 959	if (block_bh) {
 960		struct ext4_xattr_entry *entry = BFIRST(block_bh);
 961
 962		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 963			if (entry->e_value_inum)
 964				/* Ref count update on ea_inode. */
 965				credits += 1;
 966	}
 967	return credits;
 968}
 969
 970static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 971				     int credits, struct buffer_head *bh,
 972				     bool dirty, bool block_csum)
 973{
 974	int error;
 975
 976	if (!ext4_handle_valid(handle))
 977		return 0;
 978
 979	if (handle->h_buffer_credits >= credits)
 980		return 0;
 981
 982	error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 983	if (!error)
 984		return 0;
 985	if (error < 0) {
 986		ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 987		return error;
 988	}
 989
 990	if (bh && dirty) {
 991		if (block_csum)
 992			ext4_xattr_block_csum_set(inode, bh);
 993		error = ext4_handle_dirty_metadata(handle, NULL, bh);
 994		if (error) {
 995			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 996				     error);
 997			return error;
 998		}
 999	}
1000
1001	error = ext4_journal_restart(handle, credits);
1002	if (error) {
1003		ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1004		return error;
1005	}
1006
1007	if (bh) {
1008		error = ext4_journal_get_write_access(handle, bh);
1009		if (error) {
1010			ext4_warning(inode->i_sb,
1011				     "Get write access failed (error %d)",
1012				     error);
1013			return error;
1014		}
1015	}
1016	return 0;
1017}
1018
1019static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1020				       int ref_change)
1021{
1022	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1023	struct ext4_iloc iloc;
1024	s64 ref_count;
1025	u32 hash;
1026	int ret;
1027
1028	inode_lock(ea_inode);
1029
1030	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1031	if (ret)
1032		goto out;
1033
1034	ref_count = ext4_xattr_inode_get_ref(ea_inode);
1035	ref_count += ref_change;
1036	ext4_xattr_inode_set_ref(ea_inode, ref_count);
1037
1038	if (ref_change > 0) {
1039		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1040			  ea_inode->i_ino, ref_count);
1041
1042		if (ref_count == 1) {
1043			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1044				  ea_inode->i_ino, ea_inode->i_nlink);
1045
1046			set_nlink(ea_inode, 1);
1047			ext4_orphan_del(handle, ea_inode);
1048
1049			if (ea_inode_cache) {
1050				hash = ext4_xattr_inode_get_hash(ea_inode);
1051				mb_cache_entry_create(ea_inode_cache,
1052						      GFP_NOFS, hash,
1053						      ea_inode->i_ino,
1054						      true /* reusable */);
1055			}
1056		}
1057	} else {
1058		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1059			  ea_inode->i_ino, ref_count);
1060
1061		if (ref_count == 0) {
1062			WARN_ONCE(ea_inode->i_nlink != 1,
1063				  "EA inode %lu i_nlink=%u",
1064				  ea_inode->i_ino, ea_inode->i_nlink);
1065
1066			clear_nlink(ea_inode);
1067			ext4_orphan_add(handle, ea_inode);
1068
1069			if (ea_inode_cache) {
1070				hash = ext4_xattr_inode_get_hash(ea_inode);
1071				mb_cache_entry_delete(ea_inode_cache, hash,
1072						      ea_inode->i_ino);
1073			}
1074		}
1075	}
1076
1077	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1078	if (ret)
1079		ext4_warning_inode(ea_inode,
1080				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1081out:
1082	inode_unlock(ea_inode);
1083	return ret;
1084}
1085
1086static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1087{
1088	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1089}
1090
1091static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1092{
1093	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1094}
1095
1096static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1097					struct ext4_xattr_entry *first)
1098{
1099	struct inode *ea_inode;
1100	struct ext4_xattr_entry *entry;
1101	struct ext4_xattr_entry *failed_entry;
1102	unsigned int ea_ino;
1103	int err, saved_err;
1104
1105	for (entry = first; !IS_LAST_ENTRY(entry);
1106	     entry = EXT4_XATTR_NEXT(entry)) {
1107		if (!entry->e_value_inum)
1108			continue;
1109		ea_ino = le32_to_cpu(entry->e_value_inum);
1110		err = ext4_xattr_inode_iget(parent, ea_ino,
1111					    le32_to_cpu(entry->e_hash),
1112					    &ea_inode);
1113		if (err)
1114			goto cleanup;
1115		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1116		if (err) {
1117			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1118			iput(ea_inode);
1119			goto cleanup;
1120		}
1121		iput(ea_inode);
1122	}
1123	return 0;
1124
1125cleanup:
1126	saved_err = err;
1127	failed_entry = entry;
1128
1129	for (entry = first; entry != failed_entry;
1130	     entry = EXT4_XATTR_NEXT(entry)) {
1131		if (!entry->e_value_inum)
1132			continue;
1133		ea_ino = le32_to_cpu(entry->e_value_inum);
1134		err = ext4_xattr_inode_iget(parent, ea_ino,
1135					    le32_to_cpu(entry->e_hash),
1136					    &ea_inode);
1137		if (err) {
1138			ext4_warning(parent->i_sb,
1139				     "cleanup ea_ino %u iget error %d", ea_ino,
1140				     err);
1141			continue;
1142		}
1143		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1144		if (err)
1145			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1146					   err);
1147		iput(ea_inode);
1148	}
1149	return saved_err;
1150}
1151
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1152static void
1153ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1154			     struct buffer_head *bh,
1155			     struct ext4_xattr_entry *first, bool block_csum,
1156			     struct ext4_xattr_inode_array **ea_inode_array,
1157			     int extra_credits, bool skip_quota)
1158{
1159	struct inode *ea_inode;
1160	struct ext4_xattr_entry *entry;
1161	bool dirty = false;
1162	unsigned int ea_ino;
1163	int err;
1164	int credits;
1165
1166	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1167	credits = 2 + extra_credits;
1168
1169	for (entry = first; !IS_LAST_ENTRY(entry);
1170	     entry = EXT4_XATTR_NEXT(entry)) {
1171		if (!entry->e_value_inum)
1172			continue;
1173		ea_ino = le32_to_cpu(entry->e_value_inum);
1174		err = ext4_xattr_inode_iget(parent, ea_ino,
1175					    le32_to_cpu(entry->e_hash),
1176					    &ea_inode);
1177		if (err)
1178			continue;
1179
1180		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1181		if (err) {
1182			ext4_warning_inode(ea_inode,
1183					   "Expand inode array err=%d", err);
1184			iput(ea_inode);
1185			continue;
1186		}
1187
1188		err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1189						dirty, block_csum);
1190		if (err) {
 
 
1191			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1192					   err);
1193			continue;
1194		}
 
 
 
 
 
 
 
 
 
 
1195
1196		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1197		if (err) {
1198			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1199					   err);
1200			continue;
1201		}
1202
1203		if (!skip_quota)
1204			ext4_xattr_inode_free_quota(parent, ea_inode,
1205					      le32_to_cpu(entry->e_value_size));
1206
1207		/*
1208		 * Forget about ea_inode within the same transaction that
1209		 * decrements the ref count. This avoids duplicate decrements in
1210		 * case the rest of the work spills over to subsequent
1211		 * transactions.
1212		 */
1213		entry->e_value_inum = 0;
1214		entry->e_value_size = 0;
1215
1216		dirty = true;
1217	}
1218
1219	if (dirty) {
1220		/*
1221		 * Note that we are deliberately skipping csum calculation for
1222		 * the final update because we do not expect any journal
1223		 * restarts until xattr block is freed.
1224		 */
1225
1226		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1227		if (err)
1228			ext4_warning_inode(parent,
1229					   "handle dirty metadata err=%d", err);
1230	}
1231}
1232
1233/*
1234 * Release the xattr block BH: If the reference count is > 1, decrement it;
1235 * otherwise free the block.
1236 */
1237static void
1238ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1239			 struct buffer_head *bh,
1240			 struct ext4_xattr_inode_array **ea_inode_array,
1241			 int extra_credits)
1242{
1243	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1244	u32 hash, ref;
1245	int error = 0;
1246
1247	BUFFER_TRACE(bh, "get_write_access");
1248	error = ext4_journal_get_write_access(handle, bh);
 
1249	if (error)
1250		goto out;
1251
 
1252	lock_buffer(bh);
1253	hash = le32_to_cpu(BHDR(bh)->h_hash);
1254	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1255	if (ref == 1) {
1256		ea_bdebug(bh, "refcount now=0; freeing");
1257		/*
1258		 * This must happen under buffer lock for
1259		 * ext4_xattr_block_set() to reliably detect freed block
1260		 */
1261		if (ea_block_cache)
1262			mb_cache_entry_delete(ea_block_cache, hash,
1263					      bh->b_blocknr);
 
 
 
 
 
 
 
 
 
1264		get_bh(bh);
1265		unlock_buffer(bh);
1266
1267		if (ext4_has_feature_ea_inode(inode->i_sb))
1268			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1269						     BFIRST(bh),
1270						     true /* block_csum */,
1271						     ea_inode_array,
1272						     extra_credits,
1273						     true /* skip_quota */);
1274		ext4_free_blocks(handle, inode, bh, 0, 1,
1275				 EXT4_FREE_BLOCKS_METADATA |
1276				 EXT4_FREE_BLOCKS_FORGET);
1277	} else {
1278		ref--;
1279		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1280		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1281			struct mb_cache_entry *ce;
1282
1283			if (ea_block_cache) {
1284				ce = mb_cache_entry_get(ea_block_cache, hash,
1285							bh->b_blocknr);
1286				if (ce) {
1287					ce->e_reusable = 1;
1288					mb_cache_entry_put(ea_block_cache, ce);
1289				}
1290			}
1291		}
1292
1293		ext4_xattr_block_csum_set(inode, bh);
1294		/*
1295		 * Beware of this ugliness: Releasing of xattr block references
1296		 * from different inodes can race and so we have to protect
1297		 * from a race where someone else frees the block (and releases
1298		 * its journal_head) before we are done dirtying the buffer. In
1299		 * nojournal mode this race is harmless and we actually cannot
1300		 * call ext4_handle_dirty_metadata() with locked buffer as
1301		 * that function can call sync_dirty_buffer() so for that case
1302		 * we handle the dirtying after unlocking the buffer.
1303		 */
1304		if (ext4_handle_valid(handle))
1305			error = ext4_handle_dirty_metadata(handle, inode, bh);
1306		unlock_buffer(bh);
1307		if (!ext4_handle_valid(handle))
1308			error = ext4_handle_dirty_metadata(handle, inode, bh);
1309		if (IS_SYNC(inode))
1310			ext4_handle_sync(handle);
1311		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1312		ea_bdebug(bh, "refcount now=%d; releasing",
1313			  le32_to_cpu(BHDR(bh)->h_refcount));
1314	}
1315out:
1316	ext4_std_error(inode->i_sb, error);
1317	return;
1318}
1319
1320/*
1321 * Find the available free space for EAs. This also returns the total number of
1322 * bytes used by EA entries.
1323 */
1324static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1325				    size_t *min_offs, void *base, int *total)
1326{
1327	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1328		if (!last->e_value_inum && last->e_value_size) {
1329			size_t offs = le16_to_cpu(last->e_value_offs);
1330			if (offs < *min_offs)
1331				*min_offs = offs;
1332		}
1333		if (total)
1334			*total += EXT4_XATTR_LEN(last->e_name_len);
1335	}
1336	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1337}
1338
1339/*
1340 * Write the value of the EA in an inode.
1341 */
1342static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1343				  const void *buf, int bufsize)
1344{
1345	struct buffer_head *bh = NULL;
1346	unsigned long block = 0;
1347	int blocksize = ea_inode->i_sb->s_blocksize;
1348	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1349	int csize, wsize = 0;
1350	int ret = 0;
1351	int retries = 0;
1352
1353retry:
1354	while (ret >= 0 && ret < max_blocks) {
1355		struct ext4_map_blocks map;
1356		map.m_lblk = block += ret;
1357		map.m_len = max_blocks -= ret;
1358
1359		ret = ext4_map_blocks(handle, ea_inode, &map,
1360				      EXT4_GET_BLOCKS_CREATE);
1361		if (ret <= 0) {
1362			ext4_mark_inode_dirty(handle, ea_inode);
1363			if (ret == -ENOSPC &&
1364			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1365				ret = 0;
1366				goto retry;
1367			}
1368			break;
1369		}
1370	}
1371
1372	if (ret < 0)
1373		return ret;
1374
1375	block = 0;
1376	while (wsize < bufsize) {
1377		if (bh != NULL)
1378			brelse(bh);
1379		csize = (bufsize - wsize) > blocksize ? blocksize :
1380								bufsize - wsize;
1381		bh = ext4_getblk(handle, ea_inode, block, 0);
1382		if (IS_ERR(bh))
1383			return PTR_ERR(bh);
1384		if (!bh) {
1385			WARN_ON_ONCE(1);
1386			EXT4_ERROR_INODE(ea_inode,
1387					 "ext4_getblk() return bh = NULL");
1388			return -EFSCORRUPTED;
1389		}
1390		ret = ext4_journal_get_write_access(handle, bh);
 
1391		if (ret)
1392			goto out;
1393
1394		memcpy(bh->b_data, buf, csize);
1395		set_buffer_uptodate(bh);
1396		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1397
1398		buf += csize;
1399		wsize += csize;
1400		block += 1;
1401	}
1402
1403	inode_lock(ea_inode);
1404	i_size_write(ea_inode, wsize);
1405	ext4_update_i_disksize(ea_inode, wsize);
1406	inode_unlock(ea_inode);
1407
1408	ext4_mark_inode_dirty(handle, ea_inode);
 
 
1409
1410out:
1411	brelse(bh);
1412
1413	return ret;
1414}
1415
1416/*
1417 * Create an inode to store the value of a large EA.
1418 */
1419static struct inode *ext4_xattr_inode_create(handle_t *handle,
1420					     struct inode *inode, u32 hash)
1421{
1422	struct inode *ea_inode = NULL;
1423	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1424	int err;
1425
1426	/*
1427	 * Let the next inode be the goal, so we try and allocate the EA inode
1428	 * in the same group, or nearby one.
1429	 */
1430	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1431				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1432				  EXT4_EA_INODE_FL);
1433	if (!IS_ERR(ea_inode)) {
1434		ea_inode->i_op = &ext4_file_inode_operations;
1435		ea_inode->i_fop = &ext4_file_operations;
1436		ext4_set_aops(ea_inode);
1437		ext4_xattr_inode_set_class(ea_inode);
1438		unlock_new_inode(ea_inode);
1439		ext4_xattr_inode_set_ref(ea_inode, 1);
1440		ext4_xattr_inode_set_hash(ea_inode, hash);
1441		err = ext4_mark_inode_dirty(handle, ea_inode);
1442		if (!err)
1443			err = ext4_inode_attach_jinode(ea_inode);
1444		if (err) {
 
 
 
1445			iput(ea_inode);
1446			return ERR_PTR(err);
1447		}
1448
1449		/*
1450		 * Xattr inodes are shared therefore quota charging is performed
1451		 * at a higher level.
1452		 */
1453		dquot_free_inode(ea_inode);
1454		dquot_drop(ea_inode);
1455		inode_lock(ea_inode);
1456		ea_inode->i_flags |= S_NOQUOTA;
1457		inode_unlock(ea_inode);
1458	}
1459
1460	return ea_inode;
1461}
1462
1463static struct inode *
1464ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1465			    size_t value_len, u32 hash)
1466{
1467	struct inode *ea_inode;
1468	struct mb_cache_entry *ce;
1469	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1470	void *ea_data;
1471
1472	if (!ea_inode_cache)
1473		return NULL;
1474
1475	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1476	if (!ce)
1477		return NULL;
1478
1479	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
 
 
 
1480	if (!ea_data) {
1481		mb_cache_entry_put(ea_inode_cache, ce);
1482		return NULL;
1483	}
1484
1485	while (ce) {
1486		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1487				     EXT4_IGET_NORMAL);
1488		if (!IS_ERR(ea_inode) &&
1489		    !is_bad_inode(ea_inode) &&
1490		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1491		    i_size_read(ea_inode) == value_len &&
1492		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1493		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1494						    value_len) &&
1495		    !memcmp(value, ea_data, value_len)) {
1496			mb_cache_entry_touch(ea_inode_cache, ce);
1497			mb_cache_entry_put(ea_inode_cache, ce);
1498			kvfree(ea_data);
1499			return ea_inode;
1500		}
1501
1502		if (!IS_ERR(ea_inode))
1503			iput(ea_inode);
1504		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1505	}
1506	kvfree(ea_data);
1507	return NULL;
1508}
1509
1510/*
1511 * Add value of the EA in an inode.
1512 */
1513static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1514					  const void *value, size_t value_len,
1515					  struct inode **ret_inode)
1516{
1517	struct inode *ea_inode;
1518	u32 hash;
1519	int err;
1520
1521	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1522	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1523	if (ea_inode) {
1524		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1525		if (err) {
1526			iput(ea_inode);
1527			return err;
1528		}
1529
1530		*ret_inode = ea_inode;
1531		return 0;
1532	}
1533
1534	/* Create an inode for the EA value */
1535	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1536	if (IS_ERR(ea_inode))
1537		return PTR_ERR(ea_inode);
1538
1539	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1540	if (err) {
1541		ext4_xattr_inode_dec_ref(handle, ea_inode);
 
1542		iput(ea_inode);
1543		return err;
1544	}
1545
1546	if (EA_INODE_CACHE(inode))
1547		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1548				      ea_inode->i_ino, true /* reusable */);
1549
1550	*ret_inode = ea_inode;
1551	return 0;
1552}
1553
1554/*
1555 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1556 * feature is enabled.
1557 */
1558#define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1559
1560static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1561				struct ext4_xattr_search *s,
1562				handle_t *handle, struct inode *inode,
1563				bool is_block)
1564{
1565	struct ext4_xattr_entry *last, *next;
1566	struct ext4_xattr_entry *here = s->here;
1567	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1568	int in_inode = i->in_inode;
1569	struct inode *old_ea_inode = NULL;
1570	struct inode *new_ea_inode = NULL;
1571	size_t old_size, new_size;
1572	int ret;
1573
1574	/* Space used by old and new values. */
1575	old_size = (!s->not_found && !here->e_value_inum) ?
1576			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1577	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1578
1579	/*
1580	 * Optimization for the simple case when old and new values have the
1581	 * same padded sizes. Not applicable if external inodes are involved.
1582	 */
1583	if (new_size && new_size == old_size) {
1584		size_t offs = le16_to_cpu(here->e_value_offs);
1585		void *val = s->base + offs;
1586
1587		here->e_value_size = cpu_to_le32(i->value_len);
1588		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1589			memset(val, 0, new_size);
1590		} else {
1591			memcpy(val, i->value, i->value_len);
1592			/* Clear padding bytes. */
1593			memset(val + i->value_len, 0, new_size - i->value_len);
1594		}
1595		goto update_hash;
1596	}
1597
1598	/* Compute min_offs and last. */
1599	last = s->first;
1600	for (; !IS_LAST_ENTRY(last); last = next) {
1601		next = EXT4_XATTR_NEXT(last);
1602		if ((void *)next >= s->end) {
1603			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1604			ret = -EFSCORRUPTED;
1605			goto out;
1606		}
1607		if (!last->e_value_inum && last->e_value_size) {
1608			size_t offs = le16_to_cpu(last->e_value_offs);
1609			if (offs < min_offs)
1610				min_offs = offs;
1611		}
1612	}
1613
1614	/* Check whether we have enough space. */
1615	if (i->value) {
1616		size_t free;
1617
1618		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1619		if (!s->not_found)
1620			free += EXT4_XATTR_LEN(name_len) + old_size;
1621
1622		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1623			ret = -ENOSPC;
1624			goto out;
1625		}
1626
1627		/*
1628		 * If storing the value in an external inode is an option,
1629		 * reserve space for xattr entries/names in the external
1630		 * attribute block so that a long value does not occupy the
1631		 * whole space and prevent futher entries being added.
1632		 */
1633		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1634		    new_size && is_block &&
1635		    (min_offs + old_size - new_size) <
1636					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1637			ret = -ENOSPC;
1638			goto out;
1639		}
1640	}
1641
1642	/*
1643	 * Getting access to old and new ea inodes is subject to failures.
1644	 * Finish that work before doing any modifications to the xattr data.
1645	 */
1646	if (!s->not_found && here->e_value_inum) {
1647		ret = ext4_xattr_inode_iget(inode,
1648					    le32_to_cpu(here->e_value_inum),
1649					    le32_to_cpu(here->e_hash),
1650					    &old_ea_inode);
1651		if (ret) {
1652			old_ea_inode = NULL;
1653			goto out;
1654		}
1655	}
1656	if (i->value && in_inode) {
1657		WARN_ON_ONCE(!i->value_len);
1658
1659		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1660		if (ret)
1661			goto out;
1662
1663		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1664						     i->value_len,
1665						     &new_ea_inode);
1666		if (ret) {
1667			new_ea_inode = NULL;
1668			ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1669			goto out;
1670		}
1671	}
1672
1673	if (old_ea_inode) {
1674		/* We are ready to release ref count on the old_ea_inode. */
1675		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1676		if (ret) {
1677			/* Release newly required ref count on new_ea_inode. */
1678			if (new_ea_inode) {
1679				int err;
1680
1681				err = ext4_xattr_inode_dec_ref(handle,
1682							       new_ea_inode);
1683				if (err)
1684					ext4_warning_inode(new_ea_inode,
1685						  "dec ref new_ea_inode err=%d",
1686						  err);
1687				ext4_xattr_inode_free_quota(inode, new_ea_inode,
1688							    i->value_len);
1689			}
1690			goto out;
1691		}
1692
1693		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1694					    le32_to_cpu(here->e_value_size));
1695	}
1696
1697	/* No failures allowed past this point. */
1698
1699	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1700		/* Remove the old value. */
1701		void *first_val = s->base + min_offs;
1702		size_t offs = le16_to_cpu(here->e_value_offs);
1703		void *val = s->base + offs;
1704
1705		memmove(first_val + old_size, first_val, val - first_val);
1706		memset(first_val, 0, old_size);
1707		min_offs += old_size;
1708
1709		/* Adjust all value offsets. */
1710		last = s->first;
1711		while (!IS_LAST_ENTRY(last)) {
1712			size_t o = le16_to_cpu(last->e_value_offs);
1713
1714			if (!last->e_value_inum &&
1715			    last->e_value_size && o < offs)
1716				last->e_value_offs = cpu_to_le16(o + old_size);
1717			last = EXT4_XATTR_NEXT(last);
1718		}
1719	}
1720
1721	if (!i->value) {
1722		/* Remove old name. */
1723		size_t size = EXT4_XATTR_LEN(name_len);
1724
1725		last = ENTRY((void *)last - size);
1726		memmove(here, (void *)here + size,
1727			(void *)last - (void *)here + sizeof(__u32));
1728		memset(last, 0, size);
1729	} else if (s->not_found) {
1730		/* Insert new name. */
1731		size_t size = EXT4_XATTR_LEN(name_len);
1732		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1733
1734		memmove((void *)here + size, here, rest);
1735		memset(here, 0, size);
1736		here->e_name_index = i->name_index;
1737		here->e_name_len = name_len;
1738		memcpy(here->e_name, i->name, name_len);
1739	} else {
1740		/* This is an update, reset value info. */
1741		here->e_value_inum = 0;
1742		here->e_value_offs = 0;
1743		here->e_value_size = 0;
1744	}
1745
1746	if (i->value) {
1747		/* Insert new value. */
1748		if (in_inode) {
1749			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1750		} else if (i->value_len) {
1751			void *val = s->base + min_offs - new_size;
1752
1753			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1754			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1755				memset(val, 0, new_size);
1756			} else {
1757				memcpy(val, i->value, i->value_len);
1758				/* Clear padding bytes. */
1759				memset(val + i->value_len, 0,
1760				       new_size - i->value_len);
1761			}
1762		}
1763		here->e_value_size = cpu_to_le32(i->value_len);
1764	}
1765
1766update_hash:
1767	if (i->value) {
1768		__le32 hash = 0;
1769
1770		/* Entry hash calculation. */
1771		if (in_inode) {
1772			__le32 crc32c_hash;
1773
1774			/*
1775			 * Feed crc32c hash instead of the raw value for entry
1776			 * hash calculation. This is to avoid walking
1777			 * potentially long value buffer again.
1778			 */
1779			crc32c_hash = cpu_to_le32(
1780				       ext4_xattr_inode_get_hash(new_ea_inode));
1781			hash = ext4_xattr_hash_entry(here->e_name,
1782						     here->e_name_len,
1783						     &crc32c_hash, 1);
1784		} else if (is_block) {
1785			__le32 *value = s->base + le16_to_cpu(
1786							here->e_value_offs);
1787
1788			hash = ext4_xattr_hash_entry(here->e_name,
1789						     here->e_name_len, value,
1790						     new_size >> 2);
1791		}
1792		here->e_hash = hash;
1793	}
1794
1795	if (is_block)
1796		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1797
1798	ret = 0;
1799out:
1800	iput(old_ea_inode);
1801	iput(new_ea_inode);
1802	return ret;
1803}
1804
1805struct ext4_xattr_block_find {
1806	struct ext4_xattr_search s;
1807	struct buffer_head *bh;
1808};
1809
1810static int
1811ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1812		      struct ext4_xattr_block_find *bs)
1813{
1814	struct super_block *sb = inode->i_sb;
1815	int error;
1816
1817	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1818		  i->name_index, i->name, i->value, (long)i->value_len);
1819
1820	if (EXT4_I(inode)->i_file_acl) {
1821		/* The inode already has an extended attribute block. */
1822		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1823		if (IS_ERR(bs->bh))
1824			return PTR_ERR(bs->bh);
 
 
 
1825		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1826			atomic_read(&(bs->bh->b_count)),
1827			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1828		error = ext4_xattr_check_block(inode, bs->bh);
1829		if (error)
1830			return error;
1831		/* Find the named attribute. */
1832		bs->s.base = BHDR(bs->bh);
1833		bs->s.first = BFIRST(bs->bh);
1834		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1835		bs->s.here = bs->s.first;
1836		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1837					 i->name_index, i->name, 1);
1838		if (error && error != -ENODATA)
1839			return error;
1840		bs->s.not_found = error;
1841	}
1842	return 0;
1843}
1844
1845static int
1846ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1847		     struct ext4_xattr_info *i,
1848		     struct ext4_xattr_block_find *bs)
1849{
1850	struct super_block *sb = inode->i_sb;
1851	struct buffer_head *new_bh = NULL;
1852	struct ext4_xattr_search s_copy = bs->s;
1853	struct ext4_xattr_search *s = &s_copy;
1854	struct mb_cache_entry *ce = NULL;
1855	int error = 0;
1856	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1857	struct inode *ea_inode = NULL, *tmp_inode;
1858	size_t old_ea_inode_quota = 0;
1859	unsigned int ea_ino;
1860
1861
1862#define header(x) ((struct ext4_xattr_header *)(x))
1863
1864	if (s->base) {
 
 
1865		BUFFER_TRACE(bs->bh, "get_write_access");
1866		error = ext4_journal_get_write_access(handle, bs->bh);
 
1867		if (error)
1868			goto cleanup;
1869		lock_buffer(bs->bh);
1870
1871		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1872			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1873
1874			/*
1875			 * This must happen under buffer lock for
1876			 * ext4_xattr_block_set() to reliably detect modified
1877			 * block
1878			 */
1879			if (ea_block_cache)
1880				mb_cache_entry_delete(ea_block_cache, hash,
1881						      bs->bh->b_blocknr);
 
 
 
 
 
 
 
 
 
 
 
1882			ea_bdebug(bs->bh, "modifying in-place");
1883			error = ext4_xattr_set_entry(i, s, handle, inode,
1884						     true /* is_block */);
1885			ext4_xattr_block_csum_set(inode, bs->bh);
1886			unlock_buffer(bs->bh);
1887			if (error == -EFSCORRUPTED)
1888				goto bad_block;
1889			if (!error)
1890				error = ext4_handle_dirty_metadata(handle,
1891								   inode,
1892								   bs->bh);
1893			if (error)
1894				goto cleanup;
1895			goto inserted;
1896		} else {
1897			int offset = (char *)s->here - bs->bh->b_data;
 
 
 
 
 
 
 
 
 
 
1898
1899			unlock_buffer(bs->bh);
1900			ea_bdebug(bs->bh, "cloning");
1901			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1902			error = -ENOMEM;
1903			if (s->base == NULL)
 
 
 
 
 
 
 
 
1904				goto cleanup;
1905			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1906			s->first = ENTRY(header(s->base)+1);
1907			header(s->base)->h_refcount = cpu_to_le32(1);
1908			s->here = ENTRY(s->base + offset);
1909			s->end = s->base + bs->bh->b_size;
1910
1911			/*
1912			 * If existing entry points to an xattr inode, we need
1913			 * to prevent ext4_xattr_set_entry() from decrementing
1914			 * ref count on it because the reference belongs to the
1915			 * original block. In this case, make the entry look
1916			 * like it has an empty value.
1917			 */
1918			if (!s->not_found && s->here->e_value_inum) {
1919				ea_ino = le32_to_cpu(s->here->e_value_inum);
1920				error = ext4_xattr_inode_iget(inode, ea_ino,
1921					      le32_to_cpu(s->here->e_hash),
1922					      &tmp_inode);
1923				if (error)
1924					goto cleanup;
1925
1926				if (!ext4_test_inode_state(tmp_inode,
1927						EXT4_STATE_LUSTRE_EA_INODE)) {
1928					/*
1929					 * Defer quota free call for previous
1930					 * inode until success is guaranteed.
1931					 */
1932					old_ea_inode_quota = le32_to_cpu(
1933							s->here->e_value_size);
1934				}
1935				iput(tmp_inode);
1936
1937				s->here->e_value_inum = 0;
1938				s->here->e_value_size = 0;
1939			}
1940		}
1941	} else {
1942		/* Allocate a buffer where we construct the new block. */
1943		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1944		/* assert(header == s->base) */
1945		error = -ENOMEM;
1946		if (s->base == NULL)
1947			goto cleanup;
1948		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1949		header(s->base)->h_blocks = cpu_to_le32(1);
1950		header(s->base)->h_refcount = cpu_to_le32(1);
1951		s->first = ENTRY(header(s->base)+1);
1952		s->here = ENTRY(header(s->base)+1);
1953		s->end = s->base + sb->s_blocksize;
1954	}
1955
1956	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1957	if (error == -EFSCORRUPTED)
1958		goto bad_block;
1959	if (error)
1960		goto cleanup;
1961
1962	if (i->value && s->here->e_value_inum) {
1963		/*
1964		 * A ref count on ea_inode has been taken as part of the call to
1965		 * ext4_xattr_set_entry() above. We would like to drop this
1966		 * extra ref but we have to wait until the xattr block is
1967		 * initialized and has its own ref count on the ea_inode.
1968		 */
1969		ea_ino = le32_to_cpu(s->here->e_value_inum);
1970		error = ext4_xattr_inode_iget(inode, ea_ino,
1971					      le32_to_cpu(s->here->e_hash),
1972					      &ea_inode);
1973		if (error) {
1974			ea_inode = NULL;
1975			goto cleanup;
1976		}
1977	}
1978
1979inserted:
1980	if (!IS_LAST_ENTRY(s->first)) {
1981		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1982						     &ce);
1983		if (new_bh) {
1984			/* We found an identical block in the cache. */
1985			if (new_bh == bs->bh)
1986				ea_bdebug(new_bh, "keeping");
1987			else {
1988				u32 ref;
1989
1990				WARN_ON_ONCE(dquot_initialize_needed(inode));
1991
1992				/* The old block is released after updating
1993				   the inode. */
1994				error = dquot_alloc_block(inode,
1995						EXT4_C2B(EXT4_SB(sb), 1));
1996				if (error)
1997					goto cleanup;
1998				BUFFER_TRACE(new_bh, "get_write_access");
1999				error = ext4_journal_get_write_access(handle,
2000								      new_bh);
 
2001				if (error)
2002					goto cleanup_dquot;
2003				lock_buffer(new_bh);
2004				/*
2005				 * We have to be careful about races with
2006				 * freeing, rehashing or adding references to
2007				 * xattr block. Once we hold buffer lock xattr
2008				 * block's state is stable so we can check
2009				 * whether the block got freed / rehashed or
2010				 * not.  Since we unhash mbcache entry under
2011				 * buffer lock when freeing / rehashing xattr
2012				 * block, checking whether entry is still
2013				 * hashed is reliable. Same rules hold for
2014				 * e_reusable handling.
2015				 */
2016				if (hlist_bl_unhashed(&ce->e_hash_list) ||
2017				    !ce->e_reusable) {
2018					/*
2019					 * Undo everything and check mbcache
2020					 * again.
2021					 */
2022					unlock_buffer(new_bh);
2023					dquot_free_block(inode,
2024							 EXT4_C2B(EXT4_SB(sb),
2025								  1));
2026					brelse(new_bh);
2027					mb_cache_entry_put(ea_block_cache, ce);
2028					ce = NULL;
2029					new_bh = NULL;
2030					goto inserted;
2031				}
2032				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2033				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2034				if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2035					ce->e_reusable = 0;
2036				ea_bdebug(new_bh, "reusing; refcount now=%d",
2037					  ref);
2038				ext4_xattr_block_csum_set(inode, new_bh);
2039				unlock_buffer(new_bh);
2040				error = ext4_handle_dirty_metadata(handle,
2041								   inode,
2042								   new_bh);
2043				if (error)
2044					goto cleanup_dquot;
2045			}
2046			mb_cache_entry_touch(ea_block_cache, ce);
2047			mb_cache_entry_put(ea_block_cache, ce);
2048			ce = NULL;
2049		} else if (bs->bh && s->base == bs->bh->b_data) {
2050			/* We were modifying this block in-place. */
2051			ea_bdebug(bs->bh, "keeping this block");
2052			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2053			new_bh = bs->bh;
2054			get_bh(new_bh);
2055		} else {
2056			/* We need to allocate a new block */
2057			ext4_fsblk_t goal, block;
2058
2059			WARN_ON_ONCE(dquot_initialize_needed(inode));
2060
2061			goal = ext4_group_first_block_no(sb,
2062						EXT4_I(inode)->i_block_group);
2063
2064			/* non-extent files can't have physical blocks past 2^32 */
2065			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2066				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2067
2068			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2069						     NULL, &error);
2070			if (error)
2071				goto cleanup;
2072
2073			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2074				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2075
2076			ea_idebug(inode, "creating block %llu",
2077				  (unsigned long long)block);
2078
2079			new_bh = sb_getblk(sb, block);
2080			if (unlikely(!new_bh)) {
2081				error = -ENOMEM;
2082getblk_failed:
2083				ext4_free_blocks(handle, inode, NULL, block, 1,
2084						 EXT4_FREE_BLOCKS_METADATA);
2085				goto cleanup;
2086			}
2087			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2088						      ENTRY(header(s->base)+1));
2089			if (error)
2090				goto getblk_failed;
2091			if (ea_inode) {
2092				/* Drop the extra ref on ea_inode. */
2093				error = ext4_xattr_inode_dec_ref(handle,
2094								 ea_inode);
2095				if (error)
2096					ext4_warning_inode(ea_inode,
2097							   "dec ref error=%d",
2098							   error);
2099				iput(ea_inode);
2100				ea_inode = NULL;
2101			}
2102
2103			lock_buffer(new_bh);
2104			error = ext4_journal_get_create_access(handle, new_bh);
 
2105			if (error) {
2106				unlock_buffer(new_bh);
2107				error = -EIO;
2108				goto getblk_failed;
2109			}
2110			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2111			ext4_xattr_block_csum_set(inode, new_bh);
2112			set_buffer_uptodate(new_bh);
2113			unlock_buffer(new_bh);
2114			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2115			error = ext4_handle_dirty_metadata(handle, inode,
2116							   new_bh);
2117			if (error)
2118				goto cleanup;
2119		}
2120	}
2121
2122	if (old_ea_inode_quota)
2123		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2124
2125	/* Update the inode. */
2126	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2127
2128	/* Drop the previous xattr block. */
2129	if (bs->bh && bs->bh != new_bh) {
2130		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2131
2132		ext4_xattr_release_block(handle, inode, bs->bh,
2133					 &ea_inode_array,
2134					 0 /* extra_credits */);
2135		ext4_xattr_inode_array_free(ea_inode_array);
2136	}
2137	error = 0;
2138
2139cleanup:
2140	if (ea_inode) {
2141		int error2;
2142
2143		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2144		if (error2)
2145			ext4_warning_inode(ea_inode, "dec ref error=%d",
2146					   error2);
2147
2148		/* If there was an error, revert the quota charge. */
2149		if (error)
2150			ext4_xattr_inode_free_quota(inode, ea_inode,
2151						    i_size_read(ea_inode));
2152		iput(ea_inode);
2153	}
2154	if (ce)
2155		mb_cache_entry_put(ea_block_cache, ce);
2156	brelse(new_bh);
2157	if (!(bs->bh && s->base == bs->bh->b_data))
2158		kfree(s->base);
2159
2160	return error;
2161
2162cleanup_dquot:
2163	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2164	goto cleanup;
2165
2166bad_block:
2167	EXT4_ERROR_INODE(inode, "bad block %llu",
2168			 EXT4_I(inode)->i_file_acl);
2169	goto cleanup;
2170
2171#undef header
2172}
2173
2174int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2175			  struct ext4_xattr_ibody_find *is)
2176{
2177	struct ext4_xattr_ibody_header *header;
2178	struct ext4_inode *raw_inode;
2179	int error;
2180
2181	if (EXT4_I(inode)->i_extra_isize == 0)
2182		return 0;
 
2183	raw_inode = ext4_raw_inode(&is->iloc);
2184	header = IHDR(inode, raw_inode);
2185	is->s.base = is->s.first = IFIRST(header);
2186	is->s.here = is->s.first;
2187	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2188	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2189		error = xattr_check_inode(inode, header, is->s.end);
2190		if (error)
2191			return error;
2192		/* Find the named attribute. */
2193		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2194					 i->name_index, i->name, 0);
2195		if (error && error != -ENODATA)
2196			return error;
2197		is->s.not_found = error;
2198	}
2199	return 0;
2200}
2201
2202int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2203				struct ext4_xattr_info *i,
2204				struct ext4_xattr_ibody_find *is)
2205{
2206	struct ext4_xattr_ibody_header *header;
2207	struct ext4_xattr_search *s = &is->s;
2208	int error;
2209
2210	if (EXT4_I(inode)->i_extra_isize == 0)
2211		return -ENOSPC;
2212	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2213	if (error)
2214		return error;
2215	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2216	if (!IS_LAST_ENTRY(s->first)) {
2217		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2218		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2219	} else {
2220		header->h_magic = cpu_to_le32(0);
2221		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2222	}
2223	return 0;
2224}
2225
2226static int 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_I(inode)->i_extra_isize == 0)
2235		return -ENOSPC;
2236	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2237	if (error)
2238		return error;
2239	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2240	if (!IS_LAST_ENTRY(s->first)) {
2241		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2242		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2243	} else {
2244		header->h_magic = cpu_to_le32(0);
2245		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2246	}
2247	return 0;
2248}
2249
2250static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2251				 struct ext4_xattr_info *i)
2252{
2253	void *value;
2254
2255	/* When e_value_inum is set the value is stored externally. */
2256	if (s->here->e_value_inum)
2257		return 0;
2258	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2259		return 0;
2260	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2261	return !memcmp(value, i->value, i->value_len);
2262}
2263
2264static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2265{
2266	struct buffer_head *bh;
2267	int error;
2268
2269	if (!EXT4_I(inode)->i_file_acl)
2270		return NULL;
2271	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2272	if (IS_ERR(bh))
2273		return bh;
2274	error = ext4_xattr_check_block(inode, bh);
2275	if (error) {
2276		brelse(bh);
2277		return ERR_PTR(error);
2278	}
2279	return bh;
2280}
2281
2282/*
2283 * ext4_xattr_set_handle()
2284 *
2285 * Create, replace or remove an extended attribute for this inode.  Value
2286 * is NULL to remove an existing extended attribute, and non-NULL to
2287 * either replace an existing extended attribute, or create a new extended
2288 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2289 * specify that an extended attribute must exist and must not exist
2290 * previous to the call, respectively.
2291 *
2292 * Returns 0, or a negative error number on failure.
2293 */
2294int
2295ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2296		      const char *name, const void *value, size_t value_len,
2297		      int flags)
2298{
2299	struct ext4_xattr_info i = {
2300		.name_index = name_index,
2301		.name = name,
2302		.value = value,
2303		.value_len = value_len,
2304		.in_inode = 0,
2305	};
2306	struct ext4_xattr_ibody_find is = {
2307		.s = { .not_found = -ENODATA, },
2308	};
2309	struct ext4_xattr_block_find bs = {
2310		.s = { .not_found = -ENODATA, },
2311	};
2312	int no_expand;
2313	int error;
2314
2315	if (!name)
2316		return -EINVAL;
2317	if (strlen(name) > 255)
2318		return -ERANGE;
2319
2320	ext4_write_lock_xattr(inode, &no_expand);
2321
2322	/* Check journal credits under write lock. */
2323	if (ext4_handle_valid(handle)) {
2324		struct buffer_head *bh;
2325		int credits;
2326
2327		bh = ext4_xattr_get_block(inode);
2328		if (IS_ERR(bh)) {
2329			error = PTR_ERR(bh);
2330			goto cleanup;
2331		}
2332
2333		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2334						   value_len,
2335						   flags & XATTR_CREATE);
2336		brelse(bh);
2337
2338		if (!ext4_handle_has_enough_credits(handle, credits)) {
2339			error = -ENOSPC;
2340			goto cleanup;
2341		}
 
2342	}
2343
2344	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2345	if (error)
2346		goto cleanup;
2347
2348	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2349		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2350		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2351		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2352	}
2353
2354	error = ext4_xattr_ibody_find(inode, &i, &is);
2355	if (error)
2356		goto cleanup;
2357	if (is.s.not_found)
2358		error = ext4_xattr_block_find(inode, &i, &bs);
2359	if (error)
2360		goto cleanup;
2361	if (is.s.not_found && bs.s.not_found) {
2362		error = -ENODATA;
2363		if (flags & XATTR_REPLACE)
2364			goto cleanup;
2365		error = 0;
2366		if (!value)
2367			goto cleanup;
2368	} else {
2369		error = -EEXIST;
2370		if (flags & XATTR_CREATE)
2371			goto cleanup;
2372	}
2373
2374	if (!value) {
2375		if (!is.s.not_found)
2376			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2377		else if (!bs.s.not_found)
2378			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2379	} else {
2380		error = 0;
2381		/* Xattr value did not change? Save us some work and bail out */
2382		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2383			goto cleanup;
2384		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2385			goto cleanup;
2386
2387		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2388		    (EXT4_XATTR_SIZE(i.value_len) >
2389			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2390			i.in_inode = 1;
2391retry_inode:
2392		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2393		if (!error && !bs.s.not_found) {
2394			i.value = NULL;
2395			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2396		} else if (error == -ENOSPC) {
2397			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2398				brelse(bs.bh);
2399				bs.bh = NULL;
2400				error = ext4_xattr_block_find(inode, &i, &bs);
2401				if (error)
2402					goto cleanup;
2403			}
2404			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2405			if (!error && !is.s.not_found) {
2406				i.value = NULL;
2407				error = ext4_xattr_ibody_set(handle, inode, &i,
2408							     &is);
2409			} else if (error == -ENOSPC) {
2410				/*
2411				 * Xattr does not fit in the block, store at
2412				 * external inode if possible.
2413				 */
2414				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2415				    !i.in_inode) {
2416					i.in_inode = 1;
2417					goto retry_inode;
2418				}
2419			}
2420		}
2421	}
2422	if (!error) {
2423		ext4_xattr_update_super_block(handle, inode->i_sb);
2424		inode->i_ctime = current_time(inode);
 
2425		if (!value)
2426			no_expand = 0;
2427		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2428		/*
2429		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2430		 * error != 0.
2431		 */
2432		is.iloc.bh = NULL;
2433		if (IS_SYNC(inode))
2434			ext4_handle_sync(handle);
2435	}
 
2436
2437cleanup:
2438	brelse(is.iloc.bh);
2439	brelse(bs.bh);
2440	ext4_write_unlock_xattr(inode, &no_expand);
2441	return error;
2442}
2443
2444int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2445			   bool is_create, int *credits)
2446{
2447	struct buffer_head *bh;
2448	int err;
2449
2450	*credits = 0;
2451
2452	if (!EXT4_SB(inode->i_sb)->s_journal)
2453		return 0;
2454
2455	down_read(&EXT4_I(inode)->xattr_sem);
2456
2457	bh = ext4_xattr_get_block(inode);
2458	if (IS_ERR(bh)) {
2459		err = PTR_ERR(bh);
2460	} else {
2461		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2462						    value_len, is_create);
2463		brelse(bh);
2464		err = 0;
2465	}
2466
2467	up_read(&EXT4_I(inode)->xattr_sem);
2468	return err;
2469}
2470
2471/*
2472 * ext4_xattr_set()
2473 *
2474 * Like ext4_xattr_set_handle, but start from an inode. This extended
2475 * attribute modification is a filesystem transaction by itself.
2476 *
2477 * Returns 0, or a negative error number on failure.
2478 */
2479int
2480ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2481	       const void *value, size_t value_len, int flags)
2482{
2483	handle_t *handle;
2484	struct super_block *sb = inode->i_sb;
2485	int error, retries = 0;
2486	int credits;
2487
2488	error = dquot_initialize(inode);
2489	if (error)
2490		return error;
2491
2492retry:
2493	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2494				       &credits);
2495	if (error)
2496		return error;
2497
2498	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2499	if (IS_ERR(handle)) {
2500		error = PTR_ERR(handle);
2501	} else {
2502		int error2;
2503
2504		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2505					      value, value_len, flags);
2506		error2 = ext4_journal_stop(handle);
2507		if (error == -ENOSPC &&
2508		    ext4_should_retry_alloc(sb, &retries))
2509			goto retry;
2510		if (error == 0)
2511			error = error2;
2512	}
 
2513
2514	return error;
2515}
2516
2517/*
2518 * Shift the EA entries in the inode to create space for the increased
2519 * i_extra_isize.
2520 */
2521static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2522				     int value_offs_shift, void *to,
2523				     void *from, size_t n)
2524{
2525	struct ext4_xattr_entry *last = entry;
2526	int new_offs;
2527
2528	/* We always shift xattr headers further thus offsets get lower */
2529	BUG_ON(value_offs_shift > 0);
2530
2531	/* Adjust the value offsets of the entries */
2532	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2533		if (!last->e_value_inum && last->e_value_size) {
2534			new_offs = le16_to_cpu(last->e_value_offs) +
2535							value_offs_shift;
2536			last->e_value_offs = cpu_to_le16(new_offs);
2537		}
2538	}
2539	/* Shift the entries by n bytes */
2540	memmove(to, from, n);
2541}
2542
2543/*
2544 * Move xattr pointed to by 'entry' from inode into external xattr block
2545 */
2546static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2547				    struct ext4_inode *raw_inode,
2548				    struct ext4_xattr_entry *entry)
2549{
2550	struct ext4_xattr_ibody_find *is = NULL;
2551	struct ext4_xattr_block_find *bs = NULL;
2552	char *buffer = NULL, *b_entry_name = NULL;
2553	size_t value_size = le32_to_cpu(entry->e_value_size);
2554	struct ext4_xattr_info i = {
2555		.value = NULL,
2556		.value_len = 0,
2557		.name_index = entry->e_name_index,
2558		.in_inode = !!entry->e_value_inum,
2559	};
2560	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2561	int error;
2562
2563	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2564	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2565	buffer = kmalloc(value_size, GFP_NOFS);
2566	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2567	if (!is || !bs || !buffer || !b_entry_name) {
2568		error = -ENOMEM;
2569		goto out;
2570	}
2571
2572	is->s.not_found = -ENODATA;
2573	bs->s.not_found = -ENODATA;
2574	is->iloc.bh = NULL;
2575	bs->bh = NULL;
2576
2577	/* Save the entry name and the entry value */
2578	if (entry->e_value_inum) {
2579		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2580		if (error)
2581			goto out;
2582	} else {
2583		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2584		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2585	}
2586
2587	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2588	b_entry_name[entry->e_name_len] = '\0';
2589	i.name = b_entry_name;
2590
2591	error = ext4_get_inode_loc(inode, &is->iloc);
2592	if (error)
2593		goto out;
2594
2595	error = ext4_xattr_ibody_find(inode, &i, is);
2596	if (error)
2597		goto out;
2598
2599	/* Remove the chosen entry from the inode */
2600	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2601	if (error)
2602		goto out;
2603
2604	i.value = buffer;
2605	i.value_len = value_size;
2606	error = ext4_xattr_block_find(inode, &i, bs);
2607	if (error)
2608		goto out;
2609
2610	/* Add entry which was removed from the inode into the block */
2611	error = ext4_xattr_block_set(handle, inode, &i, bs);
2612	if (error)
2613		goto out;
2614	error = 0;
2615out:
2616	kfree(b_entry_name);
2617	kfree(buffer);
2618	if (is)
2619		brelse(is->iloc.bh);
2620	if (bs)
2621		brelse(bs->bh);
2622	kfree(is);
2623	kfree(bs);
2624
2625	return error;
2626}
2627
2628static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2629				       struct ext4_inode *raw_inode,
2630				       int isize_diff, size_t ifree,
2631				       size_t bfree, int *total_ino)
2632{
2633	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2634	struct ext4_xattr_entry *small_entry;
2635	struct ext4_xattr_entry *entry;
2636	struct ext4_xattr_entry *last;
2637	unsigned int entry_size;	/* EA entry size */
2638	unsigned int total_size;	/* EA entry size + value size */
2639	unsigned int min_total_size;
2640	int error;
2641
2642	while (isize_diff > ifree) {
2643		entry = NULL;
2644		small_entry = NULL;
2645		min_total_size = ~0U;
2646		last = IFIRST(header);
2647		/* Find the entry best suited to be pushed into EA block */
2648		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2649			/* never move system.data out of the inode */
2650			if ((last->e_name_len == 4) &&
2651			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2652			    !memcmp(last->e_name, "data", 4))
2653				continue;
2654			total_size = EXT4_XATTR_LEN(last->e_name_len);
2655			if (!last->e_value_inum)
2656				total_size += EXT4_XATTR_SIZE(
2657					       le32_to_cpu(last->e_value_size));
2658			if (total_size <= bfree &&
2659			    total_size < min_total_size) {
2660				if (total_size + ifree < isize_diff) {
2661					small_entry = last;
2662				} else {
2663					entry = last;
2664					min_total_size = total_size;
2665				}
2666			}
2667		}
2668
2669		if (entry == NULL) {
2670			if (small_entry == NULL)
2671				return -ENOSPC;
2672			entry = small_entry;
2673		}
2674
2675		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2676		total_size = entry_size;
2677		if (!entry->e_value_inum)
2678			total_size += EXT4_XATTR_SIZE(
2679					      le32_to_cpu(entry->e_value_size));
2680		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2681						 entry);
2682		if (error)
2683			return error;
2684
2685		*total_ino -= entry_size;
2686		ifree += total_size;
2687		bfree -= total_size;
2688	}
2689
2690	return 0;
2691}
2692
2693/*
2694 * Expand an inode by new_extra_isize bytes when EAs are present.
2695 * Returns 0 on success or negative error number on failure.
2696 */
2697int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2698			       struct ext4_inode *raw_inode, handle_t *handle)
2699{
2700	struct ext4_xattr_ibody_header *header;
2701	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2702	static unsigned int mnt_count;
2703	size_t min_offs;
2704	size_t ifree, bfree;
2705	int total_ino;
2706	void *base, *end;
2707	int error = 0, tried_min_extra_isize = 0;
2708	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2709	int isize_diff;	/* How much do we need to grow i_extra_isize */
2710
2711retry:
2712	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2713	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2714		return 0;
2715
2716	header = IHDR(inode, raw_inode);
2717
2718	/*
2719	 * Check if enough free space is available in the inode to shift the
2720	 * entries ahead by new_extra_isize.
2721	 */
2722
2723	base = IFIRST(header);
2724	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2725	min_offs = end - base;
2726	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2727
2728	error = xattr_check_inode(inode, header, end);
2729	if (error)
2730		goto cleanup;
2731
2732	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2733	if (ifree >= isize_diff)
2734		goto shift;
2735
2736	/*
2737	 * Enough free space isn't available in the inode, check if
2738	 * EA block can hold new_extra_isize bytes.
2739	 */
2740	if (EXT4_I(inode)->i_file_acl) {
2741		struct buffer_head *bh;
2742
2743		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2744		if (IS_ERR(bh)) {
2745			error = PTR_ERR(bh);
2746			goto cleanup;
2747		}
2748		error = ext4_xattr_check_block(inode, bh);
2749		if (error) {
2750			brelse(bh);
2751			goto cleanup;
2752		}
2753		base = BHDR(bh);
2754		end = bh->b_data + bh->b_size;
2755		min_offs = end - base;
2756		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2757					      NULL);
2758		brelse(bh);
2759		if (bfree + ifree < isize_diff) {
2760			if (!tried_min_extra_isize && s_min_extra_isize) {
2761				tried_min_extra_isize++;
2762				new_extra_isize = s_min_extra_isize;
2763				goto retry;
2764			}
2765			error = -ENOSPC;
2766			goto cleanup;
2767		}
2768	} else {
2769		bfree = inode->i_sb->s_blocksize;
2770	}
2771
2772	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2773					    isize_diff, ifree, bfree,
2774					    &total_ino);
2775	if (error) {
2776		if (error == -ENOSPC && !tried_min_extra_isize &&
2777		    s_min_extra_isize) {
2778			tried_min_extra_isize++;
2779			new_extra_isize = s_min_extra_isize;
2780			goto retry;
2781		}
2782		goto cleanup;
2783	}
2784shift:
2785	/* Adjust the offsets and shift the remaining entries ahead */
2786	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2787			- new_extra_isize, (void *)raw_inode +
2788			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2789			(void *)header, total_ino);
2790	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2791
2792cleanup:
2793	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2794		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2795			     inode->i_ino);
2796		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2797	}
2798	return error;
2799}
2800
2801#define EIA_INCR 16 /* must be 2^n */
2802#define EIA_MASK (EIA_INCR - 1)
2803
2804/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2805 * If @ea_inode_array is new or full it will be grown and the old
2806 * contents copied over.
2807 */
2808static int
2809ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2810			struct inode *inode)
2811{
2812	if (*ea_inode_array == NULL) {
2813		/*
2814		 * Start with 15 inodes, so it fits into a power-of-two size.
2815		 * If *ea_inode_array is NULL, this is essentially offsetof()
2816		 */
2817		(*ea_inode_array) =
2818			kmalloc(offsetof(struct ext4_xattr_inode_array,
2819					 inodes[EIA_MASK]),
2820				GFP_NOFS);
2821		if (*ea_inode_array == NULL)
2822			return -ENOMEM;
2823		(*ea_inode_array)->count = 0;
2824	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2825		/* expand the array once all 15 + n * 16 slots are full */
2826		struct ext4_xattr_inode_array *new_array = NULL;
2827		int count = (*ea_inode_array)->count;
2828
2829		/* if new_array is NULL, this is essentially offsetof() */
2830		new_array = kmalloc(
2831				offsetof(struct ext4_xattr_inode_array,
2832					 inodes[count + EIA_INCR]),
2833				GFP_NOFS);
2834		if (new_array == NULL)
2835			return -ENOMEM;
2836		memcpy(new_array, *ea_inode_array,
2837		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2838		kfree(*ea_inode_array);
2839		*ea_inode_array = new_array;
2840	}
2841	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2842	return 0;
2843}
2844
2845/*
2846 * ext4_xattr_delete_inode()
2847 *
2848 * Free extended attribute resources associated with this inode. Traverse
2849 * all entries and decrement reference on any xattr inodes associated with this
2850 * inode. This is called immediately before an inode is freed. We have exclusive
2851 * access to the inode. If an orphan inode is deleted it will also release its
2852 * references on xattr block and xattr inodes.
2853 */
2854int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2855			    struct ext4_xattr_inode_array **ea_inode_array,
2856			    int extra_credits)
2857{
2858	struct buffer_head *bh = NULL;
2859	struct ext4_xattr_ibody_header *header;
2860	struct ext4_iloc iloc = { .bh = NULL };
2861	struct ext4_xattr_entry *entry;
2862	struct inode *ea_inode;
2863	int error;
2864
2865	error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2866					  NULL /* bh */,
2867					  false /* dirty */,
2868					  false /* block_csum */);
2869	if (error) {
2870		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2871		goto cleanup;
2872	}
2873
2874	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2875	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2876
2877		error = ext4_get_inode_loc(inode, &iloc);
2878		if (error) {
2879			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2880			goto cleanup;
2881		}
2882
2883		error = ext4_journal_get_write_access(handle, iloc.bh);
 
2884		if (error) {
2885			EXT4_ERROR_INODE(inode, "write access (error %d)",
2886					 error);
2887			goto cleanup;
2888		}
2889
2890		header = IHDR(inode, ext4_raw_inode(&iloc));
2891		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2892			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2893						     IFIRST(header),
2894						     false /* block_csum */,
2895						     ea_inode_array,
2896						     extra_credits,
2897						     false /* skip_quota */);
2898	}
2899
2900	if (EXT4_I(inode)->i_file_acl) {
2901		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2902		if (IS_ERR(bh)) {
2903			error = PTR_ERR(bh);
2904			if (error == -EIO)
2905				EXT4_ERROR_INODE(inode, "block %llu read error",
2906						 EXT4_I(inode)->i_file_acl);
 
 
2907			bh = NULL;
2908			goto cleanup;
2909		}
2910		error = ext4_xattr_check_block(inode, bh);
2911		if (error)
2912			goto cleanup;
2913
2914		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2915			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2916			     entry = EXT4_XATTR_NEXT(entry)) {
2917				if (!entry->e_value_inum)
2918					continue;
2919				error = ext4_xattr_inode_iget(inode,
2920					      le32_to_cpu(entry->e_value_inum),
2921					      le32_to_cpu(entry->e_hash),
2922					      &ea_inode);
2923				if (error)
2924					continue;
2925				ext4_xattr_inode_free_quota(inode, ea_inode,
2926					      le32_to_cpu(entry->e_value_size));
2927				iput(ea_inode);
2928			}
2929
2930		}
2931
2932		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2933					 extra_credits);
2934		/*
2935		 * Update i_file_acl value in the same transaction that releases
2936		 * block.
2937		 */
2938		EXT4_I(inode)->i_file_acl = 0;
2939		error = ext4_mark_inode_dirty(handle, inode);
2940		if (error) {
2941			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2942					 error);
2943			goto cleanup;
2944		}
 
2945	}
2946	error = 0;
2947cleanup:
2948	brelse(iloc.bh);
2949	brelse(bh);
2950	return error;
2951}
2952
2953void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2954{
2955	int idx;
2956
2957	if (ea_inode_array == NULL)
2958		return;
2959
2960	for (idx = 0; idx < ea_inode_array->count; ++idx)
2961		iput(ea_inode_array->inodes[idx]);
2962	kfree(ea_inode_array);
2963}
2964
2965/*
2966 * ext4_xattr_block_cache_insert()
2967 *
2968 * Create a new entry in the extended attribute block cache, and insert
2969 * it unless such an entry is already in the cache.
2970 *
2971 * Returns 0, or a negative error number on failure.
2972 */
2973static void
2974ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2975			      struct buffer_head *bh)
2976{
2977	struct ext4_xattr_header *header = BHDR(bh);
2978	__u32 hash = le32_to_cpu(header->h_hash);
2979	int reusable = le32_to_cpu(header->h_refcount) <
2980		       EXT4_XATTR_REFCOUNT_MAX;
2981	int error;
2982
2983	if (!ea_block_cache)
2984		return;
2985	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2986				      bh->b_blocknr, reusable);
2987	if (error) {
2988		if (error == -EBUSY)
2989			ea_bdebug(bh, "already in cache");
2990	} else
2991		ea_bdebug(bh, "inserting [%x]", (int)hash);
2992}
2993
2994/*
2995 * ext4_xattr_cmp()
2996 *
2997 * Compare two extended attribute blocks for equality.
2998 *
2999 * Returns 0 if the blocks are equal, 1 if they differ, and
3000 * a negative error number on errors.
3001 */
3002static int
3003ext4_xattr_cmp(struct ext4_xattr_header *header1,
3004	       struct ext4_xattr_header *header2)
3005{
3006	struct ext4_xattr_entry *entry1, *entry2;
3007
3008	entry1 = ENTRY(header1+1);
3009	entry2 = ENTRY(header2+1);
3010	while (!IS_LAST_ENTRY(entry1)) {
3011		if (IS_LAST_ENTRY(entry2))
3012			return 1;
3013		if (entry1->e_hash != entry2->e_hash ||
3014		    entry1->e_name_index != entry2->e_name_index ||
3015		    entry1->e_name_len != entry2->e_name_len ||
3016		    entry1->e_value_size != entry2->e_value_size ||
3017		    entry1->e_value_inum != entry2->e_value_inum ||
3018		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3019			return 1;
3020		if (!entry1->e_value_inum &&
3021		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3022			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3023			   le32_to_cpu(entry1->e_value_size)))
3024			return 1;
3025
3026		entry1 = EXT4_XATTR_NEXT(entry1);
3027		entry2 = EXT4_XATTR_NEXT(entry2);
3028	}
3029	if (!IS_LAST_ENTRY(entry2))
3030		return 1;
3031	return 0;
3032}
3033
3034/*
3035 * ext4_xattr_block_cache_find()
3036 *
3037 * Find an identical extended attribute block.
3038 *
3039 * Returns a pointer to the block found, or NULL if such a block was
3040 * not found or an error occurred.
3041 */
3042static struct buffer_head *
3043ext4_xattr_block_cache_find(struct inode *inode,
3044			    struct ext4_xattr_header *header,
3045			    struct mb_cache_entry **pce)
3046{
3047	__u32 hash = le32_to_cpu(header->h_hash);
3048	struct mb_cache_entry *ce;
3049	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3050
3051	if (!ea_block_cache)
3052		return NULL;
3053	if (!header->h_hash)
3054		return NULL;  /* never share */
3055	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3056	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3057	while (ce) {
3058		struct buffer_head *bh;
3059
3060		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3061		if (IS_ERR(bh)) {
3062			if (PTR_ERR(bh) == -ENOMEM)
3063				return NULL;
3064			bh = NULL;
3065			EXT4_ERROR_INODE(inode, "block %lu read error",
3066					 (unsigned long)ce->e_value);
3067		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3068			*pce = ce;
3069			return bh;
3070		}
3071		brelse(bh);
3072		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3073	}
3074	return NULL;
3075}
3076
3077#define NAME_HASH_SHIFT 5
3078#define VALUE_HASH_SHIFT 16
3079
3080/*
3081 * ext4_xattr_hash_entry()
3082 *
3083 * Compute the hash of an extended attribute.
3084 */
3085static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3086				    size_t value_count)
3087{
3088	__u32 hash = 0;
3089
3090	while (name_len--) {
3091		hash = (hash << NAME_HASH_SHIFT) ^
3092		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3093		       *name++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3094	}
3095	while (value_count--) {
3096		hash = (hash << VALUE_HASH_SHIFT) ^
3097		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3098		       le32_to_cpu(*value++);
3099	}
3100	return cpu_to_le32(hash);
3101}
3102
3103#undef NAME_HASH_SHIFT
3104#undef VALUE_HASH_SHIFT
3105
3106#define BLOCK_HASH_SHIFT 16
3107
3108/*
3109 * ext4_xattr_rehash()
3110 *
3111 * Re-compute the extended attribute hash value after an entry has changed.
3112 */
3113static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3114{
3115	struct ext4_xattr_entry *here;
3116	__u32 hash = 0;
3117
3118	here = ENTRY(header+1);
3119	while (!IS_LAST_ENTRY(here)) {
3120		if (!here->e_hash) {
3121			/* Block is not shared if an entry's hash value == 0 */
3122			hash = 0;
3123			break;
3124		}
3125		hash = (hash << BLOCK_HASH_SHIFT) ^
3126		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3127		       le32_to_cpu(here->e_hash);
3128		here = EXT4_XATTR_NEXT(here);
3129	}
3130	header->h_hash = cpu_to_le32(hash);
3131}
3132
3133#undef BLOCK_HASH_SHIFT
3134
3135#define	HASH_BUCKET_BITS	10
3136
3137struct mb_cache *
3138ext4_xattr_create_cache(void)
3139{
3140	return mb_cache_create(HASH_BUCKET_BITS);
3141}
3142
3143void ext4_xattr_destroy_cache(struct mb_cache *cache)
3144{
3145	if (cache)
3146		mb_cache_destroy(cache);
3147}
3148