xattr.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * xattr.c
   4 *
   5 * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   6 *
   7 * CREDITS:
   8 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
   9 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  10 */
  11
  12#include <linux/capability.h>
  13#include <linux/fs.h>
  14#include <linux/types.h>
  15#include <linux/slab.h>
  16#include <linux/highmem.h>
  17#include <linux/pagemap.h>
  18#include <linux/uio.h>
  19#include <linux/sched.h>
  20#include <linux/splice.h>
  21#include <linux/mount.h>
  22#include <linux/writeback.h>
  23#include <linux/falloc.h>
  24#include <linux/sort.h>
  25#include <linux/init.h>
  26#include <linux/module.h>
  27#include <linux/string.h>
  28#include <linux/security.h>
  29
  30#include <cluster/masklog.h>
  31
  32#include "ocfs2.h"
  33#include "alloc.h"
  34#include "blockcheck.h"
  35#include "dlmglue.h"
  36#include "file.h"
  37#include "symlink.h"
  38#include "sysfile.h"
  39#include "inode.h"
  40#include "journal.h"
  41#include "ocfs2_fs.h"
  42#include "suballoc.h"
  43#include "uptodate.h"
  44#include "buffer_head_io.h"
  45#include "super.h"
  46#include "xattr.h"
  47#include "refcounttree.h"
  48#include "acl.h"
  49#include "ocfs2_trace.h"
  50
  51struct ocfs2_xattr_def_value_root {
  52	/* Must be last as it ends in a flexible-array member. */
  53	TRAILING_OVERLAP(struct ocfs2_xattr_value_root, xv, xr_list.l_recs,
  54		struct ocfs2_extent_rec		er;
  55	);
  56};
  57static_assert(offsetof(struct ocfs2_xattr_def_value_root, xv.xr_list.l_recs) ==
  58	      offsetof(struct ocfs2_xattr_def_value_root, er));
  59
  60struct ocfs2_xattr_bucket {
  61	/* The inode these xattrs are associated with */
  62	struct inode *bu_inode;
  63
  64	/* The actual buffers that make up the bucket */
  65	struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  66
  67	/* How many blocks make up one bucket for this filesystem */
  68	int bu_blocks;
  69};
  70
  71struct ocfs2_xattr_set_ctxt {
  72	handle_t *handle;
  73	struct ocfs2_alloc_context *meta_ac;
  74	struct ocfs2_alloc_context *data_ac;
  75	struct ocfs2_cached_dealloc_ctxt dealloc;
  76	int set_abort;
  77};
  78
  79#define OCFS2_XATTR_ROOT_SIZE	(sizeof(struct ocfs2_xattr_def_value_root))
  80#define OCFS2_XATTR_INLINE_SIZE	80
  81#define OCFS2_XATTR_HEADER_GAP	4
  82#define OCFS2_XATTR_FREE_IN_IBODY	(OCFS2_MIN_XATTR_INLINE_SIZE \
  83					 - sizeof(struct ocfs2_xattr_header) \
  84					 - OCFS2_XATTR_HEADER_GAP)
  85#define OCFS2_XATTR_FREE_IN_BLOCK(ptr)	((ptr)->i_sb->s_blocksize \
  86					 - sizeof(struct ocfs2_xattr_block) \
  87					 - sizeof(struct ocfs2_xattr_header) \
  88					 - OCFS2_XATTR_HEADER_GAP)
  89
  90static struct ocfs2_xattr_def_value_root def_xv = {
  91	.xv.xr_list.l_count = cpu_to_le16(1),
  92};
  93
  94const struct xattr_handler * const ocfs2_xattr_handlers[] = {
  95	&ocfs2_xattr_user_handler,
  96	&ocfs2_xattr_trusted_handler,
  97	&ocfs2_xattr_security_handler,
  98	NULL
  99};
 100
 101static const struct xattr_handler * const ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 102	[OCFS2_XATTR_INDEX_USER]		= &ocfs2_xattr_user_handler,
 103	[OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]	= &nop_posix_acl_access,
 104	[OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]	= &nop_posix_acl_default,
 105	[OCFS2_XATTR_INDEX_TRUSTED]		= &ocfs2_xattr_trusted_handler,
 106	[OCFS2_XATTR_INDEX_SECURITY]		= &ocfs2_xattr_security_handler,
 107};
 108
 109struct ocfs2_xattr_info {
 110	int		xi_name_index;
 111	const char	*xi_name;
 112	int		xi_name_len;
 113	const void	*xi_value;
 114	size_t		xi_value_len;
 115};
 116
 117struct ocfs2_xattr_search {
 118	struct buffer_head *inode_bh;
 119	/*
 120	 * xattr_bh point to the block buffer head which has extended attribute
 121	 * when extended attribute in inode, xattr_bh is equal to inode_bh.
 122	 */
 123	struct buffer_head *xattr_bh;
 124	struct ocfs2_xattr_header *header;
 125	struct ocfs2_xattr_bucket *bucket;
 126	void *base;
 127	void *end;
 128	struct ocfs2_xattr_entry *here;
 129	int not_found;
 130};
 131
 132/* Operations on struct ocfs2_xa_entry */
 133struct ocfs2_xa_loc;
 134struct ocfs2_xa_loc_operations {
 135	/*
 136	 * Journal functions
 137	 */
 138	int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 139				  int type);
 140	void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 141
 142	/*
 143	 * Return a pointer to the appropriate buffer in loc->xl_storage
 144	 * at the given offset from loc->xl_header.
 145	 */
 146	void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 147
 148	/* Can we reuse the existing entry for the new value? */
 149	int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 150			     struct ocfs2_xattr_info *xi);
 151
 152	/* How much space is needed for the new value? */
 153	int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 154			       struct ocfs2_xattr_info *xi);
 155
 156	/*
 157	 * Return the offset of the first name+value pair.  This is
 158	 * the start of our downward-filling free space.
 159	 */
 160	int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 161
 162	/*
 163	 * Remove the name+value at this location.  Do whatever is
 164	 * appropriate with the remaining name+value pairs.
 165	 */
 166	void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 167
 168	/* Fill xl_entry with a new entry */
 169	void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 170
 171	/* Add name+value storage to an entry */
 172	void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 173
 174	/*
 175	 * Initialize the value buf's access and bh fields for this entry.
 176	 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 177	 */
 178	void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 179				   struct ocfs2_xattr_value_buf *vb);
 180};
 181
 182/*
 183 * Describes an xattr entry location.  This is a memory structure
 184 * tracking the on-disk structure.
 185 */
 186struct ocfs2_xa_loc {
 187	/* This xattr belongs to this inode */
 188	struct inode *xl_inode;
 189
 190	/* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 191	struct ocfs2_xattr_header *xl_header;
 192
 193	/* Bytes from xl_header to the end of the storage */
 194	int xl_size;
 195
 196	/*
 197	 * The ocfs2_xattr_entry this location describes.  If this is
 198	 * NULL, this location describes the on-disk structure where it
 199	 * would have been.
 200	 */
 201	struct ocfs2_xattr_entry *xl_entry;
 202
 203	/*
 204	 * Internal housekeeping
 205	 */
 206
 207	/* Buffer(s) containing this entry */
 208	void *xl_storage;
 209
 210	/* Operations on the storage backing this location */
 211	const struct ocfs2_xa_loc_operations *xl_ops;
 212};
 213
 214/*
 215 * Convenience functions to calculate how much space is needed for a
 216 * given name+value pair
 217 */
 218static int namevalue_size(int name_len, uint64_t value_len)
 219{
 220	if (value_len > OCFS2_XATTR_INLINE_SIZE)
 221		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 222	else
 223		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 224}
 225
 226static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 227{
 228	return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 229}
 230
 231static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 232{
 233	u64 value_len = le64_to_cpu(xe->xe_value_size);
 234
 235	BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 236	       ocfs2_xattr_is_local(xe));
 237	return namevalue_size(xe->xe_name_len, value_len);
 238}
 239
 240
 241static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 242					     struct ocfs2_xattr_header *xh,
 243					     int index,
 244					     int *block_off,
 245					     int *new_offset);
 246
 247static int ocfs2_xattr_block_find(struct inode *inode,
 248				  int name_index,
 249				  const char *name,
 250				  struct ocfs2_xattr_search *xs);
 251static int ocfs2_xattr_index_block_find(struct inode *inode,
 252					struct buffer_head *root_bh,
 253					int name_index,
 254					const char *name,
 255					struct ocfs2_xattr_search *xs);
 256
 257static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 258					struct buffer_head *blk_bh,
 259					char *buffer,
 260					size_t buffer_size);
 261
 262static int ocfs2_xattr_create_index_block(struct inode *inode,
 263					  struct ocfs2_xattr_search *xs,
 264					  struct ocfs2_xattr_set_ctxt *ctxt);
 265
 266static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 267					     struct ocfs2_xattr_info *xi,
 268					     struct ocfs2_xattr_search *xs,
 269					     struct ocfs2_xattr_set_ctxt *ctxt);
 270
 271typedef int (xattr_tree_rec_func)(struct inode *inode,
 272				  struct buffer_head *root_bh,
 273				  u64 blkno, u32 cpos, u32 len, void *para);
 274static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 275					   struct buffer_head *root_bh,
 276					   xattr_tree_rec_func *rec_func,
 277					   void *para);
 278static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 279					struct ocfs2_xattr_bucket *bucket,
 280					void *para);
 281static int ocfs2_rm_xattr_cluster(struct inode *inode,
 282				  struct buffer_head *root_bh,
 283				  u64 blkno,
 284				  u32 cpos,
 285				  u32 len,
 286				  void *para);
 287
 288static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 289				  u64 src_blk, u64 last_blk, u64 to_blk,
 290				  unsigned int start_bucket,
 291				  u32 *first_hash);
 292static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 293					struct ocfs2_dinode *di,
 294					struct ocfs2_xattr_info *xi,
 295					struct ocfs2_xattr_search *xis,
 296					struct ocfs2_xattr_search *xbs,
 297					struct ocfs2_refcount_tree **ref_tree,
 298					int *meta_need,
 299					int *credits);
 300static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 301					   struct ocfs2_xattr_bucket *bucket,
 302					   int offset,
 303					   struct ocfs2_xattr_value_root **xv,
 304					   struct buffer_head **bh);
 305
 306static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 307{
 308	return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 309}
 310
 311static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 312{
 313	return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 314}
 315
 316#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 317#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 318#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 319
 320static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 321{
 322	struct ocfs2_xattr_bucket *bucket;
 323	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 324
 325	BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 326
 327	bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 328	if (bucket) {
 329		bucket->bu_inode = inode;
 330		bucket->bu_blocks = blks;
 331	}
 332
 333	return bucket;
 334}
 335
 336static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 337{
 338	int i;
 339
 340	for (i = 0; i < bucket->bu_blocks; i++) {
 341		brelse(bucket->bu_bhs[i]);
 342		bucket->bu_bhs[i] = NULL;
 343	}
 344}
 345
 346static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 347{
 348	if (bucket) {
 349		ocfs2_xattr_bucket_relse(bucket);
 350		bucket->bu_inode = NULL;
 351		kfree(bucket);
 352	}
 353}
 354
 355/*
 356 * A bucket that has never been written to disk doesn't need to be
 357 * read.  We just need the buffer_heads.  Don't call this for
 358 * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 359 * them fully.
 360 */
 361static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 362				   u64 xb_blkno, int new)
 363{
 364	int i, rc = 0;
 365
 366	for (i = 0; i < bucket->bu_blocks; i++) {
 367		bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 368					      xb_blkno + i);
 369		if (!bucket->bu_bhs[i]) {
 370			rc = -ENOMEM;
 371			mlog_errno(rc);
 372			break;
 373		}
 374
 375		if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 376					   bucket->bu_bhs[i])) {
 377			if (new)
 378				ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 379							      bucket->bu_bhs[i]);
 380			else {
 381				set_buffer_uptodate(bucket->bu_bhs[i]);
 382				ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 383							  bucket->bu_bhs[i]);
 384			}
 385		}
 386	}
 387
 388	if (rc)
 389		ocfs2_xattr_bucket_relse(bucket);
 390	return rc;
 391}
 392
 393/* Read the xattr bucket at xb_blkno */
 394static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 395				   u64 xb_blkno)
 396{
 397	int rc;
 398
 399	rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 400			       bucket->bu_blocks, bucket->bu_bhs, 0,
 401			       NULL);
 402	if (!rc) {
 403		spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 404		rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 405						 bucket->bu_bhs,
 406						 bucket->bu_blocks,
 407						 &bucket_xh(bucket)->xh_check);
 408		spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 409		if (rc)
 410			mlog_errno(rc);
 411	}
 412
 413	if (rc)
 414		ocfs2_xattr_bucket_relse(bucket);
 415	return rc;
 416}
 417
 418static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 419					     struct ocfs2_xattr_bucket *bucket,
 420					     int type)
 421{
 422	int i, rc = 0;
 423
 424	for (i = 0; i < bucket->bu_blocks; i++) {
 425		rc = ocfs2_journal_access(handle,
 426					  INODE_CACHE(bucket->bu_inode),
 427					  bucket->bu_bhs[i], type);
 428		if (rc) {
 429			mlog_errno(rc);
 430			break;
 431		}
 432	}
 433
 434	return rc;
 435}
 436
 437static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 438					     struct ocfs2_xattr_bucket *bucket)
 439{
 440	int i;
 441
 442	spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 443	ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 444				   bucket->bu_bhs, bucket->bu_blocks,
 445				   &bucket_xh(bucket)->xh_check);
 446	spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 447
 448	for (i = 0; i < bucket->bu_blocks; i++)
 449		ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 450}
 451
 452static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 453					 struct ocfs2_xattr_bucket *src)
 454{
 455	int i;
 456	int blocksize = src->bu_inode->i_sb->s_blocksize;
 457
 458	BUG_ON(dest->bu_blocks != src->bu_blocks);
 459	BUG_ON(dest->bu_inode != src->bu_inode);
 460
 461	for (i = 0; i < src->bu_blocks; i++) {
 462		memcpy(bucket_block(dest, i), bucket_block(src, i),
 463		       blocksize);
 464	}
 465}
 466
 467static int ocfs2_validate_xattr_block(struct super_block *sb,
 468				      struct buffer_head *bh)
 469{
 470	int rc;
 471	struct ocfs2_xattr_block *xb =
 472		(struct ocfs2_xattr_block *)bh->b_data;
 473
 474	trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 475
 476	BUG_ON(!buffer_uptodate(bh));
 477
 478	/*
 479	 * If the ecc fails, we return the error but otherwise
 480	 * leave the filesystem running.  We know any error is
 481	 * local to this block.
 482	 */
 483	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 484	if (rc)
 485		return rc;
 486
 487	/*
 488	 * Errors after here are fatal
 489	 */
 490
 491	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 492		return ocfs2_error(sb,
 493				   "Extended attribute block #%llu has bad signature %.*s\n",
 494				   (unsigned long long)bh->b_blocknr, 7,
 495				   xb->xb_signature);
 496	}
 497
 498	if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 499		return ocfs2_error(sb,
 500				   "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 501				   (unsigned long long)bh->b_blocknr,
 502				   (unsigned long long)le64_to_cpu(xb->xb_blkno));
 503	}
 504
 505	if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 506		return ocfs2_error(sb,
 507				   "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 508				   (unsigned long long)bh->b_blocknr,
 509				   le32_to_cpu(xb->xb_fs_generation));
 510	}
 511
 512	return 0;
 513}
 514
 515static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 516				  struct buffer_head **bh)
 517{
 518	int rc;
 519	struct buffer_head *tmp = *bh;
 520
 521	rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 522			      ocfs2_validate_xattr_block);
 523
 524	/* If ocfs2_read_block() got us a new bh, pass it up. */
 525	if (!rc && !*bh)
 526		*bh = tmp;
 527
 528	return rc;
 529}
 530
 531static inline const char *ocfs2_xattr_prefix(int name_index)
 532{
 533	const struct xattr_handler *handler = NULL;
 534
 535	if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 536		handler = ocfs2_xattr_handler_map[name_index];
 537	return handler ? xattr_prefix(handler) : NULL;
 538}
 539
 540static u32 ocfs2_xattr_name_hash(struct inode *inode,
 541				 const char *name,
 542				 int name_len)
 543{
 544	/* Get hash value of uuid from super block */
 545	u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 546	int i;
 547
 548	/* hash extended attribute name */
 549	for (i = 0; i < name_len; i++) {
 550		hash = (hash << OCFS2_HASH_SHIFT) ^
 551		       (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 552		       *name++;
 553	}
 554
 555	return hash;
 556}
 557
 558static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 559{
 560	return namevalue_size(name_len, value_len) +
 561		sizeof(struct ocfs2_xattr_entry);
 562}
 563
 564static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 565{
 566	return namevalue_size_xi(xi) +
 567		sizeof(struct ocfs2_xattr_entry);
 568}
 569
 570static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 571{
 572	return namevalue_size_xe(xe) +
 573		sizeof(struct ocfs2_xattr_entry);
 574}
 575
 576int ocfs2_calc_security_init(struct inode *dir,
 577			     struct ocfs2_security_xattr_info *si,
 578			     int *want_clusters,
 579			     int *xattr_credits,
 580			     struct ocfs2_alloc_context **xattr_ac)
 581{
 582	int ret = 0;
 583	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 584	int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 585						 si->value_len);
 586
 587	/*
 588	 * The max space of security xattr taken inline is
 589	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
 590	 * So reserve one metadata block for it is ok.
 591	 */
 592	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 593	    s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 594		ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 595		if (ret) {
 596			mlog_errno(ret);
 597			return ret;
 598		}
 599		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 600	}
 601
 602	/* reserve clusters for xattr value which will be set in B tree*/
 603	if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 604		int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 605							    si->value_len);
 606
 607		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 608							   new_clusters);
 609		*want_clusters += new_clusters;
 610	}
 611	return ret;
 612}
 613
 614int ocfs2_calc_xattr_init(struct inode *dir,
 615			  struct buffer_head *dir_bh,
 616			  umode_t mode,
 617			  struct ocfs2_security_xattr_info *si,
 618			  int *want_clusters,
 619			  int *xattr_credits,
 620			  int *want_meta)
 621{
 622	int ret = 0;
 623	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 624	int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 625
 626	if (si->enable)
 627		s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 628						     si->value_len);
 629
 630	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 631		down_read(&OCFS2_I(dir)->ip_xattr_sem);
 632		acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 633					OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 634					"", NULL, 0);
 635		up_read(&OCFS2_I(dir)->ip_xattr_sem);
 636		if (acl_len > 0) {
 637			a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 638			if (S_ISDIR(mode))
 639				a_size <<= 1;
 640		} else if (acl_len != 0 && acl_len != -ENODATA) {
 641			ret = acl_len;
 642			mlog_errno(ret);
 643			return ret;
 644		}
 645	}
 646
 647	if (!(s_size + a_size))
 648		return ret;
 649
 650	/*
 651	 * The max space of security xattr taken inline is
 652	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
 653	 * The max space of acl xattr taken inline is
 654	 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 655	 * when blocksize = 512, may reserve one more cluster for
 656	 * xattr bucket, otherwise reserve one metadata block
 657	 * for them is ok.
 658	 * If this is a new directory with inline data,
 659	 * we choose to reserve the entire inline area for
 660	 * directory contents and force an external xattr block.
 661	 */
 662	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 663	    (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 664	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 665		*want_meta = *want_meta + 1;
 666		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 667	}
 668
 669	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 670	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 671		*want_clusters += 1;
 672		*xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 673	}
 674
 675	/*
 676	 * reserve credits and clusters for xattrs which has large value
 677	 * and have to be set outside
 678	 */
 679	if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 680		new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 681							si->value_len);
 682		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 683							   new_clusters);
 684		*want_clusters += new_clusters;
 685	}
 686	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 687	    acl_len > OCFS2_XATTR_INLINE_SIZE) {
 688		/* for directory, it has DEFAULT and ACCESS two types of acls */
 689		new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 690				ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 691		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 692							   new_clusters);
 693		*want_clusters += new_clusters;
 694	}
 695
 696	return ret;
 697}
 698
 699static int ocfs2_xattr_extend_allocation(struct inode *inode,
 700					 u32 clusters_to_add,
 701					 struct ocfs2_xattr_value_buf *vb,
 702					 struct ocfs2_xattr_set_ctxt *ctxt)
 703{
 704	int status = 0, credits;
 705	handle_t *handle = ctxt->handle;
 706	enum ocfs2_alloc_restarted why;
 707	u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 708	struct ocfs2_extent_tree et;
 709
 710	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 711
 712	while (clusters_to_add) {
 713		trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 714
 715		status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 716				       OCFS2_JOURNAL_ACCESS_WRITE);
 717		if (status < 0) {
 718			mlog_errno(status);
 719			break;
 720		}
 721
 722		prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 723		status = ocfs2_add_clusters_in_btree(handle,
 724						     &et,
 725						     &logical_start,
 726						     clusters_to_add,
 727						     0,
 728						     ctxt->data_ac,
 729						     ctxt->meta_ac,
 730						     &why);
 731		if ((status < 0) && (status != -EAGAIN)) {
 732			if (status != -ENOSPC)
 733				mlog_errno(status);
 734			break;
 735		}
 736
 737		ocfs2_journal_dirty(handle, vb->vb_bh);
 738
 739		clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 740					 prev_clusters;
 741
 742		if (why != RESTART_NONE && clusters_to_add) {
 743			/*
 744			 * We can only fail in case the alloc file doesn't give
 745			 * up enough clusters.
 746			 */
 747			BUG_ON(why == RESTART_META);
 748
 749			credits = ocfs2_calc_extend_credits(inode->i_sb,
 750							    &vb->vb_xv->xr_list);
 751			status = ocfs2_extend_trans(handle, credits);
 752			if (status < 0) {
 753				status = -ENOMEM;
 754				mlog_errno(status);
 755				break;
 756			}
 757		}
 758	}
 759
 760	return status;
 761}
 762
 763static int __ocfs2_remove_xattr_range(struct inode *inode,
 764				      struct ocfs2_xattr_value_buf *vb,
 765				      u32 cpos, u32 phys_cpos, u32 len,
 766				      unsigned int ext_flags,
 767				      struct ocfs2_xattr_set_ctxt *ctxt)
 768{
 769	int ret;
 770	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 771	handle_t *handle = ctxt->handle;
 772	struct ocfs2_extent_tree et;
 773
 774	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 775
 776	ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 777			    OCFS2_JOURNAL_ACCESS_WRITE);
 778	if (ret) {
 779		mlog_errno(ret);
 780		goto out;
 781	}
 782
 783	ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 784				  &ctxt->dealloc);
 785	if (ret) {
 786		mlog_errno(ret);
 787		goto out;
 788	}
 789
 790	le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 791	ocfs2_journal_dirty(handle, vb->vb_bh);
 792
 793	if (ext_flags & OCFS2_EXT_REFCOUNTED)
 794		ret = ocfs2_decrease_refcount(inode, handle,
 795					ocfs2_blocks_to_clusters(inode->i_sb,
 796								 phys_blkno),
 797					len, ctxt->meta_ac, &ctxt->dealloc, 1);
 798	else
 799		ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 800						  phys_blkno, len);
 801	if (ret)
 802		mlog_errno(ret);
 803
 804out:
 805	return ret;
 806}
 807
 808static int ocfs2_xattr_shrink_size(struct inode *inode,
 809				   u32 old_clusters,
 810				   u32 new_clusters,
 811				   struct ocfs2_xattr_value_buf *vb,
 812				   struct ocfs2_xattr_set_ctxt *ctxt)
 813{
 814	int ret = 0;
 815	unsigned int ext_flags;
 816	u32 trunc_len, cpos, phys_cpos, alloc_size;
 817	u64 block;
 818
 819	if (old_clusters <= new_clusters)
 820		return 0;
 821
 822	cpos = new_clusters;
 823	trunc_len = old_clusters - new_clusters;
 824	while (trunc_len) {
 825		ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 826					       &alloc_size,
 827					       &vb->vb_xv->xr_list, &ext_flags);
 828		if (ret) {
 829			mlog_errno(ret);
 830			goto out;
 831		}
 832
 833		if (alloc_size > trunc_len)
 834			alloc_size = trunc_len;
 835
 836		ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 837						 phys_cpos, alloc_size,
 838						 ext_flags, ctxt);
 839		if (ret) {
 840			mlog_errno(ret);
 841			goto out;
 842		}
 843
 844		block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 845		ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 846						       block, alloc_size);
 847		cpos += alloc_size;
 848		trunc_len -= alloc_size;
 849	}
 850
 851out:
 852	return ret;
 853}
 854
 855static int ocfs2_xattr_value_truncate(struct inode *inode,
 856				      struct ocfs2_xattr_value_buf *vb,
 857				      int len,
 858				      struct ocfs2_xattr_set_ctxt *ctxt)
 859{
 860	int ret;
 861	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 862	u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 863
 864	if (new_clusters == old_clusters)
 865		return 0;
 866
 867	if (new_clusters > old_clusters)
 868		ret = ocfs2_xattr_extend_allocation(inode,
 869						    new_clusters - old_clusters,
 870						    vb, ctxt);
 871	else
 872		ret = ocfs2_xattr_shrink_size(inode,
 873					      old_clusters, new_clusters,
 874					      vb, ctxt);
 875
 876	return ret;
 877}
 878
 879static int ocfs2_xattr_list_entry(struct super_block *sb,
 880				  char *buffer, size_t size,
 881				  size_t *result, int type,
 882				  const char *name, int name_len)
 883{
 884	char *p = buffer + *result;
 885	const char *prefix;
 886	int prefix_len;
 887	int total_len;
 888
 889	switch(type) {
 890	case OCFS2_XATTR_INDEX_USER:
 891		if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 892			return 0;
 893		break;
 894
 895	case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 896	case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 897		if (!(sb->s_flags & SB_POSIXACL))
 898			return 0;
 899		break;
 900
 901	case OCFS2_XATTR_INDEX_TRUSTED:
 902		if (!capable(CAP_SYS_ADMIN))
 903			return 0;
 904		break;
 905	}
 906
 907	prefix = ocfs2_xattr_prefix(type);
 908	if (!prefix)
 909		return 0;
 910	prefix_len = strlen(prefix);
 911	total_len = prefix_len + name_len + 1;
 912	*result += total_len;
 913
 914	/* we are just looking for how big our buffer needs to be */
 915	if (!size)
 916		return 0;
 917
 918	if (*result > size)
 919		return -ERANGE;
 920
 921	memcpy(p, prefix, prefix_len);
 922	memcpy(p + prefix_len, name, name_len);
 923	p[prefix_len + name_len] = '\0';
 924
 925	return 0;
 926}
 927
 928static int ocfs2_xattr_list_entries(struct inode *inode,
 929				    struct ocfs2_xattr_header *header,
 930				    char *buffer, size_t buffer_size)
 931{
 932	size_t result = 0;
 933	int i, type, ret;
 934	const char *name;
 935
 936	for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 937		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 938		type = ocfs2_xattr_get_type(entry);
 939		name = (const char *)header +
 940			le16_to_cpu(entry->xe_name_offset);
 941
 942		ret = ocfs2_xattr_list_entry(inode->i_sb,
 943					     buffer, buffer_size,
 944					     &result, type, name,
 945					     entry->xe_name_len);
 946		if (ret)
 947			return ret;
 948	}
 949
 950	return result;
 951}
 952
 953int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 954					 struct ocfs2_dinode *di)
 955{
 956	struct ocfs2_xattr_header *xh;
 957	int i;
 958
 959	xh = (struct ocfs2_xattr_header *)
 960		 ((void *)di + inode->i_sb->s_blocksize -
 961		 le16_to_cpu(di->i_xattr_inline_size));
 962
 963	for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 964		if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 965			return 1;
 966
 967	return 0;
 968}
 969
 970static int ocfs2_xattr_ibody_list(struct inode *inode,
 971				  struct ocfs2_dinode *di,
 972				  char *buffer,
 973				  size_t buffer_size)
 974{
 975	struct ocfs2_xattr_header *header = NULL;
 976	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 977	int ret = 0;
 978	u16 xattr_count;
 979	size_t max_entries;
 980	u16 inline_size;
 981
 982	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 983		return ret;
 984
 985	inline_size = le16_to_cpu(di->i_xattr_inline_size);
 986
 987	/* Validate inline size is reasonable */
 988	if (inline_size > inode->i_sb->s_blocksize ||
 989	    inline_size < sizeof(struct ocfs2_xattr_header)) {
 990		ocfs2_error(inode->i_sb,
 991			    "Invalid xattr inline size %u in inode %llu\n",
 992			    inline_size,
 993			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 994		return -EFSCORRUPTED;
 995	}
 996
 997	header = (struct ocfs2_xattr_header *)
 998		 ((void *)di + inode->i_sb->s_blocksize - inline_size);
 999
1000	xattr_count = le16_to_cpu(header->xh_count);
1001	max_entries = (inline_size - sizeof(struct ocfs2_xattr_header)) /
1002		       sizeof(struct ocfs2_xattr_entry);
1003
1004	if (xattr_count > max_entries) {
1005		ocfs2_error(inode->i_sb,
1006			    "xattr entry count %u exceeds maximum %zu in inode %llu\n",
1007			    xattr_count, max_entries,
1008			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
1009		return -EFSCORRUPTED;
1010	}
1011
1012	ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
1013
1014	return ret;
1015}
1016
1017static int ocfs2_xattr_block_list(struct inode *inode,
1018				  struct ocfs2_dinode *di,
1019				  char *buffer,
1020				  size_t buffer_size)
1021{
1022	struct buffer_head *blk_bh = NULL;
1023	struct ocfs2_xattr_block *xb;
1024	int ret = 0;
1025
1026	if (!di->i_xattr_loc)
1027		return ret;
1028
1029	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1030				     &blk_bh);
1031	if (ret < 0) {
1032		mlog_errno(ret);
1033		return ret;
1034	}
1035
1036	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1037	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1038		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1039		ret = ocfs2_xattr_list_entries(inode, header,
1040					       buffer, buffer_size);
1041	} else
1042		ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1043						   buffer, buffer_size);
1044
1045	brelse(blk_bh);
1046
1047	return ret;
1048}
1049
1050ssize_t ocfs2_listxattr(struct dentry *dentry,
1051			char *buffer,
1052			size_t size)
1053{
1054	int ret = 0, i_ret = 0, b_ret = 0;
1055	struct buffer_head *di_bh = NULL;
1056	struct ocfs2_dinode *di = NULL;
1057	struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1058
1059	if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1060		return -EOPNOTSUPP;
1061
1062	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1063		return ret;
1064
1065	ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1066	if (ret < 0) {
1067		mlog_errno(ret);
1068		return ret;
1069	}
1070
1071	di = (struct ocfs2_dinode *)di_bh->b_data;
1072
1073	down_read(&oi->ip_xattr_sem);
1074	i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1075	if (i_ret < 0)
1076		b_ret = 0;
1077	else {
1078		if (buffer) {
1079			buffer += i_ret;
1080			size -= i_ret;
1081		}
1082		b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1083					       buffer, size);
1084		if (b_ret < 0)
1085			i_ret = 0;
1086	}
1087	up_read(&oi->ip_xattr_sem);
1088	ocfs2_inode_unlock(d_inode(dentry), 0);
1089
1090	brelse(di_bh);
1091
1092	return i_ret + b_ret;
1093}
1094
1095static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
1096				  const char *name,
1097				  struct ocfs2_xattr_search *xs)
1098{
1099	struct ocfs2_xattr_entry *entry;
1100	size_t name_len;
1101	int i, name_offset, cmp = 1;
1102
1103	if (name == NULL)
1104		return -EINVAL;
1105
1106	name_len = strlen(name);
1107	entry = xs->here;
1108	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1109		if ((void *)entry >= xs->end) {
1110			ocfs2_error(inode->i_sb, "corrupted xattr entries");
1111			return -EFSCORRUPTED;
1112		}
1113		cmp = name_index - ocfs2_xattr_get_type(entry);
1114		if (!cmp)
1115			cmp = name_len - entry->xe_name_len;
1116		if (!cmp) {
1117			name_offset = le16_to_cpu(entry->xe_name_offset);
1118			if ((xs->base + name_offset + name_len) > xs->end) {
1119				ocfs2_error(inode->i_sb,
1120					    "corrupted xattr entries");
1121				return -EFSCORRUPTED;
1122			}
1123			cmp = memcmp(name, (xs->base + name_offset), name_len);
1124		}
1125		if (cmp == 0)
1126			break;
1127		entry += 1;
1128	}
1129	xs->here = entry;
1130
1131	return cmp ? -ENODATA : 0;
1132}
1133
1134static int ocfs2_xattr_get_value_outside(struct inode *inode,
1135					 struct ocfs2_xattr_value_root *xv,
1136					 void *buffer,
1137					 size_t len)
1138{
1139	u32 cpos, p_cluster, num_clusters, bpc, clusters;
1140	u64 blkno;
1141	int i, ret = 0;
1142	size_t cplen, blocksize;
1143	struct buffer_head *bh = NULL;
1144	struct ocfs2_extent_list *el;
1145
1146	el = &xv->xr_list;
1147	clusters = le32_to_cpu(xv->xr_clusters);
1148	bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1149	blocksize = inode->i_sb->s_blocksize;
1150
1151	cpos = 0;
1152	while (cpos < clusters) {
1153		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1154					       &num_clusters, el, NULL);
1155		if (ret) {
1156			mlog_errno(ret);
1157			goto out;
1158		}
1159
1160		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1161		/* Copy ocfs2_xattr_value */
1162		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1163			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1164					       &bh, NULL);
1165			if (ret) {
1166				mlog_errno(ret);
1167				goto out;
1168			}
1169
1170			cplen = len >= blocksize ? blocksize : len;
1171			memcpy(buffer, bh->b_data, cplen);
1172			len -= cplen;
1173			buffer += cplen;
1174
1175			brelse(bh);
1176			bh = NULL;
1177			if (len == 0)
1178				break;
1179		}
1180		cpos += num_clusters;
1181	}
1182out:
1183	return ret;
1184}
1185
1186static int ocfs2_xattr_ibody_get(struct inode *inode,
1187				 int name_index,
1188				 const char *name,
1189				 void *buffer,
1190				 size_t buffer_size,
1191				 struct ocfs2_xattr_search *xs)
1192{
1193	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1194	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1195	struct ocfs2_xattr_value_root *xv;
1196	size_t size;
1197	int ret = 0;
1198
1199	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1200		return -ENODATA;
1201
1202	xs->end = (void *)di + inode->i_sb->s_blocksize;
1203	xs->header = (struct ocfs2_xattr_header *)
1204			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
1205	xs->base = (void *)xs->header;
1206	xs->here = xs->header->xh_entries;
1207
1208	ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
1209	if (ret)
1210		return ret;
1211	size = le64_to_cpu(xs->here->xe_value_size);
1212	if (buffer) {
1213		if (size > buffer_size)
1214			return -ERANGE;
1215		if (ocfs2_xattr_is_local(xs->here)) {
1216			memcpy(buffer, (void *)xs->base +
1217			       le16_to_cpu(xs->here->xe_name_offset) +
1218			       OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1219		} else {
1220			xv = (struct ocfs2_xattr_value_root *)
1221				(xs->base + le16_to_cpu(
1222				 xs->here->xe_name_offset) +
1223				OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1224			ret = ocfs2_xattr_get_value_outside(inode, xv,
1225							    buffer, size);
1226			if (ret < 0) {
1227				mlog_errno(ret);
1228				return ret;
1229			}
1230		}
1231	}
1232
1233	return size;
1234}
1235
1236static int ocfs2_xattr_block_get(struct inode *inode,
1237				 int name_index,
1238				 const char *name,
1239				 void *buffer,
1240				 size_t buffer_size,
1241				 struct ocfs2_xattr_search *xs)
1242{
1243	struct ocfs2_xattr_block *xb;
1244	struct ocfs2_xattr_value_root *xv;
1245	size_t size;
1246	int ret = -ENODATA, name_offset, name_len, i;
1247	int block_off;
1248
1249	xs->bucket = ocfs2_xattr_bucket_new(inode);
1250	if (!xs->bucket) {
1251		ret = -ENOMEM;
1252		mlog_errno(ret);
1253		goto cleanup;
1254	}
1255
1256	ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1257	if (ret) {
1258		mlog_errno(ret);
1259		goto cleanup;
1260	}
1261
1262	if (xs->not_found) {
1263		ret = -ENODATA;
1264		goto cleanup;
1265	}
1266
1267	xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1268	size = le64_to_cpu(xs->here->xe_value_size);
1269	if (buffer) {
1270		ret = -ERANGE;
1271		if (size > buffer_size)
1272			goto cleanup;
1273
1274		name_offset = le16_to_cpu(xs->here->xe_name_offset);
1275		name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1276		i = xs->here - xs->header->xh_entries;
1277
1278		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1279			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1280								bucket_xh(xs->bucket),
1281								i,
1282								&block_off,
1283								&name_offset);
1284			if (ret) {
1285				mlog_errno(ret);
1286				goto cleanup;
1287			}
1288			xs->base = bucket_block(xs->bucket, block_off);
1289		}
1290		if (ocfs2_xattr_is_local(xs->here)) {
1291			memcpy(buffer, (void *)xs->base +
1292			       name_offset + name_len, size);
1293		} else {
1294			xv = (struct ocfs2_xattr_value_root *)
1295				(xs->base + name_offset + name_len);
1296			ret = ocfs2_xattr_get_value_outside(inode, xv,
1297							    buffer, size);
1298			if (ret < 0) {
1299				mlog_errno(ret);
1300				goto cleanup;
1301			}
1302		}
1303	}
1304	ret = size;
1305cleanup:
1306	ocfs2_xattr_bucket_free(xs->bucket);
1307
1308	brelse(xs->xattr_bh);
1309	xs->xattr_bh = NULL;
1310	return ret;
1311}
1312
1313int ocfs2_xattr_get_nolock(struct inode *inode,
1314			   struct buffer_head *di_bh,
1315			   int name_index,
1316			   const char *name,
1317			   void *buffer,
1318			   size_t buffer_size)
1319{
1320	int ret;
1321	struct ocfs2_dinode *di = NULL;
1322	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1323	struct ocfs2_xattr_search xis = {
1324		.not_found = -ENODATA,
1325	};
1326	struct ocfs2_xattr_search xbs = {
1327		.not_found = -ENODATA,
1328	};
1329
1330	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1331		return -EOPNOTSUPP;
1332
1333	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1334		return -ENODATA;
1335
1336	xis.inode_bh = xbs.inode_bh = di_bh;
1337	di = (struct ocfs2_dinode *)di_bh->b_data;
1338
1339	ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1340				    buffer_size, &xis);
1341	if (ret == -ENODATA && di->i_xattr_loc)
1342		ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1343					    buffer_size, &xbs);
1344
1345	return ret;
1346}
1347
1348/* ocfs2_xattr_get()
1349 *
1350 * Copy an extended attribute into the buffer provided.
1351 * Buffer is NULL to compute the size of buffer required.
1352 */
1353static int ocfs2_xattr_get(struct inode *inode,
1354			   int name_index,
1355			   const char *name,
1356			   void *buffer,
1357			   size_t buffer_size)
1358{
1359	int ret, had_lock;
1360	struct buffer_head *di_bh = NULL;
1361	struct ocfs2_lock_holder oh;
1362
1363	had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1364	if (had_lock < 0) {
1365		mlog_errno(had_lock);
1366		return had_lock;
1367	}
1368	down_read(&OCFS2_I(inode)->ip_xattr_sem);
1369	ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1370				     name, buffer, buffer_size);
1371	up_read(&OCFS2_I(inode)->ip_xattr_sem);
1372
1373	ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1374
1375	brelse(di_bh);
1376
1377	return ret;
1378}
1379
1380static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1381					   handle_t *handle,
1382					   struct ocfs2_xattr_value_buf *vb,
1383					   const void *value,
1384					   int value_len)
1385{
1386	int ret = 0, i, cp_len;
1387	u16 blocksize = inode->i_sb->s_blocksize;
1388	u32 p_cluster, num_clusters;
1389	u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1390	u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1391	u64 blkno;
1392	struct buffer_head *bh = NULL;
1393	unsigned int ext_flags;
1394	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1395
1396	BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1397
1398	while (cpos < clusters) {
1399		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1400					       &num_clusters, &xv->xr_list,
1401					       &ext_flags);
1402		if (ret) {
1403			mlog_errno(ret);
1404			goto out;
1405		}
1406
1407		BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1408
1409		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1410
1411		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1412			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1413					       &bh, NULL);
1414			if (ret) {
1415				mlog_errno(ret);
1416				goto out;
1417			}
1418
1419			ret = ocfs2_journal_access(handle,
1420						   INODE_CACHE(inode),
1421						   bh,
1422						   OCFS2_JOURNAL_ACCESS_WRITE);
1423			if (ret < 0) {
1424				mlog_errno(ret);
1425				goto out;
1426			}
1427
1428			cp_len = value_len > blocksize ? blocksize : value_len;
1429			memcpy(bh->b_data, value, cp_len);
1430			value_len -= cp_len;
1431			value += cp_len;
1432			if (cp_len < blocksize)
1433				memset(bh->b_data + cp_len, 0,
1434				       blocksize - cp_len);
1435
1436			ocfs2_journal_dirty(handle, bh);
1437			brelse(bh);
1438			bh = NULL;
1439
1440			/*
1441			 * XXX: do we need to empty all the following
1442			 * blocks in this cluster?
1443			 */
1444			if (!value_len)
1445				break;
1446		}
1447		cpos += num_clusters;
1448	}
1449out:
1450	brelse(bh);
1451
1452	return ret;
1453}
1454
1455static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1456				       int num_entries)
1457{
1458	int free_space;
1459
1460	if (!needed_space)
1461		return 0;
1462
1463	free_space = free_start -
1464		sizeof(struct ocfs2_xattr_header) -
1465		(num_entries * sizeof(struct ocfs2_xattr_entry)) -
1466		OCFS2_XATTR_HEADER_GAP;
1467	if (free_space < 0)
1468		return -EIO;
1469	if (free_space < needed_space)
1470		return -ENOSPC;
1471
1472	return 0;
1473}
1474
1475static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1476				   int type)
1477{
1478	return loc->xl_ops->xlo_journal_access(handle, loc, type);
1479}
1480
1481static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1482{
1483	loc->xl_ops->xlo_journal_dirty(handle, loc);
1484}
1485
1486/* Give a pointer into the storage for the given offset */
1487static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1488{
1489	BUG_ON(offset >= loc->xl_size);
1490	return loc->xl_ops->xlo_offset_pointer(loc, offset);
1491}
1492
1493/*
1494 * Wipe the name+value pair and allow the storage to reclaim it.  This
1495 * must be followed by either removal of the entry or a call to
1496 * ocfs2_xa_add_namevalue().
1497 */
1498static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1499{
1500	loc->xl_ops->xlo_wipe_namevalue(loc);
1501}
1502
1503/*
1504 * Find lowest offset to a name+value pair.  This is the start of our
1505 * downward-growing free space.
1506 */
1507static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1508{
1509	return loc->xl_ops->xlo_get_free_start(loc);
1510}
1511
1512/* Can we reuse loc->xl_entry for xi? */
1513static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1514				    struct ocfs2_xattr_info *xi)
1515{
1516	return loc->xl_ops->xlo_can_reuse(loc, xi);
1517}
1518
1519/* How much free space is needed to set the new value */
1520static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1521				struct ocfs2_xattr_info *xi)
1522{
1523	return loc->xl_ops->xlo_check_space(loc, xi);
1524}
1525
1526static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1527{
1528	loc->xl_ops->xlo_add_entry(loc, name_hash);
1529	loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1530	/*
1531	 * We can't leave the new entry's xe_name_offset at zero or
1532	 * add_namevalue() will go nuts.  We set it to the size of our
1533	 * storage so that it can never be less than any other entry.
1534	 */
1535	loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1536}
1537
1538static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1539				   struct ocfs2_xattr_info *xi)
1540{
1541	int size = namevalue_size_xi(xi);
1542	int nameval_offset;
1543	char *nameval_buf;
1544
1545	loc->xl_ops->xlo_add_namevalue(loc, size);
1546	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1547	loc->xl_entry->xe_name_len = xi->xi_name_len;
1548	ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1549	ocfs2_xattr_set_local(loc->xl_entry,
1550			      xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1551
1552	nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1553	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1554	memset(nameval_buf, 0, size);
1555	memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1556}
1557
1558static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1559				    struct ocfs2_xattr_value_buf *vb)
1560{
1561	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1562	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1563
1564	/* Value bufs are for value trees */
1565	BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1566	BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1567	       (name_size + OCFS2_XATTR_ROOT_SIZE));
1568
1569	loc->xl_ops->xlo_fill_value_buf(loc, vb);
1570	vb->vb_xv =
1571		(struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1572							nameval_offset +
1573							name_size);
1574}
1575
1576static int ocfs2_xa_block_journal_access(handle_t *handle,
1577					 struct ocfs2_xa_loc *loc, int type)
1578{
1579	struct buffer_head *bh = loc->xl_storage;
1580	ocfs2_journal_access_func access;
1581
1582	if (loc->xl_size == (bh->b_size -
1583			     offsetof(struct ocfs2_xattr_block,
1584				      xb_attrs.xb_header)))
1585		access = ocfs2_journal_access_xb;
1586	else
1587		access = ocfs2_journal_access_di;
1588	return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1589}
1590
1591static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1592					 struct ocfs2_xa_loc *loc)
1593{
1594	struct buffer_head *bh = loc->xl_storage;
1595
1596	ocfs2_journal_dirty(handle, bh);
1597}
1598
1599static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1600					   int offset)
1601{
1602	return (char *)loc->xl_header + offset;
1603}
1604
1605static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1606				    struct ocfs2_xattr_info *xi)
1607{
1608	/*
1609	 * Block storage is strict.  If the sizes aren't exact, we will
1610	 * remove the old one and reinsert the new.
1611	 */
1612	return namevalue_size_xe(loc->xl_entry) ==
1613		namevalue_size_xi(xi);
1614}
1615
1616static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1617{
1618	struct ocfs2_xattr_header *xh = loc->xl_header;
1619	int i, count = le16_to_cpu(xh->xh_count);
1620	int offset, free_start = loc->xl_size;
1621
1622	for (i = 0; i < count; i++) {
1623		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1624		if (offset < free_start)
1625			free_start = offset;
1626	}
1627
1628	return free_start;
1629}
1630
1631static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1632				      struct ocfs2_xattr_info *xi)
1633{
1634	int count = le16_to_cpu(loc->xl_header->xh_count);
1635	int free_start = ocfs2_xa_get_free_start(loc);
1636	int needed_space = ocfs2_xi_entry_usage(xi);
1637
1638	/*
1639	 * Block storage will reclaim the original entry before inserting
1640	 * the new value, so we only need the difference.  If the new
1641	 * entry is smaller than the old one, we don't need anything.
1642	 */
1643	if (loc->xl_entry) {
1644		/* Don't need space if we're reusing! */
1645		if (ocfs2_xa_can_reuse_entry(loc, xi))
1646			needed_space = 0;
1647		else
1648			needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1649	}
1650	if (needed_space < 0)
1651		needed_space = 0;
1652	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1653}
1654
1655/*
1656 * Block storage for xattrs keeps the name+value pairs compacted.  When
1657 * we remove one, we have to shift any that preceded it towards the end.
1658 */
1659static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1660{
1661	int i, offset;
1662	int namevalue_offset, first_namevalue_offset, namevalue_size;
1663	struct ocfs2_xattr_entry *entry = loc->xl_entry;
1664	struct ocfs2_xattr_header *xh = loc->xl_header;
1665	int count = le16_to_cpu(xh->xh_count);
1666
1667	namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1668	namevalue_size = namevalue_size_xe(entry);
1669	first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1670
1671	/* Shift the name+value pairs */
1672	memmove((char *)xh + first_namevalue_offset + namevalue_size,
1673		(char *)xh + first_namevalue_offset,
1674		namevalue_offset - first_namevalue_offset);
1675	memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1676
1677	/* Now tell xh->xh_entries about it */
1678	for (i = 0; i < count; i++) {
1679		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1680		if (offset <= namevalue_offset)
1681			le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1682				     namevalue_size);
1683	}
1684
1685	/*
1686	 * Note that we don't update xh_free_start or xh_name_value_len
1687	 * because they're not used in block-stored xattrs.
1688	 */
1689}
1690
1691static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1692{
1693	int count = le16_to_cpu(loc->xl_header->xh_count);
1694	loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1695	le16_add_cpu(&loc->xl_header->xh_count, 1);
1696	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1697}
1698
1699static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1700{
1701	int free_start = ocfs2_xa_get_free_start(loc);
1702
1703	loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1704}
1705
1706static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1707					  struct ocfs2_xattr_value_buf *vb)
1708{
1709	struct buffer_head *bh = loc->xl_storage;
1710
1711	if (loc->xl_size == (bh->b_size -
1712			     offsetof(struct ocfs2_xattr_block,
1713				      xb_attrs.xb_header)))
1714		vb->vb_access = ocfs2_journal_access_xb;
1715	else
1716		vb->vb_access = ocfs2_journal_access_di;
1717	vb->vb_bh = bh;
1718}
1719
1720/*
1721 * Operations for xattrs stored in blocks.  This includes inline inode
1722 * storage and unindexed ocfs2_xattr_blocks.
1723 */
1724static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1725	.xlo_journal_access	= ocfs2_xa_block_journal_access,
1726	.xlo_journal_dirty	= ocfs2_xa_block_journal_dirty,
1727	.xlo_offset_pointer	= ocfs2_xa_block_offset_pointer,
1728	.xlo_check_space	= ocfs2_xa_block_check_space,
1729	.xlo_can_reuse		= ocfs2_xa_block_can_reuse,
1730	.xlo_get_free_start	= ocfs2_xa_block_get_free_start,
1731	.xlo_wipe_namevalue	= ocfs2_xa_block_wipe_namevalue,
1732	.xlo_add_entry		= ocfs2_xa_block_add_entry,
1733	.xlo_add_namevalue	= ocfs2_xa_block_add_namevalue,
1734	.xlo_fill_value_buf	= ocfs2_xa_block_fill_value_buf,
1735};
1736
1737static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1738					  struct ocfs2_xa_loc *loc, int type)
1739{
1740	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1741
1742	return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1743}
1744
1745static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1746					  struct ocfs2_xa_loc *loc)
1747{
1748	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1749
1750	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1751}
1752
1753static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1754					    int offset)
1755{
1756	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1757	int block, block_offset;
1758
1759	/* The header is at the front of the bucket */
1760	block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1761	block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1762
1763	return bucket_block(bucket, block) + block_offset;
1764}
1765
1766static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1767				     struct ocfs2_xattr_info *xi)
1768{
1769	return namevalue_size_xe(loc->xl_entry) >=
1770		namevalue_size_xi(xi);
1771}
1772
1773static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1774{
1775	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1776	return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1777}
1778
1779static int ocfs2_bucket_align_free_start(struct super_block *sb,
1780					 int free_start, int size)
1781{
1782	/*
1783	 * We need to make sure that the name+value pair fits within
1784	 * one block.
1785	 */
1786	if (((free_start - size) >> sb->s_blocksize_bits) !=
1787	    ((free_start - 1) >> sb->s_blocksize_bits))
1788		free_start -= free_start % sb->s_blocksize;
1789
1790	return free_start;
1791}
1792
1793static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1794				       struct ocfs2_xattr_info *xi)
1795{
1796	int rc;
1797	int count = le16_to_cpu(loc->xl_header->xh_count);
1798	int free_start = ocfs2_xa_get_free_start(loc);
1799	int needed_space = ocfs2_xi_entry_usage(xi);
1800	int size = namevalue_size_xi(xi);
1801	struct super_block *sb = loc->xl_inode->i_sb;
1802
1803	/*
1804	 * Bucket storage does not reclaim name+value pairs it cannot
1805	 * reuse.  They live as holes until the bucket fills, and then
1806	 * the bucket is defragmented.  However, the bucket can reclaim
1807	 * the ocfs2_xattr_entry.
1808	 */
1809	if (loc->xl_entry) {
1810		/* Don't need space if we're reusing! */
1811		if (ocfs2_xa_can_reuse_entry(loc, xi))
1812			needed_space = 0;
1813		else
1814			needed_space -= sizeof(struct ocfs2_xattr_entry);
1815	}
1816	BUG_ON(needed_space < 0);
1817
1818	if (free_start < size) {
1819		if (needed_space)
1820			return -ENOSPC;
1821	} else {
1822		/*
1823		 * First we check if it would fit in the first place.
1824		 * Below, we align the free start to a block.  This may
1825		 * slide us below the minimum gap.  By checking unaligned
1826		 * first, we avoid that error.
1827		 */
1828		rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1829						 count);
1830		if (rc)
1831			return rc;
1832		free_start = ocfs2_bucket_align_free_start(sb, free_start,
1833							   size);
1834	}
1835	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1836}
1837
1838static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1839{
1840	le16_add_cpu(&loc->xl_header->xh_name_value_len,
1841		     -namevalue_size_xe(loc->xl_entry));
1842}
1843
1844static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1845{
1846	struct ocfs2_xattr_header *xh = loc->xl_header;
1847	int count = le16_to_cpu(xh->xh_count);
1848	int low = 0, high = count - 1, tmp;
1849	struct ocfs2_xattr_entry *tmp_xe;
1850
1851	/*
1852	 * We keep buckets sorted by name_hash, so we need to find
1853	 * our insert place.
1854	 */
1855	while (low <= high && count) {
1856		tmp = (low + high) / 2;
1857		tmp_xe = &xh->xh_entries[tmp];
1858
1859		if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1860			low = tmp + 1;
1861		else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1862			high = tmp - 1;
1863		else {
1864			low = tmp;
1865			break;
1866		}
1867	}
1868
1869	if (low != count)
1870		memmove(&xh->xh_entries[low + 1],
1871			&xh->xh_entries[low],
1872			((count - low) * sizeof(struct ocfs2_xattr_entry)));
1873
1874	le16_add_cpu(&xh->xh_count, 1);
1875	loc->xl_entry = &xh->xh_entries[low];
1876	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1877}
1878
1879static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1880{
1881	int free_start = ocfs2_xa_get_free_start(loc);
1882	struct ocfs2_xattr_header *xh = loc->xl_header;
1883	struct super_block *sb = loc->xl_inode->i_sb;
1884	int nameval_offset;
1885
1886	free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1887	nameval_offset = free_start - size;
1888	loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1889	xh->xh_free_start = cpu_to_le16(nameval_offset);
1890	le16_add_cpu(&xh->xh_name_value_len, size);
1891
1892}
1893
1894static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1895					   struct ocfs2_xattr_value_buf *vb)
1896{
1897	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1898	struct super_block *sb = loc->xl_inode->i_sb;
1899	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1900	int size = namevalue_size_xe(loc->xl_entry);
1901	int block_offset = nameval_offset >> sb->s_blocksize_bits;
1902
1903	/* Values are not allowed to straddle block boundaries */
1904	BUG_ON(block_offset !=
1905	       ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1906	/* We expect the bucket to be filled in */
1907	BUG_ON(!bucket->bu_bhs[block_offset]);
1908
1909	vb->vb_access = ocfs2_journal_access;
1910	vb->vb_bh = bucket->bu_bhs[block_offset];
1911}
1912
1913/* Operations for xattrs stored in buckets. */
1914static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1915	.xlo_journal_access	= ocfs2_xa_bucket_journal_access,
1916	.xlo_journal_dirty	= ocfs2_xa_bucket_journal_dirty,
1917	.xlo_offset_pointer	= ocfs2_xa_bucket_offset_pointer,
1918	.xlo_check_space	= ocfs2_xa_bucket_check_space,
1919	.xlo_can_reuse		= ocfs2_xa_bucket_can_reuse,
1920	.xlo_get_free_start	= ocfs2_xa_bucket_get_free_start,
1921	.xlo_wipe_namevalue	= ocfs2_xa_bucket_wipe_namevalue,
1922	.xlo_add_entry		= ocfs2_xa_bucket_add_entry,
1923	.xlo_add_namevalue	= ocfs2_xa_bucket_add_namevalue,
1924	.xlo_fill_value_buf	= ocfs2_xa_bucket_fill_value_buf,
1925};
1926
1927static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1928{
1929	struct ocfs2_xattr_value_buf vb;
1930
1931	if (ocfs2_xattr_is_local(loc->xl_entry))
1932		return 0;
1933
1934	ocfs2_xa_fill_value_buf(loc, &vb);
1935	return le32_to_cpu(vb.vb_xv->xr_clusters);
1936}
1937
1938static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1939				   struct ocfs2_xattr_set_ctxt *ctxt)
1940{
1941	int trunc_rc, access_rc;
1942	struct ocfs2_xattr_value_buf vb;
1943
1944	ocfs2_xa_fill_value_buf(loc, &vb);
1945	trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1946					      ctxt);
1947
1948	/*
1949	 * The caller of ocfs2_xa_value_truncate() has already called
1950	 * ocfs2_xa_journal_access on the loc.  However, The truncate code
1951	 * calls ocfs2_extend_trans().  This may commit the previous
1952	 * transaction and open a new one.  If this is a bucket, truncate
1953	 * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1954	 * the caller is expecting to dirty the entire bucket.  So we must
1955	 * reset the journal work.  We do this even if truncate has failed,
1956	 * as it could have failed after committing the extend.
1957	 */
1958	access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1959					    OCFS2_JOURNAL_ACCESS_WRITE);
1960
1961	/* Errors in truncate take precedence */
1962	return trunc_rc ? trunc_rc : access_rc;
1963}
1964
1965static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1966{
1967	int index, count;
1968	struct ocfs2_xattr_header *xh = loc->xl_header;
1969	struct ocfs2_xattr_entry *entry = loc->xl_entry;
1970
1971	ocfs2_xa_wipe_namevalue(loc);
1972	loc->xl_entry = NULL;
1973
1974	le16_add_cpu(&xh->xh_count, -1);
1975	count = le16_to_cpu(xh->xh_count);
1976
1977	/*
1978	 * Only zero out the entry if there are more remaining.  This is
1979	 * important for an empty bucket, as it keeps track of the
1980	 * bucket's hash value.  It doesn't hurt empty block storage.
1981	 */
1982	if (count) {
1983		index = ((char *)entry - (char *)&xh->xh_entries) /
1984			sizeof(struct ocfs2_xattr_entry);
1985		memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1986			(count - index) * sizeof(struct ocfs2_xattr_entry));
1987		memset(&xh->xh_entries[count], 0,
1988		       sizeof(struct ocfs2_xattr_entry));
1989	}
1990}
1991
1992/*
1993 * If we have a problem adjusting the size of an external value during
1994 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1995 * in an intermediate state.  For example, the value may be partially
1996 * truncated.
1997 *
1998 * If the value tree hasn't changed, the extend/truncate went nowhere.
1999 * We have nothing to do.  The caller can treat it as a straight error.
2000 *
2001 * If the value tree got partially truncated, we now have a corrupted
2002 * extended attribute.  We're going to wipe its entry and leak the
2003 * clusters.  Better to leak some storage than leave a corrupt entry.
2004 *
2005 * If the value tree grew, it obviously didn't grow enough for the
2006 * new entry.  We're not going to try and reclaim those clusters either.
2007 * If there was already an external value there (orig_clusters != 0),
2008 * the new clusters are attached safely and we can just leave the old
2009 * value in place.  If there was no external value there, we remove
2010 * the entry.
2011 *
2012 * This way, the xattr block we store in the journal will be consistent.
2013 * If the size change broke because of the journal, no changes will hit
2014 * disk anyway.
2015 */
2016static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
2017					    const char *what,
2018					    unsigned int orig_clusters)
2019{
2020	unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
2021	char *nameval_buf = ocfs2_xa_offset_pointer(loc,
2022				le16_to_cpu(loc->xl_entry->xe_name_offset));
2023
2024	if (new_clusters < orig_clusters) {
2025		mlog(ML_ERROR,
2026		     "Partial truncate while %s xattr %.*s.  Leaking "
2027		     "%u clusters and removing the entry\n",
2028		     what, loc->xl_entry->xe_name_len, nameval_buf,
2029		     orig_clusters - new_clusters);
2030		ocfs2_xa_remove_entry(loc);
2031	} else if (!orig_clusters) {
2032		mlog(ML_ERROR,
2033		     "Unable to allocate an external value for xattr "
2034		     "%.*s safely.  Leaking %u clusters and removing the "
2035		     "entry\n",
2036		     loc->xl_entry->xe_name_len, nameval_buf,
2037		     new_clusters - orig_clusters);
2038		ocfs2_xa_remove_entry(loc);
2039	} else if (new_clusters > orig_clusters)
2040		mlog(ML_ERROR,
2041		     "Unable to grow xattr %.*s safely.  %u new clusters "
2042		     "have been added, but the value will not be "
2043		     "modified\n",
2044		     loc->xl_entry->xe_name_len, nameval_buf,
2045		     new_clusters - orig_clusters);
2046}
2047
2048static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2049			   struct ocfs2_xattr_set_ctxt *ctxt)
2050{
2051	int rc = 0;
2052	unsigned int orig_clusters;
2053
2054	if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2055		orig_clusters = ocfs2_xa_value_clusters(loc);
2056		rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2057		if (rc) {
2058			mlog_errno(rc);
2059			/*
2060			 * Since this is remove, we can return 0 if
2061			 * ocfs2_xa_cleanup_value_truncate() is going to
2062			 * wipe the entry anyway.  So we check the
2063			 * cluster count as well.
2064			 */
2065			if (orig_clusters != ocfs2_xa_value_clusters(loc))
2066				rc = 0;
2067			ocfs2_xa_cleanup_value_truncate(loc, "removing",
2068							orig_clusters);
2069			goto out;
2070		}
2071	}
2072
2073	ocfs2_xa_remove_entry(loc);
2074
2075out:
2076	return rc;
2077}
2078
2079static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2080{
2081	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2082	char *nameval_buf;
2083
2084	nameval_buf = ocfs2_xa_offset_pointer(loc,
2085				le16_to_cpu(loc->xl_entry->xe_name_offset));
2086	memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2087}
2088
2089/*
2090 * Take an existing entry and make it ready for the new value.  This
2091 * won't allocate space, but it may free space.  It should be ready for
2092 * ocfs2_xa_prepare_entry() to finish the work.
2093 */
2094static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2095				struct ocfs2_xattr_info *xi,
2096				struct ocfs2_xattr_set_ctxt *ctxt)
2097{
2098	int rc = 0;
2099	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2100	unsigned int orig_clusters;
2101	char *nameval_buf;
2102	int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2103	int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2104
2105	BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2106	       name_size);
2107
2108	nameval_buf = ocfs2_xa_offset_pointer(loc,
2109				le16_to_cpu(loc->xl_entry->xe_name_offset));
2110	if (xe_local) {
2111		memset(nameval_buf + name_size, 0,
2112		       namevalue_size_xe(loc->xl_entry) - name_size);
2113		if (!xi_local)
2114			ocfs2_xa_install_value_root(loc);
2115	} else {
2116		orig_clusters = ocfs2_xa_value_clusters(loc);
2117		if (xi_local) {
2118			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2119			if (rc < 0)
2120				mlog_errno(rc);
2121			else
2122				memset(nameval_buf + name_size, 0,
2123				       namevalue_size_xe(loc->xl_entry) -
2124				       name_size);
2125		} else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2126			   xi->xi_value_len) {
2127			rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2128						     ctxt);
2129			if (rc < 0)
2130				mlog_errno(rc);
2131		}
2132
2133		if (rc) {
2134			ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2135							orig_clusters);
2136			goto out;
2137		}
2138	}
2139
2140	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2141	ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2142
2143out:
2144	return rc;
2145}
2146
2147/*
2148 * Prepares loc->xl_entry to receive the new xattr.  This includes
2149 * properly setting up the name+value pair region.  If loc->xl_entry
2150 * already exists, it will take care of modifying it appropriately.
2151 *
2152 * Note that this modifies the data.  You did journal_access already,
2153 * right?
2154 */
2155static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2156				  struct ocfs2_xattr_info *xi,
2157				  u32 name_hash,
2158				  struct ocfs2_xattr_set_ctxt *ctxt)
2159{
2160	int rc = 0;
2161	unsigned int orig_clusters;
2162	__le64 orig_value_size = 0;
2163
2164	rc = ocfs2_xa_check_space(loc, xi);
2165	if (rc)
2166		goto out;
2167
2168	if (loc->xl_entry) {
2169		if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2170			orig_value_size = loc->xl_entry->xe_value_size;
2171			rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2172			if (rc)
2173				goto out;
2174			goto alloc_value;
2175		}
2176
2177		if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2178			orig_clusters = ocfs2_xa_value_clusters(loc);
2179			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2180			if (rc) {
2181				mlog_errno(rc);
2182				ocfs2_xa_cleanup_value_truncate(loc,
2183								"overwriting",
2184								orig_clusters);
2185				goto out;
2186			}
2187		}
2188		ocfs2_xa_wipe_namevalue(loc);
2189	} else
2190		ocfs2_xa_add_entry(loc, name_hash);
2191
2192	/*
2193	 * If we get here, we have a blank entry.  Fill it.  We grow our
2194	 * name+value pair back from the end.
2195	 */
2196	ocfs2_xa_add_namevalue(loc, xi);
2197	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2198		ocfs2_xa_install_value_root(loc);
2199
2200alloc_value:
2201	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2202		orig_clusters = ocfs2_xa_value_clusters(loc);
2203		rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2204		if (rc < 0) {
2205			ctxt->set_abort = 1;
2206			ocfs2_xa_cleanup_value_truncate(loc, "growing",
2207							orig_clusters);
2208			/*
2209			 * If we were growing an existing value,
2210			 * ocfs2_xa_cleanup_value_truncate() won't remove
2211			 * the entry. We need to restore the original value
2212			 * size.
2213			 */
2214			if (loc->xl_entry) {
2215				BUG_ON(!orig_value_size);
2216				loc->xl_entry->xe_value_size = orig_value_size;
2217			}
2218			mlog_errno(rc);
2219		}
2220	}
2221
2222out:
2223	return rc;
2224}
2225
2226/*
2227 * Store the value portion of the name+value pair.  This will skip
2228 * values that are stored externally.  Their tree roots were set up
2229 * by ocfs2_xa_prepare_entry().
2230 */
2231static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2232				struct ocfs2_xattr_info *xi,
2233				struct ocfs2_xattr_set_ctxt *ctxt)
2234{
2235	int rc = 0;
2236	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2237	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2238	char *nameval_buf;
2239	struct ocfs2_xattr_value_buf vb;
2240
2241	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2242	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2243		ocfs2_xa_fill_value_buf(loc, &vb);
2244		rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2245						     ctxt->handle, &vb,
2246						     xi->xi_value,
2247						     xi->xi_value_len);
2248	} else
2249		memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2250
2251	return rc;
2252}
2253
2254static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2255			struct ocfs2_xattr_info *xi,
2256			struct ocfs2_xattr_set_ctxt *ctxt)
2257{
2258	int ret;
2259	u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2260					      xi->xi_name_len);
2261
2262	ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2263				      OCFS2_JOURNAL_ACCESS_WRITE);
2264	if (ret) {
2265		mlog_errno(ret);
2266		goto out;
2267	}
2268
2269	/*
2270	 * From here on out, everything is going to modify the buffer a
2271	 * little.  Errors are going to leave the xattr header in a
2272	 * sane state.  Thus, even with errors we dirty the sucker.
2273	 */
2274
2275	/* Don't worry, we are never called with !xi_value and !xl_entry */
2276	if (!xi->xi_value) {
2277		ret = ocfs2_xa_remove(loc, ctxt);
2278		goto out_dirty;
2279	}
2280
2281	ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2282	if (ret) {
2283		if (ret != -ENOSPC)
2284			mlog_errno(ret);
2285		goto out_dirty;
2286	}
2287
2288	ret = ocfs2_xa_store_value(loc, xi, ctxt);
2289	if (ret)
2290		mlog_errno(ret);
2291
2292out_dirty:
2293	ocfs2_xa_journal_dirty(ctxt->handle, loc);
2294
2295out:
2296	return ret;
2297}
2298
2299static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2300				     struct inode *inode,
2301				     struct buffer_head *bh,
2302				     struct ocfs2_xattr_entry *entry)
2303{
2304	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2305
2306	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2307
2308	loc->xl_inode = inode;
2309	loc->xl_ops = &ocfs2_xa_block_loc_ops;
2310	loc->xl_storage = bh;
2311	loc->xl_entry = entry;
2312	loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2313	loc->xl_header =
2314		(struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2315					      loc->xl_size);
2316}
2317
2318static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2319					  struct inode *inode,
2320					  struct buffer_head *bh,
2321					  struct ocfs2_xattr_entry *entry)
2322{
2323	struct ocfs2_xattr_block *xb =
2324		(struct ocfs2_xattr_block *)bh->b_data;
2325
2326	BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2327
2328	loc->xl_inode = inode;
2329	loc->xl_ops = &ocfs2_xa_block_loc_ops;
2330	loc->xl_storage = bh;
2331	loc->xl_header = &(xb->xb_attrs.xb_header);
2332	loc->xl_entry = entry;
2333	loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2334					     xb_attrs.xb_header);
2335}
2336
2337static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2338					   struct ocfs2_xattr_bucket *bucket,
2339					   struct ocfs2_xattr_entry *entry)
2340{
2341	loc->xl_inode = bucket->bu_inode;
2342	loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2343	loc->xl_storage = bucket;
2344	loc->xl_header = bucket_xh(bucket);
2345	loc->xl_entry = entry;
2346	loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2347}
2348
2349/*
2350 * In xattr remove, if it is stored outside and refcounted, we may have
2351 * the chance to split the refcount tree. So need the allocators.
2352 */
2353static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2354					struct ocfs2_xattr_value_root *xv,
2355					struct ocfs2_caching_info *ref_ci,
2356					struct buffer_head *ref_root_bh,
2357					struct ocfs2_alloc_context **meta_ac,
2358					int *ref_credits)
2359{
2360	int ret, meta_add = 0;
2361	u32 p_cluster, num_clusters;
2362	unsigned int ext_flags;
2363
2364	*ref_credits = 0;
2365	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2366				       &num_clusters,
2367				       &xv->xr_list,
2368				       &ext_flags);
2369	if (ret) {
2370		mlog_errno(ret);
2371		goto out;
2372	}
2373
2374	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2375		goto out;
2376
2377	ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2378						 ref_root_bh, xv,
2379						 &meta_add, ref_credits);
2380	if (ret) {
2381		mlog_errno(ret);
2382		goto out;
2383	}
2384
2385	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2386						meta_add, meta_ac);
2387	if (ret)
2388		mlog_errno(ret);
2389
2390out:
2391	return ret;
2392}
2393
2394static int ocfs2_remove_value_outside(struct inode*inode,
2395				      struct ocfs2_xattr_value_buf *vb,
2396				      struct ocfs2_xattr_header *header,
2397				      struct ocfs2_caching_info *ref_ci,
2398				      struct buffer_head *ref_root_bh)
2399{
2400	int ret = 0, i, ref_credits;
2401	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2402	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2403	void *val;
2404
2405	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2406
2407	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2408		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2409
2410		if (ocfs2_xattr_is_local(entry))
2411			continue;
2412
2413		val = (void *)header +
2414			le16_to_cpu(entry->xe_name_offset);
2415		vb->vb_xv = (struct ocfs2_xattr_value_root *)
2416			(val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2417
2418		ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2419							 ref_ci, ref_root_bh,
2420							 &ctxt.meta_ac,
2421							 &ref_credits);
2422
2423		ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2424					ocfs2_remove_extent_credits(osb->sb));
2425		if (IS_ERR(ctxt.handle)) {
2426			ret = PTR_ERR(ctxt.handle);
2427			mlog_errno(ret);
2428			break;
2429		}
2430
2431		ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2432
2433		ocfs2_commit_trans(osb, ctxt.handle);
2434		if (ctxt.meta_ac) {
2435			ocfs2_free_alloc_context(ctxt.meta_ac);
2436			ctxt.meta_ac = NULL;
2437		}
2438
2439		if (ret < 0) {
2440			mlog_errno(ret);
2441			break;
2442		}
2443
2444	}
2445
2446	if (ctxt.meta_ac)
2447		ocfs2_free_alloc_context(ctxt.meta_ac);
2448	ocfs2_schedule_truncate_log_flush(osb, 1);
2449	ocfs2_run_deallocs(osb, &ctxt.dealloc);
2450	return ret;
2451}
2452
2453static int ocfs2_xattr_ibody_remove(struct inode *inode,
2454				    struct buffer_head *di_bh,
2455				    struct ocfs2_caching_info *ref_ci,
2456				    struct buffer_head *ref_root_bh)
2457{
2458
2459	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2460	struct ocfs2_xattr_header *header;
2461	int ret;
2462	struct ocfs2_xattr_value_buf vb = {
2463		.vb_bh = di_bh,
2464		.vb_access = ocfs2_journal_access_di,
2465	};
2466
2467	header = (struct ocfs2_xattr_header *)
2468		 ((void *)di + inode->i_sb->s_blocksize -
2469		 le16_to_cpu(di->i_xattr_inline_size));
2470
2471	ret = ocfs2_remove_value_outside(inode, &vb, header,
2472					 ref_ci, ref_root_bh);
2473
2474	return ret;
2475}
2476
2477struct ocfs2_rm_xattr_bucket_para {
2478	struct ocfs2_caching_info *ref_ci;
2479	struct buffer_head *ref_root_bh;
2480};
2481
2482static int ocfs2_xattr_block_remove(struct inode *inode,
2483				    struct buffer_head *blk_bh,
2484				    struct ocfs2_caching_info *ref_ci,
2485				    struct buffer_head *ref_root_bh)
2486{
2487	struct ocfs2_xattr_block *xb;
2488	int ret = 0;
2489	struct ocfs2_xattr_value_buf vb = {
2490		.vb_bh = blk_bh,
2491		.vb_access = ocfs2_journal_access_xb,
2492	};
2493	struct ocfs2_rm_xattr_bucket_para args = {
2494		.ref_ci = ref_ci,
2495		.ref_root_bh = ref_root_bh,
2496	};
2497
2498	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2499	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2500		struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2501		ret = ocfs2_remove_value_outside(inode, &vb, header,
2502						 ref_ci, ref_root_bh);
2503	} else
2504		ret = ocfs2_iterate_xattr_index_block(inode,
2505						blk_bh,
2506						ocfs2_rm_xattr_cluster,
2507						&args);
2508
2509	return ret;
2510}
2511
2512static int ocfs2_xattr_free_block(struct inode *inode,
2513				  u64 block,
2514				  struct ocfs2_caching_info *ref_ci,
2515				  struct buffer_head *ref_root_bh)
2516{
2517	struct inode *xb_alloc_inode;
2518	struct buffer_head *xb_alloc_bh = NULL;
2519	struct buffer_head *blk_bh = NULL;
2520	struct ocfs2_xattr_block *xb;
2521	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2522	handle_t *handle;
2523	int ret = 0;
2524	u64 blk, bg_blkno;
2525	u16 bit;
2526
2527	ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2528	if (ret < 0) {
2529		mlog_errno(ret);
2530		goto out;
2531	}
2532
2533	ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2534	if (ret < 0) {
2535		mlog_errno(ret);
2536		goto out;
2537	}
2538
2539	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2540	blk = le64_to_cpu(xb->xb_blkno);
2541	bit = le16_to_cpu(xb->xb_suballoc_bit);
2542	if (xb->xb_suballoc_loc)
2543		bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2544	else
2545		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2546
2547	xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2548				EXTENT_ALLOC_SYSTEM_INODE,
2549				le16_to_cpu(xb->xb_suballoc_slot));
2550	if (!xb_alloc_inode) {
2551		ret = -ENOMEM;
2552		mlog_errno(ret);
2553		goto out;
2554	}
2555	inode_lock(xb_alloc_inode);
2556
2557	ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2558	if (ret < 0) {
2559		mlog_errno(ret);
2560		goto out_mutex;
2561	}
2562
2563	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2564	if (IS_ERR(handle)) {
2565		ret = PTR_ERR(handle);
2566		mlog_errno(ret);
2567		goto out_unlock;
2568	}
2569
2570	ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2571				       bit, bg_blkno, 1);
2572	if (ret < 0)
2573		mlog_errno(ret);
2574
2575	ocfs2_commit_trans(osb, handle);
2576out_unlock:
2577	ocfs2_inode_unlock(xb_alloc_inode, 1);
2578	brelse(xb_alloc_bh);
2579out_mutex:
2580	inode_unlock(xb_alloc_inode);
2581	iput(xb_alloc_inode);
2582out:
2583	brelse(blk_bh);
2584	return ret;
2585}
2586
2587/*
2588 * ocfs2_xattr_remove()
2589 *
2590 * Free extended attribute resources associated with this inode.
2591 */
2592int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2593{
2594	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2595	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2596	struct ocfs2_refcount_tree *ref_tree = NULL;
2597	struct buffer_head *ref_root_bh = NULL;
2598	struct ocfs2_caching_info *ref_ci = NULL;
2599	handle_t *handle;
2600	int ret;
2601
2602	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2603		return 0;
2604
2605	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2606		return 0;
2607
2608	if (ocfs2_is_refcount_inode(inode)) {
2609		ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2610					       le64_to_cpu(di->i_refcount_loc),
2611					       1, &ref_tree, &ref_root_bh);
2612		if (ret) {
2613			mlog_errno(ret);
2614			goto out;
2615		}
2616		ref_ci = &ref_tree->rf_ci;
2617
2618	}
2619
2620	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2621		ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2622					       ref_ci, ref_root_bh);
2623		if (ret < 0) {
2624			mlog_errno(ret);
2625			goto out;
2626		}
2627	}
2628
2629	if (di->i_xattr_loc) {
2630		ret = ocfs2_xattr_free_block(inode,
2631					     le64_to_cpu(di->i_xattr_loc),
2632					     ref_ci, ref_root_bh);
2633		if (ret < 0) {
2634			mlog_errno(ret);
2635			goto out;
2636		}
2637	}
2638
2639	handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2640				   OCFS2_INODE_UPDATE_CREDITS);
2641	if (IS_ERR(handle)) {
2642		ret = PTR_ERR(handle);
2643		mlog_errno(ret);
2644		goto out;
2645	}
2646	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2647				      OCFS2_JOURNAL_ACCESS_WRITE);
2648	if (ret) {
2649		mlog_errno(ret);
2650		goto out_commit;
2651	}
2652
2653	di->i_xattr_loc = 0;
2654
2655	spin_lock(&oi->ip_lock);
2656	oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2657	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2658	spin_unlock(&oi->ip_lock);
2659	ocfs2_update_inode_fsync_trans(handle, inode, 0);
2660
2661	ocfs2_journal_dirty(handle, di_bh);
2662out_commit:
2663	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2664out:
2665	if (ref_tree)
2666		ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2667	brelse(ref_root_bh);
2668	return ret;
2669}
2670
2671static int ocfs2_xattr_has_space_inline(struct inode *inode,
2672					struct ocfs2_dinode *di)
2673{
2674	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2675	unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2676	int free;
2677
2678	if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2679		return 0;
2680
2681	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2682		struct ocfs2_inline_data *idata = &di->id2.i_data;
2683		free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2684	} else if (ocfs2_inode_is_fast_symlink(inode)) {
2685		free = ocfs2_fast_symlink_chars(inode->i_sb) -
2686			le64_to_cpu(di->i_size);
2687	} else {
2688		struct ocfs2_extent_list *el = &di->id2.i_list;
2689		free = (le16_to_cpu(el->l_count) -
2690			le16_to_cpu(el->l_next_free_rec)) *
2691			sizeof(struct ocfs2_extent_rec);
2692	}
2693	if (free >= xattrsize)
2694		return 1;
2695
2696	return 0;
2697}
2698
2699/*
2700 * ocfs2_xattr_ibody_find()
2701 *
2702 * Find extended attribute in inode block and
2703 * fill search info into struct ocfs2_xattr_search.
2704 */
2705static int ocfs2_xattr_ibody_find(struct inode *inode,
2706				  int name_index,
2707				  const char *name,
2708				  struct ocfs2_xattr_search *xs)
2709{
2710	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2711	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2712	int ret;
2713	int has_space = 0;
2714
2715	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2716		return 0;
2717
2718	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2719		down_read(&oi->ip_alloc_sem);
2720		has_space = ocfs2_xattr_has_space_inline(inode, di);
2721		up_read(&oi->ip_alloc_sem);
2722		if (!has_space)
2723			return 0;
2724	}
2725
2726	xs->xattr_bh = xs->inode_bh;
2727	xs->end = (void *)di + inode->i_sb->s_blocksize;
2728	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2729		xs->header = (struct ocfs2_xattr_header *)
2730			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
2731	else
2732		xs->header = (struct ocfs2_xattr_header *)
2733			(xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2734	xs->base = (void *)xs->header;
2735	xs->here = xs->header->xh_entries;
2736
2737	/* Find the named attribute. */
2738	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2739		ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2740		if (ret && ret != -ENODATA)
2741			return ret;
2742		xs->not_found = ret;
2743	}
2744
2745	return 0;
2746}
2747
2748static int ocfs2_xattr_ibody_init(struct inode *inode,
2749				  struct buffer_head *di_bh,
2750				  struct ocfs2_xattr_set_ctxt *ctxt)
2751{
2752	int ret;
2753	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2754	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2755	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2756	unsigned int xattrsize = osb->s_xattr_inline_size;
2757
2758	if (!ocfs2_xattr_has_space_inline(inode, di)) {
2759		ret = -ENOSPC;
2760		goto out;
2761	}
2762
2763	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2764				      OCFS2_JOURNAL_ACCESS_WRITE);
2765	if (ret) {
2766		mlog_errno(ret);
2767		goto out;
2768	}
2769
2770	/*
2771	 * Adjust extent record count or inline data size
2772	 * to reserve space for extended attribute.
2773	 */
2774	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2775		struct ocfs2_inline_data *idata = &di->id2.i_data;
2776		le16_add_cpu(&idata->id_count, -xattrsize);
2777	} else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2778		struct ocfs2_extent_list *el = &di->id2.i_list;
2779		le16_add_cpu(&el->l_count, -(xattrsize /
2780					     sizeof(struct ocfs2_extent_rec)));
2781	}
2782	di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2783
2784	spin_lock(&oi->ip_lock);
2785	oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2786	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2787	spin_unlock(&oi->ip_lock);
2788
2789	ocfs2_journal_dirty(ctxt->handle, di_bh);
2790
2791out:
2792	return ret;
2793}
2794
2795/*
2796 * ocfs2_xattr_ibody_set()
2797 *
2798 * Set, replace or remove an extended attribute into inode block.
2799 *
2800 */
2801static int ocfs2_xattr_ibody_set(struct inode *inode,
2802				 struct ocfs2_xattr_info *xi,
2803				 struct ocfs2_xattr_search *xs,
2804				 struct ocfs2_xattr_set_ctxt *ctxt)
2805{
2806	int ret;
2807	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2808	struct ocfs2_xa_loc loc;
2809
2810	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2811		return -ENOSPC;
2812
2813	down_write(&oi->ip_alloc_sem);
2814	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2815		ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2816		if (ret) {
2817			if (ret != -ENOSPC)
2818				mlog_errno(ret);
2819			goto out;
2820		}
2821	}
2822
2823	ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2824				 xs->not_found ? NULL : xs->here);
2825	ret = ocfs2_xa_set(&loc, xi, ctxt);
2826	if (ret) {
2827		if (ret != -ENOSPC)
2828			mlog_errno(ret);
2829		goto out;
2830	}
2831	xs->here = loc.xl_entry;
2832
2833out:
2834	up_write(&oi->ip_alloc_sem);
2835
2836	return ret;
2837}
2838
2839/*
2840 * ocfs2_xattr_block_find()
2841 *
2842 * Find extended attribute in external block and
2843 * fill search info into struct ocfs2_xattr_search.
2844 */
2845static int ocfs2_xattr_block_find(struct inode *inode,
2846				  int name_index,
2847				  const char *name,
2848				  struct ocfs2_xattr_search *xs)
2849{
2850	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2851	struct buffer_head *blk_bh = NULL;
2852	struct ocfs2_xattr_block *xb;
2853	int ret = 0;
2854
2855	if (!di->i_xattr_loc)
2856		return ret;
2857
2858	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2859				     &blk_bh);
2860	if (ret < 0) {
2861		mlog_errno(ret);
2862		return ret;
2863	}
2864
2865	xs->xattr_bh = blk_bh;
2866	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2867
2868	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2869		xs->header = &xb->xb_attrs.xb_header;
2870		xs->base = (void *)xs->header;
2871		xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2872		xs->here = xs->header->xh_entries;
2873
2874		ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2875	} else
2876		ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2877						   name_index,
2878						   name, xs);
2879
2880	if (ret && ret != -ENODATA) {
2881		xs->xattr_bh = NULL;
2882		goto cleanup;
2883	}
2884	xs->not_found = ret;
2885	return 0;
2886cleanup:
2887	brelse(blk_bh);
2888
2889	return ret;
2890}
2891
2892static int ocfs2_create_xattr_block(struct inode *inode,
2893				    struct buffer_head *inode_bh,
2894				    struct ocfs2_xattr_set_ctxt *ctxt,
2895				    int indexed,
2896				    struct buffer_head **ret_bh)
2897{
2898	int ret;
2899	u16 suballoc_bit_start;
2900	u32 num_got;
2901	u64 suballoc_loc, first_blkno;
2902	struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2903	struct buffer_head *new_bh = NULL;
2904	struct ocfs2_xattr_block *xblk;
2905
2906	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2907				      inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2908	if (ret < 0) {
2909		mlog_errno(ret);
2910		goto end;
2911	}
2912
2913	ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2914				   &suballoc_loc, &suballoc_bit_start,
2915				   &num_got, &first_blkno);
2916	if (ret < 0) {
2917		mlog_errno(ret);
2918		goto end;
2919	}
2920
2921	new_bh = sb_getblk(inode->i_sb, first_blkno);
2922	if (!new_bh) {
2923		ret = -ENOMEM;
2924		mlog_errno(ret);
2925		goto end;
2926	}
2927
2928	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2929
2930	ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2931				      new_bh,
2932				      OCFS2_JOURNAL_ACCESS_CREATE);
2933	if (ret < 0) {
2934		mlog_errno(ret);
2935		goto end;
2936	}
2937
2938	/* Initialize ocfs2_xattr_block */
2939	xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2940	memset(xblk, 0, inode->i_sb->s_blocksize);
2941	strscpy(xblk->xb_signature, OCFS2_XATTR_BLOCK_SIGNATURE);
2942	xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2943	xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2944	xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2945	xblk->xb_fs_generation =
2946		cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2947	xblk->xb_blkno = cpu_to_le64(first_blkno);
2948	if (indexed) {
2949		struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2950		xr->xt_clusters = cpu_to_le32(1);
2951		xr->xt_last_eb_blk = 0;
2952		xr->xt_list.l_tree_depth = 0;
2953		xr->xt_list.l_count = cpu_to_le16(
2954					ocfs2_xattr_recs_per_xb(inode->i_sb));
2955		xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2956		xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2957	}
2958	ocfs2_journal_dirty(ctxt->handle, new_bh);
2959
2960	/* Add it to the inode */
2961	di->i_xattr_loc = cpu_to_le64(first_blkno);
2962
2963	spin_lock(&OCFS2_I(inode)->ip_lock);
2964	OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2965	di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2966	spin_unlock(&OCFS2_I(inode)->ip_lock);
2967
2968	ocfs2_journal_dirty(ctxt->handle, inode_bh);
2969
2970	*ret_bh = new_bh;
2971	new_bh = NULL;
2972
2973end:
2974	brelse(new_bh);
2975	return ret;
2976}
2977
2978/*
2979 * ocfs2_xattr_block_set()
2980 *
2981 * Set, replace or remove an extended attribute into external block.
2982 *
2983 */
2984static int ocfs2_xattr_block_set(struct inode *inode,
2985				 struct ocfs2_xattr_info *xi,
2986				 struct ocfs2_xattr_search *xs,
2987				 struct ocfs2_xattr_set_ctxt *ctxt)
2988{
2989	struct buffer_head *new_bh = NULL;
2990	struct ocfs2_xattr_block *xblk = NULL;
2991	int ret;
2992	struct ocfs2_xa_loc loc;
2993
2994	if (!xs->xattr_bh) {
2995		ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2996					       0, &new_bh);
2997		if (ret) {
2998			mlog_errno(ret);
2999			goto end;
3000		}
3001
3002		xs->xattr_bh = new_bh;
3003		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
3004		xs->header = &xblk->xb_attrs.xb_header;
3005		xs->base = (void *)xs->header;
3006		xs->end = (void *)xblk + inode->i_sb->s_blocksize;
3007		xs->here = xs->header->xh_entries;
3008	} else
3009		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
3010
3011	if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
3012		ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
3013					      xs->not_found ? NULL : xs->here);
3014
3015		ret = ocfs2_xa_set(&loc, xi, ctxt);
3016		if (!ret)
3017			xs->here = loc.xl_entry;
3018		else if ((ret != -ENOSPC) || ctxt->set_abort)
3019			goto end;
3020		else {
3021			ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
3022			if (ret)
3023				goto end;
3024		}
3025	}
3026
3027	if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
3028		ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
3029
3030end:
3031	return ret;
3032}
3033
3034/* Check whether the new xattr can be inserted into the inode. */
3035static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3036				       struct ocfs2_xattr_info *xi,
3037				       struct ocfs2_xattr_search *xs)
3038{
3039	struct ocfs2_xattr_entry *last;
3040	int free, i;
3041	size_t min_offs = xs->end - xs->base;
3042
3043	if (!xs->header)
3044		return 0;
3045
3046	last = xs->header->xh_entries;
3047
3048	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3049		size_t offs = le16_to_cpu(last->xe_name_offset);
3050		if (offs < min_offs)
3051			min_offs = offs;
3052		last += 1;
3053	}
3054
3055	free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3056	if (free < 0)
3057		return 0;
3058
3059	BUG_ON(!xs->not_found);
3060
3061	if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3062		return 1;
3063
3064	return 0;
3065}
3066
3067static int ocfs2_calc_xattr_set_need(struct inode *inode,
3068				     struct ocfs2_dinode *di,
3069				     struct ocfs2_xattr_info *xi,
3070				     struct ocfs2_xattr_search *xis,
3071				     struct ocfs2_xattr_search *xbs,
3072				     int *clusters_need,
3073				     int *meta_need,
3074				     int *credits_need)
3075{
3076	int ret = 0, old_in_xb = 0;
3077	int clusters_add = 0, meta_add = 0, credits = 0;
3078	struct buffer_head *bh = NULL;
3079	struct ocfs2_xattr_block *xb = NULL;
3080	struct ocfs2_xattr_entry *xe = NULL;
3081	struct ocfs2_xattr_value_root *xv = NULL;
3082	char *base = NULL;
3083	int name_offset, name_len = 0;
3084	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3085						    xi->xi_value_len);
3086	u64 value_size;
3087
3088	/*
3089	 * Calculate the clusters we need to write.
3090	 * No matter whether we replace an old one or add a new one,
3091	 * we need this for writing.
3092	 */
3093	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3094		credits += new_clusters *
3095			   ocfs2_clusters_to_blocks(inode->i_sb, 1);
3096
3097	if (xis->not_found && xbs->not_found) {
3098		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3099
3100		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3101			clusters_add += new_clusters;
3102			credits += ocfs2_calc_extend_credits(inode->i_sb,
3103							&def_xv.xv.xr_list);
3104		}
3105
3106		goto meta_guess;
3107	}
3108
3109	if (!xis->not_found) {
3110		xe = xis->here;
3111		name_offset = le16_to_cpu(xe->xe_name_offset);
3112		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3113		base = xis->base;
3114		credits += OCFS2_INODE_UPDATE_CREDITS;
3115	} else {
3116		int i, block_off = 0;
3117		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3118		xe = xbs->here;
3119		name_offset = le16_to_cpu(xe->xe_name_offset);
3120		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3121		i = xbs->here - xbs->header->xh_entries;
3122		old_in_xb = 1;
3123
3124		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3125			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3126							bucket_xh(xbs->bucket),
3127							i, &block_off,
3128							&name_offset);
3129			base = bucket_block(xbs->bucket, block_off);
3130			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3131		} else {
3132			base = xbs->base;
3133			credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3134		}
3135	}
3136
3137	/*
3138	 * delete a xattr doesn't need metadata and cluster allocation.
3139	 * so just calculate the credits and return.
3140	 *
3141	 * The credits for removing the value tree will be extended
3142	 * by ocfs2_remove_extent itself.
3143	 */
3144	if (!xi->xi_value) {
3145		if (!ocfs2_xattr_is_local(xe))
3146			credits += ocfs2_remove_extent_credits(inode->i_sb);
3147
3148		goto out;
3149	}
3150
3151	/* do cluster allocation guess first. */
3152	value_size = le64_to_cpu(xe->xe_value_size);
3153
3154	if (old_in_xb) {
3155		/*
3156		 * In xattr set, we always try to set the xe in inode first,
3157		 * so if it can be inserted into inode successfully, the old
3158		 * one will be removed from the xattr block, and this xattr
3159		 * will be inserted into inode as a new xattr in inode.
3160		 */
3161		if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3162			clusters_add += new_clusters;
3163			credits += ocfs2_remove_extent_credits(inode->i_sb) +
3164				    OCFS2_INODE_UPDATE_CREDITS;
3165			if (!ocfs2_xattr_is_local(xe))
3166				credits += ocfs2_calc_extend_credits(
3167							inode->i_sb,
3168							&def_xv.xv.xr_list);
3169			goto out;
3170		}
3171	}
3172
3173	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3174		/* the new values will be stored outside. */
3175		u32 old_clusters = 0;
3176
3177		if (!ocfs2_xattr_is_local(xe)) {
3178			old_clusters =	ocfs2_clusters_for_bytes(inode->i_sb,
3179								 value_size);
3180			xv = (struct ocfs2_xattr_value_root *)
3181			     (base + name_offset + name_len);
3182			value_size = OCFS2_XATTR_ROOT_SIZE;
3183		} else
3184			xv = &def_xv.xv;
3185
3186		if (old_clusters >= new_clusters) {
3187			credits += ocfs2_remove_extent_credits(inode->i_sb);
3188			goto out;
3189		} else {
3190			meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3191			clusters_add += new_clusters - old_clusters;
3192			credits += ocfs2_calc_extend_credits(inode->i_sb,
3193							     &xv->xr_list);
3194			if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3195				goto out;
3196		}
3197	} else {
3198		/*
3199		 * Now the new value will be stored inside. So if the new
3200		 * value is smaller than the size of value root or the old
3201		 * value, we don't need any allocation, otherwise we have
3202		 * to guess metadata allocation.
3203		 */
3204		if ((ocfs2_xattr_is_local(xe) &&
3205		     (value_size >= xi->xi_value_len)) ||
3206		    (!ocfs2_xattr_is_local(xe) &&
3207		     OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3208			goto out;
3209	}
3210
3211meta_guess:
3212	/* calculate metadata allocation. */
3213	if (di->i_xattr_loc) {
3214		if (!xbs->xattr_bh) {
3215			ret = ocfs2_read_xattr_block(inode,
3216						     le64_to_cpu(di->i_xattr_loc),
3217						     &bh);
3218			if (ret) {
3219				mlog_errno(ret);
3220				goto out;
3221			}
3222
3223			xb = (struct ocfs2_xattr_block *)bh->b_data;
3224		} else
3225			xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3226
3227		/*
3228		 * If there is already an xattr tree, good, we can calculate
3229		 * like other b-trees. Otherwise we may have the chance of
3230		 * create a tree, the credit calculation is borrowed from
3231		 * ocfs2_calc_extend_credits with root_el = NULL. And the
3232		 * new tree will be cluster based, so no meta is needed.
3233		 */
3234		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3235			struct ocfs2_extent_list *el =
3236				 &xb->xb_attrs.xb_root.xt_list;
3237			meta_add += ocfs2_extend_meta_needed(el);
3238			credits += ocfs2_calc_extend_credits(inode->i_sb,
3239							     el);
3240		} else
3241			credits += OCFS2_SUBALLOC_ALLOC + 1;
3242
3243		/*
3244		 * This cluster will be used either for new bucket or for
3245		 * new xattr block.
3246		 * If the cluster size is the same as the bucket size, one
3247		 * more is needed since we may need to extend the bucket
3248		 * also.
3249		 */
3250		clusters_add += 1;
3251		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3252		if (OCFS2_XATTR_BUCKET_SIZE ==
3253			OCFS2_SB(inode->i_sb)->s_clustersize) {
3254			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3255			clusters_add += 1;
3256		}
3257	} else {
3258		credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3259		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3260			struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3261			meta_add += ocfs2_extend_meta_needed(el);
3262			credits += ocfs2_calc_extend_credits(inode->i_sb,
3263							     el);
3264		} else {
3265			meta_add += 1;
3266		}
3267	}
3268out:
3269	if (clusters_need)
3270		*clusters_need = clusters_add;
3271	if (meta_need)
3272		*meta_need = meta_add;
3273	if (credits_need)
3274		*credits_need = credits;
3275	brelse(bh);
3276	return ret;
3277}
3278
3279static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3280				     struct ocfs2_dinode *di,
3281				     struct ocfs2_xattr_info *xi,
3282				     struct ocfs2_xattr_search *xis,
3283				     struct ocfs2_xattr_search *xbs,
3284				     struct ocfs2_xattr_set_ctxt *ctxt,
3285				     int extra_meta,
3286				     int *credits)
3287{
3288	int clusters_add, meta_add, ret;
3289	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3290
3291	memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3292
3293	ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3294
3295	ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3296					&clusters_add, &meta_add, credits);
3297	if (ret) {
3298		mlog_errno(ret);
3299		return ret;
3300	}
3301
3302	meta_add += extra_meta;
3303	trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3304					clusters_add, *credits);
3305
3306	if (meta_add) {
3307		ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3308							&ctxt->meta_ac);
3309		if (ret) {
3310			mlog_errno(ret);
3311			goto out;
3312		}
3313	}
3314
3315	if (clusters_add) {
3316		ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3317		if (ret)
3318			mlog_errno(ret);
3319	}
3320out:
3321	if (ret) {
3322		if (ctxt->meta_ac) {
3323			ocfs2_free_alloc_context(ctxt->meta_ac);
3324			ctxt->meta_ac = NULL;
3325		}
3326
3327		/*
3328		 * We cannot have an error and a non null ctxt->data_ac.
3329		 */
3330	}
3331
3332	return ret;
3333}
3334
3335static int __ocfs2_xattr_set_handle(struct inode *inode,
3336				    struct ocfs2_dinode *di,
3337				    struct ocfs2_xattr_info *xi,
3338				    struct ocfs2_xattr_search *xis,
3339				    struct ocfs2_xattr_search *xbs,
3340				    struct ocfs2_xattr_set_ctxt *ctxt)
3341{
3342	int ret = 0, credits, old_found;
3343
3344	if (!xi->xi_value) {
3345		/* Remove existing extended attribute */
3346		if (!xis->not_found)
3347			ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3348		else if (!xbs->not_found)
3349			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3350	} else {
3351		/* We always try to set extended attribute into inode first*/
3352		ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3353		if (!ret && !xbs->not_found) {
3354			/*
3355			 * If succeed and that extended attribute existing in
3356			 * external block, then we will remove it.
3357			 */
3358			xi->xi_value = NULL;
3359			xi->xi_value_len = 0;
3360
3361			old_found = xis->not_found;
3362			xis->not_found = -ENODATA;
3363			ret = ocfs2_calc_xattr_set_need(inode,
3364							di,
3365							xi,
3366							xis,
3367							xbs,
3368							NULL,
3369							NULL,
3370							&credits);
3371			xis->not_found = old_found;
3372			if (ret) {
3373				mlog_errno(ret);
3374				goto out;
3375			}
3376
3377			ret = ocfs2_extend_trans(ctxt->handle, credits);
3378			if (ret) {
3379				mlog_errno(ret);
3380				goto out;
3381			}
3382			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3383		} else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3384			if (di->i_xattr_loc && !xbs->xattr_bh) {
3385				ret = ocfs2_xattr_block_find(inode,
3386							     xi->xi_name_index,
3387							     xi->xi_name, xbs);
3388				if (ret)
3389					goto out;
3390
3391				old_found = xis->not_found;
3392				xis->not_found = -ENODATA;
3393				ret = ocfs2_calc_xattr_set_need(inode,
3394								di,
3395								xi,
3396								xis,
3397								xbs,
3398								NULL,
3399								NULL,
3400								&credits);
3401				xis->not_found = old_found;
3402				if (ret) {
3403					mlog_errno(ret);
3404					goto out;
3405				}
3406
3407				ret = ocfs2_extend_trans(ctxt->handle, credits);
3408				if (ret) {
3409					mlog_errno(ret);
3410					goto out;
3411				}
3412			}
3413			/*
3414			 * If no space in inode, we will set extended attribute
3415			 * into external block.
3416			 */
3417			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3418			if (ret)
3419				goto out;
3420			if (!xis->not_found) {
3421				/*
3422				 * If succeed and that extended attribute
3423				 * existing in inode, we will remove it.
3424				 */
3425				xi->xi_value = NULL;
3426				xi->xi_value_len = 0;
3427				xbs->not_found = -ENODATA;
3428				ret = ocfs2_calc_xattr_set_need(inode,
3429								di,
3430								xi,
3431								xis,
3432								xbs,
3433								NULL,
3434								NULL,
3435								&credits);
3436				if (ret) {
3437					mlog_errno(ret);
3438					goto out;
3439				}
3440
3441				ret = ocfs2_extend_trans(ctxt->handle, credits);
3442				if (ret) {
3443					mlog_errno(ret);
3444					goto out;
3445				}
3446				ret = ocfs2_xattr_ibody_set(inode, xi,
3447							    xis, ctxt);
3448			}
3449		}
3450	}
3451
3452	if (!ret) {
3453		/* Update inode ctime. */
3454		ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3455					      xis->inode_bh,
3456					      OCFS2_JOURNAL_ACCESS_WRITE);
3457		if (ret) {
3458			mlog_errno(ret);
3459			goto out;
3460		}
3461
3462		inode_set_ctime_current(inode);
3463		di->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
3464		di->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
3465		ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3466	}
3467out:
3468	return ret;
3469}
3470
3471/*
3472 * This function only called duing creating inode
3473 * for init security/acl xattrs of the new inode.
3474 * All transanction credits have been reserved in mknod.
3475 */
3476int ocfs2_xattr_set_handle(handle_t *handle,
3477			   struct inode *inode,
3478			   struct buffer_head *di_bh,
3479			   int name_index,
3480			   const char *name,
3481			   const void *value,
3482			   size_t value_len,
3483			   int flags,
3484			   struct ocfs2_alloc_context *meta_ac,
3485			   struct ocfs2_alloc_context *data_ac)
3486{
3487	struct ocfs2_dinode *di;
3488	int ret;
3489
3490	struct ocfs2_xattr_info xi = {
3491		.xi_name_index = name_index,
3492		.xi_name = name,
3493		.xi_name_len = strlen(name),
3494		.xi_value = value,
3495		.xi_value_len = value_len,
3496	};
3497
3498	struct ocfs2_xattr_search xis = {
3499		.not_found = -ENODATA,
3500	};
3501
3502	struct ocfs2_xattr_search xbs = {
3503		.not_found = -ENODATA,
3504	};
3505
3506	struct ocfs2_xattr_set_ctxt ctxt = {
3507		.handle = handle,
3508		.meta_ac = meta_ac,
3509		.data_ac = data_ac,
3510	};
3511
3512	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3513		return -EOPNOTSUPP;
3514
3515	/*
3516	 * In extreme situation, may need xattr bucket when
3517	 * block size is too small. And we have already reserved
3518	 * the credits for bucket in mknod.
3519	 */
3520	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3521		xbs.bucket = ocfs2_xattr_bucket_new(inode);
3522		if (!xbs.bucket) {
3523			mlog_errno(-ENOMEM);
3524			return -ENOMEM;
3525		}
3526	}
3527
3528	xis.inode_bh = xbs.inode_bh = di_bh;
3529	di = (struct ocfs2_dinode *)di_bh->b_data;
3530
3531	down_write(&OCFS2_I(inode)->ip_xattr_sem);
3532
3533	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3534	if (ret)
3535		goto cleanup;
3536	if (xis.not_found) {
3537		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3538		if (ret)
3539			goto cleanup;
3540	}
3541
3542	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3543
3544cleanup:
3545	up_write(&OCFS2_I(inode)->ip_xattr_sem);
3546	brelse(xbs.xattr_bh);
3547	ocfs2_xattr_bucket_free(xbs.bucket);
3548
3549	return ret;
3550}
3551
3552/*
3553 * ocfs2_xattr_set()
3554 *
3555 * Set, replace or remove an extended attribute for this inode.
3556 * value is NULL to remove an existing extended attribute, else either
3557 * create or replace an extended attribute.
3558 */
3559int ocfs2_xattr_set(struct inode *inode,
3560		    int name_index,
3561		    const char *name,
3562		    const void *value,
3563		    size_t value_len,
3564		    int flags)
3565{
3566	struct buffer_head *di_bh = NULL;
3567	struct ocfs2_dinode *di;
3568	int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3569	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3570	struct inode *tl_inode = osb->osb_tl_inode;
3571	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3572	struct ocfs2_refcount_tree *ref_tree = NULL;
3573	struct ocfs2_lock_holder oh;
3574
3575	struct ocfs2_xattr_info xi = {
3576		.xi_name_index = name_index,
3577		.xi_name = name,
3578		.xi_name_len = strlen(name),
3579		.xi_value = value,
3580		.xi_value_len = value_len,
3581	};
3582
3583	struct ocfs2_xattr_search xis = {
3584		.not_found = -ENODATA,
3585	};
3586
3587	struct ocfs2_xattr_search xbs = {
3588		.not_found = -ENODATA,
3589	};
3590
3591	if (!ocfs2_supports_xattr(osb))
3592		return -EOPNOTSUPP;
3593
3594	/*
3595	 * Only xbs will be used on indexed trees.  xis doesn't need a
3596	 * bucket.
3597	 */
3598	xbs.bucket = ocfs2_xattr_bucket_new(inode);
3599	if (!xbs.bucket) {
3600		mlog_errno(-ENOMEM);
3601		return -ENOMEM;
3602	}
3603
3604	had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3605	if (had_lock < 0) {
3606		ret = had_lock;
3607		mlog_errno(ret);
3608		goto cleanup_nolock;
3609	}
3610	xis.inode_bh = xbs.inode_bh = di_bh;
3611	di = (struct ocfs2_dinode *)di_bh->b_data;
3612
3613	down_write(&OCFS2_I(inode)->ip_xattr_sem);
3614	/*
3615	 * Scan inode and external block to find the same name
3616	 * extended attribute and collect search information.
3617	 */
3618	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3619	if (ret)
3620		goto cleanup;
3621	if (xis.not_found) {
3622		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3623		if (ret)
3624			goto cleanup;
3625	}
3626
3627	if (xis.not_found && xbs.not_found) {
3628		ret = -ENODATA;
3629		if (flags & XATTR_REPLACE)
3630			goto cleanup;
3631		ret = 0;
3632		if (!value)
3633			goto cleanup;
3634	} else {
3635		ret = -EEXIST;
3636		if (flags & XATTR_CREATE)
3637			goto cleanup;
3638	}
3639
3640	/* Check whether the value is refcounted and do some preparation. */
3641	if (ocfs2_is_refcount_inode(inode) &&
3642	    (!xis.not_found || !xbs.not_found)) {
3643		ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3644						   &xis, &xbs, &ref_tree,
3645						   &ref_meta, &ref_credits);
3646		if (ret) {
3647			mlog_errno(ret);
3648			goto cleanup;
3649		}
3650	}
3651
3652	inode_lock(tl_inode);
3653
3654	if (ocfs2_truncate_log_needs_flush(osb)) {
3655		ret = __ocfs2_flush_truncate_log(osb);
3656		if (ret < 0) {
3657			inode_unlock(tl_inode);
3658			mlog_errno(ret);
3659			goto cleanup;
3660		}
3661	}
3662	inode_unlock(tl_inode);
3663
3664	ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3665					&xbs, &ctxt, ref_meta, &credits);
3666	if (ret) {
3667		mlog_errno(ret);
3668		goto cleanup;
3669	}
3670
3671	/* we need to update inode's ctime field, so add credit for it. */
3672	credits += OCFS2_INODE_UPDATE_CREDITS;
3673	ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3674	if (IS_ERR(ctxt.handle)) {
3675		ret = PTR_ERR(ctxt.handle);
3676		mlog_errno(ret);
3677		goto out_free_ac;
3678	}
3679
3680	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3681	ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3682
3683	ocfs2_commit_trans(osb, ctxt.handle);
3684
3685out_free_ac:
3686	if (ctxt.data_ac)
3687		ocfs2_free_alloc_context(ctxt.data_ac);
3688	if (ctxt.meta_ac)
3689		ocfs2_free_alloc_context(ctxt.meta_ac);
3690	if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3691		ocfs2_schedule_truncate_log_flush(osb, 1);
3692	ocfs2_run_deallocs(osb, &ctxt.dealloc);
3693
3694cleanup:
3695	if (ref_tree)
3696		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3697	up_write(&OCFS2_I(inode)->ip_xattr_sem);
3698	if (!value && !ret) {
3699		ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3700		if (ret)
3701			mlog_errno(ret);
3702	}
3703	ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3704cleanup_nolock:
3705	brelse(di_bh);
3706	brelse(xbs.xattr_bh);
3707	ocfs2_xattr_bucket_free(xbs.bucket);
3708
3709	return ret;
3710}
3711
3712/*
3713 * Find the xattr extent rec which may contains name_hash.
3714 * e_cpos will be the first name hash of the xattr rec.
3715 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3716 */
3717static int ocfs2_xattr_get_rec(struct inode *inode,
3718			       u32 name_hash,
3719			       u64 *p_blkno,
3720			       u32 *e_cpos,
3721			       u32 *num_clusters,
3722			       struct ocfs2_extent_list *el)
3723{
3724	int ret = 0, i;
3725	struct buffer_head *eb_bh = NULL;
3726	struct ocfs2_extent_block *eb;
3727	struct ocfs2_extent_rec *rec = NULL;
3728	u64 e_blkno = 0;
3729
3730	if (el->l_tree_depth) {
3731		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3732				      &eb_bh);
3733		if (ret) {
3734			mlog_errno(ret);
3735			goto out;
3736		}
3737
3738		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3739		el = &eb->h_list;
3740
3741		if (el->l_tree_depth) {
3742			ret = ocfs2_error(inode->i_sb,
3743					  "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3744					  inode->i_ino,
3745					  (unsigned long long)eb_bh->b_blocknr);
3746			goto out;
3747		}
3748	}
3749
3750	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3751		rec = &el->l_recs[i];
3752
3753		if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3754			e_blkno = le64_to_cpu(rec->e_blkno);
3755			break;
3756		}
3757	}
3758
3759	if (!e_blkno) {
3760		ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3761				  inode->i_ino,
3762				  le32_to_cpu(rec->e_cpos),
3763				  ocfs2_rec_clusters(el, rec));
3764		goto out;
3765	}
3766
3767	*p_blkno = le64_to_cpu(rec->e_blkno);
3768	*num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3769	if (e_cpos)
3770		*e_cpos = le32_to_cpu(rec->e_cpos);
3771out:
3772	brelse(eb_bh);
3773	return ret;
3774}
3775
3776typedef int (xattr_bucket_func)(struct inode *inode,
3777				struct ocfs2_xattr_bucket *bucket,
3778				void *para);
3779
3780static int ocfs2_find_xe_in_bucket(struct inode *inode,
3781				   struct ocfs2_xattr_bucket *bucket,
3782				   int name_index,
3783				   const char *name,
3784				   u32 name_hash,
3785				   u16 *xe_index,
3786				   int *found)
3787{
3788	int i, ret = 0, cmp = 1, block_off, new_offset;
3789	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3790	size_t name_len = strlen(name);
3791	struct ocfs2_xattr_entry *xe = NULL;
3792	char *xe_name;
3793
3794	/*
3795	 * We don't use binary search in the bucket because there
3796	 * may be multiple entries with the same name hash.
3797	 */
3798	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3799		xe = &xh->xh_entries[i];
3800
3801		if (name_hash > le32_to_cpu(xe->xe_name_hash))
3802			continue;
3803		else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3804			break;
3805
3806		cmp = name_index - ocfs2_xattr_get_type(xe);
3807		if (!cmp)
3808			cmp = name_len - xe->xe_name_len;
3809		if (cmp)
3810			continue;
3811
3812		ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3813							xh,
3814							i,
3815							&block_off,
3816							&new_offset);
3817		if (ret) {
3818			mlog_errno(ret);
3819			break;
3820		}
3821
3822
3823		xe_name = bucket_block(bucket, block_off) + new_offset;
3824		if (!memcmp(name, xe_name, name_len)) {
3825			*xe_index = i;
3826			*found = 1;
3827			ret = 0;
3828			break;
3829		}
3830	}
3831
3832	return ret;
3833}
3834
3835/*
3836 * Find the specified xattr entry in a series of buckets.
3837 * This series start from p_blkno and last for num_clusters.
3838 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3839 * the num of the valid buckets.
3840 *
3841 * Return the buffer_head this xattr should reside in. And if the xattr's
3842 * hash is in the gap of 2 buckets, return the lower bucket.
3843 */
3844static int ocfs2_xattr_bucket_find(struct inode *inode,
3845				   int name_index,
3846				   const char *name,
3847				   u32 name_hash,
3848				   u64 p_blkno,
3849				   u32 first_hash,
3850				   u32 num_clusters,
3851				   struct ocfs2_xattr_search *xs)
3852{
3853	int ret, found = 0;
3854	struct ocfs2_xattr_header *xh = NULL;
3855	struct ocfs2_xattr_entry *xe = NULL;
3856	u16 index = 0;
3857	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3858	int low_bucket = 0, bucket, high_bucket;
3859	struct ocfs2_xattr_bucket *search;
3860	u64 blkno, lower_blkno = 0;
3861
3862	search = ocfs2_xattr_bucket_new(inode);
3863	if (!search) {
3864		ret = -ENOMEM;
3865		mlog_errno(ret);
3866		goto out;
3867	}
3868
3869	ret = ocfs2_read_xattr_bucket(search, p_blkno);
3870	if (ret) {
3871		mlog_errno(ret);
3872		goto out;
3873	}
3874
3875	xh = bucket_xh(search);
3876	high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3877	while (low_bucket <= high_bucket) {
3878		ocfs2_xattr_bucket_relse(search);
3879
3880		bucket = (low_bucket + high_bucket) / 2;
3881		blkno = p_blkno + bucket * blk_per_bucket;
3882		ret = ocfs2_read_xattr_bucket(search, blkno);
3883		if (ret) {
3884			mlog_errno(ret);
3885			goto out;
3886		}
3887
3888		xh = bucket_xh(search);
3889		xe = &xh->xh_entries[0];
3890		if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3891			high_bucket = bucket - 1;
3892			continue;
3893		}
3894
3895		/*
3896		 * Check whether the hash of the last entry in our
3897		 * bucket is larger than the search one. for an empty
3898		 * bucket, the last one is also the first one.
3899		 */
3900		if (xh->xh_count)
3901			xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3902
3903		/* record lower_blkno which may be the insert place. */
3904		lower_blkno = blkno;
3905
3906		if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3907			low_bucket = bucket + 1;
3908			continue;
3909		}
3910
3911		/* the searched xattr should reside in this bucket if exists. */
3912		ret = ocfs2_find_xe_in_bucket(inode, search,
3913					      name_index, name, name_hash,
3914					      &index, &found);
3915		if (ret) {
3916			mlog_errno(ret);
3917			goto out;
3918		}
3919		break;
3920	}
3921
3922	/*
3923	 * Record the bucket we have found.
3924	 * When the xattr's hash value is in the gap of 2 buckets, we will
3925	 * always set it to the previous bucket.
3926	 */
3927	if (!lower_blkno)
3928		lower_blkno = p_blkno;
3929
3930	/* This should be in cache - we just read it during the search */
3931	ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3932	if (ret) {
3933		mlog_errno(ret);
3934		goto out;
3935	}
3936
3937	xs->header = bucket_xh(xs->bucket);
3938	xs->base = bucket_block(xs->bucket, 0);
3939	xs->end = xs->base + inode->i_sb->s_blocksize;
3940
3941	if (found) {
3942		xs->here = &xs->header->xh_entries[index];
3943		trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3944			name, name_index, name_hash,
3945			(unsigned long long)bucket_blkno(xs->bucket),
3946			index);
3947	} else
3948		ret = -ENODATA;
3949
3950out:
3951	ocfs2_xattr_bucket_free(search);
3952	return ret;
3953}
3954
3955static int ocfs2_xattr_index_block_find(struct inode *inode,
3956					struct buffer_head *root_bh,
3957					int name_index,
3958					const char *name,
3959					struct ocfs2_xattr_search *xs)
3960{
3961	int ret;
3962	struct ocfs2_xattr_block *xb =
3963			(struct ocfs2_xattr_block *)root_bh->b_data;
3964	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3965	struct ocfs2_extent_list *el = &xb_root->xt_list;
3966	u64 p_blkno = 0;
3967	u32 first_hash, num_clusters = 0;
3968	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3969
3970	if (le16_to_cpu(el->l_next_free_rec) == 0)
3971		return -ENODATA;
3972
3973	trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3974					name, name_index, name_hash,
3975					(unsigned long long)root_bh->b_blocknr,
3976					-1);
3977
3978	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3979				  &num_clusters, el);
3980	if (ret) {
3981		mlog_errno(ret);
3982		goto out;
3983	}
3984
3985	BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3986
3987	trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3988					name, name_index, first_hash,
3989					(unsigned long long)p_blkno,
3990					num_clusters);
3991
3992	ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3993				      p_blkno, first_hash, num_clusters, xs);
3994
3995out:
3996	return ret;
3997}
3998
3999static int ocfs2_iterate_xattr_buckets(struct inode *inode,
4000				       u64 blkno,
4001				       u32 clusters,
4002				       xattr_bucket_func *func,
4003				       void *para)
4004{
4005	int i, ret = 0;
4006	u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
4007	u32 num_buckets = clusters * bpc;
4008	struct ocfs2_xattr_bucket *bucket;
4009
4010	bucket = ocfs2_xattr_bucket_new(inode);
4011	if (!bucket) {
4012		mlog_errno(-ENOMEM);
4013		return -ENOMEM;
4014	}
4015
4016	trace_ocfs2_iterate_xattr_buckets(
4017		(unsigned long long)OCFS2_I(inode)->ip_blkno,
4018		(unsigned long long)blkno, clusters);
4019
4020	for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
4021		ret = ocfs2_read_xattr_bucket(bucket, blkno);
4022		if (ret) {
4023			mlog_errno(ret);
4024			break;
4025		}
4026
4027		/*
4028		 * The real bucket num in this series of blocks is stored
4029		 * in the 1st bucket.
4030		 */
4031		if (i == 0)
4032			num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4033
4034		trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4035		     le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4036		if (func) {
4037			ret = func(inode, bucket, para);
4038			if (ret && ret != -ERANGE)
4039				mlog_errno(ret);
4040			/* Fall through to bucket_relse() */
4041		}
4042
4043		ocfs2_xattr_bucket_relse(bucket);
4044		if (ret)
4045			break;
4046	}
4047
4048	ocfs2_xattr_bucket_free(bucket);
4049	return ret;
4050}
4051
4052struct ocfs2_xattr_tree_list {
4053	char *buffer;
4054	size_t buffer_size;
4055	size_t result;
4056};
4057
4058static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4059					     struct ocfs2_xattr_header *xh,
4060					     int index,
4061					     int *block_off,
4062					     int *new_offset)
4063{
4064	u16 name_offset;
4065
4066	if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4067		return -EINVAL;
4068
4069	name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4070
4071	*block_off = name_offset >> sb->s_blocksize_bits;
4072	*new_offset = name_offset % sb->s_blocksize;
4073
4074	return 0;
4075}
4076
4077static int ocfs2_list_xattr_bucket(struct inode *inode,
4078				   struct ocfs2_xattr_bucket *bucket,
4079				   void *para)
4080{
4081	int ret = 0, type;
4082	struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4083	int i, block_off, new_offset;
4084	const char *name;
4085
4086	for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4087		struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4088		type = ocfs2_xattr_get_type(entry);
4089
4090		ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4091							bucket_xh(bucket),
4092							i,
4093							&block_off,
4094							&new_offset);
4095		if (ret)
4096			break;
4097
4098		name = (const char *)bucket_block(bucket, block_off) +
4099			new_offset;
4100		ret = ocfs2_xattr_list_entry(inode->i_sb,
4101					     xl->buffer,
4102					     xl->buffer_size,
4103					     &xl->result,
4104					     type, name,
4105					     entry->xe_name_len);
4106		if (ret)
4107			break;
4108	}
4109
4110	return ret;
4111}
4112
4113static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4114					   struct buffer_head *blk_bh,
4115					   xattr_tree_rec_func *rec_func,
4116					   void *para)
4117{
4118	struct ocfs2_xattr_block *xb =
4119			(struct ocfs2_xattr_block *)blk_bh->b_data;
4120	struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4121	int ret = 0;
4122	u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4123	u64 p_blkno = 0;
4124
4125	if (!el->l_next_free_rec || !rec_func)
4126		return 0;
4127
4128	while (name_hash > 0) {
4129		ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4130					  &e_cpos, &num_clusters, el);
4131		if (ret) {
4132			mlog_errno(ret);
4133			break;
4134		}
4135
4136		ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4137			       num_clusters, para);
4138		if (ret) {
4139			if (ret != -ERANGE)
4140				mlog_errno(ret);
4141			break;
4142		}
4143
4144		if (e_cpos == 0)
4145			break;
4146
4147		name_hash = e_cpos - 1;
4148	}
4149
4150	return ret;
4151
4152}
4153
4154static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4155				     struct buffer_head *root_bh,
4156				     u64 blkno, u32 cpos, u32 len, void *para)
4157{
4158	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4159					   ocfs2_list_xattr_bucket, para);
4160}
4161
4162static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4163					     struct buffer_head *blk_bh,
4164					     char *buffer,
4165					     size_t buffer_size)
4166{
4167	int ret;
4168	struct ocfs2_xattr_tree_list xl = {
4169		.buffer = buffer,
4170		.buffer_size = buffer_size,
4171		.result = 0,
4172	};
4173
4174	ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4175					      ocfs2_list_xattr_tree_rec, &xl);
4176	if (ret) {
4177		mlog_errno(ret);
4178		goto out;
4179	}
4180
4181	ret = xl.result;
4182out:
4183	return ret;
4184}
4185
4186static int cmp_xe(const void *a, const void *b)
4187{
4188	const struct ocfs2_xattr_entry *l = a, *r = b;
4189	u32 l_hash = le32_to_cpu(l->xe_name_hash);
4190	u32 r_hash = le32_to_cpu(r->xe_name_hash);
4191
4192	if (l_hash > r_hash)
4193		return 1;
4194	if (l_hash < r_hash)
4195		return -1;
4196	return 0;
4197}
4198
4199/*
4200 * When the ocfs2_xattr_block is filled up, new bucket will be created
4201 * and all the xattr entries will be moved to the new bucket.
4202 * The header goes at the start of the bucket, and the names+values are
4203 * filled from the end.  This is why *target starts as the last buffer.
4204 * Note: we need to sort the entries since they are not saved in order
4205 * in the ocfs2_xattr_block.
4206 */
4207static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4208					   struct buffer_head *xb_bh,
4209					   struct ocfs2_xattr_bucket *bucket)
4210{
4211	int i, blocksize = inode->i_sb->s_blocksize;
4212	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4213	u16 offset, size, off_change;
4214	struct ocfs2_xattr_entry *xe;
4215	struct ocfs2_xattr_block *xb =
4216				(struct ocfs2_xattr_block *)xb_bh->b_data;
4217	struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4218	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4219	u16 count = le16_to_cpu(xb_xh->xh_count);
4220	char *src = xb_bh->b_data;
4221	char *target = bucket_block(bucket, blks - 1);
4222
4223	trace_ocfs2_cp_xattr_block_to_bucket_begin(
4224				(unsigned long long)xb_bh->b_blocknr,
4225				(unsigned long long)bucket_blkno(bucket));
4226
4227	for (i = 0; i < blks; i++)
4228		memset(bucket_block(bucket, i), 0, blocksize);
4229
4230	/*
4231	 * Since the xe_name_offset is based on ocfs2_xattr_header,
4232	 * there is a offset change corresponding to the change of
4233	 * ocfs2_xattr_header's position.
4234	 */
4235	off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4236	xe = &xb_xh->xh_entries[count - 1];
4237	offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4238	size = blocksize - offset;
4239
4240	/* copy all the names and values. */
4241	memcpy(target + offset, src + offset, size);
4242
4243	/* Init new header now. */
4244	xh->xh_count = xb_xh->xh_count;
4245	xh->xh_num_buckets = cpu_to_le16(1);
4246	xh->xh_name_value_len = cpu_to_le16(size);
4247	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4248
4249	/* copy all the entries. */
4250	target = bucket_block(bucket, 0);
4251	offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4252	size = count * sizeof(struct ocfs2_xattr_entry);
4253	memcpy(target + offset, (char *)xb_xh + offset, size);
4254
4255	/* Change the xe offset for all the xe because of the move. */
4256	off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4257		 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4258	for (i = 0; i < count; i++)
4259		le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4260
4261	trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4262
4263	sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4264	     cmp_xe, NULL);
4265}
4266
4267/*
4268 * After we move xattr from block to index btree, we have to
4269 * update ocfs2_xattr_search to the new xe and base.
4270 *
4271 * When the entry is in xattr block, xattr_bh indicates the storage place.
4272 * While if the entry is in index b-tree, "bucket" indicates the
4273 * real place of the xattr.
4274 */
4275static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4276					    struct ocfs2_xattr_search *xs,
4277					    struct buffer_head *old_bh)
4278{
4279	char *buf = old_bh->b_data;
4280	struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4281	struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4282	int i;
4283
4284	xs->header = bucket_xh(xs->bucket);
4285	xs->base = bucket_block(xs->bucket, 0);
4286	xs->end = xs->base + inode->i_sb->s_blocksize;
4287
4288	if (xs->not_found)
4289		return;
4290
4291	i = xs->here - old_xh->xh_entries;
4292	xs->here = &xs->header->xh_entries[i];
4293}
4294
4295static int ocfs2_xattr_create_index_block(struct inode *inode,
4296					  struct ocfs2_xattr_search *xs,
4297					  struct ocfs2_xattr_set_ctxt *ctxt)
4298{
4299	int ret;
4300	u32 bit_off, len;
4301	u64 blkno;
4302	handle_t *handle = ctxt->handle;
4303	struct ocfs2_inode_info *oi = OCFS2_I(inode);
4304	struct buffer_head *xb_bh = xs->xattr_bh;
4305	struct ocfs2_xattr_block *xb =
4306			(struct ocfs2_xattr_block *)xb_bh->b_data;
4307	struct ocfs2_xattr_tree_root *xr;
4308	u16 xb_flags = le16_to_cpu(xb->xb_flags);
4309
4310	trace_ocfs2_xattr_create_index_block_begin(
4311				(unsigned long long)xb_bh->b_blocknr);
4312
4313	BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4314	BUG_ON(!xs->bucket);
4315
4316	/*
4317	 * XXX:
4318	 * We can use this lock for now, and maybe move to a dedicated mutex
4319	 * if performance becomes a problem later.
4320	 */
4321	down_write(&oi->ip_alloc_sem);
4322
4323	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4324				      OCFS2_JOURNAL_ACCESS_WRITE);
4325	if (ret) {
4326		mlog_errno(ret);
4327		goto out;
4328	}
4329
4330	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4331				     1, 1, &bit_off, &len);
4332	if (ret) {
4333		mlog_errno(ret);
4334		goto out;
4335	}
4336
4337	/*
4338	 * The bucket may spread in many blocks, and
4339	 * we will only touch the 1st block and the last block
4340	 * in the whole bucket(one for entry and one for data).
4341	 */
4342	blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4343
4344	trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4345
4346	ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4347	if (ret) {
4348		mlog_errno(ret);
4349		goto out;
4350	}
4351
4352	ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4353						OCFS2_JOURNAL_ACCESS_CREATE);
4354	if (ret) {
4355		mlog_errno(ret);
4356		goto out;
4357	}
4358
4359	ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4360	ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4361
4362	ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4363
4364	/* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4365	memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4366	       offsetof(struct ocfs2_xattr_block, xb_attrs));
4367
4368	xr = &xb->xb_attrs.xb_root;
4369	xr->xt_clusters = cpu_to_le32(1);
4370	xr->xt_last_eb_blk = 0;
4371	xr->xt_list.l_tree_depth = 0;
4372	xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4373	xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4374
4375	xr->xt_list.l_recs[0].e_cpos = 0;
4376	xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4377	xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4378
4379	xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4380
4381	ocfs2_journal_dirty(handle, xb_bh);
4382
4383out:
4384	up_write(&oi->ip_alloc_sem);
4385
4386	return ret;
4387}
4388
4389static int cmp_xe_offset(const void *a, const void *b)
4390{
4391	const struct ocfs2_xattr_entry *l = a, *r = b;
4392	u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4393	u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4394
4395	if (l_name_offset < r_name_offset)
4396		return 1;
4397	if (l_name_offset > r_name_offset)
4398		return -1;
4399	return 0;
4400}
4401
4402/*
4403 * defrag a xattr bucket if we find that the bucket has some
4404 * holes between name/value pairs.
4405 * We will move all the name/value pairs to the end of the bucket
4406 * so that we can spare some space for insertion.
4407 */
4408static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4409				     handle_t *handle,
4410				     struct ocfs2_xattr_bucket *bucket)
4411{
4412	int ret, i;
4413	size_t end, offset, len;
4414	struct ocfs2_xattr_header *xh;
4415	char *entries, *buf, *bucket_buf = NULL;
4416	u64 blkno = bucket_blkno(bucket);
4417	u16 xh_free_start;
4418	size_t blocksize = inode->i_sb->s_blocksize;
4419	struct ocfs2_xattr_entry *xe;
4420
4421	/*
4422	 * In order to make the operation more efficient and generic,
4423	 * we copy all the blocks into a contiguous memory and do the
4424	 * defragment there, so if anything is error, we will not touch
4425	 * the real block.
4426	 */
4427	bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4428	if (!bucket_buf) {
4429		ret = -EIO;
4430		goto out;
4431	}
4432
4433	buf = bucket_buf;
4434	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4435		memcpy(buf, bucket_block(bucket, i), blocksize);
4436
4437	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4438						OCFS2_JOURNAL_ACCESS_WRITE);
4439	if (ret < 0) {
4440		mlog_errno(ret);
4441		goto out;
4442	}
4443
4444	xh = (struct ocfs2_xattr_header *)bucket_buf;
4445	entries = (char *)xh->xh_entries;
4446	xh_free_start = le16_to_cpu(xh->xh_free_start);
4447
4448	trace_ocfs2_defrag_xattr_bucket(
4449	     (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4450	     xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4451
4452	/*
4453	 * sort all the entries by their offset.
4454	 * the largest will be the first, so that we can
4455	 * move them to the end one by one.
4456	 */
4457	sort(entries, le16_to_cpu(xh->xh_count),
4458	     sizeof(struct ocfs2_xattr_entry),
4459	     cmp_xe_offset, NULL);
4460
4461	/* Move all name/values to the end of the bucket. */
4462	xe = xh->xh_entries;
4463	end = OCFS2_XATTR_BUCKET_SIZE;
4464	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4465		offset = le16_to_cpu(xe->xe_name_offset);
4466		len = namevalue_size_xe(xe);
4467
4468		/*
4469		 * We must make sure that the name/value pair
4470		 * exist in the same block. So adjust end to
4471		 * the previous block end if needed.
4472		 */
4473		if (((end - len) / blocksize !=
4474			(end - 1) / blocksize))
4475			end = end - end % blocksize;
4476
4477		if (end > offset + len) {
4478			memmove(bucket_buf + end - len,
4479				bucket_buf + offset, len);
4480			xe->xe_name_offset = cpu_to_le16(end - len);
4481		}
4482
4483		mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4484				"bucket %llu\n", (unsigned long long)blkno);
4485
4486		end -= len;
4487	}
4488
4489	mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4490			"bucket %llu\n", (unsigned long long)blkno);
4491
4492	if (xh_free_start == end)
4493		goto out;
4494
4495	memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4496	xh->xh_free_start = cpu_to_le16(end);
4497
4498	/* sort the entries by their name_hash. */
4499	sort(entries, le16_to_cpu(xh->xh_count),
4500	     sizeof(struct ocfs2_xattr_entry),
4501	     cmp_xe, NULL);
4502
4503	buf = bucket_buf;
4504	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4505		memcpy(bucket_block(bucket, i), buf, blocksize);
4506	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4507
4508out:
4509	kfree(bucket_buf);
4510	return ret;
4511}
4512
4513/*
4514 * prev_blkno points to the start of an existing extent.  new_blkno
4515 * points to a newly allocated extent.  Because we know each of our
4516 * clusters contains more than bucket, we can easily split one cluster
4517 * at a bucket boundary.  So we take the last cluster of the existing
4518 * extent and split it down the middle.  We move the last half of the
4519 * buckets in the last cluster of the existing extent over to the new
4520 * extent.
4521 *
4522 * first_bh is the buffer at prev_blkno so we can update the existing
4523 * extent's bucket count.  header_bh is the bucket were we were hoping
4524 * to insert our xattr.  If the bucket move places the target in the new
4525 * extent, we'll update first_bh and header_bh after modifying the old
4526 * extent.
4527 *
4528 * first_hash will be set as the 1st xe's name_hash in the new extent.
4529 */
4530static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4531					       handle_t *handle,
4532					       struct ocfs2_xattr_bucket *first,
4533					       struct ocfs2_xattr_bucket *target,
4534					       u64 new_blkno,
4535					       u32 num_clusters,
4536					       u32 *first_hash)
4537{
4538	int ret;
4539	struct super_block *sb = inode->i_sb;
4540	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4541	int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4542	int to_move = num_buckets / 2;
4543	u64 src_blkno;
4544	u64 last_cluster_blkno = bucket_blkno(first) +
4545		((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4546
4547	BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4548	BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4549
4550	trace_ocfs2_mv_xattr_bucket_cross_cluster(
4551				(unsigned long long)last_cluster_blkno,
4552				(unsigned long long)new_blkno);
4553
4554	ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4555				     last_cluster_blkno, new_blkno,
4556				     to_move, first_hash);
4557	if (ret) {
4558		mlog_errno(ret);
4559		goto out;
4560	}
4561
4562	/* This is the first bucket that got moved */
4563	src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4564
4565	/*
4566	 * If the target bucket was part of the moved buckets, we need to
4567	 * update first and target.
4568	 */
4569	if (bucket_blkno(target) >= src_blkno) {
4570		/* Find the block for the new target bucket */
4571		src_blkno = new_blkno +
4572			(bucket_blkno(target) - src_blkno);
4573
4574		ocfs2_xattr_bucket_relse(first);
4575		ocfs2_xattr_bucket_relse(target);
4576
4577		/*
4578		 * These shouldn't fail - the buffers are in the
4579		 * journal from ocfs2_cp_xattr_bucket().
4580		 */
4581		ret = ocfs2_read_xattr_bucket(first, new_blkno);
4582		if (ret) {
4583			mlog_errno(ret);
4584			goto out;
4585		}
4586		ret = ocfs2_read_xattr_bucket(target, src_blkno);
4587		if (ret)
4588			mlog_errno(ret);
4589
4590	}
4591
4592out:
4593	return ret;
4594}
4595
4596/*
4597 * Find the suitable pos when we divide a bucket into 2.
4598 * We have to make sure the xattrs with the same hash value exist
4599 * in the same bucket.
4600 *
4601 * If this ocfs2_xattr_header covers more than one hash value, find a
4602 * place where the hash value changes.  Try to find the most even split.
4603 * The most common case is that all entries have different hash values,
4604 * and the first check we make will find a place to split.
4605 */
4606static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4607{
4608	struct ocfs2_xattr_entry *entries = xh->xh_entries;
4609	int count = le16_to_cpu(xh->xh_count);
4610	int delta, middle = count / 2;
4611
4612	/*
4613	 * We start at the middle.  Each step gets farther away in both
4614	 * directions.  We therefore hit the change in hash value
4615	 * nearest to the middle.  Note that this loop does not execute for
4616	 * count < 2.
4617	 */
4618	for (delta = 0; delta < middle; delta++) {
4619		/* Let's check delta earlier than middle */
4620		if (cmp_xe(&entries[middle - delta - 1],
4621			   &entries[middle - delta]))
4622			return middle - delta;
4623
4624		/* For even counts, don't walk off the end */
4625		if ((middle + delta + 1) == count)
4626			continue;
4627
4628		/* Now try delta past middle */
4629		if (cmp_xe(&entries[middle + delta],
4630			   &entries[middle + delta + 1]))
4631			return middle + delta + 1;
4632	}
4633
4634	/* Every entry had the same hash */
4635	return count;
4636}
4637
4638/*
4639 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4640 * first_hash will record the 1st hash of the new bucket.
4641 *
4642 * Normally half of the xattrs will be moved.  But we have to make
4643 * sure that the xattrs with the same hash value are stored in the
4644 * same bucket. If all the xattrs in this bucket have the same hash
4645 * value, the new bucket will be initialized as an empty one and the
4646 * first_hash will be initialized as (hash_value+1).
4647 */
4648static int ocfs2_divide_xattr_bucket(struct inode *inode,
4649				    handle_t *handle,
4650				    u64 blk,
4651				    u64 new_blk,
4652				    u32 *first_hash,
4653				    int new_bucket_head)
4654{
4655	int ret, i;
4656	int count, start, len, name_value_len = 0, name_offset = 0;
4657	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4658	struct ocfs2_xattr_header *xh;
4659	struct ocfs2_xattr_entry *xe;
4660	int blocksize = inode->i_sb->s_blocksize;
4661
4662	trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4663					      (unsigned long long)new_blk);
4664
4665	s_bucket = ocfs2_xattr_bucket_new(inode);
4666	t_bucket = ocfs2_xattr_bucket_new(inode);
4667	if (!s_bucket || !t_bucket) {
4668		ret = -ENOMEM;
4669		mlog_errno(ret);
4670		goto out;
4671	}
4672
4673	ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4674	if (ret) {
4675		mlog_errno(ret);
4676		goto out;
4677	}
4678
4679	ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4680						OCFS2_JOURNAL_ACCESS_WRITE);
4681	if (ret) {
4682		mlog_errno(ret);
4683		goto out;
4684	}
4685
4686	/*
4687	 * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4688	 * there's no need to read it.
4689	 */
4690	ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4691	if (ret) {
4692		mlog_errno(ret);
4693		goto out;
4694	}
4695
4696	/*
4697	 * Hey, if we're overwriting t_bucket, what difference does
4698	 * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4699	 * same part of ocfs2_cp_xattr_bucket().
4700	 */
4701	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4702						new_bucket_head ?
4703						OCFS2_JOURNAL_ACCESS_CREATE :
4704						OCFS2_JOURNAL_ACCESS_WRITE);
4705	if (ret) {
4706		mlog_errno(ret);
4707		goto out;
4708	}
4709
4710	xh = bucket_xh(s_bucket);
4711	count = le16_to_cpu(xh->xh_count);
4712	start = ocfs2_xattr_find_divide_pos(xh);
4713
4714	if (start == count) {
4715		xe = &xh->xh_entries[start-1];
4716
4717		/*
4718		 * initialized a new empty bucket here.
4719		 * The hash value is set as one larger than
4720		 * that of the last entry in the previous bucket.
4721		 */
4722		for (i = 0; i < t_bucket->bu_blocks; i++)
4723			memset(bucket_block(t_bucket, i), 0, blocksize);
4724
4725		xh = bucket_xh(t_bucket);
4726		xh->xh_free_start = cpu_to_le16(blocksize);
4727		xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4728		le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4729
4730		goto set_num_buckets;
4731	}
4732
4733	/* copy the whole bucket to the new first. */
4734	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4735
4736	/* update the new bucket. */
4737	xh = bucket_xh(t_bucket);
4738
4739	/*
4740	 * Calculate the total name/value len and xh_free_start for
4741	 * the old bucket first.
4742	 */
4743	name_offset = OCFS2_XATTR_BUCKET_SIZE;
4744	name_value_len = 0;
4745	for (i = 0; i < start; i++) {
4746		xe = &xh->xh_entries[i];
4747		name_value_len += namevalue_size_xe(xe);
4748		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4749			name_offset = le16_to_cpu(xe->xe_name_offset);
4750	}
4751
4752	/*
4753	 * Now begin the modification to the new bucket.
4754	 *
4755	 * In the new bucket, We just move the xattr entry to the beginning
4756	 * and don't touch the name/value. So there will be some holes in the
4757	 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4758	 * called.
4759	 */
4760	xe = &xh->xh_entries[start];
4761	len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4762	trace_ocfs2_divide_xattr_bucket_move(len,
4763			(int)((char *)xe - (char *)xh),
4764			(int)((char *)xh->xh_entries - (char *)xh));
4765	memmove((char *)xh->xh_entries, (char *)xe, len);
4766	xe = &xh->xh_entries[count - start];
4767	len = sizeof(struct ocfs2_xattr_entry) * start;
4768	memset((char *)xe, 0, len);
4769
4770	le16_add_cpu(&xh->xh_count, -start);
4771	le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4772
4773	/* Calculate xh_free_start for the new bucket. */
4774	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4775	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4776		xe = &xh->xh_entries[i];
4777		if (le16_to_cpu(xe->xe_name_offset) <
4778		    le16_to_cpu(xh->xh_free_start))
4779			xh->xh_free_start = xe->xe_name_offset;
4780	}
4781
4782set_num_buckets:
4783	/* set xh->xh_num_buckets for the new xh. */
4784	if (new_bucket_head)
4785		xh->xh_num_buckets = cpu_to_le16(1);
4786	else
4787		xh->xh_num_buckets = 0;
4788
4789	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4790
4791	/* store the first_hash of the new bucket. */
4792	if (first_hash)
4793		*first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4794
4795	/*
4796	 * Now only update the 1st block of the old bucket.  If we
4797	 * just added a new empty bucket, there is no need to modify
4798	 * it.
4799	 */
4800	if (start == count)
4801		goto out;
4802
4803	xh = bucket_xh(s_bucket);
4804	memset(&xh->xh_entries[start], 0,
4805	       sizeof(struct ocfs2_xattr_entry) * (count - start));
4806	xh->xh_count = cpu_to_le16(start);
4807	xh->xh_free_start = cpu_to_le16(name_offset);
4808	xh->xh_name_value_len = cpu_to_le16(name_value_len);
4809
4810	ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4811
4812out:
4813	ocfs2_xattr_bucket_free(s_bucket);
4814	ocfs2_xattr_bucket_free(t_bucket);
4815
4816	return ret;
4817}
4818
4819/*
4820 * Copy xattr from one bucket to another bucket.
4821 *
4822 * The caller must make sure that the journal transaction
4823 * has enough space for journaling.
4824 */
4825static int ocfs2_cp_xattr_bucket(struct inode *inode,
4826				 handle_t *handle,
4827				 u64 s_blkno,
4828				 u64 t_blkno,
4829				 int t_is_new)
4830{
4831	int ret;
4832	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4833
4834	BUG_ON(s_blkno == t_blkno);
4835
4836	trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4837				    (unsigned long long)t_blkno,
4838				    t_is_new);
4839
4840	s_bucket = ocfs2_xattr_bucket_new(inode);
4841	t_bucket = ocfs2_xattr_bucket_new(inode);
4842	if (!s_bucket || !t_bucket) {
4843		ret = -ENOMEM;
4844		mlog_errno(ret);
4845		goto out;
4846	}
4847
4848	ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4849	if (ret)
4850		goto out;
4851
4852	/*
4853	 * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4854	 * there's no need to read it.
4855	 */
4856	ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4857	if (ret)
4858		goto out;
4859
4860	/*
4861	 * Hey, if we're overwriting t_bucket, what difference does
4862	 * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4863	 * cluster to fill, we came here from
4864	 * ocfs2_mv_xattr_buckets(), and it is really new -
4865	 * ACCESS_CREATE is required.  But we also might have moved data
4866	 * out of t_bucket before extending back into it.
4867	 * ocfs2_add_new_xattr_bucket() can do this - its call to
4868	 * ocfs2_add_new_xattr_cluster() may have created a new extent
4869	 * and copied out the end of the old extent.  Then it re-extends
4870	 * the old extent back to create space for new xattrs.  That's
4871	 * how we get here, and the bucket isn't really new.
4872	 */
4873	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4874						t_is_new ?
4875						OCFS2_JOURNAL_ACCESS_CREATE :
4876						OCFS2_JOURNAL_ACCESS_WRITE);
4877	if (ret)
4878		goto out;
4879
4880	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4881	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4882
4883out:
4884	ocfs2_xattr_bucket_free(t_bucket);
4885	ocfs2_xattr_bucket_free(s_bucket);
4886
4887	return ret;
4888}
4889
4890/*
4891 * src_blk points to the start of an existing extent.  last_blk points to
4892 * last cluster in that extent.  to_blk points to a newly allocated
4893 * extent.  We copy the buckets from the cluster at last_blk to the new
4894 * extent.  If start_bucket is non-zero, we skip that many buckets before
4895 * we start copying.  The new extent's xh_num_buckets gets set to the
4896 * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4897 * by the same amount.
4898 */
4899static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4900				  u64 src_blk, u64 last_blk, u64 to_blk,
4901				  unsigned int start_bucket,
4902				  u32 *first_hash)
4903{
4904	int i, ret, credits;
4905	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4906	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4907	int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4908	struct ocfs2_xattr_bucket *old_first, *new_first;
4909
4910	trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4911				     (unsigned long long)to_blk);
4912
4913	BUG_ON(start_bucket >= num_buckets);
4914	if (start_bucket) {
4915		num_buckets -= start_bucket;
4916		last_blk += (start_bucket * blks_per_bucket);
4917	}
4918
4919	/* The first bucket of the original extent */
4920	old_first = ocfs2_xattr_bucket_new(inode);
4921	/* The first bucket of the new extent */
4922	new_first = ocfs2_xattr_bucket_new(inode);
4923	if (!old_first || !new_first) {
4924		ret = -ENOMEM;
4925		mlog_errno(ret);
4926		goto out;
4927	}
4928
4929	ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4930	if (ret) {
4931		mlog_errno(ret);
4932		goto out;
4933	}
4934
4935	/*
4936	 * We need to update the first bucket of the old extent and all
4937	 * the buckets going to the new extent.
4938	 */
4939	credits = ((num_buckets + 1) * blks_per_bucket);
4940	ret = ocfs2_extend_trans(handle, credits);
4941	if (ret) {
4942		mlog_errno(ret);
4943		goto out;
4944	}
4945
4946	ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4947						OCFS2_JOURNAL_ACCESS_WRITE);
4948	if (ret) {
4949		mlog_errno(ret);
4950		goto out;
4951	}
4952
4953	for (i = 0; i < num_buckets; i++) {
4954		ret = ocfs2_cp_xattr_bucket(inode, handle,
4955					    last_blk + (i * blks_per_bucket),
4956					    to_blk + (i * blks_per_bucket),
4957					    1);
4958		if (ret) {
4959			mlog_errno(ret);
4960			goto out;
4961		}
4962	}
4963
4964	/*
4965	 * Get the new bucket ready before we dirty anything
4966	 * (This actually shouldn't fail, because we already dirtied
4967	 * it once in ocfs2_cp_xattr_bucket()).
4968	 */
4969	ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4970	if (ret) {
4971		mlog_errno(ret);
4972		goto out;
4973	}
4974	ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4975						OCFS2_JOURNAL_ACCESS_WRITE);
4976	if (ret) {
4977		mlog_errno(ret);
4978		goto out;
4979	}
4980
4981	/* Now update the headers */
4982	le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4983	ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4984
4985	bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4986	ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4987
4988	if (first_hash)
4989		*first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4990
4991out:
4992	ocfs2_xattr_bucket_free(new_first);
4993	ocfs2_xattr_bucket_free(old_first);
4994	return ret;
4995}
4996
4997/*
4998 * Move some xattrs in this cluster to the new cluster.
4999 * This function should only be called when bucket size == cluster size.
5000 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
5001 */
5002static int ocfs2_divide_xattr_cluster(struct inode *inode,
5003				      handle_t *handle,
5004				      u64 prev_blk,
5005				      u64 new_blk,
5006				      u32 *first_hash)
5007{
5008	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5009	int ret, credits = 2 * blk_per_bucket;
5010
5011	BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
5012
5013	ret = ocfs2_extend_trans(handle, credits);
5014	if (ret) {
5015		mlog_errno(ret);
5016		return ret;
5017	}
5018
5019	/* Move half of the xattr in start_blk to the next bucket. */
5020	return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
5021					  new_blk, first_hash, 1);
5022}
5023
5024/*
5025 * Move some xattrs from the old cluster to the new one since they are not
5026 * contiguous in ocfs2 xattr tree.
5027 *
5028 * new_blk starts a new separate cluster, and we will move some xattrs from
5029 * prev_blk to it. v_start will be set as the first name hash value in this
5030 * new cluster so that it can be used as e_cpos during tree insertion and
5031 * don't collide with our original b-tree operations. first_bh and header_bh
5032 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5033 * to extend the insert bucket.
5034 *
5035 * The problem is how much xattr should we move to the new one and when should
5036 * we update first_bh and header_bh?
5037 * 1. If cluster size > bucket size, that means the previous cluster has more
5038 *    than 1 bucket, so just move half nums of bucket into the new cluster and
5039 *    update the first_bh and header_bh if the insert bucket has been moved
5040 *    to the new cluster.
5041 * 2. If cluster_size == bucket_size:
5042 *    a) If the previous extent rec has more than one cluster and the insert
5043 *       place isn't in the last cluster, copy the entire last cluster to the
5044 *       new one. This time, we don't need to update the first_bh and header_bh
5045 *       since they will not be moved into the new cluster.
5046 *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5047 *       the new one. And we set the extend flag to zero if the insert place is
5048 *       moved into the new allocated cluster since no extend is needed.
5049 */
5050static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5051					    handle_t *handle,
5052					    struct ocfs2_xattr_bucket *first,
5053					    struct ocfs2_xattr_bucket *target,
5054					    u64 new_blk,
5055					    u32 prev_clusters,
5056					    u32 *v_start,
5057					    int *extend)
5058{
5059	int ret;
5060
5061	trace_ocfs2_adjust_xattr_cross_cluster(
5062			(unsigned long long)bucket_blkno(first),
5063			(unsigned long long)new_blk, prev_clusters);
5064
5065	if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5066		ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5067							  handle,
5068							  first, target,
5069							  new_blk,
5070							  prev_clusters,
5071							  v_start);
5072		if (ret)
5073			mlog_errno(ret);
5074	} else {
5075		/* The start of the last cluster in the first extent */
5076		u64 last_blk = bucket_blkno(first) +
5077			((prev_clusters - 1) *
5078			 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5079
5080		if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5081			ret = ocfs2_mv_xattr_buckets(inode, handle,
5082						     bucket_blkno(first),
5083						     last_blk, new_blk, 0,
5084						     v_start);
5085			if (ret)
5086				mlog_errno(ret);
5087		} else {
5088			ret = ocfs2_divide_xattr_cluster(inode, handle,
5089							 last_blk, new_blk,
5090							 v_start);
5091			if (ret)
5092				mlog_errno(ret);
5093
5094			if ((bucket_blkno(target) == last_blk) && extend)
5095				*extend = 0;
5096		}
5097	}
5098
5099	return ret;
5100}
5101
5102/*
5103 * Add a new cluster for xattr storage.
5104 *
5105 * If the new cluster is contiguous with the previous one, it will be
5106 * appended to the same extent record, and num_clusters will be updated.
5107 * If not, we will insert a new extent for it and move some xattrs in
5108 * the last cluster into the new allocated one.
5109 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5110 * lose the benefits of hashing because we'll have to search large leaves.
5111 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5112 * if it's bigger).
5113 *
5114 * first_bh is the first block of the previous extent rec and header_bh
5115 * indicates the bucket we will insert the new xattrs. They will be updated
5116 * when the header_bh is moved into the new cluster.
5117 */
5118static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5119				       struct buffer_head *root_bh,
5120				       struct ocfs2_xattr_bucket *first,
5121				       struct ocfs2_xattr_bucket *target,
5122				       u32 *num_clusters,
5123				       u32 prev_cpos,
5124				       int *extend,
5125				       struct ocfs2_xattr_set_ctxt *ctxt)
5126{
5127	int ret;
5128	u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5129	u32 prev_clusters = *num_clusters;
5130	u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5131	u64 block;
5132	handle_t *handle = ctxt->handle;
5133	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5134	struct ocfs2_extent_tree et;
5135
5136	trace_ocfs2_add_new_xattr_cluster_begin(
5137		(unsigned long long)OCFS2_I(inode)->ip_blkno,
5138		(unsigned long long)bucket_blkno(first),
5139		prev_cpos, prev_clusters);
5140
5141	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5142
5143	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5144				      OCFS2_JOURNAL_ACCESS_WRITE);
5145	if (ret < 0) {
5146		mlog_errno(ret);
5147		goto leave;
5148	}
5149
5150	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5151				     clusters_to_add, &bit_off, &num_bits);
5152	if (ret < 0) {
5153		if (ret != -ENOSPC)
5154			mlog_errno(ret);
5155		goto leave;
5156	}
5157
5158	BUG_ON(num_bits > clusters_to_add);
5159
5160	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5161	trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5162
5163	if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5164	    (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5165	     OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5166		/*
5167		 * If this cluster is contiguous with the old one and
5168		 * adding this new cluster, we don't surpass the limit of
5169		 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5170		 * initialized and used like other buckets in the previous
5171		 * cluster.
5172		 * So add it as a contiguous one. The caller will handle
5173		 * its init process.
5174		 */
5175		v_start = prev_cpos + prev_clusters;
5176		*num_clusters = prev_clusters + num_bits;
5177	} else {
5178		ret = ocfs2_adjust_xattr_cross_cluster(inode,
5179						       handle,
5180						       first,
5181						       target,
5182						       block,
5183						       prev_clusters,
5184						       &v_start,
5185						       extend);
5186		if (ret) {
5187			mlog_errno(ret);
5188			goto leave;
5189		}
5190	}
5191
5192	trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5193						 v_start, num_bits);
5194	ret = ocfs2_insert_extent(handle, &et, v_start, block,
5195				  num_bits, 0, ctxt->meta_ac);
5196	if (ret < 0) {
5197		mlog_errno(ret);
5198		goto leave;
5199	}
5200
5201	ocfs2_journal_dirty(handle, root_bh);
5202
5203leave:
5204	return ret;
5205}
5206
5207/*
5208 * We are given an extent.  'first' is the bucket at the very front of
5209 * the extent.  The extent has space for an additional bucket past
5210 * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5211 * of the target bucket.  We wish to shift every bucket past the target
5212 * down one, filling in that additional space.  When we get back to the
5213 * target, we split the target between itself and the now-empty bucket
5214 * at target+1 (aka, target_blkno + blks_per_bucket).
5215 */
5216static int ocfs2_extend_xattr_bucket(struct inode *inode,
5217				     handle_t *handle,
5218				     struct ocfs2_xattr_bucket *first,
5219				     u64 target_blk,
5220				     u32 num_clusters)
5221{
5222	int ret, credits;
5223	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5224	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5225	u64 end_blk;
5226	u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5227
5228	trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5229					(unsigned long long)bucket_blkno(first),
5230					num_clusters, new_bucket);
5231
5232	/* The extent must have room for an additional bucket */
5233	BUG_ON(new_bucket >=
5234	       (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5235
5236	/* end_blk points to the last existing bucket */
5237	end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5238
5239	/*
5240	 * end_blk is the start of the last existing bucket.
5241	 * Thus, (end_blk - target_blk) covers the target bucket and
5242	 * every bucket after it up to, but not including, the last
5243	 * existing bucket.  Then we add the last existing bucket, the
5244	 * new bucket, and the first bucket (3 * blk_per_bucket).
5245	 */
5246	credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5247	ret = ocfs2_extend_trans(handle, credits);
5248	if (ret) {
5249		mlog_errno(ret);
5250		goto out;
5251	}
5252
5253	ret = ocfs2_xattr_bucket_journal_access(handle, first,
5254						OCFS2_JOURNAL_ACCESS_WRITE);
5255	if (ret) {
5256		mlog_errno(ret);
5257		goto out;
5258	}
5259
5260	while (end_blk != target_blk) {
5261		ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5262					    end_blk + blk_per_bucket, 0);
5263		if (ret)
5264			goto out;
5265		end_blk -= blk_per_bucket;
5266	}
5267
5268	/* Move half of the xattr in target_blkno to the next bucket. */
5269	ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5270					target_blk + blk_per_bucket, NULL, 0);
5271
5272	le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5273	ocfs2_xattr_bucket_journal_dirty(handle, first);
5274
5275out:
5276	return ret;
5277}
5278
5279/*
5280 * Add new xattr bucket in an extent record and adjust the buckets
5281 * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5282 * bucket we want to insert into.
5283 *
5284 * In the easy case, we will move all the buckets after target down by
5285 * one. Half of target's xattrs will be moved to the next bucket.
5286 *
5287 * If current cluster is full, we'll allocate a new one.  This may not
5288 * be contiguous.  The underlying calls will make sure that there is
5289 * space for the insert, shifting buckets around if necessary.
5290 * 'target' may be moved by those calls.
5291 */
5292static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5293				      struct buffer_head *xb_bh,
5294				      struct ocfs2_xattr_bucket *target,
5295				      struct ocfs2_xattr_set_ctxt *ctxt)
5296{
5297	struct ocfs2_xattr_block *xb =
5298			(struct ocfs2_xattr_block *)xb_bh->b_data;
5299	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5300	struct ocfs2_extent_list *el = &xb_root->xt_list;
5301	u32 name_hash =
5302		le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5303	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5304	int ret, num_buckets, extend = 1;
5305	u64 p_blkno;
5306	u32 e_cpos, num_clusters;
5307	/* The bucket at the front of the extent */
5308	struct ocfs2_xattr_bucket *first;
5309
5310	trace_ocfs2_add_new_xattr_bucket(
5311				(unsigned long long)bucket_blkno(target));
5312
5313	/* The first bucket of the original extent */
5314	first = ocfs2_xattr_bucket_new(inode);
5315	if (!first) {
5316		ret = -ENOMEM;
5317		mlog_errno(ret);
5318		goto out;
5319	}
5320
5321	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5322				  &num_clusters, el);
5323	if (ret) {
5324		mlog_errno(ret);
5325		goto out;
5326	}
5327
5328	ret = ocfs2_read_xattr_bucket(first, p_blkno);
5329	if (ret) {
5330		mlog_errno(ret);
5331		goto out;
5332	}
5333
5334	num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5335	if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5336		/*
5337		 * This can move first+target if the target bucket moves
5338		 * to the new extent.
5339		 */
5340		ret = ocfs2_add_new_xattr_cluster(inode,
5341						  xb_bh,
5342						  first,
5343						  target,
5344						  &num_clusters,
5345						  e_cpos,
5346						  &extend,
5347						  ctxt);
5348		if (ret) {
5349			mlog_errno(ret);
5350			goto out;
5351		}
5352	}
5353
5354	if (extend) {
5355		ret = ocfs2_extend_xattr_bucket(inode,
5356						ctxt->handle,
5357						first,
5358						bucket_blkno(target),
5359						num_clusters);
5360		if (ret)
5361			mlog_errno(ret);
5362	}
5363
5364out:
5365	ocfs2_xattr_bucket_free(first);
5366
5367	return ret;
5368}
5369
5370/*
5371 * Truncate the specified xe_off entry in xattr bucket.
5372 * bucket is indicated by header_bh and len is the new length.
5373 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5374 *
5375 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5376 */
5377static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5378					     struct ocfs2_xattr_bucket *bucket,
5379					     int xe_off,
5380					     int len,
5381					     struct ocfs2_xattr_set_ctxt *ctxt)
5382{
5383	int ret, offset;
5384	u64 value_blk;
5385	struct ocfs2_xattr_entry *xe;
5386	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5387	size_t blocksize = inode->i_sb->s_blocksize;
5388	struct ocfs2_xattr_value_buf vb = {
5389		.vb_access = ocfs2_journal_access,
5390	};
5391
5392	xe = &xh->xh_entries[xe_off];
5393
5394	BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5395
5396	offset = le16_to_cpu(xe->xe_name_offset) +
5397		 OCFS2_XATTR_SIZE(xe->xe_name_len);
5398
5399	value_blk = offset / blocksize;
5400
5401	/* We don't allow ocfs2_xattr_value to be stored in different block. */
5402	BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5403
5404	vb.vb_bh = bucket->bu_bhs[value_blk];
5405	BUG_ON(!vb.vb_bh);
5406
5407	vb.vb_xv = (struct ocfs2_xattr_value_root *)
5408		(vb.vb_bh->b_data + offset % blocksize);
5409
5410	/*
5411	 * From here on out we have to dirty the bucket.  The generic
5412	 * value calls only modify one of the bucket's bhs, but we need
5413	 * to send the bucket at once.  So if they error, they *could* have
5414	 * modified something.  We have to assume they did, and dirty
5415	 * the whole bucket.  This leaves us in a consistent state.
5416	 */
5417	trace_ocfs2_xattr_bucket_value_truncate(
5418			(unsigned long long)bucket_blkno(bucket), xe_off, len);
5419	ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5420	if (ret) {
5421		mlog_errno(ret);
5422		goto out;
5423	}
5424
5425	ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5426						OCFS2_JOURNAL_ACCESS_WRITE);
5427	if (ret) {
5428		mlog_errno(ret);
5429		goto out;
5430	}
5431
5432	xe->xe_value_size = cpu_to_le64(len);
5433
5434	ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5435
5436out:
5437	return ret;
5438}
5439
5440static int ocfs2_rm_xattr_cluster(struct inode *inode,
5441				  struct buffer_head *root_bh,
5442				  u64 blkno,
5443				  u32 cpos,
5444				  u32 len,
5445				  void *para)
5446{
5447	int ret;
5448	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5449	struct inode *tl_inode = osb->osb_tl_inode;
5450	handle_t *handle;
5451	struct ocfs2_xattr_block *xb =
5452			(struct ocfs2_xattr_block *)root_bh->b_data;
5453	struct ocfs2_alloc_context *meta_ac = NULL;
5454	struct ocfs2_cached_dealloc_ctxt dealloc;
5455	struct ocfs2_extent_tree et;
5456
5457	ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5458					  ocfs2_delete_xattr_in_bucket, para);
5459	if (ret) {
5460		mlog_errno(ret);
5461		return ret;
5462	}
5463
5464	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5465
5466	ocfs2_init_dealloc_ctxt(&dealloc);
5467
5468	trace_ocfs2_rm_xattr_cluster(
5469			(unsigned long long)OCFS2_I(inode)->ip_blkno,
5470			(unsigned long long)blkno, cpos, len);
5471
5472	ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5473					       len);
5474
5475	ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5476	if (ret) {
5477		mlog_errno(ret);
5478		return ret;
5479	}
5480
5481	inode_lock(tl_inode);
5482
5483	if (ocfs2_truncate_log_needs_flush(osb)) {
5484		ret = __ocfs2_flush_truncate_log(osb);
5485		if (ret < 0) {
5486			mlog_errno(ret);
5487			goto out;
5488		}
5489	}
5490
5491	handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5492	if (IS_ERR(handle)) {
5493		ret = -ENOMEM;
5494		mlog_errno(ret);
5495		goto out;
5496	}
5497
5498	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5499				      OCFS2_JOURNAL_ACCESS_WRITE);
5500	if (ret) {
5501		mlog_errno(ret);
5502		goto out_commit;
5503	}
5504
5505	ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5506				  &dealloc);
5507	if (ret) {
5508		mlog_errno(ret);
5509		goto out_commit;
5510	}
5511
5512	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5513	ocfs2_journal_dirty(handle, root_bh);
5514
5515	ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5516	if (ret)
5517		mlog_errno(ret);
5518	ocfs2_update_inode_fsync_trans(handle, inode, 0);
5519
5520out_commit:
5521	ocfs2_commit_trans(osb, handle);
5522out:
5523	ocfs2_schedule_truncate_log_flush(osb, 1);
5524
5525	inode_unlock(tl_inode);
5526
5527	if (meta_ac)
5528		ocfs2_free_alloc_context(meta_ac);
5529
5530	ocfs2_run_deallocs(osb, &dealloc);
5531
5532	return ret;
5533}
5534
5535/*
5536 * check whether the xattr bucket is filled up with the same hash value.
5537 * If we want to insert the xattr with the same hash, return -ENOSPC.
5538 * If we want to insert a xattr with different hash value, go ahead
5539 * and ocfs2_divide_xattr_bucket will handle this.
5540 */
5541static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5542					      struct ocfs2_xattr_bucket *bucket,
5543					      const char *name)
5544{
5545	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5546	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5547
5548	if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5549		return 0;
5550
5551	if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5552	    xh->xh_entries[0].xe_name_hash) {
5553		mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5554		     "hash = %u\n",
5555		     (unsigned long long)bucket_blkno(bucket),
5556		     le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5557		return -ENOSPC;
5558	}
5559
5560	return 0;
5561}
5562
5563/*
5564 * Try to set the entry in the current bucket.  If we fail, the caller
5565 * will handle getting us another bucket.
5566 */
5567static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5568					struct ocfs2_xattr_info *xi,
5569					struct ocfs2_xattr_search *xs,
5570					struct ocfs2_xattr_set_ctxt *ctxt)
5571{
5572	int ret;
5573	struct ocfs2_xa_loc loc;
5574
5575	trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5576
5577	ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5578				       xs->not_found ? NULL : xs->here);
5579	ret = ocfs2_xa_set(&loc, xi, ctxt);
5580	if (!ret) {
5581		xs->here = loc.xl_entry;
5582		goto out;
5583	}
5584	if (ret != -ENOSPC) {
5585		mlog_errno(ret);
5586		goto out;
5587	}
5588
5589	/* Ok, we need space.  Let's try defragmenting the bucket. */
5590	ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5591					xs->bucket);
5592	if (ret) {
5593		mlog_errno(ret);
5594		goto out;
5595	}
5596
5597	ret = ocfs2_xa_set(&loc, xi, ctxt);
5598	if (!ret) {
5599		xs->here = loc.xl_entry;
5600		goto out;
5601	}
5602	if (ret != -ENOSPC)
5603		mlog_errno(ret);
5604
5605
5606out:
5607	return ret;
5608}
5609
5610static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5611					     struct ocfs2_xattr_info *xi,
5612					     struct ocfs2_xattr_search *xs,
5613					     struct ocfs2_xattr_set_ctxt *ctxt)
5614{
5615	int ret;
5616
5617	trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5618
5619	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5620	if (!ret)
5621		goto out;
5622	if (ret != -ENOSPC) {
5623		mlog_errno(ret);
5624		goto out;
5625	}
5626
5627	/* Ack, need more space.  Let's try to get another bucket! */
5628
5629	/*
5630	 * We do not allow for overlapping ranges between buckets. And
5631	 * the maximum number of collisions we will allow for then is
5632	 * one bucket's worth, so check it here whether we need to
5633	 * add a new bucket for the insert.
5634	 */
5635	ret = ocfs2_check_xattr_bucket_collision(inode,
5636						 xs->bucket,
5637						 xi->xi_name);
5638	if (ret) {
5639		mlog_errno(ret);
5640		goto out;
5641	}
5642
5643	ret = ocfs2_add_new_xattr_bucket(inode,
5644					 xs->xattr_bh,
5645					 xs->bucket,
5646					 ctxt);
5647	if (ret) {
5648		mlog_errno(ret);
5649		goto out;
5650	}
5651
5652	/*
5653	 * ocfs2_add_new_xattr_bucket() will have updated
5654	 * xs->bucket if it moved, but it will not have updated
5655	 * any of the other search fields.  Thus, we drop it and
5656	 * re-search.  Everything should be cached, so it'll be
5657	 * quick.
5658	 */
5659	ocfs2_xattr_bucket_relse(xs->bucket);
5660	ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5661					   xi->xi_name_index,
5662					   xi->xi_name, xs);
5663	if (ret && ret != -ENODATA)
5664		goto out;
5665	xs->not_found = ret;
5666
5667	/* Ok, we have a new bucket, let's try again */
5668	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5669	if (ret && (ret != -ENOSPC))
5670		mlog_errno(ret);
5671
5672out:
5673	return ret;
5674}
5675
5676static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5677					struct ocfs2_xattr_bucket *bucket,
5678					void *para)
5679{
5680	int ret = 0, ref_credits;
5681	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5682	u16 i;
5683	struct ocfs2_xattr_entry *xe;
5684	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5685	struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5686	int credits = ocfs2_remove_extent_credits(osb->sb) +
5687		ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5688	struct ocfs2_xattr_value_root *xv;
5689	struct ocfs2_rm_xattr_bucket_para *args =
5690			(struct ocfs2_rm_xattr_bucket_para *)para;
5691
5692	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5693
5694	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5695		xe = &xh->xh_entries[i];
5696		if (ocfs2_xattr_is_local(xe))
5697			continue;
5698
5699		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5700						      i, &xv, NULL);
5701		if (ret) {
5702			mlog_errno(ret);
5703			break;
5704		}
5705
5706		ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5707							 args->ref_ci,
5708							 args->ref_root_bh,
5709							 &ctxt.meta_ac,
5710							 &ref_credits);
5711
5712		ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5713		if (IS_ERR(ctxt.handle)) {
5714			ret = PTR_ERR(ctxt.handle);
5715			mlog_errno(ret);
5716			break;
5717		}
5718
5719		ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5720							i, 0, &ctxt);
5721
5722		ocfs2_commit_trans(osb, ctxt.handle);
5723		if (ctxt.meta_ac) {
5724			ocfs2_free_alloc_context(ctxt.meta_ac);
5725			ctxt.meta_ac = NULL;
5726		}
5727		if (ret) {
5728			mlog_errno(ret);
5729			break;
5730		}
5731	}
5732
5733	if (ctxt.meta_ac)
5734		ocfs2_free_alloc_context(ctxt.meta_ac);
5735	ocfs2_schedule_truncate_log_flush(osb, 1);
5736	ocfs2_run_deallocs(osb, &ctxt.dealloc);
5737	return ret;
5738}
5739
5740/*
5741 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5742 * or change the extent record flag), we need to recalculate
5743 * the metaecc for the whole bucket. So it is done here.
5744 *
5745 * Note:
5746 * We have to give the extra credits for the caller.
5747 */
5748static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5749					    handle_t *handle,
5750					    void *para)
5751{
5752	int ret;
5753	struct ocfs2_xattr_bucket *bucket =
5754			(struct ocfs2_xattr_bucket *)para;
5755
5756	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5757						OCFS2_JOURNAL_ACCESS_WRITE);
5758	if (ret) {
5759		mlog_errno(ret);
5760		return ret;
5761	}
5762
5763	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5764
5765	return 0;
5766}
5767
5768/*
5769 * Special action we need if the xattr value is refcounted.
5770 *
5771 * 1. If the xattr is refcounted, lock the tree.
5772 * 2. CoW the xattr if we are setting the new value and the value
5773 *    will be stored outside.
5774 * 3. In other case, decrease_refcount will work for us, so just
5775 *    lock the refcount tree, calculate the meta and credits is OK.
5776 *
5777 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5778 * currently CoW is a completed transaction, while this function
5779 * will also lock the allocators and let us deadlock. So we will
5780 * CoW the whole xattr value.
5781 */
5782static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5783					struct ocfs2_dinode *di,
5784					struct ocfs2_xattr_info *xi,
5785					struct ocfs2_xattr_search *xis,
5786					struct ocfs2_xattr_search *xbs,
5787					struct ocfs2_refcount_tree **ref_tree,
5788					int *meta_add,
5789					int *credits)
5790{
5791	int ret = 0;
5792	struct ocfs2_xattr_block *xb;
5793	struct ocfs2_xattr_entry *xe;
5794	char *base;
5795	u32 p_cluster, num_clusters;
5796	unsigned int ext_flags;
5797	int name_offset, name_len;
5798	struct ocfs2_xattr_value_buf vb;
5799	struct ocfs2_xattr_bucket *bucket = NULL;
5800	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5801	struct ocfs2_post_refcount refcount;
5802	struct ocfs2_post_refcount *p = NULL;
5803	struct buffer_head *ref_root_bh = NULL;
5804
5805	if (!xis->not_found) {
5806		xe = xis->here;
5807		name_offset = le16_to_cpu(xe->xe_name_offset);
5808		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5809		base = xis->base;
5810		vb.vb_bh = xis->inode_bh;
5811		vb.vb_access = ocfs2_journal_access_di;
5812	} else {
5813		int i, block_off = 0;
5814		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5815		xe = xbs->here;
5816		name_offset = le16_to_cpu(xe->xe_name_offset);
5817		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5818		i = xbs->here - xbs->header->xh_entries;
5819
5820		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5821			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5822							bucket_xh(xbs->bucket),
5823							i, &block_off,
5824							&name_offset);
5825			if (ret) {
5826				mlog_errno(ret);
5827				goto out;
5828			}
5829			base = bucket_block(xbs->bucket, block_off);
5830			vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5831			vb.vb_access = ocfs2_journal_access;
5832
5833			if (ocfs2_meta_ecc(osb)) {
5834				/*create parameters for ocfs2_post_refcount. */
5835				bucket = xbs->bucket;
5836				refcount.credits = bucket->bu_blocks;
5837				refcount.para = bucket;
5838				refcount.func =
5839					ocfs2_xattr_bucket_post_refcount;
5840				p = &refcount;
5841			}
5842		} else {
5843			base = xbs->base;
5844			vb.vb_bh = xbs->xattr_bh;
5845			vb.vb_access = ocfs2_journal_access_xb;
5846		}
5847	}
5848
5849	if (ocfs2_xattr_is_local(xe))
5850		goto out;
5851
5852	vb.vb_xv = (struct ocfs2_xattr_value_root *)
5853				(base + name_offset + name_len);
5854
5855	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5856				       &num_clusters, &vb.vb_xv->xr_list,
5857				       &ext_flags);
5858	if (ret) {
5859		mlog_errno(ret);
5860		goto out;
5861	}
5862
5863	/*
5864	 * We just need to check the 1st extent record, since we always
5865	 * CoW the whole xattr. So there shouldn't be a xattr with
5866	 * some REFCOUNT extent recs after the 1st one.
5867	 */
5868	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5869		goto out;
5870
5871	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5872				       1, ref_tree, &ref_root_bh);
5873	if (ret) {
5874		mlog_errno(ret);
5875		goto out;
5876	}
5877
5878	/*
5879	 * If we are deleting the xattr or the new size will be stored inside,
5880	 * cool, leave it there, the xattr truncate process will remove them
5881	 * for us(it still needs the refcount tree lock and the meta, credits).
5882	 * And the worse case is that every cluster truncate will split the
5883	 * refcount tree, and make the original extent become 3. So we will need
5884	 * 2 * cluster more extent recs at most.
5885	 */
5886	if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5887
5888		ret = ocfs2_refcounted_xattr_delete_need(inode,
5889							 &(*ref_tree)->rf_ci,
5890							 ref_root_bh, vb.vb_xv,
5891							 meta_add, credits);
5892		if (ret)
5893			mlog_errno(ret);
5894		goto out;
5895	}
5896
5897	ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5898				       *ref_tree, ref_root_bh, 0,
5899				       le32_to_cpu(vb.vb_xv->xr_clusters), p);
5900	if (ret)
5901		mlog_errno(ret);
5902
5903out:
5904	brelse(ref_root_bh);
5905	return ret;
5906}
5907
5908/*
5909 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5910 * The physical clusters will be added to refcount tree.
5911 */
5912static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5913				struct ocfs2_xattr_value_root *xv,
5914				struct ocfs2_extent_tree *value_et,
5915				struct ocfs2_caching_info *ref_ci,
5916				struct buffer_head *ref_root_bh,
5917				struct ocfs2_cached_dealloc_ctxt *dealloc,
5918				struct ocfs2_post_refcount *refcount)
5919{
5920	int ret = 0;
5921	u32 clusters = le32_to_cpu(xv->xr_clusters);
5922	u32 cpos, p_cluster, num_clusters;
5923	struct ocfs2_extent_list *el = &xv->xr_list;
5924	unsigned int ext_flags;
5925
5926	cpos = 0;
5927	while (cpos < clusters) {
5928		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5929					       &num_clusters, el, &ext_flags);
5930		if (ret) {
5931			mlog_errno(ret);
5932			break;
5933		}
5934
5935		cpos += num_clusters;
5936		if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5937			continue;
5938
5939		BUG_ON(!p_cluster);
5940
5941		ret = ocfs2_add_refcount_flag(inode, value_et,
5942					      ref_ci, ref_root_bh,
5943					      cpos - num_clusters,
5944					      p_cluster, num_clusters,
5945					      dealloc, refcount);
5946		if (ret) {
5947			mlog_errno(ret);
5948			break;
5949		}
5950	}
5951
5952	return ret;
5953}
5954
5955/*
5956 * Given a normal ocfs2_xattr_header, refcount all the entries which
5957 * have value stored outside.
5958 * Used for xattrs stored in inode and ocfs2_xattr_block.
5959 */
5960static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5961				struct ocfs2_xattr_value_buf *vb,
5962				struct ocfs2_xattr_header *header,
5963				struct ocfs2_caching_info *ref_ci,
5964				struct buffer_head *ref_root_bh,
5965				struct ocfs2_cached_dealloc_ctxt *dealloc)
5966{
5967
5968	struct ocfs2_xattr_entry *xe;
5969	struct ocfs2_xattr_value_root *xv;
5970	struct ocfs2_extent_tree et;
5971	int i, ret = 0;
5972
5973	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5974		xe = &header->xh_entries[i];
5975
5976		if (ocfs2_xattr_is_local(xe))
5977			continue;
5978
5979		xv = (struct ocfs2_xattr_value_root *)((void *)header +
5980			le16_to_cpu(xe->xe_name_offset) +
5981			OCFS2_XATTR_SIZE(xe->xe_name_len));
5982
5983		vb->vb_xv = xv;
5984		ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5985
5986		ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5987							ref_ci, ref_root_bh,
5988							dealloc, NULL);
5989		if (ret) {
5990			mlog_errno(ret);
5991			break;
5992		}
5993	}
5994
5995	return ret;
5996}
5997
5998static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5999				struct buffer_head *fe_bh,
6000				struct ocfs2_caching_info *ref_ci,
6001				struct buffer_head *ref_root_bh,
6002				struct ocfs2_cached_dealloc_ctxt *dealloc)
6003{
6004	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6005	struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
6006				(fe_bh->b_data + inode->i_sb->s_blocksize -
6007				le16_to_cpu(di->i_xattr_inline_size));
6008	struct ocfs2_xattr_value_buf vb = {
6009		.vb_bh = fe_bh,
6010		.vb_access = ocfs2_journal_access_di,
6011	};
6012
6013	return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6014						  ref_ci, ref_root_bh, dealloc);
6015}
6016
6017struct ocfs2_xattr_tree_value_refcount_para {
6018	struct ocfs2_caching_info *ref_ci;
6019	struct buffer_head *ref_root_bh;
6020	struct ocfs2_cached_dealloc_ctxt *dealloc;
6021};
6022
6023static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6024					   struct ocfs2_xattr_bucket *bucket,
6025					   int offset,
6026					   struct ocfs2_xattr_value_root **xv,
6027					   struct buffer_head **bh)
6028{
6029	int ret, block_off, name_offset;
6030	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6031	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6032	void *base;
6033
6034	ret = ocfs2_xattr_bucket_get_name_value(sb,
6035						bucket_xh(bucket),
6036						offset,
6037						&block_off,
6038						&name_offset);
6039	if (ret) {
6040		mlog_errno(ret);
6041		goto out;
6042	}
6043
6044	base = bucket_block(bucket, block_off);
6045
6046	*xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6047			 OCFS2_XATTR_SIZE(xe->xe_name_len));
6048
6049	if (bh)
6050		*bh = bucket->bu_bhs[block_off];
6051out:
6052	return ret;
6053}
6054
6055/*
6056 * For a given xattr bucket, refcount all the entries which
6057 * have value stored outside.
6058 */
6059static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6060					     struct ocfs2_xattr_bucket *bucket,
6061					     void *para)
6062{
6063	int i, ret = 0;
6064	struct ocfs2_extent_tree et;
6065	struct ocfs2_xattr_tree_value_refcount_para *ref =
6066			(struct ocfs2_xattr_tree_value_refcount_para *)para;
6067	struct ocfs2_xattr_header *xh =
6068			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6069	struct ocfs2_xattr_entry *xe;
6070	struct ocfs2_xattr_value_buf vb = {
6071		.vb_access = ocfs2_journal_access,
6072	};
6073	struct ocfs2_post_refcount refcount = {
6074		.credits = bucket->bu_blocks,
6075		.para = bucket,
6076		.func = ocfs2_xattr_bucket_post_refcount,
6077	};
6078	struct ocfs2_post_refcount *p = NULL;
6079
6080	/* We only need post_refcount if we support metaecc. */
6081	if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6082		p = &refcount;
6083
6084	trace_ocfs2_xattr_bucket_value_refcount(
6085				(unsigned long long)bucket_blkno(bucket),
6086				le16_to_cpu(xh->xh_count));
6087	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6088		xe = &xh->xh_entries[i];
6089
6090		if (ocfs2_xattr_is_local(xe))
6091			continue;
6092
6093		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6094						      &vb.vb_xv, &vb.vb_bh);
6095		if (ret) {
6096			mlog_errno(ret);
6097			break;
6098		}
6099
6100		ocfs2_init_xattr_value_extent_tree(&et,
6101						   INODE_CACHE(inode), &vb);
6102
6103		ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6104							&et, ref->ref_ci,
6105							ref->ref_root_bh,
6106							ref->dealloc, p);
6107		if (ret) {
6108			mlog_errno(ret);
6109			break;
6110		}
6111	}
6112
6113	return ret;
6114
6115}
6116
6117static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6118				     struct buffer_head *root_bh,
6119				     u64 blkno, u32 cpos, u32 len, void *para)
6120{
6121	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6122					   ocfs2_xattr_bucket_value_refcount,
6123					   para);
6124}
6125
6126static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6127				struct buffer_head *blk_bh,
6128				struct ocfs2_caching_info *ref_ci,
6129				struct buffer_head *ref_root_bh,
6130				struct ocfs2_cached_dealloc_ctxt *dealloc)
6131{
6132	int ret = 0;
6133	struct ocfs2_xattr_block *xb =
6134				(struct ocfs2_xattr_block *)blk_bh->b_data;
6135
6136	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6137		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6138		struct ocfs2_xattr_value_buf vb = {
6139			.vb_bh = blk_bh,
6140			.vb_access = ocfs2_journal_access_xb,
6141		};
6142
6143		ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6144							 ref_ci, ref_root_bh,
6145							 dealloc);
6146	} else {
6147		struct ocfs2_xattr_tree_value_refcount_para para = {
6148			.ref_ci = ref_ci,
6149			.ref_root_bh = ref_root_bh,
6150			.dealloc = dealloc,
6151		};
6152
6153		ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6154						ocfs2_refcount_xattr_tree_rec,
6155						&para);
6156	}
6157
6158	return ret;
6159}
6160
6161int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6162				     struct buffer_head *fe_bh,
6163				     struct ocfs2_caching_info *ref_ci,
6164				     struct buffer_head *ref_root_bh,
6165				     struct ocfs2_cached_dealloc_ctxt *dealloc)
6166{
6167	int ret = 0;
6168	struct ocfs2_inode_info *oi = OCFS2_I(inode);
6169	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6170	struct buffer_head *blk_bh = NULL;
6171
6172	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6173		ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6174							 ref_ci, ref_root_bh,
6175							 dealloc);
6176		if (ret) {
6177			mlog_errno(ret);
6178			goto out;
6179		}
6180	}
6181
6182	if (!di->i_xattr_loc)
6183		goto out;
6184
6185	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6186				     &blk_bh);
6187	if (ret < 0) {
6188		mlog_errno(ret);
6189		goto out;
6190	}
6191
6192	ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6193						ref_root_bh, dealloc);
6194	if (ret)
6195		mlog_errno(ret);
6196
6197	brelse(blk_bh);
6198out:
6199
6200	return ret;
6201}
6202
6203typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6204/*
6205 * Store the information we need in xattr reflink.
6206 * old_bh and new_bh are inode bh for the old and new inode.
6207 */
6208struct ocfs2_xattr_reflink {
6209	struct inode *old_inode;
6210	struct inode *new_inode;
6211	struct buffer_head *old_bh;
6212	struct buffer_head *new_bh;
6213	struct ocfs2_caching_info *ref_ci;
6214	struct buffer_head *ref_root_bh;
6215	struct ocfs2_cached_dealloc_ctxt *dealloc;
6216	should_xattr_reflinked *xattr_reflinked;
6217};
6218
6219/*
6220 * Given a xattr header and xe offset,
6221 * return the proper xv and the corresponding bh.
6222 * xattr in inode, block and xattr tree have different implementations.
6223 */
6224typedef int (get_xattr_value_root)(struct super_block *sb,
6225				   struct buffer_head *bh,
6226				   struct ocfs2_xattr_header *xh,
6227				   int offset,
6228				   struct ocfs2_xattr_value_root **xv,
6229				   struct buffer_head **ret_bh,
6230				   void *para);
6231
6232/*
6233 * Calculate all the xattr value root metadata stored in this xattr header and
6234 * credits we need if we create them from the scratch.
6235 * We use get_xattr_value_root so that all types of xattr container can use it.
6236 */
6237static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6238					     struct buffer_head *bh,
6239					     struct ocfs2_xattr_header *xh,
6240					     int *metas, int *credits,
6241					     int *num_recs,
6242					     get_xattr_value_root *func,
6243					     void *para)
6244{
6245	int i, ret = 0;
6246	struct ocfs2_xattr_value_root *xv;
6247	struct ocfs2_xattr_entry *xe;
6248
6249	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6250		xe = &xh->xh_entries[i];
6251		if (ocfs2_xattr_is_local(xe))
6252			continue;
6253
6254		ret = func(sb, bh, xh, i, &xv, NULL, para);
6255		if (ret) {
6256			mlog_errno(ret);
6257			break;
6258		}
6259
6260		*metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6261			  le16_to_cpu(xv->xr_list.l_next_free_rec);
6262
6263		*credits += ocfs2_calc_extend_credits(sb,
6264						&def_xv.xv.xr_list);
6265
6266		/*
6267		 * If the value is a tree with depth > 1, We don't go deep
6268		 * to the extent block, so just calculate a maximum record num.
6269		 */
6270		if (!xv->xr_list.l_tree_depth)
6271			*num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6272		else
6273			*num_recs += ocfs2_clusters_for_bytes(sb,
6274							      XATTR_SIZE_MAX);
6275	}
6276
6277	return ret;
6278}
6279
6280/* Used by xattr inode and block to return the right xv and buffer_head. */
6281static int ocfs2_get_xattr_value_root(struct super_block *sb,
6282				      struct buffer_head *bh,
6283				      struct ocfs2_xattr_header *xh,
6284				      int offset,
6285				      struct ocfs2_xattr_value_root **xv,
6286				      struct buffer_head **ret_bh,
6287				      void *para)
6288{
6289	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6290
6291	*xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6292		le16_to_cpu(xe->xe_name_offset) +
6293		OCFS2_XATTR_SIZE(xe->xe_name_len));
6294
6295	if (ret_bh)
6296		*ret_bh = bh;
6297
6298	return 0;
6299}
6300
6301/*
6302 * Lock the meta_ac and calculate how much credits we need for reflink xattrs.
6303 * It is only used for inline xattr and xattr block.
6304 */
6305static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6306					struct ocfs2_xattr_header *xh,
6307					struct buffer_head *ref_root_bh,
6308					int *credits,
6309					struct ocfs2_alloc_context **meta_ac)
6310{
6311	int ret, meta_add = 0, num_recs = 0;
6312	struct ocfs2_refcount_block *rb =
6313			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
6314
6315	*credits = 0;
6316
6317	ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6318						&meta_add, credits, &num_recs,
6319						ocfs2_get_xattr_value_root,
6320						NULL);
6321	if (ret) {
6322		mlog_errno(ret);
6323		goto out;
6324	}
6325
6326	/*
6327	 * We need to add/modify num_recs in refcount tree, so just calculate
6328	 * an approximate number we need for refcount tree change.
6329	 * Sometimes we need to split the tree, and after split,  half recs
6330	 * will be moved to the new block, and a new block can only provide
6331	 * half number of recs. So we multiple new blocks by 2.
6332	 */
6333	num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6334	meta_add += num_recs;
6335	*credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6336	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6337		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6338			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6339	else
6340		*credits += 1;
6341
6342	ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6343	if (ret)
6344		mlog_errno(ret);
6345
6346out:
6347	return ret;
6348}
6349
6350/*
6351 * Given a xattr header, reflink all the xattrs in this container.
6352 * It can be used for inode, block and bucket.
6353 *
6354 * NOTE:
6355 * Before we call this function, the caller has memcpy the xattr in
6356 * old_xh to the new_xh.
6357 *
6358 * If args.xattr_reflinked is set, call it to decide whether the xe should
6359 * be reflinked or not. If not, remove it from the new xattr header.
6360 */
6361static int ocfs2_reflink_xattr_header(handle_t *handle,
6362				      struct ocfs2_xattr_reflink *args,
6363				      struct buffer_head *old_bh,
6364				      struct ocfs2_xattr_header *xh,
6365				      struct buffer_head *new_bh,
6366				      struct ocfs2_xattr_header *new_xh,
6367				      struct ocfs2_xattr_value_buf *vb,
6368				      struct ocfs2_alloc_context *meta_ac,
6369				      get_xattr_value_root *func,
6370				      void *para)
6371{
6372	int ret = 0, i, j;
6373	struct super_block *sb = args->old_inode->i_sb;
6374	struct buffer_head *value_bh;
6375	struct ocfs2_xattr_entry *xe, *last;
6376	struct ocfs2_xattr_value_root *xv, *new_xv;
6377	struct ocfs2_extent_tree data_et;
6378	u32 clusters, cpos, p_cluster, num_clusters;
6379	unsigned int ext_flags = 0;
6380
6381	trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6382					 le16_to_cpu(xh->xh_count));
6383
6384	last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)] - 1;
6385	for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6386		xe = &xh->xh_entries[i];
6387
6388		if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6389			xe = &new_xh->xh_entries[j];
6390
6391			le16_add_cpu(&new_xh->xh_count, -1);
6392			if (new_xh->xh_count) {
6393				memmove(xe, xe + 1,
6394					(void *)last - (void *)xe);
6395				memset(last, 0,
6396				       sizeof(struct ocfs2_xattr_entry));
6397				last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)] - 1;
6398			} else {
6399				memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
6400				last = NULL;
6401			}
6402
6403			/*
6404			 * We don't want j to increase in the next round since
6405			 * it is already moved ahead.
6406			 */
6407			j--;
6408			continue;
6409		}
6410
6411		if (ocfs2_xattr_is_local(xe))
6412			continue;
6413
6414		ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6415		if (ret) {
6416			mlog_errno(ret);
6417			break;
6418		}
6419
6420		ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6421		if (ret) {
6422			mlog_errno(ret);
6423			break;
6424		}
6425
6426		/*
6427		 * For the xattr which has l_tree_depth = 0, all the extent
6428		 * recs have already be copied to the new xh with the
6429		 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6430		 * increase the refount count int the refcount tree.
6431		 *
6432		 * For the xattr which has l_tree_depth > 0, we need
6433		 * to initialize it to the empty default value root,
6434		 * and then insert the extents one by one.
6435		 */
6436		if (xv->xr_list.l_tree_depth) {
6437			memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6438			vb->vb_xv = new_xv;
6439			vb->vb_bh = value_bh;
6440			ocfs2_init_xattr_value_extent_tree(&data_et,
6441					INODE_CACHE(args->new_inode), vb);
6442		}
6443
6444		clusters = le32_to_cpu(xv->xr_clusters);
6445		cpos = 0;
6446		while (cpos < clusters) {
6447			ret = ocfs2_xattr_get_clusters(args->old_inode,
6448						       cpos,
6449						       &p_cluster,
6450						       &num_clusters,
6451						       &xv->xr_list,
6452						       &ext_flags);
6453			if (ret) {
6454				mlog_errno(ret);
6455				goto out;
6456			}
6457
6458			BUG_ON(!p_cluster);
6459
6460			if (xv->xr_list.l_tree_depth) {
6461				ret = ocfs2_insert_extent(handle,
6462						&data_et, cpos,
6463						ocfs2_clusters_to_blocks(
6464							args->old_inode->i_sb,
6465							p_cluster),
6466						num_clusters, ext_flags,
6467						meta_ac);
6468				if (ret) {
6469					mlog_errno(ret);
6470					goto out;
6471				}
6472			}
6473
6474			ret = ocfs2_increase_refcount(handle, args->ref_ci,
6475						      args->ref_root_bh,
6476						      p_cluster, num_clusters,
6477						      meta_ac, args->dealloc);
6478			if (ret) {
6479				mlog_errno(ret);
6480				goto out;
6481			}
6482
6483			cpos += num_clusters;
6484		}
6485	}
6486
6487out:
6488	return ret;
6489}
6490
6491static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6492{
6493	int ret = 0, credits = 0;
6494	handle_t *handle;
6495	struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6496	struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6497	int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6498	int header_off = osb->sb->s_blocksize - inline_size;
6499	struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6500					(args->old_bh->b_data + header_off);
6501	struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6502					(args->new_bh->b_data + header_off);
6503	struct ocfs2_alloc_context *meta_ac = NULL;
6504	struct ocfs2_inode_info *new_oi;
6505	struct ocfs2_dinode *new_di;
6506	struct ocfs2_xattr_value_buf vb = {
6507		.vb_bh = args->new_bh,
6508		.vb_access = ocfs2_journal_access_di,
6509	};
6510
6511	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6512						  &credits, &meta_ac);
6513	if (ret) {
6514		mlog_errno(ret);
6515		goto out;
6516	}
6517
6518	handle = ocfs2_start_trans(osb, credits);
6519	if (IS_ERR(handle)) {
6520		ret = PTR_ERR(handle);
6521		mlog_errno(ret);
6522		goto out;
6523	}
6524
6525	ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6526				      args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6527	if (ret) {
6528		mlog_errno(ret);
6529		goto out_commit;
6530	}
6531
6532	memcpy(args->new_bh->b_data + header_off,
6533	       args->old_bh->b_data + header_off, inline_size);
6534
6535	new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6536	new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6537
6538	ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6539					 args->new_bh, new_xh, &vb, meta_ac,
6540					 ocfs2_get_xattr_value_root, NULL);
6541	if (ret) {
6542		mlog_errno(ret);
6543		goto out_commit;
6544	}
6545
6546	new_oi = OCFS2_I(args->new_inode);
6547
6548	spin_lock(&new_oi->ip_lock);
6549	new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6550	new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6551	spin_unlock(&new_oi->ip_lock);
6552
6553	ocfs2_journal_dirty(handle, args->new_bh);
6554
6555out_commit:
6556	ocfs2_commit_trans(osb, handle);
6557
6558out:
6559	if (meta_ac)
6560		ocfs2_free_alloc_context(meta_ac);
6561	return ret;
6562}
6563
6564static int ocfs2_create_empty_xattr_block(struct inode *inode,
6565					  struct buffer_head *fe_bh,
6566					  struct buffer_head **ret_bh,
6567					  int indexed)
6568{
6569	int ret;
6570	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6571	struct ocfs2_xattr_set_ctxt ctxt;
6572
6573	memset(&ctxt, 0, sizeof(ctxt));
6574	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6575	if (ret < 0) {
6576		mlog_errno(ret);
6577		return ret;
6578	}
6579
6580	ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6581	if (IS_ERR(ctxt.handle)) {
6582		ret = PTR_ERR(ctxt.handle);
6583		mlog_errno(ret);
6584		goto out;
6585	}
6586
6587	trace_ocfs2_create_empty_xattr_block(
6588				(unsigned long long)fe_bh->b_blocknr, indexed);
6589	ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6590				       ret_bh);
6591	if (ret)
6592		mlog_errno(ret);
6593
6594	ocfs2_commit_trans(osb, ctxt.handle);
6595out:
6596	ocfs2_free_alloc_context(ctxt.meta_ac);
6597	return ret;
6598}
6599
6600static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6601				     struct buffer_head *blk_bh,
6602				     struct buffer_head *new_blk_bh)
6603{
6604	int ret = 0, credits = 0;
6605	handle_t *handle;
6606	struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6607	struct ocfs2_dinode *new_di;
6608	struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6609	int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6610	struct ocfs2_xattr_block *xb =
6611			(struct ocfs2_xattr_block *)blk_bh->b_data;
6612	struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6613	struct ocfs2_xattr_block *new_xb =
6614			(struct ocfs2_xattr_block *)new_blk_bh->b_data;
6615	struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6616	struct ocfs2_alloc_context *meta_ac;
6617	struct ocfs2_xattr_value_buf vb = {
6618		.vb_bh = new_blk_bh,
6619		.vb_access = ocfs2_journal_access_xb,
6620	};
6621
6622	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6623						  &credits, &meta_ac);
6624	if (ret) {
6625		mlog_errno(ret);
6626		return ret;
6627	}
6628
6629	/* One more credits in case we need to add xattr flags in new inode. */
6630	handle = ocfs2_start_trans(osb, credits + 1);
6631	if (IS_ERR(handle)) {
6632		ret = PTR_ERR(handle);
6633		mlog_errno(ret);
6634		goto out;
6635	}
6636
6637	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6638		ret = ocfs2_journal_access_di(handle,
6639					      INODE_CACHE(args->new_inode),
6640					      args->new_bh,
6641					      OCFS2_JOURNAL_ACCESS_WRITE);
6642		if (ret) {
6643			mlog_errno(ret);
6644			goto out_commit;
6645		}
6646	}
6647
6648	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6649				      new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6650	if (ret) {
6651		mlog_errno(ret);
6652		goto out_commit;
6653	}
6654
6655	memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6656	       osb->sb->s_blocksize - header_off);
6657
6658	ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6659					 new_blk_bh, new_xh, &vb, meta_ac,
6660					 ocfs2_get_xattr_value_root, NULL);
6661	if (ret) {
6662		mlog_errno(ret);
6663		goto out_commit;
6664	}
6665
6666	ocfs2_journal_dirty(handle, new_blk_bh);
6667
6668	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6669		new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6670		spin_lock(&new_oi->ip_lock);
6671		new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6672		new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6673		spin_unlock(&new_oi->ip_lock);
6674
6675		ocfs2_journal_dirty(handle, args->new_bh);
6676	}
6677
6678out_commit:
6679	ocfs2_commit_trans(osb, handle);
6680
6681out:
6682	ocfs2_free_alloc_context(meta_ac);
6683	return ret;
6684}
6685
6686struct ocfs2_reflink_xattr_tree_args {
6687	struct ocfs2_xattr_reflink *reflink;
6688	struct buffer_head *old_blk_bh;
6689	struct buffer_head *new_blk_bh;
6690	struct ocfs2_xattr_bucket *old_bucket;
6691	struct ocfs2_xattr_bucket *new_bucket;
6692};
6693
6694/*
6695 * NOTE:
6696 * We have to handle the case that both old bucket and new bucket
6697 * will call this function to get the right ret_bh.
6698 * So The caller must give us the right bh.
6699 */
6700static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6701					struct buffer_head *bh,
6702					struct ocfs2_xattr_header *xh,
6703					int offset,
6704					struct ocfs2_xattr_value_root **xv,
6705					struct buffer_head **ret_bh,
6706					void *para)
6707{
6708	struct ocfs2_reflink_xattr_tree_args *args =
6709			(struct ocfs2_reflink_xattr_tree_args *)para;
6710	struct ocfs2_xattr_bucket *bucket;
6711
6712	if (bh == args->old_bucket->bu_bhs[0])
6713		bucket = args->old_bucket;
6714	else
6715		bucket = args->new_bucket;
6716
6717	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6718					       xv, ret_bh);
6719}
6720
6721struct ocfs2_value_tree_metas {
6722	int num_metas;
6723	int credits;
6724	int num_recs;
6725};
6726
6727static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6728					struct buffer_head *bh,
6729					struct ocfs2_xattr_header *xh,
6730					int offset,
6731					struct ocfs2_xattr_value_root **xv,
6732					struct buffer_head **ret_bh,
6733					void *para)
6734{
6735	struct ocfs2_xattr_bucket *bucket =
6736				(struct ocfs2_xattr_bucket *)para;
6737
6738	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6739					       xv, ret_bh);
6740}
6741
6742static int ocfs2_calc_value_tree_metas(struct inode *inode,
6743				      struct ocfs2_xattr_bucket *bucket,
6744				      void *para)
6745{
6746	struct ocfs2_value_tree_metas *metas =
6747			(struct ocfs2_value_tree_metas *)para;
6748	struct ocfs2_xattr_header *xh =
6749			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6750
6751	/* Add the credits for this bucket first. */
6752	metas->credits += bucket->bu_blocks;
6753	return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6754					xh, &metas->num_metas,
6755					&metas->credits, &metas->num_recs,
6756					ocfs2_value_tree_metas_in_bucket,
6757					bucket);
6758}
6759
6760/*
6761 * Given a xattr extent rec starting from blkno and having len clusters,
6762 * iterate all the buckets calculate how much metadata we need for reflinking
6763 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6764 */
6765static int ocfs2_lock_reflink_xattr_rec_allocators(
6766				struct ocfs2_reflink_xattr_tree_args *args,
6767				struct ocfs2_extent_tree *xt_et,
6768				u64 blkno, u32 len, int *credits,
6769				struct ocfs2_alloc_context **meta_ac,
6770				struct ocfs2_alloc_context **data_ac)
6771{
6772	int ret, num_free_extents;
6773	struct ocfs2_value_tree_metas metas;
6774	struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6775	struct ocfs2_refcount_block *rb;
6776
6777	memset(&metas, 0, sizeof(metas));
6778
6779	ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6780					  ocfs2_calc_value_tree_metas, &metas);
6781	if (ret) {
6782		mlog_errno(ret);
6783		goto out;
6784	}
6785
6786	*credits = metas.credits;
6787
6788	/*
6789	 * Calculate we need for refcount tree change.
6790	 *
6791	 * We need to add/modify num_recs in refcount tree, so just calculate
6792	 * an approximate number we need for refcount tree change.
6793	 * Sometimes we need to split the tree, and after split,  half recs
6794	 * will be moved to the new block, and a new block can only provide
6795	 * half number of recs. So we multiple new blocks by 2.
6796	 * In the end, we have to add credits for modifying the already
6797	 * existed refcount block.
6798	 */
6799	rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6800	metas.num_recs =
6801		(metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6802		 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6803	metas.num_metas += metas.num_recs;
6804	*credits += metas.num_recs +
6805		    metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6806	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6807		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6808			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6809	else
6810		*credits += 1;
6811
6812	/* count in the xattr tree change. */
6813	num_free_extents = ocfs2_num_free_extents(xt_et);
6814	if (num_free_extents < 0) {
6815		ret = num_free_extents;
6816		mlog_errno(ret);
6817		goto out;
6818	}
6819
6820	if (num_free_extents < len)
6821		metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6822
6823	*credits += ocfs2_calc_extend_credits(osb->sb,
6824					      xt_et->et_root_el);
6825
6826	if (metas.num_metas) {
6827		ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6828							meta_ac);
6829		if (ret) {
6830			mlog_errno(ret);
6831			goto out;
6832		}
6833	}
6834
6835	if (len) {
6836		ret = ocfs2_reserve_clusters(osb, len, data_ac);
6837		if (ret)
6838			mlog_errno(ret);
6839	}
6840out:
6841	if (ret) {
6842		if (*meta_ac) {
6843			ocfs2_free_alloc_context(*meta_ac);
6844			*meta_ac = NULL;
6845		}
6846	}
6847
6848	return ret;
6849}
6850
6851static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6852				u64 blkno, u64 new_blkno, u32 clusters,
6853				u32 *cpos, int num_buckets,
6854				struct ocfs2_alloc_context *meta_ac,
6855				struct ocfs2_alloc_context *data_ac,
6856				struct ocfs2_reflink_xattr_tree_args *args)
6857{
6858	int i, j, ret = 0;
6859	struct super_block *sb = args->reflink->old_inode->i_sb;
6860	int bpb = args->old_bucket->bu_blocks;
6861	struct ocfs2_xattr_value_buf vb = {
6862		.vb_access = ocfs2_journal_access,
6863	};
6864
6865	for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6866		ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6867		if (ret) {
6868			mlog_errno(ret);
6869			break;
6870		}
6871
6872		ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6873		if (ret) {
6874			mlog_errno(ret);
6875			break;
6876		}
6877
6878		ret = ocfs2_xattr_bucket_journal_access(handle,
6879						args->new_bucket,
6880						OCFS2_JOURNAL_ACCESS_CREATE);
6881		if (ret) {
6882			mlog_errno(ret);
6883			break;
6884		}
6885
6886		for (j = 0; j < bpb; j++)
6887			memcpy(bucket_block(args->new_bucket, j),
6888			       bucket_block(args->old_bucket, j),
6889			       sb->s_blocksize);
6890
6891		/*
6892		 * Record the start cpos so that we can use it to initialize
6893		 * our xattr tree we also set the xh_num_bucket for the new
6894		 * bucket.
6895		 */
6896		if (i == 0) {
6897			*cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6898					    xh_entries[0].xe_name_hash);
6899			bucket_xh(args->new_bucket)->xh_num_buckets =
6900				cpu_to_le16(num_buckets);
6901		}
6902
6903		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6904
6905		ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6906					args->old_bucket->bu_bhs[0],
6907					bucket_xh(args->old_bucket),
6908					args->new_bucket->bu_bhs[0],
6909					bucket_xh(args->new_bucket),
6910					&vb, meta_ac,
6911					ocfs2_get_reflink_xattr_value_root,
6912					args);
6913		if (ret) {
6914			mlog_errno(ret);
6915			break;
6916		}
6917
6918		/*
6919		 * Re-access and dirty the bucket to calculate metaecc.
6920		 * Because we may extend the transaction in reflink_xattr_header
6921		 * which will let the already accessed block gone.
6922		 */
6923		ret = ocfs2_xattr_bucket_journal_access(handle,
6924						args->new_bucket,
6925						OCFS2_JOURNAL_ACCESS_WRITE);
6926		if (ret) {
6927			mlog_errno(ret);
6928			break;
6929		}
6930
6931		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6932
6933		ocfs2_xattr_bucket_relse(args->old_bucket);
6934		ocfs2_xattr_bucket_relse(args->new_bucket);
6935	}
6936
6937	ocfs2_xattr_bucket_relse(args->old_bucket);
6938	ocfs2_xattr_bucket_relse(args->new_bucket);
6939	return ret;
6940}
6941
6942static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6943				struct inode *inode,
6944				struct ocfs2_reflink_xattr_tree_args *args,
6945				struct ocfs2_extent_tree *et,
6946				struct ocfs2_alloc_context *meta_ac,
6947				struct ocfs2_alloc_context *data_ac,
6948				u64 blkno, u32 cpos, u32 len)
6949{
6950	int ret, first_inserted = 0;
6951	u32 p_cluster, num_clusters, reflink_cpos = 0;
6952	u64 new_blkno;
6953	unsigned int num_buckets, reflink_buckets;
6954	unsigned int bpc =
6955		ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6956
6957	ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6958	if (ret) {
6959		mlog_errno(ret);
6960		goto out;
6961	}
6962	num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6963	ocfs2_xattr_bucket_relse(args->old_bucket);
6964
6965	while (len && num_buckets) {
6966		ret = ocfs2_claim_clusters(handle, data_ac,
6967					   1, &p_cluster, &num_clusters);
6968		if (ret) {
6969			mlog_errno(ret);
6970			goto out;
6971		}
6972
6973		new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6974		reflink_buckets = min(num_buckets, bpc * num_clusters);
6975
6976		ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6977						 new_blkno, num_clusters,
6978						 &reflink_cpos, reflink_buckets,
6979						 meta_ac, data_ac, args);
6980		if (ret) {
6981			mlog_errno(ret);
6982			goto out;
6983		}
6984
6985		/*
6986		 * For the 1st allocated cluster, we make it use the same cpos
6987		 * so that the xattr tree looks the same as the original one
6988		 * in the most case.
6989		 */
6990		if (!first_inserted) {
6991			reflink_cpos = cpos;
6992			first_inserted = 1;
6993		}
6994		ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6995					  num_clusters, 0, meta_ac);
6996		if (ret)
6997			mlog_errno(ret);
6998
6999		trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
7000						  num_clusters, reflink_cpos);
7001
7002		len -= num_clusters;
7003		blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
7004		num_buckets -= reflink_buckets;
7005	}
7006out:
7007	return ret;
7008}
7009
7010/*
7011 * Create the same xattr extent record in the new inode's xattr tree.
7012 */
7013static int ocfs2_reflink_xattr_rec(struct inode *inode,
7014				   struct buffer_head *root_bh,
7015				   u64 blkno,
7016				   u32 cpos,
7017				   u32 len,
7018				   void *para)
7019{
7020	int ret, credits = 0;
7021	handle_t *handle;
7022	struct ocfs2_reflink_xattr_tree_args *args =
7023			(struct ocfs2_reflink_xattr_tree_args *)para;
7024	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7025	struct ocfs2_alloc_context *meta_ac = NULL;
7026	struct ocfs2_alloc_context *data_ac = NULL;
7027	struct ocfs2_extent_tree et;
7028
7029	trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7030
7031	ocfs2_init_xattr_tree_extent_tree(&et,
7032					  INODE_CACHE(args->reflink->new_inode),
7033					  args->new_blk_bh);
7034
7035	ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7036						      len, &credits,
7037						      &meta_ac, &data_ac);
7038	if (ret) {
7039		mlog_errno(ret);
7040		goto out;
7041	}
7042
7043	handle = ocfs2_start_trans(osb, credits);
7044	if (IS_ERR(handle)) {
7045		ret = PTR_ERR(handle);
7046		mlog_errno(ret);
7047		goto out;
7048	}
7049
7050	ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7051					  meta_ac, data_ac,
7052					  blkno, cpos, len);
7053	if (ret)
7054		mlog_errno(ret);
7055
7056	ocfs2_commit_trans(osb, handle);
7057
7058out:
7059	if (meta_ac)
7060		ocfs2_free_alloc_context(meta_ac);
7061	if (data_ac)
7062		ocfs2_free_alloc_context(data_ac);
7063	return ret;
7064}
7065
7066/*
7067 * Create reflinked xattr buckets.
7068 * We will add bucket one by one, and refcount all the xattrs in the bucket
7069 * if they are stored outside.
7070 */
7071static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7072				    struct buffer_head *blk_bh,
7073				    struct buffer_head *new_blk_bh)
7074{
7075	int ret;
7076	struct ocfs2_reflink_xattr_tree_args para;
7077
7078	memset(&para, 0, sizeof(para));
7079	para.reflink = args;
7080	para.old_blk_bh = blk_bh;
7081	para.new_blk_bh = new_blk_bh;
7082
7083	para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7084	if (!para.old_bucket) {
7085		mlog_errno(-ENOMEM);
7086		return -ENOMEM;
7087	}
7088
7089	para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7090	if (!para.new_bucket) {
7091		ret = -ENOMEM;
7092		mlog_errno(ret);
7093		goto out;
7094	}
7095
7096	ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7097					      ocfs2_reflink_xattr_rec,
7098					      &para);
7099	if (ret)
7100		mlog_errno(ret);
7101
7102out:
7103	ocfs2_xattr_bucket_free(para.old_bucket);
7104	ocfs2_xattr_bucket_free(para.new_bucket);
7105	return ret;
7106}
7107
7108static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7109					struct buffer_head *blk_bh)
7110{
7111	int ret, indexed = 0;
7112	struct buffer_head *new_blk_bh = NULL;
7113	struct ocfs2_xattr_block *xb =
7114			(struct ocfs2_xattr_block *)blk_bh->b_data;
7115
7116
7117	if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7118		indexed = 1;
7119
7120	ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7121					     &new_blk_bh, indexed);
7122	if (ret) {
7123		mlog_errno(ret);
7124		goto out;
7125	}
7126
7127	if (!indexed)
7128		ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7129	else
7130		ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7131	if (ret)
7132		mlog_errno(ret);
7133
7134out:
7135	brelse(new_blk_bh);
7136	return ret;
7137}
7138
7139static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7140{
7141	int type = ocfs2_xattr_get_type(xe);
7142
7143	return type != OCFS2_XATTR_INDEX_SECURITY &&
7144	       type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7145	       type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7146}
7147
7148int ocfs2_reflink_xattrs(struct inode *old_inode,
7149			 struct buffer_head *old_bh,
7150			 struct inode *new_inode,
7151			 struct buffer_head *new_bh,
7152			 bool preserve_security)
7153{
7154	int ret;
7155	struct ocfs2_xattr_reflink args;
7156	struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7157	struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7158	struct buffer_head *blk_bh = NULL;
7159	struct ocfs2_cached_dealloc_ctxt dealloc;
7160	struct ocfs2_refcount_tree *ref_tree;
7161	struct buffer_head *ref_root_bh = NULL;
7162
7163	ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7164				       le64_to_cpu(di->i_refcount_loc),
7165				       1, &ref_tree, &ref_root_bh);
7166	if (ret) {
7167		mlog_errno(ret);
7168		goto out;
7169	}
7170
7171	ocfs2_init_dealloc_ctxt(&dealloc);
7172
7173	args.old_inode = old_inode;
7174	args.new_inode = new_inode;
7175	args.old_bh = old_bh;
7176	args.new_bh = new_bh;
7177	args.ref_ci = &ref_tree->rf_ci;
7178	args.ref_root_bh = ref_root_bh;
7179	args.dealloc = &dealloc;
7180	if (preserve_security)
7181		args.xattr_reflinked = NULL;
7182	else
7183		args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7184
7185	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7186		ret = ocfs2_reflink_xattr_inline(&args);
7187		if (ret) {
7188			mlog_errno(ret);
7189			goto out_unlock;
7190		}
7191	}
7192
7193	if (!di->i_xattr_loc)
7194		goto out_unlock;
7195
7196	ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7197				     &blk_bh);
7198	if (ret < 0) {
7199		mlog_errno(ret);
7200		goto out_unlock;
7201	}
7202
7203	ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7204	if (ret)
7205		mlog_errno(ret);
7206
7207	brelse(blk_bh);
7208
7209out_unlock:
7210	ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7211				   ref_tree, 1);
7212	brelse(ref_root_bh);
7213
7214	if (ocfs2_dealloc_has_cluster(&dealloc)) {
7215		ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7216		ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7217	}
7218
7219out:
7220	return ret;
7221}
7222
7223/*
7224 * Initialize security and acl for a already created inode.
7225 * Used for reflink a non-preserve-security file.
7226 *
7227 * It uses common api like ocfs2_xattr_set, so the caller
7228 * must not hold any lock expect i_rwsem.
7229 */
7230int ocfs2_init_security_and_acl(struct inode *dir,
7231				struct inode *inode,
7232				const struct qstr *qstr)
7233{
7234	int ret = 0;
7235	struct buffer_head *dir_bh = NULL;
7236
7237	ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7238	if (ret) {
7239		mlog_errno(ret);
7240		goto leave;
7241	}
7242
7243	ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7244	if (ret) {
7245		mlog_errno(ret);
7246		goto leave;
7247	}
7248	ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7249	if (ret)
7250		mlog_errno(ret);
7251
7252	ocfs2_inode_unlock(dir, 0);
7253	brelse(dir_bh);
7254leave:
7255	return ret;
7256}
7257
7258/*
7259 * 'security' attributes support
7260 */
7261static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7262				    struct dentry *unused, struct inode *inode,
7263				    const char *name, void *buffer, size_t size)
7264{
7265	return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7266			       name, buffer, size);
7267}
7268
7269static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7270				    struct mnt_idmap *idmap,
7271				    struct dentry *unused, struct inode *inode,
7272				    const char *name, const void *value,
7273				    size_t size, int flags)
7274{
7275	return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7276			       name, value, size, flags);
7277}
7278
7279static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7280		     void *fs_info)
7281{
7282	struct ocfs2_security_xattr_info *si = fs_info;
7283	const struct xattr *xattr;
7284	int err = 0;
7285
7286	if (si) {
7287		si->value = kmemdup(xattr_array->value, xattr_array->value_len,
7288				    GFP_KERNEL);
7289		if (!si->value)
7290			return -ENOMEM;
7291
7292		si->name = xattr_array->name;
7293		si->value_len = xattr_array->value_len;
7294		return 0;
7295	}
7296
7297	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7298		err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7299				      xattr->name, xattr->value,
7300				      xattr->value_len, XATTR_CREATE);
7301		if (err)
7302			break;
7303	}
7304	return err;
7305}
7306
7307int ocfs2_init_security_get(struct inode *inode,
7308			    struct inode *dir,
7309			    const struct qstr *qstr,
7310			    struct ocfs2_security_xattr_info *si)
7311{
7312	int ret;
7313
7314	/* check whether ocfs2 support feature xattr */
7315	if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7316		return -EOPNOTSUPP;
7317	if (si) {
7318		ret = security_inode_init_security(inode, dir, qstr,
7319						   &ocfs2_initxattrs, si);
7320		/*
7321		 * security_inode_init_security() does not return -EOPNOTSUPP,
7322		 * we have to check the xattr ourselves.
7323		 */
7324		if (!ret && !si->name)
7325			si->enable = 0;
7326
7327		return ret;
7328	}
7329
7330	return security_inode_init_security(inode, dir, qstr,
7331					    &ocfs2_initxattrs, NULL);
7332}
7333
7334int ocfs2_init_security_set(handle_t *handle,
7335			    struct inode *inode,
7336			    struct buffer_head *di_bh,
7337			    struct ocfs2_security_xattr_info *si,
7338			    struct ocfs2_alloc_context *xattr_ac,
7339			    struct ocfs2_alloc_context *data_ac)
7340{
7341	return ocfs2_xattr_set_handle(handle, inode, di_bh,
7342				     OCFS2_XATTR_INDEX_SECURITY,
7343				     si->name, si->value, si->value_len, 0,
7344				     xattr_ac, data_ac);
7345}
7346
7347const struct xattr_handler ocfs2_xattr_security_handler = {
7348	.prefix	= XATTR_SECURITY_PREFIX,
7349	.get	= ocfs2_xattr_security_get,
7350	.set	= ocfs2_xattr_security_set,
7351};
7352
7353/*
7354 * 'trusted' attributes support
7355 */
7356static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7357				   struct dentry *unused, struct inode *inode,
7358				   const char *name, void *buffer, size_t size)
7359{
7360	return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7361			       name, buffer, size);
7362}
7363
7364static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7365				   struct mnt_idmap *idmap,
7366				   struct dentry *unused, struct inode *inode,
7367				   const char *name, const void *value,
7368				   size_t size, int flags)
7369{
7370	return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7371			       name, value, size, flags);
7372}
7373
7374const struct xattr_handler ocfs2_xattr_trusted_handler = {
7375	.prefix	= XATTR_TRUSTED_PREFIX,
7376	.get	= ocfs2_xattr_trusted_get,
7377	.set	= ocfs2_xattr_trusted_set,
7378};
7379
7380/*
7381 * 'user' attributes support
7382 */
7383static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7384				struct dentry *unused, struct inode *inode,
7385				const char *name, void *buffer, size_t size)
7386{
7387	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7388
7389	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7390		return -EOPNOTSUPP;
7391	return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7392			       buffer, size);
7393}
7394
7395static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7396				struct mnt_idmap *idmap,
7397				struct dentry *unused, struct inode *inode,
7398				const char *name, const void *value,
7399				size_t size, int flags)
7400{
7401	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7402
7403	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7404		return -EOPNOTSUPP;
7405
7406	return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7407			       name, value, size, flags);
7408}
7409
7410const struct xattr_handler ocfs2_xattr_user_handler = {
7411	.prefix	= XATTR_USER_PREFIX,
7412	.get	= ocfs2_xattr_user_get,
7413	.set	= ocfs2_xattr_user_set,
7414};