fs/ocfs2/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * xattr.c
   6  *
   7  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   8  *
   9  * CREDITS:
  10  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  11  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  12  */
  13
  14 #include <linux/capability.h>
  15 #include <linux/fs.h>
  16 #include <linux/types.h>
  17 #include <linux/slab.h>
  18 #include <linux/highmem.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/uio.h>
  21 #include <linux/sched.h>
  22 #include <linux/splice.h>
  23 #include <linux/mount.h>
  24 #include <linux/writeback.h>
  25 #include <linux/falloc.h>
  26 #include <linux/sort.h>
  27 #include <linux/init.h>
  28 #include <linux/module.h>
  29 #include <linux/string.h>
  30 #include <linux/security.h>
  31
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35 #include "alloc.h"
  36 #include "blockcheck.h"
  37 #include "dlmglue.h"
  38 #include "file.h"
  39 #include "symlink.h"
  40 #include "sysfile.h"
  41 #include "inode.h"
  42 #include "journal.h"
  43 #include "ocfs2_fs.h"
  44 #include "suballoc.h"
  45 #include "uptodate.h"
  46 #include "buffer_head_io.h"
  47 #include "super.h"
  48 #include "xattr.h"
  49 #include "refcounttree.h"
  50 #include "acl.h"
  51 #include "ocfs2_trace.h"
  52
  53 struct ocfs2_xattr_def_value_root {
  54         struct ocfs2_xattr_value_root   xv;
  55         struct ocfs2_extent_rec         er;
  56 };
  57
  58 struct ocfs2_xattr_bucket {
  59         /* The inode these xattrs are associated with */
  60         struct inode *bu_inode;
  61
  62         /* The actual buffers that make up the bucket */
  63         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  64
  65         /* How many blocks make up one bucket for this filesystem */
  66         int bu_blocks;
  67 };
  68
  69 struct ocfs2_xattr_set_ctxt {
  70         handle_t *handle;
  71         struct ocfs2_alloc_context *meta_ac;
  72         struct ocfs2_alloc_context *data_ac;
  73         struct ocfs2_cached_dealloc_ctxt dealloc;
  74         int set_abort;
  75 };
  76
  77 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  78 #define OCFS2_XATTR_INLINE_SIZE 80
  79 #define OCFS2_XATTR_HEADER_GAP  4
  80 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  81                                          - sizeof(struct ocfs2_xattr_header) \
  82                                          - OCFS2_XATTR_HEADER_GAP)
  83 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  84                                          - sizeof(struct ocfs2_xattr_block) \
  85                                          - sizeof(struct ocfs2_xattr_header) \
  86                                          - OCFS2_XATTR_HEADER_GAP)
  87
  88 static struct ocfs2_xattr_def_value_root def_xv = {
  89         .xv.xr_list.l_count = cpu_to_le16(1),
  90 };
  91
  92 const struct xattr_handler *ocfs2_xattr_handlers[] = {
  93         &ocfs2_xattr_user_handler,
  94         &posix_acl_access_xattr_handler,
  95         &posix_acl_default_xattr_handler,
  96         &ocfs2_xattr_trusted_handler,
  97         &ocfs2_xattr_security_handler,
  98         NULL
  99 };
 100
 101 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 102         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 103         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 104                                         = &posix_acl_access_xattr_handler,
 105         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 106                                         = &posix_acl_default_xattr_handler,
 107         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 108         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 109 };
 110
 111 struct ocfs2_xattr_info {
 112         int             xi_name_index;
 113         const char      *xi_name;
 114         int             xi_name_len;
 115         const void      *xi_value;
 116         size_t          xi_value_len;
 117 };
 118
 119 struct ocfs2_xattr_search {
 120         struct buffer_head *inode_bh;
 121         /*
 122          * xattr_bh point to the block buffer head which has extended attribute
 123          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 124          */
 125         struct buffer_head *xattr_bh;
 126         struct ocfs2_xattr_header *header;
 127         struct ocfs2_xattr_bucket *bucket;
 128         void *base;
 129         void *end;
 130         struct ocfs2_xattr_entry *here;
 131         int not_found;
 132 };
 133
 134 /* Operations on struct ocfs2_xa_entry */
 135 struct ocfs2_xa_loc;
 136 struct ocfs2_xa_loc_operations {
 137         /*
 138          * Journal functions
 139          */
 140         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 141                                   int type);
 142         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 143
 144         /*
 145          * Return a pointer to the appropriate buffer in loc->xl_storage
 146          * at the given offset from loc->xl_header.
 147          */
 148         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 149
 150         /* Can we reuse the existing entry for the new value? */
 151         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 152                              struct ocfs2_xattr_info *xi);
 153
 154         /* How much space is needed for the new value? */
 155         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 156                                struct ocfs2_xattr_info *xi);
 157
 158         /*
 159          * Return the offset of the first name+value pair.  This is
 160          * the start of our downward-filling free space.
 161          */
 162         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 163
 164         /*
 165          * Remove the name+value at this location.  Do whatever is
 166          * appropriate with the remaining name+value pairs.
 167          */
 168         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 169
 170         /* Fill xl_entry with a new entry */
 171         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 172
 173         /* Add name+value storage to an entry */
 174         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 175
 176         /*
 177          * Initialize the value buf's access and bh fields for this entry.
 178          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 179          */
 180         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 181                                    struct ocfs2_xattr_value_buf *vb);
 182 };
 183
 184 /*
 185  * Describes an xattr entry location.  This is a memory structure
 186  * tracking the on-disk structure.
 187  */
 188 struct ocfs2_xa_loc {
 189         /* This xattr belongs to this inode */
 190         struct inode *xl_inode;
 191
 192         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 193         struct ocfs2_xattr_header *xl_header;
 194
 195         /* Bytes from xl_header to the end of the storage */
 196         int xl_size;
 197
 198         /*
 199          * The ocfs2_xattr_entry this location describes.  If this is
 200          * NULL, this location describes the on-disk structure where it
 201          * would have been.
 202          */
 203         struct ocfs2_xattr_entry *xl_entry;
 204
 205         /*
 206          * Internal housekeeping
 207          */
 208
 209         /* Buffer(s) containing this entry */
 210         void *xl_storage;
 211
 212         /* Operations on the storage backing this location */
 213         const struct ocfs2_xa_loc_operations *xl_ops;
 214 };
 215
 216 /*
 217  * Convenience functions to calculate how much space is needed for a
 218  * given name+value pair
 219  */
 220 static int namevalue_size(int name_len, uint64_t value_len)
 221 {
 222         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 223                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 224         else
 225                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 226 }
 227
 228 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 229 {
 230         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 231 }
 232
 233 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 234 {
 235         u64 value_len = le64_to_cpu(xe->xe_value_size);
 236
 237         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 238                ocfs2_xattr_is_local(xe));
 239         return namevalue_size(xe->xe_name_len, value_len);
 240 }
 241
 242
 243 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 244                                              struct ocfs2_xattr_header *xh,
 245                                              int index,
 246                                              int *block_off,
 247                                              int *new_offset);
 248
 249 static int ocfs2_xattr_block_find(struct inode *inode,
 250                                   int name_index,
 251                                   const char *name,
 252                                   struct ocfs2_xattr_search *xs);
 253 static int ocfs2_xattr_index_block_find(struct inode *inode,
 254                                         struct buffer_head *root_bh,
 255                                         int name_index,
 256                                         const char *name,
 257                                         struct ocfs2_xattr_search *xs);
 258
 259 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 260                                         struct buffer_head *blk_bh,
 261                                         char *buffer,
 262                                         size_t buffer_size);
 263
 264 static int ocfs2_xattr_create_index_block(struct inode *inode,
 265                                           struct ocfs2_xattr_search *xs,
 266                                           struct ocfs2_xattr_set_ctxt *ctxt);
 267
 268 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 269                                              struct ocfs2_xattr_info *xi,
 270                                              struct ocfs2_xattr_search *xs,
 271                                              struct ocfs2_xattr_set_ctxt *ctxt);
 272
 273 typedef int (xattr_tree_rec_func)(struct inode *inode,
 274                                   struct buffer_head *root_bh,
 275                                   u64 blkno, u32 cpos, u32 len, void *para);
 276 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 277                                            struct buffer_head *root_bh,
 278                                            xattr_tree_rec_func *rec_func,
 279                                            void *para);
 280 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 281                                         struct ocfs2_xattr_bucket *bucket,
 282                                         void *para);
 283 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 284                                   struct buffer_head *root_bh,
 285                                   u64 blkno,
 286                                   u32 cpos,
 287                                   u32 len,
 288                                   void *para);
 289
 290 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 291                                   u64 src_blk, u64 last_blk, u64 to_blk,
 292                                   unsigned int start_bucket,
 293                                   u32 *first_hash);
 294 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 295                                         struct ocfs2_dinode *di,
 296                                         struct ocfs2_xattr_info *xi,
 297                                         struct ocfs2_xattr_search *xis,
 298                                         struct ocfs2_xattr_search *xbs,
 299                                         struct ocfs2_refcount_tree **ref_tree,
 300                                         int *meta_need,
 301                                         int *credits);
 302 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 303                                            struct ocfs2_xattr_bucket *bucket,
 304                                            int offset,
 305                                            struct ocfs2_xattr_value_root **xv,
 306                                            struct buffer_head **bh);
 307
 308 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 309 {
 310         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 311 }
 312
 313 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 314 {
 315         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 316 }
 317
 318 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 319 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 320 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 321
 322 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 323 {
 324         struct ocfs2_xattr_bucket *bucket;
 325         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 326
 327         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 328
 329         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 330         if (bucket) {
 331                 bucket->bu_inode = inode;
 332                 bucket->bu_blocks = blks;
 333         }
 334
 335         return bucket;
 336 }
 337
 338 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 339 {
 340         int i;
 341
 342         for (i = 0; i < bucket->bu_blocks; i++) {
 343                 brelse(bucket->bu_bhs[i]);
 344                 bucket->bu_bhs[i] = NULL;
 345         }
 346 }
 347
 348 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 349 {
 350         if (bucket) {
 351                 ocfs2_xattr_bucket_relse(bucket);
 352                 bucket->bu_inode = NULL;
 353                 kfree(bucket);
 354         }
 355 }
 356
 357 /*
 358  * A bucket that has never been written to disk doesn't need to be
 359  * read.  We just need the buffer_heads.  Don't call this for
 360  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 361  * them fully.
 362  */
 363 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 364                                    u64 xb_blkno, int new)
 365 {
 366         int i, rc = 0;
 367
 368         for (i = 0; i < bucket->bu_blocks; i++) {
 369                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 370                                               xb_blkno + i);
 371                 if (!bucket->bu_bhs[i]) {
 372                         rc = -ENOMEM;
 373                         mlog_errno(rc);
 374                         break;
 375                 }
 376
 377                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 378                                            bucket->bu_bhs[i])) {
 379                         if (new)
 380                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 381                                                               bucket->bu_bhs[i]);
 382                         else {
 383                                 set_buffer_uptodate(bucket->bu_bhs[i]);
 384                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 385                                                           bucket->bu_bhs[i]);
 386                         }
 387                 }
 388         }
 389
 390         if (rc)
 391                 ocfs2_xattr_bucket_relse(bucket);
 392         return rc;
 393 }
 394
 395 /* Read the xattr bucket at xb_blkno */
 396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 397                                    u64 xb_blkno)
 398 {
 399         int rc;
 400
 401         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 402                                bucket->bu_blocks, bucket->bu_bhs, 0,
 403                                NULL);
 404         if (!rc) {
 405                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 406                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 407                                                  bucket->bu_bhs,
 408                                                  bucket->bu_blocks,
 409                                                  &bucket_xh(bucket)->xh_check);
 410                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 411                 if (rc)
 412                         mlog_errno(rc);
 413         }
 414
 415         if (rc)
 416                 ocfs2_xattr_bucket_relse(bucket);
 417         return rc;
 418 }
 419
 420 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 421                                              struct ocfs2_xattr_bucket *bucket,
 422                                              int type)
 423 {
 424         int i, rc = 0;
 425
 426         for (i = 0; i < bucket->bu_blocks; i++) {
 427                 rc = ocfs2_journal_access(handle,
 428                                           INODE_CACHE(bucket->bu_inode),
 429                                           bucket->bu_bhs[i], type);
 430                 if (rc) {
 431                         mlog_errno(rc);
 432                         break;
 433                 }
 434         }
 435
 436         return rc;
 437 }
 438
 439 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 440                                              struct ocfs2_xattr_bucket *bucket)
 441 {
 442         int i;
 443
 444         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 445         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 446                                    bucket->bu_bhs, bucket->bu_blocks,
 447                                    &bucket_xh(bucket)->xh_check);
 448         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 449
 450         for (i = 0; i < bucket->bu_blocks; i++)
 451                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 452 }
 453
 454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 455                                          struct ocfs2_xattr_bucket *src)
 456 {
 457         int i;
 458         int blocksize = src->bu_inode->i_sb->s_blocksize;
 459
 460         BUG_ON(dest->bu_blocks != src->bu_blocks);
 461         BUG_ON(dest->bu_inode != src->bu_inode);
 462
 463         for (i = 0; i < src->bu_blocks; i++) {
 464                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 465                        blocksize);
 466         }
 467 }
 468
 469 static int ocfs2_validate_xattr_block(struct super_block *sb,
 470                                       struct buffer_head *bh)
 471 {
 472         int rc;
 473         struct ocfs2_xattr_block *xb =
 474                 (struct ocfs2_xattr_block *)bh->b_data;
 475
 476         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 477
 478         BUG_ON(!buffer_uptodate(bh));
 479
 480         /*
 481          * If the ecc fails, we return the error but otherwise
 482          * leave the filesystem running.  We know any error is
 483          * local to this block.
 484          */
 485         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 486         if (rc)
 487                 return rc;
 488
 489         /*
 490          * Errors after here are fatal
 491          */
 492
 493         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 494                 return ocfs2_error(sb,
 495                                    "Extended attribute block #%llu has bad signature %.*s\n",
 496                                    (unsigned long long)bh->b_blocknr, 7,
 497                                    xb->xb_signature);
 498         }
 499
 500         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 501                 return ocfs2_error(sb,
 502                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 503                                    (unsigned long long)bh->b_blocknr,
 504                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
 505         }
 506
 507         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 508                 return ocfs2_error(sb,
 509                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 510                                    (unsigned long long)bh->b_blocknr,
 511                                    le32_to_cpu(xb->xb_fs_generation));
 512         }
 513
 514         return 0;
 515 }
 516
 517 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 518                                   struct buffer_head **bh)
 519 {
 520         int rc;
 521         struct buffer_head *tmp = *bh;
 522
 523         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 524                               ocfs2_validate_xattr_block);
 525
 526         /* If ocfs2_read_block() got us a new bh, pass it up. */
 527         if (!rc && !*bh)
 528                 *bh = tmp;
 529
 530         return rc;
 531 }
 532
 533 static inline const char *ocfs2_xattr_prefix(int name_index)
 534 {
 535         const struct xattr_handler *handler = NULL;
 536
 537         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 538                 handler = ocfs2_xattr_handler_map[name_index];
 539         return handler ? xattr_prefix(handler) : NULL;
 540 }
 541
 542 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 543                                  const char *name,
 544                                  int name_len)
 545 {
 546         /* Get hash value of uuid from super block */
 547         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 548         int i;
 549
 550         /* hash extended attribute name */
 551         for (i = 0; i < name_len; i++) {
 552                 hash = (hash << OCFS2_HASH_SHIFT) ^
 553                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 554                        *name++;
 555         }
 556
 557         return hash;
 558 }
 559
 560 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 561 {
 562         return namevalue_size(name_len, value_len) +
 563                 sizeof(struct ocfs2_xattr_entry);
 564 }
 565
 566 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 567 {
 568         return namevalue_size_xi(xi) +
 569                 sizeof(struct ocfs2_xattr_entry);
 570 }
 571
 572 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 573 {
 574         return namevalue_size_xe(xe) +
 575                 sizeof(struct ocfs2_xattr_entry);
 576 }
 577
 578 int ocfs2_calc_security_init(struct inode *dir,
 579                              struct ocfs2_security_xattr_info *si,
 580                              int *want_clusters,
 581                              int *xattr_credits,
 582                              struct ocfs2_alloc_context **xattr_ac)
 583 {
 584         int ret = 0;
 585         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 586         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 587                                                  si->value_len);
 588
 589         /*
 590          * The max space of security xattr taken inline is
 591          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 592          * So reserve one metadata block for it is ok.
 593          */
 594         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 595             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 596                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 597                 if (ret) {
 598                         mlog_errno(ret);
 599                         return ret;
 600                 }
 601                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 602         }
 603
 604         /* reserve clusters for xattr value which will be set in B tree*/
 605         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 606                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 607                                                             si->value_len);
 608
 609                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 610                                                            new_clusters);
 611                 *want_clusters += new_clusters;
 612         }
 613         return ret;
 614 }
 615
 616 int ocfs2_calc_xattr_init(struct inode *dir,
 617                           struct buffer_head *dir_bh,
 618                           umode_t mode,
 619                           struct ocfs2_security_xattr_info *si,
 620                           int *want_clusters,
 621                           int *xattr_credits,
 622                           int *want_meta)
 623 {
 624         int ret = 0;
 625         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 626         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 627
 628         if (si->enable)
 629                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 630                                                      si->value_len);
 631
 632         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 633                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
 634                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 635                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 636                                         "", NULL, 0);
 637                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
 638                 if (acl_len > 0) {
 639                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 640                         if (S_ISDIR(mode))
 641                                 a_size <<= 1;
 642                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 643                         ret = acl_len;
 644                         mlog_errno(ret);
 645                         return ret;
 646                 }
 647         }
 648
 649         if (!(s_size + a_size))
 650                 return ret;
 651
 652         /*
 653          * The max space of security xattr taken inline is
 654          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 655          * The max space of acl xattr taken inline is
 656          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 657          * when blocksize = 512, may reserve one more cluser for
 658          * xattr bucket, otherwise reserve one metadata block
 659          * for them is ok.
 660          * If this is a new directory with inline data,
 661          * we choose to reserve the entire inline area for
 662          * directory contents and force an external xattr block.
 663          */
 664         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 665             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 666             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 667                 *want_meta = *want_meta + 1;
 668                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 669         }
 670
 671         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 672             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 673                 *want_clusters += 1;
 674                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 675         }
 676
 677         /*
 678          * reserve credits and clusters for xattrs which has large value
 679          * and have to be set outside
 680          */
 681         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 682                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 683                                                         si->value_len);
 684                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 685                                                            new_clusters);
 686                 *want_clusters += new_clusters;
 687         }
 688         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 689             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 690                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 691                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 692                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 693                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 694                                                            new_clusters);
 695                 *want_clusters += new_clusters;
 696         }
 697
 698         return ret;
 699 }
 700
 701 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 702                                          u32 clusters_to_add,
 703                                          struct ocfs2_xattr_value_buf *vb,
 704                                          struct ocfs2_xattr_set_ctxt *ctxt)
 705 {
 706         int status = 0, credits;
 707         handle_t *handle = ctxt->handle;
 708         enum ocfs2_alloc_restarted why;
 709         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 710         struct ocfs2_extent_tree et;
 711
 712         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 713
 714         while (clusters_to_add) {
 715                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 716
 717                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 718                                        OCFS2_JOURNAL_ACCESS_WRITE);
 719                 if (status < 0) {
 720                         mlog_errno(status);
 721                         break;
 722                 }
 723
 724                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 725                 status = ocfs2_add_clusters_in_btree(handle,
 726                                                      &et,
 727                                                      &logical_start,
 728                                                      clusters_to_add,
 729                                                      0,
 730                                                      ctxt->data_ac,
 731                                                      ctxt->meta_ac,
 732                                                      &why);
 733                 if ((status < 0) && (status != -EAGAIN)) {
 734                         if (status != -ENOSPC)
 735                                 mlog_errno(status);
 736                         break;
 737                 }
 738
 739                 ocfs2_journal_dirty(handle, vb->vb_bh);
 740
 741                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 742                                          prev_clusters;
 743
 744                 if (why != RESTART_NONE && clusters_to_add) {
 745                         /*
 746                          * We can only fail in case the alloc file doesn't give
 747                          * up enough clusters.
 748                          */
 749                         BUG_ON(why == RESTART_META);
 750
 751                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 752                                                             &vb->vb_xv->xr_list);
 753                         status = ocfs2_extend_trans(handle, credits);
 754                         if (status < 0) {
 755                                 status = -ENOMEM;
 756                                 mlog_errno(status);
 757                                 break;
 758                         }
 759                 }
 760         }
 761
 762         return status;
 763 }
 764
 765 static int __ocfs2_remove_xattr_range(struct inode *inode,
 766                                       struct ocfs2_xattr_value_buf *vb,
 767                                       u32 cpos, u32 phys_cpos, u32 len,
 768                                       unsigned int ext_flags,
 769                                       struct ocfs2_xattr_set_ctxt *ctxt)
 770 {
 771         int ret;
 772         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 773         handle_t *handle = ctxt->handle;
 774         struct ocfs2_extent_tree et;
 775
 776         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 777
 778         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 779                             OCFS2_JOURNAL_ACCESS_WRITE);
 780         if (ret) {
 781                 mlog_errno(ret);
 782                 goto out;
 783         }
 784
 785         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 786                                   &ctxt->dealloc);
 787         if (ret) {
 788                 mlog_errno(ret);
 789                 goto out;
 790         }
 791
 792         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 793         ocfs2_journal_dirty(handle, vb->vb_bh);
 794
 795         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 796                 ret = ocfs2_decrease_refcount(inode, handle,
 797                                         ocfs2_blocks_to_clusters(inode->i_sb,
 798                                                                  phys_blkno),
 799                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 800         else
 801                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 802                                                   phys_blkno, len);
 803         if (ret)
 804                 mlog_errno(ret);
 805
 806 out:
 807         return ret;
 808 }
 809
 810 static int ocfs2_xattr_shrink_size(struct inode *inode,
 811                                    u32 old_clusters,
 812                                    u32 new_clusters,
 813                                    struct ocfs2_xattr_value_buf *vb,
 814                                    struct ocfs2_xattr_set_ctxt *ctxt)
 815 {
 816         int ret = 0;
 817         unsigned int ext_flags;
 818         u32 trunc_len, cpos, phys_cpos, alloc_size;
 819         u64 block;
 820
 821         if (old_clusters <= new_clusters)
 822                 return 0;
 823
 824         cpos = new_clusters;
 825         trunc_len = old_clusters - new_clusters;
 826         while (trunc_len) {
 827                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 828                                                &alloc_size,
 829                                                &vb->vb_xv->xr_list, &ext_flags);
 830                 if (ret) {
 831                         mlog_errno(ret);
 832                         goto out;
 833                 }
 834
 835                 if (alloc_size > trunc_len)
 836                         alloc_size = trunc_len;
 837
 838                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 839                                                  phys_cpos, alloc_size,
 840                                                  ext_flags, ctxt);
 841                 if (ret) {
 842                         mlog_errno(ret);
 843                         goto out;
 844                 }
 845
 846                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 847                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 848                                                        block, alloc_size);
 849                 cpos += alloc_size;
 850                 trunc_len -= alloc_size;
 851         }
 852
 853 out:
 854         return ret;
 855 }
 856
 857 static int ocfs2_xattr_value_truncate(struct inode *inode,
 858                                       struct ocfs2_xattr_value_buf *vb,
 859                                       int len,
 860                                       struct ocfs2_xattr_set_ctxt *ctxt)
 861 {
 862         int ret;
 863         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 864         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 865
 866         if (new_clusters == old_clusters)
 867                 return 0;
 868
 869         if (new_clusters > old_clusters)
 870                 ret = ocfs2_xattr_extend_allocation(inode,
 871                                                     new_clusters - old_clusters,
 872                                                     vb, ctxt);
 873         else
 874                 ret = ocfs2_xattr_shrink_size(inode,
 875                                               old_clusters, new_clusters,
 876                                               vb, ctxt);
 877
 878         return ret;
 879 }
 880
 881 static int ocfs2_xattr_list_entry(struct super_block *sb,
 882                                   char *buffer, size_t size,
 883                                   size_t *result, int type,
 884                                   const char *name, int name_len)
 885 {
 886         char *p = buffer + *result;
 887         const char *prefix;
 888         int prefix_len;
 889         int total_len;
 890
 891         switch(type) {
 892         case OCFS2_XATTR_INDEX_USER:
 893                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 894                         return 0;
 895                 break;
 896
 897         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 898         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 899                 if (!(sb->s_flags & SB_POSIXACL))
 900                         return 0;
 901                 break;
 902
 903         case OCFS2_XATTR_INDEX_TRUSTED:
 904                 if (!capable(CAP_SYS_ADMIN))
 905                         return 0;
 906                 break;
 907         }
 908
 909         prefix = ocfs2_xattr_prefix(type);
 910         if (!prefix)
 911                 return 0;
 912         prefix_len = strlen(prefix);
 913         total_len = prefix_len + name_len + 1;
 914         *result += total_len;
 915
 916         /* we are just looking for how big our buffer needs to be */
 917         if (!size)
 918                 return 0;
 919
 920         if (*result > size)
 921                 return -ERANGE;
 922
 923         memcpy(p, prefix, prefix_len);
 924         memcpy(p + prefix_len, name, name_len);
 925         p[prefix_len + name_len] = '\0';
 926
 927         return 0;
 928 }
 929
 930 static int ocfs2_xattr_list_entries(struct inode *inode,
 931                                     struct ocfs2_xattr_header *header,
 932                                     char *buffer, size_t buffer_size)
 933 {
 934         size_t result = 0;
 935         int i, type, ret;
 936         const char *name;
 937
 938         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 939                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 940                 type = ocfs2_xattr_get_type(entry);
 941                 name = (const char *)header +
 942                         le16_to_cpu(entry->xe_name_offset);
 943
 944                 ret = ocfs2_xattr_list_entry(inode->i_sb,
 945                                              buffer, buffer_size,
 946                                              &result, type, name,
 947                                              entry->xe_name_len);
 948                 if (ret)
 949                         return ret;
 950         }
 951
 952         return result;
 953 }
 954
 955 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 956                                          struct ocfs2_dinode *di)
 957 {
 958         struct ocfs2_xattr_header *xh;
 959         int i;
 960
 961         xh = (struct ocfs2_xattr_header *)
 962                  ((void *)di + inode->i_sb->s_blocksize -
 963                  le16_to_cpu(di->i_xattr_inline_size));
 964
 965         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 966                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 967                         return 1;
 968
 969         return 0;
 970 }
 971
 972 static int ocfs2_xattr_ibody_list(struct inode *inode,
 973                                   struct ocfs2_dinode *di,
 974                                   char *buffer,
 975                                   size_t buffer_size)
 976 {
 977         struct ocfs2_xattr_header *header = NULL;
 978         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 979         int ret = 0;
 980
 981         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 982                 return ret;
 983
 984         header = (struct ocfs2_xattr_header *)
 985                  ((void *)di + inode->i_sb->s_blocksize -
 986                  le16_to_cpu(di->i_xattr_inline_size));
 987
 988         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 989
 990         return ret;
 991 }
 992
 993 static int ocfs2_xattr_block_list(struct inode *inode,
 994                                   struct ocfs2_dinode *di,
 995                                   char *buffer,
 996                                   size_t buffer_size)
 997 {
 998         struct buffer_head *blk_bh = NULL;
 999         struct ocfs2_xattr_block *xb;
1000         int ret = 0;
1001
1002         if (!di->i_xattr_loc)
1003                 return ret;
1004
1005         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1006                                      &blk_bh);
1007         if (ret < 0) {
1008                 mlog_errno(ret);
1009                 return ret;
1010         }
1011
1012         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1013         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1014                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1015                 ret = ocfs2_xattr_list_entries(inode, header,
1016                                                buffer, buffer_size);
1017         } else
1018                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1019                                                    buffer, buffer_size);
1020
1021         brelse(blk_bh);
1022
1023         return ret;
1024 }
1025
1026 ssize_t ocfs2_listxattr(struct dentry *dentry,
1027                         char *buffer,
1028                         size_t size)
1029 {
1030         int ret = 0, i_ret = 0, b_ret = 0;
1031         struct buffer_head *di_bh = NULL;
1032         struct ocfs2_dinode *di = NULL;
1033         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1034
1035         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1036                 return -EOPNOTSUPP;
1037
1038         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1039                 return ret;
1040
1041         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1042         if (ret < 0) {
1043                 mlog_errno(ret);
1044                 return ret;
1045         }
1046
1047         di = (struct ocfs2_dinode *)di_bh->b_data;
1048
1049         down_read(&oi->ip_xattr_sem);
1050         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1051         if (i_ret < 0)
1052                 b_ret = 0;
1053         else {
1054                 if (buffer) {
1055                         buffer += i_ret;
1056                         size -= i_ret;
1057                 }
1058                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1059                                                buffer, size);
1060                 if (b_ret < 0)
1061                         i_ret = 0;
1062         }
1063         up_read(&oi->ip_xattr_sem);
1064         ocfs2_inode_unlock(d_inode(dentry), 0);
1065
1066         brelse(di_bh);
1067
1068         return i_ret + b_ret;
1069 }
1070
1071 static int ocfs2_xattr_find_entry(int name_index,
1072                                   const char *name,
1073                                   struct ocfs2_xattr_search *xs)
1074 {
1075         struct ocfs2_xattr_entry *entry;
1076         size_t name_len;
1077         int i, cmp = 1;
1078
1079         if (name == NULL)
1080                 return -EINVAL;
1081
1082         name_len = strlen(name);
1083         entry = xs->here;
1084         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1085                 cmp = name_index - ocfs2_xattr_get_type(entry);
1086                 if (!cmp)
1087                         cmp = name_len - entry->xe_name_len;
1088                 if (!cmp)
1089                         cmp = memcmp(name, (xs->base +
1090                                      le16_to_cpu(entry->xe_name_offset)),
1091                                      name_len);
1092                 if (cmp == 0)
1093                         break;
1094                 entry += 1;
1095         }
1096         xs->here = entry;
1097
1098         return cmp ? -ENODATA : 0;
1099 }
1100
1101 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1102                                          struct ocfs2_xattr_value_root *xv,
1103                                          void *buffer,
1104                                          size_t len)
1105 {
1106         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1107         u64 blkno;
1108         int i, ret = 0;
1109         size_t cplen, blocksize;
1110         struct buffer_head *bh = NULL;
1111         struct ocfs2_extent_list *el;
1112
1113         el = &xv->xr_list;
1114         clusters = le32_to_cpu(xv->xr_clusters);
1115         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1116         blocksize = inode->i_sb->s_blocksize;
1117
1118         cpos = 0;
1119         while (cpos < clusters) {
1120                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1121                                                &num_clusters, el, NULL);
1122                 if (ret) {
1123                         mlog_errno(ret);
1124                         goto out;
1125                 }
1126
1127                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1128                 /* Copy ocfs2_xattr_value */
1129                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1130                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1131                                                &bh, NULL);
1132                         if (ret) {
1133                                 mlog_errno(ret);
1134                                 goto out;
1135                         }
1136
1137                         cplen = len >= blocksize ? blocksize : len;
1138                         memcpy(buffer, bh->b_data, cplen);
1139                         len -= cplen;
1140                         buffer += cplen;
1141
1142                         brelse(bh);
1143                         bh = NULL;
1144                         if (len == 0)
1145                                 break;
1146                 }
1147                 cpos += num_clusters;
1148         }
1149 out:
1150         return ret;
1151 }
1152
1153 static int ocfs2_xattr_ibody_get(struct inode *inode,
1154                                  int name_index,
1155                                  const char *name,
1156                                  void *buffer,
1157                                  size_t buffer_size,
1158                                  struct ocfs2_xattr_search *xs)
1159 {
1160         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1161         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1162         struct ocfs2_xattr_value_root *xv;
1163         size_t size;
1164         int ret = 0;
1165
1166         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1167                 return -ENODATA;
1168
1169         xs->end = (void *)di + inode->i_sb->s_blocksize;
1170         xs->header = (struct ocfs2_xattr_header *)
1171                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1172         xs->base = (void *)xs->header;
1173         xs->here = xs->header->xh_entries;
1174
1175         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1176         if (ret)
1177                 return ret;
1178         size = le64_to_cpu(xs->here->xe_value_size);
1179         if (buffer) {
1180                 if (size > buffer_size)
1181                         return -ERANGE;
1182                 if (ocfs2_xattr_is_local(xs->here)) {
1183                         memcpy(buffer, (void *)xs->base +
1184                                le16_to_cpu(xs->here->xe_name_offset) +
1185                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1186                 } else {
1187                         xv = (struct ocfs2_xattr_value_root *)
1188                                 (xs->base + le16_to_cpu(
1189                                  xs->here->xe_name_offset) +
1190                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1191                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1192                                                             buffer, size);
1193                         if (ret < 0) {
1194                                 mlog_errno(ret);
1195                                 return ret;
1196                         }
1197                 }
1198         }
1199
1200         return size;
1201 }
1202
1203 static int ocfs2_xattr_block_get(struct inode *inode,
1204                                  int name_index,
1205                                  const char *name,
1206                                  void *buffer,
1207                                  size_t buffer_size,
1208                                  struct ocfs2_xattr_search *xs)
1209 {
1210         struct ocfs2_xattr_block *xb;
1211         struct ocfs2_xattr_value_root *xv;
1212         size_t size;
1213         int ret = -ENODATA, name_offset, name_len, i;
1214         int uninitialized_var(block_off);
1215
1216         xs->bucket = ocfs2_xattr_bucket_new(inode);
1217         if (!xs->bucket) {
1218                 ret = -ENOMEM;
1219                 mlog_errno(ret);
1220                 goto cleanup;
1221         }
1222
1223         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1224         if (ret) {
1225                 mlog_errno(ret);
1226                 goto cleanup;
1227         }
1228
1229         if (xs->not_found) {
1230                 ret = -ENODATA;
1231                 goto cleanup;
1232         }
1233
1234         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1235         size = le64_to_cpu(xs->here->xe_value_size);
1236         if (buffer) {
1237                 ret = -ERANGE;
1238                 if (size > buffer_size)
1239                         goto cleanup;
1240
1241                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1242                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1243                 i = xs->here - xs->header->xh_entries;
1244
1245                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1246                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1247                                                                 bucket_xh(xs->bucket),
1248                                                                 i,
1249                                                                 &block_off,
1250                                                                 &name_offset);
1251                         if (ret) {
1252                                 mlog_errno(ret);
1253                                 goto cleanup;
1254                         }
1255                         xs->base = bucket_block(xs->bucket, block_off);
1256                 }
1257                 if (ocfs2_xattr_is_local(xs->here)) {
1258                         memcpy(buffer, (void *)xs->base +
1259                                name_offset + name_len, size);
1260                 } else {
1261                         xv = (struct ocfs2_xattr_value_root *)
1262                                 (xs->base + name_offset + name_len);
1263                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1264                                                             buffer, size);
1265                         if (ret < 0) {
1266                                 mlog_errno(ret);
1267                                 goto cleanup;
1268                         }
1269                 }
1270         }
1271         ret = size;
1272 cleanup:
1273         ocfs2_xattr_bucket_free(xs->bucket);
1274
1275         brelse(xs->xattr_bh);
1276         xs->xattr_bh = NULL;
1277         return ret;
1278 }
1279
1280 int ocfs2_xattr_get_nolock(struct inode *inode,
1281                            struct buffer_head *di_bh,
1282                            int name_index,
1283                            const char *name,
1284                            void *buffer,
1285                            size_t buffer_size)
1286 {
1287         int ret;
1288         struct ocfs2_dinode *di = NULL;
1289         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1290         struct ocfs2_xattr_search xis = {
1291                 .not_found = -ENODATA,
1292         };
1293         struct ocfs2_xattr_search xbs = {
1294                 .not_found = -ENODATA,
1295         };
1296
1297         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1298                 return -EOPNOTSUPP;
1299
1300         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1301                 return -ENODATA;
1302
1303         xis.inode_bh = xbs.inode_bh = di_bh;
1304         di = (struct ocfs2_dinode *)di_bh->b_data;
1305
1306         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1307                                     buffer_size, &xis);
1308         if (ret == -ENODATA && di->i_xattr_loc)
1309                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1310                                             buffer_size, &xbs);
1311
1312         return ret;
1313 }
1314
1315 /* ocfs2_xattr_get()
1316  *
1317  * Copy an extended attribute into the buffer provided.
1318  * Buffer is NULL to compute the size of buffer required.
1319  */
1320 static int ocfs2_xattr_get(struct inode *inode,
1321                            int name_index,
1322                            const char *name,
1323                            void *buffer,
1324                            size_t buffer_size)
1325 {
1326         int ret, had_lock;
1327         struct buffer_head *di_bh = NULL;
1328         struct ocfs2_lock_holder oh;
1329
1330         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1331         if (had_lock < 0) {
1332                 mlog_errno(had_lock);
1333                 return had_lock;
1334         }
1335         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1336         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1337                                      name, buffer, buffer_size);
1338         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1339
1340         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1341
1342         brelse(di_bh);
1343
1344         return ret;
1345 }
1346
1347 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1348                                            handle_t *handle,
1349                                            struct ocfs2_xattr_value_buf *vb,
1350                                            const void *value,
1351                                            int value_len)
1352 {
1353         int ret = 0, i, cp_len;
1354         u16 blocksize = inode->i_sb->s_blocksize;
1355         u32 p_cluster, num_clusters;
1356         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1357         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1358         u64 blkno;
1359         struct buffer_head *bh = NULL;
1360         unsigned int ext_flags;
1361         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1362
1363         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1364
1365         while (cpos < clusters) {
1366                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1367                                                &num_clusters, &xv->xr_list,
1368                                                &ext_flags);
1369                 if (ret) {
1370                         mlog_errno(ret);
1371                         goto out;
1372                 }
1373
1374                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1375
1376                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1377
1378                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1379                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1380                                                &bh, NULL);
1381                         if (ret) {
1382                                 mlog_errno(ret);
1383                                 goto out;
1384                         }
1385
1386                         ret = ocfs2_journal_access(handle,
1387                                                    INODE_CACHE(inode),
1388                                                    bh,
1389                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1390                         if (ret < 0) {
1391                                 mlog_errno(ret);
1392                                 goto out;
1393                         }
1394
1395                         cp_len = value_len > blocksize ? blocksize : value_len;
1396                         memcpy(bh->b_data, value, cp_len);
1397                         value_len -= cp_len;
1398                         value += cp_len;
1399                         if (cp_len < blocksize)
1400                                 memset(bh->b_data + cp_len, 0,
1401                                        blocksize - cp_len);
1402
1403                         ocfs2_journal_dirty(handle, bh);
1404                         brelse(bh);
1405                         bh = NULL;
1406
1407                         /*
1408                          * XXX: do we need to empty all the following
1409                          * blocks in this cluster?
1410                          */
1411                         if (!value_len)
1412                                 break;
1413                 }
1414                 cpos += num_clusters;
1415         }
1416 out:
1417         brelse(bh);
1418
1419         return ret;
1420 }
1421
1422 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1423                                        int num_entries)
1424 {
1425         int free_space;
1426
1427         if (!needed_space)
1428                 return 0;
1429
1430         free_space = free_start -
1431                 sizeof(struct ocfs2_xattr_header) -
1432                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1433                 OCFS2_XATTR_HEADER_GAP;
1434         if (free_space < 0)
1435                 return -EIO;
1436         if (free_space < needed_space)
1437                 return -ENOSPC;
1438
1439         return 0;
1440 }
1441
1442 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1443                                    int type)
1444 {
1445         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1446 }
1447
1448 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1449 {
1450         loc->xl_ops->xlo_journal_dirty(handle, loc);
1451 }
1452
1453 /* Give a pointer into the storage for the given offset */
1454 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1455 {
1456         BUG_ON(offset >= loc->xl_size);
1457         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1458 }
1459
1460 /*
1461  * Wipe the name+value pair and allow the storage to reclaim it.  This
1462  * must be followed by either removal of the entry or a call to
1463  * ocfs2_xa_add_namevalue().
1464  */
1465 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1466 {
1467         loc->xl_ops->xlo_wipe_namevalue(loc);
1468 }
1469
1470 /*
1471  * Find lowest offset to a name+value pair.  This is the start of our
1472  * downward-growing free space.
1473  */
1474 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1475 {
1476         return loc->xl_ops->xlo_get_free_start(loc);
1477 }
1478
1479 /* Can we reuse loc->xl_entry for xi? */
1480 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1481                                     struct ocfs2_xattr_info *xi)
1482 {
1483         return loc->xl_ops->xlo_can_reuse(loc, xi);
1484 }
1485
1486 /* How much free space is needed to set the new value */
1487 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1488                                 struct ocfs2_xattr_info *xi)
1489 {
1490         return loc->xl_ops->xlo_check_space(loc, xi);
1491 }
1492
1493 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1494                                    struct ocfs2_xattr_info *xi)
1495 {
1496         int size = namevalue_size_xi(xi);
1497         int nameval_offset;
1498         char *nameval_buf;
1499
1500         loc->xl_ops->xlo_add_namevalue(loc, size);
1501         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1502         loc->xl_entry->xe_name_len = xi->xi_name_len;
1503         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1504         ocfs2_xattr_set_local(loc->xl_entry,
1505                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1506
1507         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1508         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1509         memset(nameval_buf, 0, size);
1510         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1511 }
1512
1513 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1514                                     struct ocfs2_xattr_value_buf *vb)
1515 {
1516         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1517         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1518
1519         /* Value bufs are for value trees */
1520         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1521         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1522                (name_size + OCFS2_XATTR_ROOT_SIZE));
1523
1524         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1525         vb->vb_xv =
1526                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1527                                                         nameval_offset +
1528                                                         name_size);
1529 }
1530
1531 static int ocfs2_xa_block_journal_access(handle_t *handle,
1532                                          struct ocfs2_xa_loc *loc, int type)
1533 {
1534         struct buffer_head *bh = loc->xl_storage;
1535         ocfs2_journal_access_func access;
1536
1537         if (loc->xl_size == (bh->b_size -
1538                              offsetof(struct ocfs2_xattr_block,
1539                                       xb_attrs.xb_header)))
1540                 access = ocfs2_journal_access_xb;
1541         else
1542                 access = ocfs2_journal_access_di;
1543         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1544 }
1545
1546 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1547                                          struct ocfs2_xa_loc *loc)
1548 {
1549         struct buffer_head *bh = loc->xl_storage;
1550
1551         ocfs2_journal_dirty(handle, bh);
1552 }
1553
1554 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1555                                            int offset)
1556 {
1557         return (char *)loc->xl_header + offset;
1558 }
1559
1560 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1561                                     struct ocfs2_xattr_info *xi)
1562 {
1563         /*
1564          * Block storage is strict.  If the sizes aren't exact, we will
1565          * remove the old one and reinsert the new.
1566          */
1567         return namevalue_size_xe(loc->xl_entry) ==
1568                 namevalue_size_xi(xi);
1569 }
1570
1571 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1572 {
1573         struct ocfs2_xattr_header *xh = loc->xl_header;
1574         int i, count = le16_to_cpu(xh->xh_count);
1575         int offset, free_start = loc->xl_size;
1576
1577         for (i = 0; i < count; i++) {
1578                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1579                 if (offset < free_start)
1580                         free_start = offset;
1581         }
1582
1583         return free_start;
1584 }
1585
1586 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1587                                       struct ocfs2_xattr_info *xi)
1588 {
1589         int count = le16_to_cpu(loc->xl_header->xh_count);
1590         int free_start = ocfs2_xa_get_free_start(loc);
1591         int needed_space = ocfs2_xi_entry_usage(xi);
1592
1593         /*
1594          * Block storage will reclaim the original entry before inserting
1595          * the new value, so we only need the difference.  If the new
1596          * entry is smaller than the old one, we don't need anything.
1597          */
1598         if (loc->xl_entry) {
1599                 /* Don't need space if we're reusing! */
1600                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1601                         needed_space = 0;
1602                 else
1603                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1604         }
1605         if (needed_space < 0)
1606                 needed_space = 0;
1607         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1608 }
1609
1610 /*
1611  * Block storage for xattrs keeps the name+value pairs compacted.  When
1612  * we remove one, we have to shift any that preceded it towards the end.
1613  */
1614 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1615 {
1616         int i, offset;
1617         int namevalue_offset, first_namevalue_offset, namevalue_size;
1618         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1619         struct ocfs2_xattr_header *xh = loc->xl_header;
1620         int count = le16_to_cpu(xh->xh_count);
1621
1622         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1623         namevalue_size = namevalue_size_xe(entry);
1624         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1625
1626         /* Shift the name+value pairs */
1627         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1628                 (char *)xh + first_namevalue_offset,
1629                 namevalue_offset - first_namevalue_offset);
1630         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1631
1632         /* Now tell xh->xh_entries about it */
1633         for (i = 0; i < count; i++) {
1634                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1635                 if (offset <= namevalue_offset)
1636                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1637                                      namevalue_size);
1638         }
1639
1640         /*
1641          * Note that we don't update xh_free_start or xh_name_value_len
1642          * because they're not used in block-stored xattrs.
1643          */
1644 }
1645
1646 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1647 {
1648         int count = le16_to_cpu(loc->xl_header->xh_count);
1649         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1650         le16_add_cpu(&loc->xl_header->xh_count, 1);
1651         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1652 }
1653
1654 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1655 {
1656         int free_start = ocfs2_xa_get_free_start(loc);
1657
1658         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1659 }
1660
1661 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1662                                           struct ocfs2_xattr_value_buf *vb)
1663 {
1664         struct buffer_head *bh = loc->xl_storage;
1665
1666         if (loc->xl_size == (bh->b_size -
1667                              offsetof(struct ocfs2_xattr_block,
1668                                       xb_attrs.xb_header)))
1669                 vb->vb_access = ocfs2_journal_access_xb;
1670         else
1671                 vb->vb_access = ocfs2_journal_access_di;
1672         vb->vb_bh = bh;
1673 }
1674
1675 /*
1676  * Operations for xattrs stored in blocks.  This includes inline inode
1677  * storage and unindexed ocfs2_xattr_blocks.
1678  */
1679 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1680         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1681         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1682         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1683         .xlo_check_space        = ocfs2_xa_block_check_space,
1684         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1685         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1686         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1687         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1688         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1689         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1690 };
1691
1692 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1693                                           struct ocfs2_xa_loc *loc, int type)
1694 {
1695         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1696
1697         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1698 }
1699
1700 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1701                                           struct ocfs2_xa_loc *loc)
1702 {
1703         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1704
1705         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1706 }
1707
1708 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1709                                             int offset)
1710 {
1711         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1712         int block, block_offset;
1713
1714         /* The header is at the front of the bucket */
1715         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1716         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1717
1718         return bucket_block(bucket, block) + block_offset;
1719 }
1720
1721 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1722                                      struct ocfs2_xattr_info *xi)
1723 {
1724         return namevalue_size_xe(loc->xl_entry) >=
1725                 namevalue_size_xi(xi);
1726 }
1727
1728 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1729 {
1730         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1731         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1732 }
1733
1734 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1735                                          int free_start, int size)
1736 {
1737         /*
1738          * We need to make sure that the name+value pair fits within
1739          * one block.
1740          */
1741         if (((free_start - size) >> sb->s_blocksize_bits) !=
1742             ((free_start - 1) >> sb->s_blocksize_bits))
1743                 free_start -= free_start % sb->s_blocksize;
1744
1745         return free_start;
1746 }
1747
1748 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1749                                        struct ocfs2_xattr_info *xi)
1750 {
1751         int rc;
1752         int count = le16_to_cpu(loc->xl_header->xh_count);
1753         int free_start = ocfs2_xa_get_free_start(loc);
1754         int needed_space = ocfs2_xi_entry_usage(xi);
1755         int size = namevalue_size_xi(xi);
1756         struct super_block *sb = loc->xl_inode->i_sb;
1757
1758         /*
1759          * Bucket storage does not reclaim name+value pairs it cannot
1760          * reuse.  They live as holes until the bucket fills, and then
1761          * the bucket is defragmented.  However, the bucket can reclaim
1762          * the ocfs2_xattr_entry.
1763          */
1764         if (loc->xl_entry) {
1765                 /* Don't need space if we're reusing! */
1766                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1767                         needed_space = 0;
1768                 else
1769                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1770         }
1771         BUG_ON(needed_space < 0);
1772
1773         if (free_start < size) {
1774                 if (needed_space)
1775                         return -ENOSPC;
1776         } else {
1777                 /*
1778                  * First we check if it would fit in the first place.
1779                  * Below, we align the free start to a block.  This may
1780                  * slide us below the minimum gap.  By checking unaligned
1781                  * first, we avoid that error.
1782                  */
1783                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1784                                                  count);
1785                 if (rc)
1786                         return rc;
1787                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1788                                                            size);
1789         }
1790         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1791 }
1792
1793 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1794 {
1795         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1796                      -namevalue_size_xe(loc->xl_entry));
1797 }
1798
1799 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1800 {
1801         struct ocfs2_xattr_header *xh = loc->xl_header;
1802         int count = le16_to_cpu(xh->xh_count);
1803         int low = 0, high = count - 1, tmp;
1804         struct ocfs2_xattr_entry *tmp_xe;
1805
1806         /*
1807          * We keep buckets sorted by name_hash, so we need to find
1808          * our insert place.
1809          */
1810         while (low <= high && count) {
1811                 tmp = (low + high) / 2;
1812                 tmp_xe = &xh->xh_entries[tmp];
1813
1814                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1815                         low = tmp + 1;
1816                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1817                         high = tmp - 1;
1818                 else {
1819                         low = tmp;
1820                         break;
1821                 }
1822         }
1823
1824         if (low != count)
1825                 memmove(&xh->xh_entries[low + 1],
1826                         &xh->xh_entries[low],
1827                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1828
1829         le16_add_cpu(&xh->xh_count, 1);
1830         loc->xl_entry = &xh->xh_entries[low];
1831         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1832 }
1833
1834 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1835 {
1836         int free_start = ocfs2_xa_get_free_start(loc);
1837         struct ocfs2_xattr_header *xh = loc->xl_header;
1838         struct super_block *sb = loc->xl_inode->i_sb;
1839         int nameval_offset;
1840
1841         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1842         nameval_offset = free_start - size;
1843         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1844         xh->xh_free_start = cpu_to_le16(nameval_offset);
1845         le16_add_cpu(&xh->xh_name_value_len, size);
1846
1847 }
1848
1849 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1850                                            struct ocfs2_xattr_value_buf *vb)
1851 {
1852         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1853         struct super_block *sb = loc->xl_inode->i_sb;
1854         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1855         int size = namevalue_size_xe(loc->xl_entry);
1856         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1857
1858         /* Values are not allowed to straddle block boundaries */
1859         BUG_ON(block_offset !=
1860                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1861         /* We expect the bucket to be filled in */
1862         BUG_ON(!bucket->bu_bhs[block_offset]);
1863
1864         vb->vb_access = ocfs2_journal_access;
1865         vb->vb_bh = bucket->bu_bhs[block_offset];
1866 }
1867
1868 /* Operations for xattrs stored in buckets. */
1869 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1870         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1871         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1872         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1873         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1874         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1875         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1876         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1877         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1878         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1879         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1880 };
1881
1882 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1883 {
1884         struct ocfs2_xattr_value_buf vb;
1885
1886         if (ocfs2_xattr_is_local(loc->xl_entry))
1887                 return 0;
1888
1889         ocfs2_xa_fill_value_buf(loc, &vb);
1890         return le32_to_cpu(vb.vb_xv->xr_clusters);
1891 }
1892
1893 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1894                                    struct ocfs2_xattr_set_ctxt *ctxt)
1895 {
1896         int trunc_rc, access_rc;
1897         struct ocfs2_xattr_value_buf vb;
1898
1899         ocfs2_xa_fill_value_buf(loc, &vb);
1900         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1901                                               ctxt);
1902
1903         /*
1904          * The caller of ocfs2_xa_value_truncate() has already called
1905          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1906          * calls ocfs2_extend_trans().  This may commit the previous
1907          * transaction and open a new one.  If this is a bucket, truncate
1908          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1909          * the caller is expecting to dirty the entire bucket.  So we must
1910          * reset the journal work.  We do this even if truncate has failed,
1911          * as it could have failed after committing the extend.
1912          */
1913         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1914                                             OCFS2_JOURNAL_ACCESS_WRITE);
1915
1916         /* Errors in truncate take precedence */
1917         return trunc_rc ? trunc_rc : access_rc;
1918 }
1919
1920 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1921 {
1922         int index, count;
1923         struct ocfs2_xattr_header *xh = loc->xl_header;
1924         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1925
1926         ocfs2_xa_wipe_namevalue(loc);
1927         loc->xl_entry = NULL;
1928
1929         le16_add_cpu(&xh->xh_count, -1);
1930         count = le16_to_cpu(xh->xh_count);
1931
1932         /*
1933          * Only zero out the entry if there are more remaining.  This is
1934          * important for an empty bucket, as it keeps track of the
1935          * bucket's hash value.  It doesn't hurt empty block storage.
1936          */
1937         if (count) {
1938                 index = ((char *)entry - (char *)&xh->xh_entries) /
1939                         sizeof(struct ocfs2_xattr_entry);
1940                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1941                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1942                 memset(&xh->xh_entries[count], 0,
1943                        sizeof(struct ocfs2_xattr_entry));
1944         }
1945 }
1946
1947 /*
1948  * If we have a problem adjusting the size of an external value during
1949  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1950  * in an intermediate state.  For example, the value may be partially
1951  * truncated.
1952  *
1953  * If the value tree hasn't changed, the extend/truncate went nowhere.
1954  * We have nothing to do.  The caller can treat it as a straight error.
1955  *
1956  * If the value tree got partially truncated, we now have a corrupted
1957  * extended attribute.  We're going to wipe its entry and leak the
1958  * clusters.  Better to leak some storage than leave a corrupt entry.
1959  *
1960  * If the value tree grew, it obviously didn't grow enough for the
1961  * new entry.  We're not going to try and reclaim those clusters either.
1962  * If there was already an external value there (orig_clusters != 0),
1963  * the new clusters are attached safely and we can just leave the old
1964  * value in place.  If there was no external value there, we remove
1965  * the entry.
1966  *
1967  * This way, the xattr block we store in the journal will be consistent.
1968  * If the size change broke because of the journal, no changes will hit
1969  * disk anyway.
1970  */
1971 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1972                                             const char *what,
1973                                             unsigned int orig_clusters)
1974 {
1975         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1976         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1977                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1978
1979         if (new_clusters < orig_clusters) {
1980                 mlog(ML_ERROR,
1981                      "Partial truncate while %s xattr %.*s.  Leaking "
1982                      "%u clusters and removing the entry\n",
1983                      what, loc->xl_entry->xe_name_len, nameval_buf,
1984                      orig_clusters - new_clusters);
1985                 ocfs2_xa_remove_entry(loc);
1986         } else if (!orig_clusters) {
1987                 mlog(ML_ERROR,
1988                      "Unable to allocate an external value for xattr "
1989                      "%.*s safely.  Leaking %u clusters and removing the "
1990                      "entry\n",
1991                      loc->xl_entry->xe_name_len, nameval_buf,
1992                      new_clusters - orig_clusters);
1993                 ocfs2_xa_remove_entry(loc);
1994         } else if (new_clusters > orig_clusters)
1995                 mlog(ML_ERROR,
1996                      "Unable to grow xattr %.*s safely.  %u new clusters "
1997                      "have been added, but the value will not be "
1998                      "modified\n",
1999                      loc->xl_entry->xe_name_len, nameval_buf,
2000                      new_clusters - orig_clusters);
2001 }
2002
2003 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2004                            struct ocfs2_xattr_set_ctxt *ctxt)
2005 {
2006         int rc = 0;
2007         unsigned int orig_clusters;
2008
2009         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2010                 orig_clusters = ocfs2_xa_value_clusters(loc);
2011                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2012                 if (rc) {
2013                         mlog_errno(rc);
2014                         /*
2015                          * Since this is remove, we can return 0 if
2016                          * ocfs2_xa_cleanup_value_truncate() is going to
2017                          * wipe the entry anyway.  So we check the
2018                          * cluster count as well.
2019                          */
2020                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2021                                 rc = 0;
2022                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2023                                                         orig_clusters);
2024                         if (rc)
2025                                 goto out;
2026                 }
2027         }
2028
2029         ocfs2_xa_remove_entry(loc);
2030
2031 out:
2032         return rc;
2033 }
2034
2035 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2036 {
2037         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2038         char *nameval_buf;
2039
2040         nameval_buf = ocfs2_xa_offset_pointer(loc,
2041                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2042         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2043 }
2044
2045 /*
2046  * Take an existing entry and make it ready for the new value.  This
2047  * won't allocate space, but it may free space.  It should be ready for
2048  * ocfs2_xa_prepare_entry() to finish the work.
2049  */
2050 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2051                                 struct ocfs2_xattr_info *xi,
2052                                 struct ocfs2_xattr_set_ctxt *ctxt)
2053 {
2054         int rc = 0;
2055         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2056         unsigned int orig_clusters;
2057         char *nameval_buf;
2058         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2059         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2060
2061         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2062                name_size);
2063
2064         nameval_buf = ocfs2_xa_offset_pointer(loc,
2065                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2066         if (xe_local) {
2067                 memset(nameval_buf + name_size, 0,
2068                        namevalue_size_xe(loc->xl_entry) - name_size);
2069                 if (!xi_local)
2070                         ocfs2_xa_install_value_root(loc);
2071         } else {
2072                 orig_clusters = ocfs2_xa_value_clusters(loc);
2073                 if (xi_local) {
2074                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2075                         if (rc < 0)
2076                                 mlog_errno(rc);
2077                         else
2078                                 memset(nameval_buf + name_size, 0,
2079                                        namevalue_size_xe(loc->xl_entry) -
2080                                        name_size);
2081                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2082                            xi->xi_value_len) {
2083                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2084                                                      ctxt);
2085                         if (rc < 0)
2086                                 mlog_errno(rc);
2087                 }
2088
2089                 if (rc) {
2090                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2091                                                         orig_clusters);
2092                         goto out;
2093                 }
2094         }
2095
2096         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2097         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2098
2099 out:
2100         return rc;
2101 }
2102
2103 /*
2104  * Prepares loc->xl_entry to receive the new xattr.  This includes
2105  * properly setting up the name+value pair region.  If loc->xl_entry
2106  * already exists, it will take care of modifying it appropriately.
2107  *
2108  * Note that this modifies the data.  You did journal_access already,
2109  * right?
2110  */
2111 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2112                                   struct ocfs2_xattr_info *xi,
2113                                   u32 name_hash,
2114                                   struct ocfs2_xattr_set_ctxt *ctxt)
2115 {
2116         int rc = 0;
2117         unsigned int orig_clusters;
2118         __le64 orig_value_size = 0;
2119
2120         rc = ocfs2_xa_check_space(loc, xi);
2121         if (rc)
2122                 goto out;
2123
2124         if (!loc->xl_entry) {
2125                 rc = -EINVAL;
2126                 goto out;
2127         }
2128
2129         if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2130                 orig_value_size = loc->xl_entry->xe_value_size;
2131                 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2132                 if (rc)
2133                         goto out;
2134                 goto alloc_value;
2135         }
2136
2137         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2138                 orig_clusters = ocfs2_xa_value_clusters(loc);
2139                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2140                 if (rc) {
2141                         mlog_errno(rc);
2142                         ocfs2_xa_cleanup_value_truncate(loc,
2143                                                         "overwriting",
2144                                                         orig_clusters);
2145                         goto out;
2146                 }
2147         }
2148         ocfs2_xa_wipe_namevalue(loc);
2149
2150         /*
2151          * If we get here, we have a blank entry.  Fill it.  We grow our
2152          * name+value pair back from the end.
2153          */
2154         ocfs2_xa_add_namevalue(loc, xi);
2155         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2156                 ocfs2_xa_install_value_root(loc);
2157
2158 alloc_value:
2159         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2160                 orig_clusters = ocfs2_xa_value_clusters(loc);
2161                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2162                 if (rc < 0) {
2163                         ctxt->set_abort = 1;
2164                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2165                                                         orig_clusters);
2166                         /*
2167                          * If we were growing an existing value,
2168                          * ocfs2_xa_cleanup_value_truncate() won't remove
2169                          * the entry. We need to restore the original value
2170                          * size.
2171                          */
2172                         if (loc->xl_entry) {
2173                                 BUG_ON(!orig_value_size);
2174                                 loc->xl_entry->xe_value_size = orig_value_size;
2175                         }
2176                         mlog_errno(rc);
2177                 }
2178         }
2179
2180 out:
2181         return rc;
2182 }
2183
2184 /*
2185  * Store the value portion of the name+value pair.  This will skip
2186  * values that are stored externally.  Their tree roots were set up
2187  * by ocfs2_xa_prepare_entry().
2188  */
2189 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2190                                 struct ocfs2_xattr_info *xi,
2191                                 struct ocfs2_xattr_set_ctxt *ctxt)
2192 {
2193         int rc = 0;
2194         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2195         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2196         char *nameval_buf;
2197         struct ocfs2_xattr_value_buf vb;
2198
2199         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2200         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2201                 ocfs2_xa_fill_value_buf(loc, &vb);
2202                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2203                                                      ctxt->handle, &vb,
2204                                                      xi->xi_value,
2205                                                      xi->xi_value_len);
2206         } else
2207                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2208
2209         return rc;
2210 }
2211
2212 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2213                         struct ocfs2_xattr_info *xi,
2214                         struct ocfs2_xattr_set_ctxt *ctxt)
2215 {
2216         int ret;
2217         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2218                                               xi->xi_name_len);
2219
2220         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2221                                       OCFS2_JOURNAL_ACCESS_WRITE);
2222         if (ret) {
2223                 mlog_errno(ret);
2224                 goto out;
2225         }
2226
2227         /*
2228          * From here on out, everything is going to modify the buffer a
2229          * little.  Errors are going to leave the xattr header in a
2230          * sane state.  Thus, even with errors we dirty the sucker.
2231          */
2232
2233         /* Don't worry, we are never called with !xi_value and !xl_entry */
2234         if (!xi->xi_value) {
2235                 ret = ocfs2_xa_remove(loc, ctxt);
2236                 goto out_dirty;
2237         }
2238
2239         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2240         if (ret) {
2241                 if (ret != -ENOSPC)
2242                         mlog_errno(ret);
2243                 goto out_dirty;
2244         }
2245
2246         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2247         if (ret)
2248                 mlog_errno(ret);
2249
2250 out_dirty:
2251         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2252
2253 out:
2254         return ret;
2255 }
2256
2257 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2258                                      struct inode *inode,
2259                                      struct buffer_head *bh,
2260                                      struct ocfs2_xattr_entry *entry)
2261 {
2262         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2263
2264         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2265
2266         loc->xl_inode = inode;
2267         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2268         loc->xl_storage = bh;
2269         loc->xl_entry = entry;
2270         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2271         loc->xl_header =
2272                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2273                                               loc->xl_size);
2274 }
2275
2276 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2277                                           struct inode *inode,
2278                                           struct buffer_head *bh,
2279                                           struct ocfs2_xattr_entry *entry)
2280 {
2281         struct ocfs2_xattr_block *xb =
2282                 (struct ocfs2_xattr_block *)bh->b_data;
2283
2284         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2285
2286         loc->xl_inode = inode;
2287         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2288         loc->xl_storage = bh;
2289         loc->xl_header = &(xb->xb_attrs.xb_header);
2290         loc->xl_entry = entry;
2291         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2292                                              xb_attrs.xb_header);
2293 }
2294
2295 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2296                                            struct ocfs2_xattr_bucket *bucket,
2297                                            struct ocfs2_xattr_entry *entry)
2298 {
2299         loc->xl_inode = bucket->bu_inode;
2300         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2301         loc->xl_storage = bucket;
2302         loc->xl_header = bucket_xh(bucket);
2303         loc->xl_entry = entry;
2304         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2305 }
2306
2307 /*
2308  * In xattr remove, if it is stored outside and refcounted, we may have
2309  * the chance to split the refcount tree. So need the allocators.
2310  */
2311 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2312                                         struct ocfs2_xattr_value_root *xv,
2313                                         struct ocfs2_caching_info *ref_ci,
2314                                         struct buffer_head *ref_root_bh,
2315                                         struct ocfs2_alloc_context **meta_ac,
2316                                         int *ref_credits)
2317 {
2318         int ret, meta_add = 0;
2319         u32 p_cluster, num_clusters;
2320         unsigned int ext_flags;
2321
2322         *ref_credits = 0;
2323         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2324                                        &num_clusters,
2325                                        &xv->xr_list,
2326                                        &ext_flags);
2327         if (ret) {
2328                 mlog_errno(ret);
2329                 goto out;
2330         }
2331
2332         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2333                 goto out;
2334
2335         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2336                                                  ref_root_bh, xv,
2337                                                  &meta_add, ref_credits);
2338         if (ret) {
2339                 mlog_errno(ret);
2340                 goto out;
2341         }
2342
2343         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2344                                                 meta_add, meta_ac);
2345         if (ret)
2346                 mlog_errno(ret);
2347
2348 out:
2349         return ret;
2350 }
2351
2352 static int ocfs2_remove_value_outside(struct inode*inode,
2353                                       struct ocfs2_xattr_value_buf *vb,
2354                                       struct ocfs2_xattr_header *header,
2355                                       struct ocfs2_caching_info *ref_ci,
2356                                       struct buffer_head *ref_root_bh)
2357 {
2358         int ret = 0, i, ref_credits;
2359         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2360         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2361         void *val;
2362
2363         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2364
2365         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2366                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2367
2368                 if (ocfs2_xattr_is_local(entry))
2369                         continue;
2370
2371                 val = (void *)header +
2372                         le16_to_cpu(entry->xe_name_offset);
2373                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2374                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2375
2376                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2377                                                          ref_ci, ref_root_bh,
2378                                                          &ctxt.meta_ac,
2379                                                          &ref_credits);
2380
2381                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2382                                         ocfs2_remove_extent_credits(osb->sb));
2383                 if (IS_ERR(ctxt.handle)) {
2384                         ret = PTR_ERR(ctxt.handle);
2385                         mlog_errno(ret);
2386                         break;
2387                 }
2388
2389                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2390
2391                 ocfs2_commit_trans(osb, ctxt.handle);
2392                 if (ctxt.meta_ac) {
2393                         ocfs2_free_alloc_context(ctxt.meta_ac);
2394                         ctxt.meta_ac = NULL;
2395                 }
2396
2397                 if (ret < 0) {
2398                         mlog_errno(ret);
2399                         break;
2400                 }
2401
2402         }
2403
2404         if (ctxt.meta_ac)
2405                 ocfs2_free_alloc_context(ctxt.meta_ac);
2406         ocfs2_schedule_truncate_log_flush(osb, 1);
2407         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2408         return ret;
2409 }
2410
2411 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2412                                     struct buffer_head *di_bh,
2413                                     struct ocfs2_caching_info *ref_ci,
2414                                     struct buffer_head *ref_root_bh)
2415 {
2416
2417         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2418         struct ocfs2_xattr_header *header;
2419         int ret;
2420         struct ocfs2_xattr_value_buf vb = {
2421                 .vb_bh = di_bh,
2422                 .vb_access = ocfs2_journal_access_di,
2423         };
2424
2425         header = (struct ocfs2_xattr_header *)
2426                  ((void *)di + inode->i_sb->s_blocksize -
2427                  le16_to_cpu(di->i_xattr_inline_size));
2428
2429         ret = ocfs2_remove_value_outside(inode, &vb, header,
2430                                          ref_ci, ref_root_bh);
2431
2432         return ret;
2433 }
2434
2435 struct ocfs2_rm_xattr_bucket_para {
2436         struct ocfs2_caching_info *ref_ci;
2437         struct buffer_head *ref_root_bh;
2438 };
2439
2440 static int ocfs2_xattr_block_remove(struct inode *inode,
2441                                     struct buffer_head *blk_bh,
2442                                     struct ocfs2_caching_info *ref_ci,
2443                                     struct buffer_head *ref_root_bh)
2444 {
2445         struct ocfs2_xattr_block *xb;
2446         int ret = 0;
2447         struct ocfs2_xattr_value_buf vb = {
2448                 .vb_bh = blk_bh,
2449                 .vb_access = ocfs2_journal_access_xb,
2450         };
2451         struct ocfs2_rm_xattr_bucket_para args = {
2452                 .ref_ci = ref_ci,
2453                 .ref_root_bh = ref_root_bh,
2454         };
2455
2456         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2457         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2458                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2459                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2460                                                  ref_ci, ref_root_bh);
2461         } else
2462                 ret = ocfs2_iterate_xattr_index_block(inode,
2463                                                 blk_bh,
2464                                                 ocfs2_rm_xattr_cluster,
2465                                                 &args);
2466
2467         return ret;
2468 }
2469
2470 static int ocfs2_xattr_free_block(struct inode *inode,
2471                                   u64 block,
2472                                   struct ocfs2_caching_info *ref_ci,
2473                                   struct buffer_head *ref_root_bh)
2474 {
2475         struct inode *xb_alloc_inode;
2476         struct buffer_head *xb_alloc_bh = NULL;
2477         struct buffer_head *blk_bh = NULL;
2478         struct ocfs2_xattr_block *xb;
2479         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2480         handle_t *handle;
2481         int ret = 0;
2482         u64 blk, bg_blkno;
2483         u16 bit;
2484
2485         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2486         if (ret < 0) {
2487                 mlog_errno(ret);
2488                 goto out;
2489         }
2490
2491         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2492         if (ret < 0) {
2493                 mlog_errno(ret);
2494                 goto out;
2495         }
2496
2497         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2498         blk = le64_to_cpu(xb->xb_blkno);
2499         bit = le16_to_cpu(xb->xb_suballoc_bit);
2500         if (xb->xb_suballoc_loc)
2501                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2502         else
2503                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2504
2505         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2506                                 EXTENT_ALLOC_SYSTEM_INODE,
2507                                 le16_to_cpu(xb->xb_suballoc_slot));
2508         if (!xb_alloc_inode) {
2509                 ret = -ENOMEM;
2510                 mlog_errno(ret);
2511                 goto out;
2512         }
2513         inode_lock(xb_alloc_inode);
2514
2515         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2516         if (ret < 0) {
2517                 mlog_errno(ret);
2518                 goto out_mutex;
2519         }
2520
2521         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2522         if (IS_ERR(handle)) {
2523                 ret = PTR_ERR(handle);
2524                 mlog_errno(ret);
2525                 goto out_unlock;
2526         }
2527
2528         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2529                                        bit, bg_blkno, 1);
2530         if (ret < 0)
2531                 mlog_errno(ret);
2532
2533         ocfs2_commit_trans(osb, handle);
2534 out_unlock:
2535         ocfs2_inode_unlock(xb_alloc_inode, 1);
2536         brelse(xb_alloc_bh);
2537 out_mutex:
2538         inode_unlock(xb_alloc_inode);
2539         iput(xb_alloc_inode);
2540 out:
2541         brelse(blk_bh);
2542         return ret;
2543 }
2544
2545 /*
2546  * ocfs2_xattr_remove()
2547  *
2548  * Free extended attribute resources associated with this inode.
2549  */
2550 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2551 {
2552         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2553         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2554         struct ocfs2_refcount_tree *ref_tree = NULL;
2555         struct buffer_head *ref_root_bh = NULL;
2556         struct ocfs2_caching_info *ref_ci = NULL;
2557         handle_t *handle;
2558         int ret;
2559
2560         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2561                 return 0;
2562
2563         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2564                 return 0;
2565
2566         if (ocfs2_is_refcount_inode(inode)) {
2567                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2568                                                le64_to_cpu(di->i_refcount_loc),
2569                                                1, &ref_tree, &ref_root_bh);
2570                 if (ret) {
2571                         mlog_errno(ret);
2572                         goto out;
2573                 }
2574                 ref_ci = &ref_tree->rf_ci;
2575
2576         }
2577
2578         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2579                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2580                                                ref_ci, ref_root_bh);
2581                 if (ret < 0) {
2582                         mlog_errno(ret);
2583                         goto out;
2584                 }
2585         }
2586
2587         if (di->i_xattr_loc) {
2588                 ret = ocfs2_xattr_free_block(inode,
2589                                              le64_to_cpu(di->i_xattr_loc),
2590                                              ref_ci, ref_root_bh);
2591                 if (ret < 0) {
2592                         mlog_errno(ret);
2593                         goto out;
2594                 }
2595         }
2596
2597         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2598                                    OCFS2_INODE_UPDATE_CREDITS);
2599         if (IS_ERR(handle)) {
2600                 ret = PTR_ERR(handle);
2601                 mlog_errno(ret);
2602                 goto out;
2603         }
2604         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2605                                       OCFS2_JOURNAL_ACCESS_WRITE);
2606         if (ret) {
2607                 mlog_errno(ret);
2608                 goto out_commit;
2609         }
2610
2611         di->i_xattr_loc = 0;
2612
2613         spin_lock(&oi->ip_lock);
2614         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2615         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2616         spin_unlock(&oi->ip_lock);
2617         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2618
2619         ocfs2_journal_dirty(handle, di_bh);
2620 out_commit:
2621         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2622 out:
2623         if (ref_tree)
2624                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2625         brelse(ref_root_bh);
2626         return ret;
2627 }
2628
2629 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2630                                         struct ocfs2_dinode *di)
2631 {
2632         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2633         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2634         int free;
2635
2636         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2637                 return 0;
2638
2639         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2640                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2641                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2642         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2643                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2644                         le64_to_cpu(di->i_size);
2645         } else {
2646                 struct ocfs2_extent_list *el = &di->id2.i_list;
2647                 free = (le16_to_cpu(el->l_count) -
2648                         le16_to_cpu(el->l_next_free_rec)) *
2649                         sizeof(struct ocfs2_extent_rec);
2650         }
2651         if (free >= xattrsize)
2652                 return 1;
2653
2654         return 0;
2655 }
2656
2657 /*
2658  * ocfs2_xattr_ibody_find()
2659  *
2660  * Find extended attribute in inode block and
2661  * fill search info into struct ocfs2_xattr_search.
2662  */
2663 static int ocfs2_xattr_ibody_find(struct inode *inode,
2664                                   int name_index,
2665                                   const char *name,
2666                                   struct ocfs2_xattr_search *xs)
2667 {
2668         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2669         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2670         int ret;
2671         int has_space = 0;
2672
2673         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2674                 return 0;
2675
2676         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2677                 down_read(&oi->ip_alloc_sem);
2678                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2679                 up_read(&oi->ip_alloc_sem);
2680                 if (!has_space)
2681                         return 0;
2682         }
2683
2684         xs->xattr_bh = xs->inode_bh;
2685         xs->end = (void *)di + inode->i_sb->s_blocksize;
2686         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2687                 xs->header = (struct ocfs2_xattr_header *)
2688                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2689         else
2690                 xs->header = (struct ocfs2_xattr_header *)
2691                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2692         xs->base = (void *)xs->header;
2693         xs->here = xs->header->xh_entries;
2694
2695         /* Find the named attribute. */
2696         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2697                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2698                 if (ret && ret != -ENODATA)
2699                         return ret;
2700                 xs->not_found = ret;
2701         }
2702
2703         return 0;
2704 }
2705
2706 static int ocfs2_xattr_ibody_init(struct inode *inode,
2707                                   struct buffer_head *di_bh,
2708                                   struct ocfs2_xattr_set_ctxt *ctxt)
2709 {
2710         int ret;
2711         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2712         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2713         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2714         unsigned int xattrsize = osb->s_xattr_inline_size;
2715
2716         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2717                 ret = -ENOSPC;
2718                 goto out;
2719         }
2720
2721         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2722                                       OCFS2_JOURNAL_ACCESS_WRITE);
2723         if (ret) {
2724                 mlog_errno(ret);
2725                 goto out;
2726         }
2727
2728         /*
2729          * Adjust extent record count or inline data size
2730          * to reserve space for extended attribute.
2731          */
2732         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2733                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2734                 le16_add_cpu(&idata->id_count, -xattrsize);
2735         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2736                 struct ocfs2_extent_list *el = &di->id2.i_list;
2737                 le16_add_cpu(&el->l_count, -(xattrsize /
2738                                              sizeof(struct ocfs2_extent_rec)));
2739         }
2740         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2741
2742         spin_lock(&oi->ip_lock);
2743         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2744         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2745         spin_unlock(&oi->ip_lock);
2746
2747         ocfs2_journal_dirty(ctxt->handle, di_bh);
2748
2749 out:
2750         return ret;
2751 }
2752
2753 /*
2754  * ocfs2_xattr_ibody_set()
2755  *
2756  * Set, replace or remove an extended attribute into inode block.
2757  *
2758  */
2759 static int ocfs2_xattr_ibody_set(struct inode *inode,
2760                                  struct ocfs2_xattr_info *xi,
2761                                  struct ocfs2_xattr_search *xs,
2762                                  struct ocfs2_xattr_set_ctxt *ctxt)
2763 {
2764         int ret;
2765         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2766         struct ocfs2_xa_loc loc;
2767
2768         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2769                 return -ENOSPC;
2770
2771         down_write(&oi->ip_alloc_sem);
2772         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2773                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2774                 if (ret) {
2775                         if (ret != -ENOSPC)
2776                                 mlog_errno(ret);
2777                         goto out;
2778                 }
2779         }
2780
2781         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2782                                  xs->not_found ? NULL : xs->here);
2783         ret = ocfs2_xa_set(&loc, xi, ctxt);
2784         if (ret) {
2785                 if (ret != -ENOSPC)
2786                         mlog_errno(ret);
2787                 goto out;
2788         }
2789         xs->here = loc.xl_entry;
2790
2791 out:
2792         up_write(&oi->ip_alloc_sem);
2793
2794         return ret;
2795 }
2796
2797 /*
2798  * ocfs2_xattr_block_find()
2799  *
2800  * Find extended attribute in external block and
2801  * fill search info into struct ocfs2_xattr_search.
2802  */
2803 static int ocfs2_xattr_block_find(struct inode *inode,
2804                                   int name_index,
2805                                   const char *name,
2806                                   struct ocfs2_xattr_search *xs)
2807 {
2808         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2809         struct buffer_head *blk_bh = NULL;
2810         struct ocfs2_xattr_block *xb;
2811         int ret = 0;
2812
2813         if (!di->i_xattr_loc)
2814                 return ret;
2815
2816         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2817                                      &blk_bh);
2818         if (ret < 0) {
2819                 mlog_errno(ret);
2820                 return ret;
2821         }
2822
2823         xs->xattr_bh = blk_bh;
2824         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2825
2826         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2827                 xs->header = &xb->xb_attrs.xb_header;
2828                 xs->base = (void *)xs->header;
2829                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2830                 xs->here = xs->header->xh_entries;
2831
2832                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2833         } else
2834                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2835                                                    name_index,
2836                                                    name, xs);
2837
2838         if (ret && ret != -ENODATA) {
2839                 xs->xattr_bh = NULL;
2840                 goto cleanup;
2841         }
2842         xs->not_found = ret;
2843         return 0;
2844 cleanup:
2845         brelse(blk_bh);
2846
2847         return ret;
2848 }
2849
2850 static int ocfs2_create_xattr_block(struct inode *inode,
2851                                     struct buffer_head *inode_bh,
2852                                     struct ocfs2_xattr_set_ctxt *ctxt,
2853                                     int indexed,
2854                                     struct buffer_head **ret_bh)
2855 {
2856         int ret;
2857         u16 suballoc_bit_start;
2858         u32 num_got;
2859         u64 suballoc_loc, first_blkno;
2860         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2861         struct buffer_head *new_bh = NULL;
2862         struct ocfs2_xattr_block *xblk;
2863
2864         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2865                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2866         if (ret < 0) {
2867                 mlog_errno(ret);
2868                 goto end;
2869         }
2870
2871         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2872                                    &suballoc_loc, &suballoc_bit_start,
2873                                    &num_got, &first_blkno);
2874         if (ret < 0) {
2875                 mlog_errno(ret);
2876                 goto end;
2877         }
2878
2879         new_bh = sb_getblk(inode->i_sb, first_blkno);
2880         if (!new_bh) {
2881                 ret = -ENOMEM;
2882                 mlog_errno(ret);
2883                 goto end;
2884         }
2885
2886         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2887
2888         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2889                                       new_bh,
2890                                       OCFS2_JOURNAL_ACCESS_CREATE);
2891         if (ret < 0) {
2892                 mlog_errno(ret);
2893                 goto end;
2894         }
2895
2896         /* Initialize ocfs2_xattr_block */
2897         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2898         memset(xblk, 0, inode->i_sb->s_blocksize);
2899         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2900         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2901         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2902         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2903         xblk->xb_fs_generation =
2904                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2905         xblk->xb_blkno = cpu_to_le64(first_blkno);
2906         if (indexed) {
2907                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2908                 xr->xt_clusters = cpu_to_le32(1);
2909                 xr->xt_last_eb_blk = 0;
2910                 xr->xt_list.l_tree_depth = 0;
2911                 xr->xt_list.l_count = cpu_to_le16(
2912                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2913                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2914                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2915         }
2916         ocfs2_journal_dirty(ctxt->handle, new_bh);
2917
2918         /* Add it to the inode */
2919         di->i_xattr_loc = cpu_to_le64(first_blkno);
2920
2921         spin_lock(&OCFS2_I(inode)->ip_lock);
2922         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2923         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2924         spin_unlock(&OCFS2_I(inode)->ip_lock);
2925
2926         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2927
2928         *ret_bh = new_bh;
2929         new_bh = NULL;
2930
2931 end:
2932         brelse(new_bh);
2933         return ret;
2934 }
2935
2936 /*
2937  * ocfs2_xattr_block_set()
2938  *
2939  * Set, replace or remove an extended attribute into external block.
2940  *
2941  */
2942 static int ocfs2_xattr_block_set(struct inode *inode,
2943                                  struct ocfs2_xattr_info *xi,
2944                                  struct ocfs2_xattr_search *xs,
2945                                  struct ocfs2_xattr_set_ctxt *ctxt)
2946 {
2947         struct buffer_head *new_bh = NULL;
2948         struct ocfs2_xattr_block *xblk = NULL;
2949         int ret;
2950         struct ocfs2_xa_loc loc;
2951
2952         if (!xs->xattr_bh) {
2953                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2954                                                0, &new_bh);
2955                 if (ret) {
2956                         mlog_errno(ret);
2957                         goto end;
2958                 }
2959
2960                 xs->xattr_bh = new_bh;
2961                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2962                 xs->header = &xblk->xb_attrs.xb_header;
2963                 xs->base = (void *)xs->header;
2964                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2965                 xs->here = xs->header->xh_entries;
2966         } else
2967                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2968
2969         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2970                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2971                                               xs->not_found ? NULL : xs->here);
2972
2973                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2974                 if (!ret)
2975                         xs->here = loc.xl_entry;
2976                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2977                         goto end;
2978                 else {
2979                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2980                         if (ret)
2981                                 goto end;
2982                 }
2983         }
2984
2985         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2986                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2987
2988 end:
2989         return ret;
2990 }
2991
2992 /* Check whether the new xattr can be inserted into the inode. */
2993 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2994                                        struct ocfs2_xattr_info *xi,
2995                                        struct ocfs2_xattr_search *xs)
2996 {
2997         struct ocfs2_xattr_entry *last;
2998         int free, i;
2999         size_t min_offs = xs->end - xs->base;
3000
3001         if (!xs->header)
3002                 return 0;
3003
3004         last = xs->header->xh_entries;
3005
3006         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3007                 size_t offs = le16_to_cpu(last->xe_name_offset);
3008                 if (offs < min_offs)
3009                         min_offs = offs;
3010                 last += 1;
3011         }
3012
3013         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3014         if (free < 0)
3015                 return 0;
3016
3017         BUG_ON(!xs->not_found);
3018
3019         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3020                 return 1;
3021
3022         return 0;
3023 }
3024
3025 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3026                                      struct ocfs2_dinode *di,
3027                                      struct ocfs2_xattr_info *xi,
3028                                      struct ocfs2_xattr_search *xis,
3029                                      struct ocfs2_xattr_search *xbs,
3030                                      int *clusters_need,
3031                                      int *meta_need,
3032                                      int *credits_need)
3033 {
3034         int ret = 0, old_in_xb = 0;
3035         int clusters_add = 0, meta_add = 0, credits = 0;
3036         struct buffer_head *bh = NULL;
3037         struct ocfs2_xattr_block *xb = NULL;
3038         struct ocfs2_xattr_entry *xe = NULL;
3039         struct ocfs2_xattr_value_root *xv = NULL;
3040         char *base = NULL;
3041         int name_offset, name_len = 0;
3042         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3043                                                     xi->xi_value_len);
3044         u64 value_size;
3045
3046         /*
3047          * Calculate the clusters we need to write.
3048          * No matter whether we replace an old one or add a new one,
3049          * we need this for writing.
3050          */
3051         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3052                 credits += new_clusters *
3053                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3054
3055         if (xis->not_found && xbs->not_found) {
3056                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3057
3058                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3059                         clusters_add += new_clusters;
3060                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3061                                                         &def_xv.xv.xr_list);
3062                 }
3063
3064                 goto meta_guess;
3065         }
3066
3067         if (!xis->not_found) {
3068                 xe = xis->here;
3069                 name_offset = le16_to_cpu(xe->xe_name_offset);
3070                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3071                 base = xis->base;
3072                 credits += OCFS2_INODE_UPDATE_CREDITS;
3073         } else {
3074                 int i, block_off = 0;
3075                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3076                 xe = xbs->here;
3077                 name_offset = le16_to_cpu(xe->xe_name_offset);
3078                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3079                 i = xbs->here - xbs->header->xh_entries;
3080                 old_in_xb = 1;
3081
3082                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3083                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3084                                                         bucket_xh(xbs->bucket),
3085                                                         i, &block_off,
3086                                                         &name_offset);
3087                         base = bucket_block(xbs->bucket, block_off);
3088                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3089                 } else {
3090                         base = xbs->base;
3091                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3092                 }
3093         }
3094
3095         /*
3096          * delete a xattr doesn't need metadata and cluster allocation.
3097          * so just calculate the credits and return.
3098          *
3099          * The credits for removing the value tree will be extended
3100          * by ocfs2_remove_extent itself.
3101          */
3102         if (!xi->xi_value) {
3103                 if (!ocfs2_xattr_is_local(xe))
3104                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3105
3106                 goto out;
3107         }
3108
3109         /* do cluster allocation guess first. */
3110         value_size = le64_to_cpu(xe->xe_value_size);
3111
3112         if (old_in_xb) {
3113                 /*
3114                  * In xattr set, we always try to set the xe in inode first,
3115                  * so if it can be inserted into inode successfully, the old
3116                  * one will be removed from the xattr block, and this xattr
3117                  * will be inserted into inode as a new xattr in inode.
3118                  */
3119                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3120                         clusters_add += new_clusters;
3121                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3122                                     OCFS2_INODE_UPDATE_CREDITS;
3123                         if (!ocfs2_xattr_is_local(xe))
3124                                 credits += ocfs2_calc_extend_credits(
3125                                                         inode->i_sb,
3126                                                         &def_xv.xv.xr_list);
3127                         goto out;
3128                 }
3129         }
3130
3131         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3132                 /* the new values will be stored outside. */
3133                 u32 old_clusters = 0;
3134
3135                 if (!ocfs2_xattr_is_local(xe)) {
3136                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3137                                                                  value_size);
3138                         xv = (struct ocfs2_xattr_value_root *)
3139                              (base + name_offset + name_len);
3140                         value_size = OCFS2_XATTR_ROOT_SIZE;
3141                 } else
3142                         xv = &def_xv.xv;
3143
3144                 if (old_clusters >= new_clusters) {
3145                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3146                         goto out;
3147                 } else {
3148                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3149                         clusters_add += new_clusters - old_clusters;
3150                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3151                                                              &xv->xr_list);
3152                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3153                                 goto out;
3154                 }
3155         } else {
3156                 /*
3157                  * Now the new value will be stored inside. So if the new
3158                  * value is smaller than the size of value root or the old
3159                  * value, we don't need any allocation, otherwise we have
3160                  * to guess metadata allocation.
3161                  */
3162                 if ((ocfs2_xattr_is_local(xe) &&
3163                      (value_size >= xi->xi_value_len)) ||
3164                     (!ocfs2_xattr_is_local(xe) &&
3165                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3166                         goto out;
3167         }
3168
3169 meta_guess:
3170         /* calculate metadata allocation. */
3171         if (di->i_xattr_loc) {
3172                 if (!xbs->xattr_bh) {
3173                         ret = ocfs2_read_xattr_block(inode,
3174                                                      le64_to_cpu(di->i_xattr_loc),
3175                                                      &bh);
3176                         if (ret) {
3177                                 mlog_errno(ret);
3178                                 goto out;
3179                         }
3180
3181                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3182                 } else
3183                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3184
3185                 /*
3186                  * If there is already an xattr tree, good, we can calculate
3187                  * like other b-trees. Otherwise we may have the chance of
3188                  * create a tree, the credit calculation is borrowed from
3189                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3190                  * new tree will be cluster based, so no meta is needed.
3191                  */
3192                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3193                         struct ocfs2_extent_list *el =
3194                                  &xb->xb_attrs.xb_root.xt_list;
3195                         meta_add += ocfs2_extend_meta_needed(el);
3196                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3197                                                              el);
3198                 } else
3199                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3200
3201                 /*
3202                  * This cluster will be used either for new bucket or for
3203                  * new xattr block.
3204                  * If the cluster size is the same as the bucket size, one
3205                  * more is needed since we may need to extend the bucket
3206                  * also.
3207                  */
3208                 clusters_add += 1;
3209                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3210                 if (OCFS2_XATTR_BUCKET_SIZE ==
3211                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3212                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3213                         clusters_add += 1;
3214                 }
3215         } else {
3216                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3217                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3218                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3219                         meta_add += ocfs2_extend_meta_needed(el);
3220                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3221                                                              el);
3222                 } else {
3223                         meta_add += 1;
3224                 }
3225         }
3226 out:
3227         if (clusters_need)
3228                 *clusters_need = clusters_add;
3229         if (meta_need)
3230                 *meta_need = meta_add;
3231         if (credits_need)
3232                 *credits_need = credits;
3233         brelse(bh);
3234         return ret;
3235 }
3236
3237 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3238                                      struct ocfs2_dinode *di,
3239                                      struct ocfs2_xattr_info *xi,
3240                                      struct ocfs2_xattr_search *xis,
3241                                      struct ocfs2_xattr_search *xbs,
3242                                      struct ocfs2_xattr_set_ctxt *ctxt,
3243                                      int extra_meta,
3244                                      int *credits)
3245 {
3246         int clusters_add, meta_add, ret;
3247         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3248
3249         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3250
3251         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3252
3253         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3254                                         &clusters_add, &meta_add, credits);
3255         if (ret) {
3256                 mlog_errno(ret);
3257                 return ret;
3258         }
3259
3260         meta_add += extra_meta;
3261         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3262                                         clusters_add, *credits);
3263
3264         if (meta_add) {
3265                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3266                                                         &ctxt->meta_ac);
3267                 if (ret) {
3268                         mlog_errno(ret);
3269                         goto out;
3270                 }
3271         }
3272
3273         if (clusters_add) {
3274                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3275                 if (ret)
3276                         mlog_errno(ret);
3277         }
3278 out:
3279         if (ret) {
3280                 if (ctxt->meta_ac) {
3281                         ocfs2_free_alloc_context(ctxt->meta_ac);
3282                         ctxt->meta_ac = NULL;
3283                 }
3284
3285                 /*
3286                  * We cannot have an error and a non null ctxt->data_ac.
3287                  */
3288         }
3289
3290         return ret;
3291 }
3292
3293 static int __ocfs2_xattr_set_handle(struct inode *inode,
3294                                     struct ocfs2_dinode *di,
3295                                     struct ocfs2_xattr_info *xi,
3296                                     struct ocfs2_xattr_search *xis,
3297                                     struct ocfs2_xattr_search *xbs,
3298                                     struct ocfs2_xattr_set_ctxt *ctxt)
3299 {
3300         int ret = 0, credits, old_found;
3301
3302         if (!xi->xi_value) {
3303                 /* Remove existing extended attribute */
3304                 if (!xis->not_found)
3305                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3306                 else if (!xbs->not_found)
3307                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3308         } else {
3309                 /* We always try to set extended attribute into inode first*/
3310                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3311                 if (!ret && !xbs->not_found) {
3312                         /*
3313                          * If succeed and that extended attribute existing in
3314                          * external block, then we will remove it.
3315                          */
3316                         xi->xi_value = NULL;
3317                         xi->xi_value_len = 0;
3318
3319                         old_found = xis->not_found;
3320                         xis->not_found = -ENODATA;
3321                         ret = ocfs2_calc_xattr_set_need(inode,
3322                                                         di,
3323                                                         xi,
3324                                                         xis,
3325                                                         xbs,
3326                                                         NULL,
3327                                                         NULL,
3328                                                         &credits);
3329                         xis->not_found = old_found;
3330                         if (ret) {
3331                                 mlog_errno(ret);
3332                                 goto out;
3333                         }
3334
3335                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3336                         if (ret) {
3337                                 mlog_errno(ret);
3338                                 goto out;
3339                         }
3340                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3341                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3342                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3343                                 ret = ocfs2_xattr_block_find(inode,
3344                                                              xi->xi_name_index,
3345                                                              xi->xi_name, xbs);
3346                                 if (ret)
3347                                         goto out;
3348
3349                                 old_found = xis->not_found;
3350                                 xis->not_found = -ENODATA;
3351                                 ret = ocfs2_calc_xattr_set_need(inode,
3352                                                                 di,
3353                                                                 xi,
3354                                                                 xis,
3355                                                                 xbs,
3356                                                                 NULL,
3357                                                                 NULL,
3358                                                                 &credits);
3359                                 xis->not_found = old_found;
3360                                 if (ret) {
3361                                         mlog_errno(ret);
3362                                         goto out;
3363                                 }
3364
3365                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3366                                 if (ret) {
3367                                         mlog_errno(ret);
3368                                         goto out;
3369                                 }
3370                         }
3371                         /*
3372                          * If no space in inode, we will set extended attribute
3373                          * into external block.
3374                          */
3375                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3376                         if (ret)
3377                                 goto out;
3378                         if (!xis->not_found) {
3379                                 /*
3380                                  * If succeed and that extended attribute
3381                                  * existing in inode, we will remove it.
3382                                  */
3383                                 xi->xi_value = NULL;
3384                                 xi->xi_value_len = 0;
3385                                 xbs->not_found = -ENODATA;
3386                                 ret = ocfs2_calc_xattr_set_need(inode,
3387                                                                 di,
3388                                                                 xi,
3389                                                                 xis,
3390                                                                 xbs,
3391                                                                 NULL,
3392                                                                 NULL,
3393                                                                 &credits);
3394                                 if (ret) {
3395                                         mlog_errno(ret);
3396                                         goto out;
3397                                 }
3398
3399                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3400                                 if (ret) {
3401                                         mlog_errno(ret);
3402                                         goto out;
3403                                 }
3404                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3405                                                             xis, ctxt);
3406                         }
3407                 }
3408         }
3409
3410         if (!ret) {
3411                 /* Update inode ctime. */
3412                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3413                                               xis->inode_bh,
3414                                               OCFS2_JOURNAL_ACCESS_WRITE);
3415                 if (ret) {
3416                         mlog_errno(ret);
3417                         goto out;
3418                 }
3419
3420                 inode->i_ctime = current_time(inode);
3421                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3422                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3423                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3424         }
3425 out:
3426         return ret;
3427 }
3428
3429 /*
3430  * This function only called duing creating inode
3431  * for init security/acl xattrs of the new inode.
3432  * All transanction credits have been reserved in mknod.
3433  */
3434 int ocfs2_xattr_set_handle(handle_t *handle,
3435                            struct inode *inode,
3436                            struct buffer_head *di_bh,
3437                            int name_index,
3438                            const char *name,
3439                            const void *value,
3440                            size_t value_len,
3441                            int flags,
3442                            struct ocfs2_alloc_context *meta_ac,
3443                            struct ocfs2_alloc_context *data_ac)
3444 {
3445         struct ocfs2_dinode *di;
3446         int ret;
3447
3448         struct ocfs2_xattr_info xi = {
3449                 .xi_name_index = name_index,
3450                 .xi_name = name,
3451                 .xi_name_len = strlen(name),
3452                 .xi_value = value,
3453                 .xi_value_len = value_len,
3454         };
3455
3456         struct ocfs2_xattr_search xis = {
3457                 .not_found = -ENODATA,
3458         };
3459
3460         struct ocfs2_xattr_search xbs = {
3461                 .not_found = -ENODATA,
3462         };
3463
3464         struct ocfs2_xattr_set_ctxt ctxt = {
3465                 .handle = handle,
3466                 .meta_ac = meta_ac,
3467                 .data_ac = data_ac,
3468         };
3469
3470         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3471                 return -EOPNOTSUPP;
3472
3473         /*
3474          * In extreme situation, may need xattr bucket when
3475          * block size is too small. And we have already reserved
3476          * the credits for bucket in mknod.
3477          */
3478         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3479                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3480                 if (!xbs.bucket) {
3481                         mlog_errno(-ENOMEM);
3482                         return -ENOMEM;
3483                 }
3484         }
3485
3486         xis.inode_bh = xbs.inode_bh = di_bh;
3487         di = (struct ocfs2_dinode *)di_bh->b_data;
3488
3489         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3490
3491         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3492         if (ret)
3493                 goto cleanup;
3494         if (xis.not_found) {
3495                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3496                 if (ret)
3497                         goto cleanup;
3498         }
3499
3500         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3501
3502 cleanup:
3503         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3504         brelse(xbs.xattr_bh);
3505         ocfs2_xattr_bucket_free(xbs.bucket);
3506
3507         return ret;
3508 }
3509
3510 /*
3511  * ocfs2_xattr_set()
3512  *
3513  * Set, replace or remove an extended attribute for this inode.
3514  * value is NULL to remove an existing extended attribute, else either
3515  * create or replace an extended attribute.
3516  */
3517 int ocfs2_xattr_set(struct inode *inode,
3518                     int name_index,
3519                     const char *name,
3520                     const void *value,
3521                     size_t value_len,
3522                     int flags)
3523 {
3524         struct buffer_head *di_bh = NULL;
3525         struct ocfs2_dinode *di;
3526         int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3527         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3528         struct inode *tl_inode = osb->osb_tl_inode;
3529         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3530         struct ocfs2_refcount_tree *ref_tree = NULL;
3531         struct ocfs2_lock_holder oh;
3532
3533         struct ocfs2_xattr_info xi = {
3534                 .xi_name_index = name_index,
3535                 .xi_name = name,
3536                 .xi_name_len = strlen(name),
3537                 .xi_value = value,
3538                 .xi_value_len = value_len,
3539         };
3540
3541         struct ocfs2_xattr_search xis = {
3542                 .not_found = -ENODATA,
3543         };
3544
3545         struct ocfs2_xattr_search xbs = {
3546                 .not_found = -ENODATA,
3547         };
3548
3549         if (!ocfs2_supports_xattr(osb))
3550                 return -EOPNOTSUPP;
3551
3552         /*
3553          * Only xbs will be used on indexed trees.  xis doesn't need a
3554          * bucket.
3555          */
3556         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3557         if (!xbs.bucket) {
3558                 mlog_errno(-ENOMEM);
3559                 return -ENOMEM;
3560         }
3561
3562         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3563         if (had_lock < 0) {
3564                 ret = had_lock;
3565                 mlog_errno(ret);
3566                 goto cleanup_nolock;
3567         }
3568         xis.inode_bh = xbs.inode_bh = di_bh;
3569         di = (struct ocfs2_dinode *)di_bh->b_data;
3570
3571         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3572         /*
3573          * Scan inode and external block to find the same name
3574          * extended attribute and collect search information.
3575          */
3576         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3577         if (ret)
3578                 goto cleanup;
3579         if (xis.not_found) {
3580                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3581                 if (ret)
3582                         goto cleanup;
3583         }
3584
3585         if (xis.not_found && xbs.not_found) {
3586                 ret = -ENODATA;
3587                 if (flags & XATTR_REPLACE)
3588                         goto cleanup;
3589                 ret = 0;
3590                 if (!value)
3591                         goto cleanup;
3592         } else {
3593                 ret = -EEXIST;
3594                 if (flags & XATTR_CREATE)
3595                         goto cleanup;
3596         }
3597
3598         /* Check whether the value is refcounted and do some preparation. */
3599         if (ocfs2_is_refcount_inode(inode) &&
3600             (!xis.not_found || !xbs.not_found)) {
3601                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3602                                                    &xis, &xbs, &ref_tree,
3603                                                    &ref_meta, &ref_credits);
3604                 if (ret) {
3605                         mlog_errno(ret);
3606                         goto cleanup;
3607                 }
3608         }
3609
3610         inode_lock(tl_inode);
3611
3612         if (ocfs2_truncate_log_needs_flush(osb)) {
3613                 ret = __ocfs2_flush_truncate_log(osb);
3614                 if (ret < 0) {
3615                         inode_unlock(tl_inode);
3616                         mlog_errno(ret);
3617                         goto cleanup;
3618                 }
3619         }
3620         inode_unlock(tl_inode);
3621
3622         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3623                                         &xbs, &ctxt, ref_meta, &credits);
3624         if (ret) {
3625                 mlog_errno(ret);
3626                 goto cleanup;
3627         }
3628
3629         /* we need to update inode's ctime field, so add credit for it. */
3630         credits += OCFS2_INODE_UPDATE_CREDITS;
3631         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3632         if (IS_ERR(ctxt.handle)) {
3633                 ret = PTR_ERR(ctxt.handle);
3634                 mlog_errno(ret);
3635                 goto out_free_ac;
3636         }
3637
3638         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3639         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3640
3641         ocfs2_commit_trans(osb, ctxt.handle);
3642
3643 out_free_ac:
3644         if (ctxt.data_ac)
3645                 ocfs2_free_alloc_context(ctxt.data_ac);
3646         if (ctxt.meta_ac)
3647                 ocfs2_free_alloc_context(ctxt.meta_ac);
3648         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3649                 ocfs2_schedule_truncate_log_flush(osb, 1);
3650         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3651
3652 cleanup:
3653         if (ref_tree)
3654                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3655         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3656         if (!value && !ret) {
3657                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3658                 if (ret)
3659                         mlog_errno(ret);
3660         }
3661         ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3662 cleanup_nolock:
3663         brelse(di_bh);
3664         brelse(xbs.xattr_bh);
3665         ocfs2_xattr_bucket_free(xbs.bucket);
3666
3667         return ret;
3668 }
3669
3670 /*
3671  * Find the xattr extent rec which may contains name_hash.
3672  * e_cpos will be the first name hash of the xattr rec.
3673  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3674  */
3675 static int ocfs2_xattr_get_rec(struct inode *inode,
3676                                u32 name_hash,
3677                                u64 *p_blkno,
3678                                u32 *e_cpos,
3679                                u32 *num_clusters,
3680                                struct ocfs2_extent_list *el)
3681 {
3682         int ret = 0, i;
3683         struct buffer_head *eb_bh = NULL;
3684         struct ocfs2_extent_block *eb;
3685         struct ocfs2_extent_rec *rec = NULL;
3686         u64 e_blkno = 0;
3687
3688         if (el->l_tree_depth) {
3689                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3690                                       &eb_bh);
3691                 if (ret) {
3692                         mlog_errno(ret);
3693                         goto out;
3694                 }
3695
3696                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3697                 el = &eb->h_list;
3698
3699                 if (el->l_tree_depth) {
3700                         ret = ocfs2_error(inode->i_sb,
3701                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3702                                           inode->i_ino,
3703                                           (unsigned long long)eb_bh->b_blocknr);
3704                         goto out;
3705                 }
3706         }
3707
3708         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3709                 rec = &el->l_recs[i];
3710
3711                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3712                         e_blkno = le64_to_cpu(rec->e_blkno);
3713                         break;
3714                 }
3715         }
3716
3717         if (!e_blkno) {
3718                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3719                                   inode->i_ino,
3720                                   le32_to_cpu(rec->e_cpos),
3721                                   ocfs2_rec_clusters(el, rec));
3722                 goto out;
3723         }
3724
3725         *p_blkno = le64_to_cpu(rec->e_blkno);
3726         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3727         if (e_cpos)
3728                 *e_cpos = le32_to_cpu(rec->e_cpos);
3729 out:
3730         brelse(eb_bh);
3731         return ret;
3732 }
3733
3734 typedef int (xattr_bucket_func)(struct inode *inode,
3735                                 struct ocfs2_xattr_bucket *bucket,
3736                                 void *para);
3737
3738 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3739                                    struct ocfs2_xattr_bucket *bucket,
3740                                    int name_index,
3741                                    const char *name,
3742                                    u32 name_hash,
3743                                    u16 *xe_index,
3744                                    int *found)
3745 {
3746         int i, ret = 0, cmp = 1, block_off, new_offset;
3747         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3748         size_t name_len = strlen(name);
3749         struct ocfs2_xattr_entry *xe = NULL;
3750         char *xe_name;
3751
3752         /*
3753          * We don't use binary search in the bucket because there
3754          * may be multiple entries with the same name hash.
3755          */
3756         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3757                 xe = &xh->xh_entries[i];
3758
3759                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3760                         continue;
3761                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3762                         break;
3763
3764                 cmp = name_index - ocfs2_xattr_get_type(xe);
3765                 if (!cmp)
3766                         cmp = name_len - xe->xe_name_len;
3767                 if (cmp)
3768                         continue;
3769
3770                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3771                                                         xh,
3772                                                         i,
3773                                                         &block_off,
3774                                                         &new_offset);
3775                 if (ret) {
3776                         mlog_errno(ret);
3777                         break;
3778                 }
3779
3780
3781                 xe_name = bucket_block(bucket, block_off) + new_offset;
3782                 if (!memcmp(name, xe_name, name_len)) {
3783                         *xe_index = i;
3784                         *found = 1;
3785                         ret = 0;
3786                         break;
3787                 }
3788         }
3789
3790         return ret;
3791 }
3792
3793 /*
3794  * Find the specified xattr entry in a series of buckets.
3795  * This series start from p_blkno and last for num_clusters.
3796  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3797  * the num of the valid buckets.
3798  *
3799  * Return the buffer_head this xattr should reside in. And if the xattr's
3800  * hash is in the gap of 2 buckets, return the lower bucket.
3801  */
3802 static int ocfs2_xattr_bucket_find(struct inode *inode,
3803                                    int name_index,
3804                                    const char *name,
3805                                    u32 name_hash,
3806                                    u64 p_blkno,
3807                                    u32 first_hash,
3808                                    u32 num_clusters,
3809                                    struct ocfs2_xattr_search *xs)
3810 {
3811         int ret, found = 0;
3812         struct ocfs2_xattr_header *xh = NULL;
3813         struct ocfs2_xattr_entry *xe = NULL;
3814         u16 index = 0;
3815         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3816         int low_bucket = 0, bucket, high_bucket;
3817         struct ocfs2_xattr_bucket *search;
3818         u64 blkno, lower_blkno = 0;
3819
3820         search = ocfs2_xattr_bucket_new(inode);
3821         if (!search) {
3822                 ret = -ENOMEM;
3823                 mlog_errno(ret);
3824                 goto out;
3825         }
3826
3827         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3828         if (ret) {
3829                 mlog_errno(ret);
3830                 goto out;
3831         }
3832
3833         xh = bucket_xh(search);
3834         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3835         while (low_bucket <= high_bucket) {
3836                 ocfs2_xattr_bucket_relse(search);
3837
3838                 bucket = (low_bucket + high_bucket) / 2;
3839                 blkno = p_blkno + bucket * blk_per_bucket;
3840                 ret = ocfs2_read_xattr_bucket(search, blkno);
3841                 if (ret) {
3842                         mlog_errno(ret);
3843                         goto out;
3844                 }
3845
3846                 xh = bucket_xh(search);
3847                 xe = &xh->xh_entries[0];
3848                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3849                         high_bucket = bucket - 1;
3850                         continue;
3851                 }
3852
3853                 /*
3854                  * Check whether the hash of the last entry in our
3855                  * bucket is larger than the search one. for an empty
3856                  * bucket, the last one is also the first one.
3857                  */
3858                 if (xh->xh_count)
3859                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3860
3861                 /* record lower_blkno which may be the insert place. */
3862                 lower_blkno = blkno;
3863
3864                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3865                         low_bucket = bucket + 1;
3866                         continue;
3867                 }
3868
3869                 /* the searched xattr should reside in this bucket if exists. */
3870                 ret = ocfs2_find_xe_in_bucket(inode, search,
3871                                               name_index, name, name_hash,
3872                                               &index, &found);
3873                 if (ret) {
3874                         mlog_errno(ret);
3875                         goto out;
3876                 }
3877                 break;
3878         }
3879
3880         /*
3881          * Record the bucket we have found.
3882          * When the xattr's hash value is in the gap of 2 buckets, we will
3883          * always set it to the previous bucket.
3884          */
3885         if (!lower_blkno)
3886                 lower_blkno = p_blkno;
3887
3888         /* This should be in cache - we just read it during the search */
3889         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3890         if (ret) {
3891                 mlog_errno(ret);
3892                 goto out;
3893         }
3894
3895         xs->header = bucket_xh(xs->bucket);
3896         xs->base = bucket_block(xs->bucket, 0);
3897         xs->end = xs->base + inode->i_sb->s_blocksize;
3898
3899         if (found) {
3900                 xs->here = &xs->header->xh_entries[index];
3901                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3902                         name, name_index, name_hash,
3903                         (unsigned long long)bucket_blkno(xs->bucket),
3904                         index);
3905         } else
3906                 ret = -ENODATA;
3907
3908 out:
3909         ocfs2_xattr_bucket_free(search);
3910         return ret;
3911 }
3912
3913 static int ocfs2_xattr_index_block_find(struct inode *inode,
3914                                         struct buffer_head *root_bh,
3915                                         int name_index,
3916                                         const char *name,
3917                                         struct ocfs2_xattr_search *xs)
3918 {
3919         int ret;
3920         struct ocfs2_xattr_block *xb =
3921                         (struct ocfs2_xattr_block *)root_bh->b_data;
3922         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3923         struct ocfs2_extent_list *el = &xb_root->xt_list;
3924         u64 p_blkno = 0;
3925         u32 first_hash, num_clusters = 0;
3926         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3927
3928         if (le16_to_cpu(el->l_next_free_rec) == 0)
3929                 return -ENODATA;
3930
3931         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3932                                         name, name_index, name_hash,
3933                                         (unsigned long long)root_bh->b_blocknr,
3934                                         -1);
3935
3936         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3937                                   &num_clusters, el);
3938         if (ret) {
3939                 mlog_errno(ret);
3940                 goto out;
3941         }
3942
3943         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3944
3945         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3946                                         name, name_index, first_hash,
3947                                         (unsigned long long)p_blkno,
3948                                         num_clusters);
3949
3950         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3951                                       p_blkno, first_hash, num_clusters, xs);
3952
3953 out:
3954         return ret;
3955 }
3956
3957 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3958                                        u64 blkno,
3959                                        u32 clusters,
3960                                        xattr_bucket_func *func,
3961                                        void *para)
3962 {
3963         int i, ret = 0;
3964         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3965         u32 num_buckets = clusters * bpc;
3966         struct ocfs2_xattr_bucket *bucket;
3967
3968         bucket = ocfs2_xattr_bucket_new(inode);
3969         if (!bucket) {
3970                 mlog_errno(-ENOMEM);
3971                 return -ENOMEM;
3972         }
3973
3974         trace_ocfs2_iterate_xattr_buckets(
3975                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3976                 (unsigned long long)blkno, clusters);
3977
3978         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3979                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3980                 if (ret) {
3981                         mlog_errno(ret);
3982                         break;
3983                 }
3984
3985                 /*
3986                  * The real bucket num in this series of blocks is stored
3987                  * in the 1st bucket.
3988                  */
3989                 if (i == 0)
3990                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3991
3992                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3993                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3994                 if (func) {
3995                         ret = func(inode, bucket, para);
3996                         if (ret && ret != -ERANGE)
3997                                 mlog_errno(ret);
3998                         /* Fall through to bucket_relse() */
3999                 }
4000
4001                 ocfs2_xattr_bucket_relse(bucket);
4002                 if (ret)
4003                         break;
4004         }
4005
4006         ocfs2_xattr_bucket_free(bucket);
4007         return ret;
4008 }
4009
4010 struct ocfs2_xattr_tree_list {
4011         char *buffer;
4012         size_t buffer_size;
4013         size_t result;
4014 };
4015
4016 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4017                                              struct ocfs2_xattr_header *xh,
4018                                              int index,
4019                                              int *block_off,
4020                                              int *new_offset)
4021 {
4022         u16 name_offset;
4023
4024         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4025                 return -EINVAL;
4026
4027         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4028
4029         *block_off = name_offset >> sb->s_blocksize_bits;
4030         *new_offset = name_offset % sb->s_blocksize;
4031
4032         return 0;
4033 }
4034
4035 static int ocfs2_list_xattr_bucket(struct inode *inode,
4036                                    struct ocfs2_xattr_bucket *bucket,
4037                                    void *para)
4038 {
4039         int ret = 0, type;
4040         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4041         int i, block_off, new_offset;
4042         const char *name;
4043
4044         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4045                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4046                 type = ocfs2_xattr_get_type(entry);
4047
4048                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4049                                                         bucket_xh(bucket),
4050                                                         i,
4051                                                         &block_off,
4052                                                         &new_offset);
4053                 if (ret)
4054                         break;
4055
4056                 name = (const char *)bucket_block(bucket, block_off) +
4057                         new_offset;
4058                 ret = ocfs2_xattr_list_entry(inode->i_sb,
4059                                              xl->buffer,
4060                                              xl->buffer_size,
4061                                              &xl->result,
4062                                              type, name,
4063                                              entry->xe_name_len);
4064                 if (ret)
4065                         break;
4066         }
4067
4068         return ret;
4069 }
4070
4071 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4072                                            struct buffer_head *blk_bh,
4073                                            xattr_tree_rec_func *rec_func,
4074                                            void *para)
4075 {
4076         struct ocfs2_xattr_block *xb =
4077                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4078         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4079         int ret = 0;
4080         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4081         u64 p_blkno = 0;
4082
4083         if (!el->l_next_free_rec || !rec_func)
4084                 return 0;
4085
4086         while (name_hash > 0) {
4087                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4088                                           &e_cpos, &num_clusters, el);
4089                 if (ret) {
4090                         mlog_errno(ret);
4091                         break;
4092                 }
4093
4094                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4095                                num_clusters, para);
4096                 if (ret) {
4097                         if (ret != -ERANGE)
4098                                 mlog_errno(ret);
4099                         break;
4100                 }
4101
4102                 if (e_cpos == 0)
4103                         break;
4104
4105                 name_hash = e_cpos - 1;
4106         }
4107
4108         return ret;
4109
4110 }
4111
4112 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4113                                      struct buffer_head *root_bh,
4114                                      u64 blkno, u32 cpos, u32 len, void *para)
4115 {
4116         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4117                                            ocfs2_list_xattr_bucket, para);
4118 }
4119
4120 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4121                                              struct buffer_head *blk_bh,
4122                                              char *buffer,
4123                                              size_t buffer_size)
4124 {
4125         int ret;
4126         struct ocfs2_xattr_tree_list xl = {
4127                 .buffer = buffer,
4128                 .buffer_size = buffer_size,
4129                 .result = 0,
4130         };
4131
4132         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4133                                               ocfs2_list_xattr_tree_rec, &xl);
4134         if (ret) {
4135                 mlog_errno(ret);
4136                 goto out;
4137         }
4138
4139         ret = xl.result;
4140 out:
4141         return ret;
4142 }
4143
4144 static int cmp_xe(const void *a, const void *b)
4145 {
4146         const struct ocfs2_xattr_entry *l = a, *r = b;
4147         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4148         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4149
4150         if (l_hash > r_hash)
4151                 return 1;
4152         if (l_hash < r_hash)
4153                 return -1;
4154         return 0;
4155 }
4156
4157 static void swap_xe(void *a, void *b, int size)
4158 {
4159         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4160
4161         tmp = *l;
4162         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4163         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4164 }
4165
4166 /*
4167  * When the ocfs2_xattr_block is filled up, new bucket will be created
4168  * and all the xattr entries will be moved to the new bucket.
4169  * The header goes at the start of the bucket, and the names+values are
4170  * filled from the end.  This is why *target starts as the last buffer.
4171  * Note: we need to sort the entries since they are not saved in order
4172  * in the ocfs2_xattr_block.
4173  */
4174 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4175                                            struct buffer_head *xb_bh,
4176                                            struct ocfs2_xattr_bucket *bucket)
4177 {
4178         int i, blocksize = inode->i_sb->s_blocksize;
4179         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4180         u16 offset, size, off_change;
4181         struct ocfs2_xattr_entry *xe;
4182         struct ocfs2_xattr_block *xb =
4183                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4184         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4185         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4186         u16 count = le16_to_cpu(xb_xh->xh_count);
4187         char *src = xb_bh->b_data;
4188         char *target = bucket_block(bucket, blks - 1);
4189
4190         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4191                                 (unsigned long long)xb_bh->b_blocknr,
4192                                 (unsigned long long)bucket_blkno(bucket));
4193
4194         for (i = 0; i < blks; i++)
4195                 memset(bucket_block(bucket, i), 0, blocksize);
4196
4197         /*
4198          * Since the xe_name_offset is based on ocfs2_xattr_header,
4199          * there is a offset change corresponding to the change of
4200          * ocfs2_xattr_header's position.
4201          */
4202         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4203         xe = &xb_xh->xh_entries[count - 1];
4204         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4205         size = blocksize - offset;
4206
4207         /* copy all the names and values. */
4208         memcpy(target + offset, src + offset, size);
4209
4210         /* Init new header now. */
4211         xh->xh_count = xb_xh->xh_count;
4212         xh->xh_num_buckets = cpu_to_le16(1);
4213         xh->xh_name_value_len = cpu_to_le16(size);
4214         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4215
4216         /* copy all the entries. */
4217         target = bucket_block(bucket, 0);
4218         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4219         size = count * sizeof(struct ocfs2_xattr_entry);
4220         memcpy(target + offset, (char *)xb_xh + offset, size);
4221
4222         /* Change the xe offset for all the xe because of the move. */
4223         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4224                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4225         for (i = 0; i < count; i++)
4226                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4227
4228         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4229
4230         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4231              cmp_xe, swap_xe);
4232 }
4233
4234 /*
4235  * After we move xattr from block to index btree, we have to
4236  * update ocfs2_xattr_search to the new xe and base.
4237  *
4238  * When the entry is in xattr block, xattr_bh indicates the storage place.
4239  * While if the entry is in index b-tree, "bucket" indicates the
4240  * real place of the xattr.
4241  */
4242 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4243                                             struct ocfs2_xattr_search *xs,
4244                                             struct buffer_head *old_bh)
4245 {
4246         char *buf = old_bh->b_data;
4247         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4248         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4249         int i;
4250
4251         xs->header = bucket_xh(xs->bucket);
4252         xs->base = bucket_block(xs->bucket, 0);
4253         xs->end = xs->base + inode->i_sb->s_blocksize;
4254
4255         if (xs->not_found)
4256                 return;
4257
4258         i = xs->here - old_xh->xh_entries;
4259         xs->here = &xs->header->xh_entries[i];
4260 }
4261
4262 static int ocfs2_xattr_create_index_block(struct inode *inode,
4263                                           struct ocfs2_xattr_search *xs,
4264                                           struct ocfs2_xattr_set_ctxt *ctxt)
4265 {
4266         int ret;
4267         u32 bit_off, len;
4268         u64 blkno;
4269         handle_t *handle = ctxt->handle;
4270         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4271         struct buffer_head *xb_bh = xs->xattr_bh;
4272         struct ocfs2_xattr_block *xb =
4273                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4274         struct ocfs2_xattr_tree_root *xr;
4275         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4276
4277         trace_ocfs2_xattr_create_index_block_begin(
4278                                 (unsigned long long)xb_bh->b_blocknr);
4279
4280         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4281         BUG_ON(!xs->bucket);
4282
4283         /*
4284          * XXX:
4285          * We can use this lock for now, and maybe move to a dedicated mutex
4286          * if performance becomes a problem later.
4287          */
4288         down_write(&oi->ip_alloc_sem);
4289
4290         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4291                                       OCFS2_JOURNAL_ACCESS_WRITE);
4292         if (ret) {
4293                 mlog_errno(ret);
4294                 goto out;
4295         }
4296
4297         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4298                                      1, 1, &bit_off, &len);
4299         if (ret) {
4300                 mlog_errno(ret);
4301                 goto out;
4302         }
4303
4304         /*
4305          * The bucket may spread in many blocks, and
4306          * we will only touch the 1st block and the last block
4307          * in the whole bucket(one for entry and one for data).
4308          */
4309         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4310
4311         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4312
4313         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4314         if (ret) {
4315                 mlog_errno(ret);
4316                 goto out;
4317         }
4318
4319         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4320                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4321         if (ret) {
4322                 mlog_errno(ret);
4323                 goto out;
4324         }
4325
4326         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4327         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4328
4329         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4330
4331         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4332         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4333                offsetof(struct ocfs2_xattr_block, xb_attrs));
4334
4335         xr = &xb->xb_attrs.xb_root;
4336         xr->xt_clusters = cpu_to_le32(1);
4337         xr->xt_last_eb_blk = 0;
4338         xr->xt_list.l_tree_depth = 0;
4339         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4340         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4341
4342         xr->xt_list.l_recs[0].e_cpos = 0;
4343         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4344         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4345
4346         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4347
4348         ocfs2_journal_dirty(handle, xb_bh);
4349
4350 out:
4351         up_write(&oi->ip_alloc_sem);
4352
4353         return ret;
4354 }
4355
4356 static int cmp_xe_offset(const void *a, const void *b)
4357 {
4358         const struct ocfs2_xattr_entry *l = a, *r = b;
4359         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4360         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4361
4362         if (l_name_offset < r_name_offset)
4363                 return 1;
4364         if (l_name_offset > r_name_offset)
4365                 return -1;
4366         return 0;
4367 }
4368
4369 /*
4370  * defrag a xattr bucket if we find that the bucket has some
4371  * holes beteen name/value pairs.
4372  * We will move all the name/value pairs to the end of the bucket
4373  * so that we can spare some space for insertion.
4374  */
4375 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4376                                      handle_t *handle,
4377                                      struct ocfs2_xattr_bucket *bucket)
4378 {
4379         int ret, i;
4380         size_t end, offset, len;
4381         struct ocfs2_xattr_header *xh;
4382         char *entries, *buf, *bucket_buf = NULL;
4383         u64 blkno = bucket_blkno(bucket);
4384         u16 xh_free_start;
4385         size_t blocksize = inode->i_sb->s_blocksize;
4386         struct ocfs2_xattr_entry *xe;
4387
4388         /*
4389          * In order to make the operation more efficient and generic,
4390          * we copy all the blocks into a contiguous memory and do the
4391          * defragment there, so if anything is error, we will not touch
4392          * the real block.
4393          */
4394         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4395         if (!bucket_buf) {
4396                 ret = -EIO;
4397                 goto out;
4398         }
4399
4400         buf = bucket_buf;
4401         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4402                 memcpy(buf, bucket_block(bucket, i), blocksize);
4403
4404         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4405                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4406         if (ret < 0) {
4407                 mlog_errno(ret);
4408                 goto out;
4409         }
4410
4411         xh = (struct ocfs2_xattr_header *)bucket_buf;
4412         entries = (char *)xh->xh_entries;
4413         xh_free_start = le16_to_cpu(xh->xh_free_start);
4414
4415         trace_ocfs2_defrag_xattr_bucket(
4416              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4417              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4418
4419         /*
4420          * sort all the entries by their offset.
4421          * the largest will be the first, so that we can
4422          * move them to the end one by one.
4423          */
4424         sort(entries, le16_to_cpu(xh->xh_count),
4425              sizeof(struct ocfs2_xattr_entry),
4426              cmp_xe_offset, swap_xe);
4427
4428         /* Move all name/values to the end of the bucket. */
4429         xe = xh->xh_entries;
4430         end = OCFS2_XATTR_BUCKET_SIZE;
4431         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4432                 offset = le16_to_cpu(xe->xe_name_offset);
4433                 len = namevalue_size_xe(xe);
4434
4435                 /*
4436                  * We must make sure that the name/value pair
4437                  * exist in the same block. So adjust end to
4438                  * the previous block end if needed.
4439                  */
4440                 if (((end - len) / blocksize !=
4441                         (end - 1) / blocksize))
4442                         end = end - end % blocksize;
4443
4444                 if (end > offset + len) {
4445                         memmove(bucket_buf + end - len,
4446                                 bucket_buf + offset, len);
4447                         xe->xe_name_offset = cpu_to_le16(end - len);
4448                 }
4449
4450                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4451                                 "bucket %llu\n", (unsigned long long)blkno);
4452
4453                 end -= len;
4454         }
4455
4456         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4457                         "bucket %llu\n", (unsigned long long)blkno);
4458
4459         if (xh_free_start == end)
4460                 goto out;
4461
4462         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4463         xh->xh_free_start = cpu_to_le16(end);
4464
4465         /* sort the entries by their name_hash. */
4466         sort(entries, le16_to_cpu(xh->xh_count),
4467              sizeof(struct ocfs2_xattr_entry),
4468              cmp_xe, swap_xe);
4469
4470         buf = bucket_buf;
4471         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4472                 memcpy(bucket_block(bucket, i), buf, blocksize);
4473         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4474
4475 out:
4476         kfree(bucket_buf);
4477         return ret;
4478 }
4479
4480 /*
4481  * prev_blkno points to the start of an existing extent.  new_blkno
4482  * points to a newly allocated extent.  Because we know each of our
4483  * clusters contains more than bucket, we can easily split one cluster
4484  * at a bucket boundary.  So we take the last cluster of the existing
4485  * extent and split it down the middle.  We move the last half of the
4486  * buckets in the last cluster of the existing extent over to the new
4487  * extent.
4488  *
4489  * first_bh is the buffer at prev_blkno so we can update the existing
4490  * extent's bucket count.  header_bh is the bucket were we were hoping
4491  * to insert our xattr.  If the bucket move places the target in the new
4492  * extent, we'll update first_bh and header_bh after modifying the old
4493  * extent.
4494  *
4495  * first_hash will be set as the 1st xe's name_hash in the new extent.
4496  */
4497 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4498                                                handle_t *handle,
4499                                                struct ocfs2_xattr_bucket *first,
4500                                                struct ocfs2_xattr_bucket *target,
4501                                                u64 new_blkno,
4502                                                u32 num_clusters,
4503                                                u32 *first_hash)
4504 {
4505         int ret;
4506         struct super_block *sb = inode->i_sb;
4507         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4508         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4509         int to_move = num_buckets / 2;
4510         u64 src_blkno;
4511         u64 last_cluster_blkno = bucket_blkno(first) +
4512                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4513
4514         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4515         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4516
4517         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4518                                 (unsigned long long)last_cluster_blkno,
4519                                 (unsigned long long)new_blkno);
4520
4521         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4522                                      last_cluster_blkno, new_blkno,
4523                                      to_move, first_hash);
4524         if (ret) {
4525                 mlog_errno(ret);
4526                 goto out;
4527         }
4528
4529         /* This is the first bucket that got moved */
4530         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4531
4532         /*
4533          * If the target bucket was part of the moved buckets, we need to
4534          * update first and target.
4535          */
4536         if (bucket_blkno(target) >= src_blkno) {
4537                 /* Find the block for the new target bucket */
4538                 src_blkno = new_blkno +
4539                         (bucket_blkno(target) - src_blkno);
4540
4541                 ocfs2_xattr_bucket_relse(first);
4542                 ocfs2_xattr_bucket_relse(target);
4543
4544                 /*
4545                  * These shouldn't fail - the buffers are in the
4546                  * journal from ocfs2_cp_xattr_bucket().
4547                  */
4548                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4549                 if (ret) {
4550                         mlog_errno(ret);
4551                         goto out;
4552                 }
4553                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4554                 if (ret)
4555                         mlog_errno(ret);
4556
4557         }
4558
4559 out:
4560         return ret;
4561 }
4562
4563 /*
4564  * Find the suitable pos when we divide a bucket into 2.
4565  * We have to make sure the xattrs with the same hash value exist
4566  * in the same bucket.
4567  *
4568  * If this ocfs2_xattr_header covers more than one hash value, find a
4569  * place where the hash value changes.  Try to find the most even split.
4570  * The most common case is that all entries have different hash values,
4571  * and the first check we make will find a place to split.
4572  */
4573 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4574 {
4575         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4576         int count = le16_to_cpu(xh->xh_count);
4577         int delta, middle = count / 2;
4578
4579         /*
4580          * We start at the middle.  Each step gets farther away in both
4581          * directions.  We therefore hit the change in hash value
4582          * nearest to the middle.  Note that this loop does not execute for
4583          * count < 2.
4584          */
4585         for (delta = 0; delta < middle; delta++) {
4586                 /* Let's check delta earlier than middle */
4587                 if (cmp_xe(&entries[middle - delta - 1],
4588                            &entries[middle - delta]))
4589                         return middle - delta;
4590
4591                 /* For even counts, don't walk off the end */
4592                 if ((middle + delta + 1) == count)
4593                         continue;
4594
4595                 /* Now try delta past middle */
4596                 if (cmp_xe(&entries[middle + delta],
4597                            &entries[middle + delta + 1]))
4598                         return middle + delta + 1;
4599         }
4600
4601         /* Every entry had the same hash */
4602         return count;
4603 }
4604
4605 /*
4606  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4607  * first_hash will record the 1st hash of the new bucket.
4608  *
4609  * Normally half of the xattrs will be moved.  But we have to make
4610  * sure that the xattrs with the same hash value are stored in the
4611  * same bucket. If all the xattrs in this bucket have the same hash
4612  * value, the new bucket will be initialized as an empty one and the
4613  * first_hash will be initialized as (hash_value+1).
4614  */
4615 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4616                                     handle_t *handle,
4617                                     u64 blk,
4618                                     u64 new_blk,
4619                                     u32 *first_hash,
4620                                     int new_bucket_head)
4621 {
4622         int ret, i;
4623         int count, start, len, name_value_len = 0, name_offset = 0;
4624         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4625         struct ocfs2_xattr_header *xh;
4626         struct ocfs2_xattr_entry *xe;
4627         int blocksize = inode->i_sb->s_blocksize;
4628
4629         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4630                                               (unsigned long long)new_blk);
4631
4632         s_bucket = ocfs2_xattr_bucket_new(inode);
4633         t_bucket = ocfs2_xattr_bucket_new(inode);
4634         if (!s_bucket || !t_bucket) {
4635                 ret = -ENOMEM;
4636                 mlog_errno(ret);
4637                 goto out;
4638         }
4639
4640         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4641         if (ret) {
4642                 mlog_errno(ret);
4643                 goto out;
4644         }
4645
4646         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4647                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4648         if (ret) {
4649                 mlog_errno(ret);
4650                 goto out;
4651         }
4652
4653         /*
4654          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4655          * there's no need to read it.
4656          */
4657         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4658         if (ret) {
4659                 mlog_errno(ret);
4660                 goto out;
4661         }
4662
4663         /*
4664          * Hey, if we're overwriting t_bucket, what difference does
4665          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4666          * same part of ocfs2_cp_xattr_bucket().
4667          */
4668         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4669                                                 new_bucket_head ?
4670                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4671                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4672         if (ret) {
4673                 mlog_errno(ret);
4674                 goto out;
4675         }
4676
4677         xh = bucket_xh(s_bucket);
4678         count = le16_to_cpu(xh->xh_count);
4679         start = ocfs2_xattr_find_divide_pos(xh);
4680
4681         if (start == count) {
4682                 xe = &xh->xh_entries[start-1];
4683
4684                 /*
4685                  * initialized a new empty bucket here.
4686                  * The hash value is set as one larger than
4687                  * that of the last entry in the previous bucket.
4688                  */
4689                 for (i = 0; i < t_bucket->bu_blocks; i++)
4690                         memset(bucket_block(t_bucket, i), 0, blocksize);
4691
4692                 xh = bucket_xh(t_bucket);
4693                 xh->xh_free_start = cpu_to_le16(blocksize);
4694                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4695                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4696
4697                 goto set_num_buckets;
4698         }
4699
4700         /* copy the whole bucket to the new first. */
4701         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4702
4703         /* update the new bucket. */
4704         xh = bucket_xh(t_bucket);
4705
4706         /*
4707          * Calculate the total name/value len and xh_free_start for
4708          * the old bucket first.
4709          */
4710         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4711         name_value_len = 0;
4712         for (i = 0; i < start; i++) {
4713                 xe = &xh->xh_entries[i];
4714                 name_value_len += namevalue_size_xe(xe);
4715                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4716                         name_offset = le16_to_cpu(xe->xe_name_offset);
4717         }
4718
4719         /*
4720          * Now begin the modification to the new bucket.
4721          *
4722          * In the new bucket, We just move the xattr entry to the beginning
4723          * and don't touch the name/value. So there will be some holes in the
4724          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4725          * called.
4726          */
4727         xe = &xh->xh_entries[start];
4728         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4729         trace_ocfs2_divide_xattr_bucket_move(len,
4730                         (int)((char *)xe - (char *)xh),
4731                         (int)((char *)xh->xh_entries - (char *)xh));
4732         memmove((char *)xh->xh_entries, (char *)xe, len);
4733         xe = &xh->xh_entries[count - start];
4734         len = sizeof(struct ocfs2_xattr_entry) * start;
4735         memset((char *)xe, 0, len);
4736
4737         le16_add_cpu(&xh->xh_count, -start);
4738         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4739
4740         /* Calculate xh_free_start for the new bucket. */
4741         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4742         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4743                 xe = &xh->xh_entries[i];
4744                 if (le16_to_cpu(xe->xe_name_offset) <
4745                     le16_to_cpu(xh->xh_free_start))
4746                         xh->xh_free_start = xe->xe_name_offset;
4747         }
4748
4749 set_num_buckets:
4750         /* set xh->xh_num_buckets for the new xh. */
4751         if (new_bucket_head)
4752                 xh->xh_num_buckets = cpu_to_le16(1);
4753         else
4754                 xh->xh_num_buckets = 0;
4755
4756         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4757
4758         /* store the first_hash of the new bucket. */
4759         if (first_hash)
4760                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4761
4762         /*
4763          * Now only update the 1st block of the old bucket.  If we
4764          * just added a new empty bucket, there is no need to modify
4765          * it.
4766          */
4767         if (start == count)
4768                 goto out;
4769
4770         xh = bucket_xh(s_bucket);
4771         memset(&xh->xh_entries[start], 0,
4772                sizeof(struct ocfs2_xattr_entry) * (count - start));
4773         xh->xh_count = cpu_to_le16(start);
4774         xh->xh_free_start = cpu_to_le16(name_offset);
4775         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4776
4777         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4778
4779 out:
4780         ocfs2_xattr_bucket_free(s_bucket);
4781         ocfs2_xattr_bucket_free(t_bucket);
4782
4783         return ret;
4784 }
4785
4786 /*
4787  * Copy xattr from one bucket to another bucket.
4788  *
4789  * The caller must make sure that the journal transaction
4790  * has enough space for journaling.
4791  */
4792 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4793                                  handle_t *handle,
4794                                  u64 s_blkno,
4795                                  u64 t_blkno,
4796                                  int t_is_new)
4797 {
4798         int ret;
4799         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4800
4801         BUG_ON(s_blkno == t_blkno);
4802
4803         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4804                                     (unsigned long long)t_blkno,
4805                                     t_is_new);
4806
4807         s_bucket = ocfs2_xattr_bucket_new(inode);
4808         t_bucket = ocfs2_xattr_bucket_new(inode);
4809         if (!s_bucket || !t_bucket) {
4810                 ret = -ENOMEM;
4811                 mlog_errno(ret);
4812                 goto out;
4813         }
4814
4815         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4816         if (ret)
4817                 goto out;
4818
4819         /*
4820          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4821          * there's no need to read it.
4822          */
4823         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4824         if (ret)
4825                 goto out;
4826
4827         /*
4828          * Hey, if we're overwriting t_bucket, what difference does
4829          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4830          * cluster to fill, we came here from
4831          * ocfs2_mv_xattr_buckets(), and it is really new -
4832          * ACCESS_CREATE is required.  But we also might have moved data
4833          * out of t_bucket before extending back into it.
4834          * ocfs2_add_new_xattr_bucket() can do this - its call to
4835          * ocfs2_add_new_xattr_cluster() may have created a new extent
4836          * and copied out the end of the old extent.  Then it re-extends
4837          * the old extent back to create space for new xattrs.  That's
4838          * how we get here, and the bucket isn't really new.
4839          */
4840         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4841                                                 t_is_new ?
4842                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4843                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4844         if (ret)
4845                 goto out;
4846
4847         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4848         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4849
4850 out:
4851         ocfs2_xattr_bucket_free(t_bucket);
4852         ocfs2_xattr_bucket_free(s_bucket);
4853
4854         return ret;
4855 }
4856
4857 /*
4858  * src_blk points to the start of an existing extent.  last_blk points to
4859  * last cluster in that extent.  to_blk points to a newly allocated
4860  * extent.  We copy the buckets from the cluster at last_blk to the new
4861  * extent.  If start_bucket is non-zero, we skip that many buckets before
4862  * we start copying.  The new extent's xh_num_buckets gets set to the
4863  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4864  * by the same amount.
4865  */
4866 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4867                                   u64 src_blk, u64 last_blk, u64 to_blk,
4868                                   unsigned int start_bucket,
4869                                   u32 *first_hash)
4870 {
4871         int i, ret, credits;
4872         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4873         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4874         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4875         struct ocfs2_xattr_bucket *old_first, *new_first;
4876
4877         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4878                                      (unsigned long long)to_blk);
4879
4880         BUG_ON(start_bucket >= num_buckets);
4881         if (start_bucket) {
4882                 num_buckets -= start_bucket;
4883                 last_blk += (start_bucket * blks_per_bucket);
4884         }
4885
4886         /* The first bucket of the original extent */
4887         old_first = ocfs2_xattr_bucket_new(inode);
4888         /* The first bucket of the new extent */
4889         new_first = ocfs2_xattr_bucket_new(inode);
4890         if (!old_first || !new_first) {
4891                 ret = -ENOMEM;
4892                 mlog_errno(ret);
4893                 goto out;
4894         }
4895
4896         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4897         if (ret) {
4898                 mlog_errno(ret);
4899                 goto out;
4900         }
4901
4902         /*
4903          * We need to update the first bucket of the old extent and all
4904          * the buckets going to the new extent.
4905          */
4906         credits = ((num_buckets + 1) * blks_per_bucket);
4907         ret = ocfs2_extend_trans(handle, credits);
4908         if (ret) {
4909                 mlog_errno(ret);
4910                 goto out;
4911         }
4912
4913         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4914                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4915         if (ret) {
4916                 mlog_errno(ret);
4917                 goto out;
4918         }
4919
4920         for (i = 0; i < num_buckets; i++) {
4921                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4922                                             last_blk + (i * blks_per_bucket),
4923                                             to_blk + (i * blks_per_bucket),
4924                                             1);
4925                 if (ret) {
4926                         mlog_errno(ret);
4927                         goto out;
4928                 }
4929         }
4930
4931         /*
4932          * Get the new bucket ready before we dirty anything
4933          * (This actually shouldn't fail, because we already dirtied
4934          * it once in ocfs2_cp_xattr_bucket()).
4935          */
4936         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4937         if (ret) {
4938                 mlog_errno(ret);
4939                 goto out;
4940         }
4941         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4942                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4943         if (ret) {
4944                 mlog_errno(ret);
4945                 goto out;
4946         }
4947
4948         /* Now update the headers */
4949         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4950         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4951
4952         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4953         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4954
4955         if (first_hash)
4956                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4957
4958 out:
4959         ocfs2_xattr_bucket_free(new_first);
4960         ocfs2_xattr_bucket_free(old_first);
4961         return ret;
4962 }
4963
4964 /*
4965  * Move some xattrs in this cluster to the new cluster.
4966  * This function should only be called when bucket size == cluster size.
4967  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4968  */
4969 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4970                                       handle_t *handle,
4971                                       u64 prev_blk,
4972                                       u64 new_blk,
4973                                       u32 *first_hash)
4974 {
4975         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4976         int ret, credits = 2 * blk_per_bucket;
4977
4978         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4979
4980         ret = ocfs2_extend_trans(handle, credits);
4981         if (ret) {
4982                 mlog_errno(ret);
4983                 return ret;
4984         }
4985
4986         /* Move half of the xattr in start_blk to the next bucket. */
4987         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4988                                           new_blk, first_hash, 1);
4989 }
4990
4991 /*
4992  * Move some xattrs from the old cluster to the new one since they are not
4993  * contiguous in ocfs2 xattr tree.
4994  *
4995  * new_blk starts a new separate cluster, and we will move some xattrs from
4996  * prev_blk to it. v_start will be set as the first name hash value in this
4997  * new cluster so that it can be used as e_cpos during tree insertion and
4998  * don't collide with our original b-tree operations. first_bh and header_bh
4999  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5000  * to extend the insert bucket.
5001  *
5002  * The problem is how much xattr should we move to the new one and when should
5003  * we update first_bh and header_bh?
5004  * 1. If cluster size > bucket size, that means the previous cluster has more
5005  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5006  *    update the first_bh and header_bh if the insert bucket has been moved
5007  *    to the new cluster.
5008  * 2. If cluster_size == bucket_size:
5009  *    a) If the previous extent rec has more than one cluster and the insert
5010  *       place isn't in the last cluster, copy the entire last cluster to the
5011  *       new one. This time, we don't need to upate the first_bh and header_bh
5012  *       since they will not be moved into the new cluster.
5013  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5014  *       the new one. And we set the extend flag to zero if the insert place is
5015  *       moved into the new allocated cluster since no extend is needed.
5016  */
5017 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5018                                             handle_t *handle,
5019                                             struct ocfs2_xattr_bucket *first,
5020                                             struct ocfs2_xattr_bucket *target,
5021                                             u64 new_blk,
5022                                             u32 prev_clusters,
5023                                             u32 *v_start,
5024                                             int *extend)
5025 {
5026         int ret;
5027
5028         trace_ocfs2_adjust_xattr_cross_cluster(
5029                         (unsigned long long)bucket_blkno(first),
5030                         (unsigned long long)new_blk, prev_clusters);
5031
5032         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5033                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5034                                                           handle,
5035                                                           first, target,
5036                                                           new_blk,
5037                                                           prev_clusters,
5038                                                           v_start);
5039                 if (ret)
5040                         mlog_errno(ret);
5041         } else {
5042                 /* The start of the last cluster in the first extent */
5043                 u64 last_blk = bucket_blkno(first) +
5044                         ((prev_clusters - 1) *
5045                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5046
5047                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5048                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5049                                                      bucket_blkno(first),
5050                                                      last_blk, new_blk, 0,
5051                                                      v_start);
5052                         if (ret)
5053                                 mlog_errno(ret);
5054                 } else {
5055                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5056                                                          last_blk, new_blk,
5057                                                          v_start);
5058                         if (ret)
5059                                 mlog_errno(ret);
5060
5061                         if ((bucket_blkno(target) == last_blk) && extend)
5062                                 *extend = 0;
5063                 }
5064         }
5065
5066         return ret;
5067 }
5068
5069 /*
5070  * Add a new cluster for xattr storage.
5071  *
5072  * If the new cluster is contiguous with the previous one, it will be
5073  * appended to the same extent record, and num_clusters will be updated.
5074  * If not, we will insert a new extent for it and move some xattrs in
5075  * the last cluster into the new allocated one.
5076  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5077  * lose the benefits of hashing because we'll have to search large leaves.
5078  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5079  * if it's bigger).
5080  *
5081  * first_bh is the first block of the previous extent rec and header_bh
5082  * indicates the bucket we will insert the new xattrs. They will be updated
5083  * when the header_bh is moved into the new cluster.
5084  */
5085 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5086                                        struct buffer_head *root_bh,
5087                                        struct ocfs2_xattr_bucket *first,
5088                                        struct ocfs2_xattr_bucket *target,
5089                                        u32 *num_clusters,
5090                                        u32 prev_cpos,
5091                                        int *extend,
5092                                        struct ocfs2_xattr_set_ctxt *ctxt)
5093 {
5094         int ret;
5095         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5096         u32 prev_clusters = *num_clusters;
5097         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5098         u64 block;
5099         handle_t *handle = ctxt->handle;
5100         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5101         struct ocfs2_extent_tree et;
5102
5103         trace_ocfs2_add_new_xattr_cluster_begin(
5104                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5105                 (unsigned long long)bucket_blkno(first),
5106                 prev_cpos, prev_clusters);
5107
5108         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5109
5110         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5111                                       OCFS2_JOURNAL_ACCESS_WRITE);
5112         if (ret < 0) {
5113                 mlog_errno(ret);
5114                 goto leave;
5115         }
5116
5117         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5118                                      clusters_to_add, &bit_off, &num_bits);
5119         if (ret < 0) {
5120                 if (ret != -ENOSPC)
5121                         mlog_errno(ret);
5122                 goto leave;
5123         }
5124
5125         BUG_ON(num_bits > clusters_to_add);
5126
5127         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5128         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5129
5130         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5131             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5132              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5133                 /*
5134                  * If this cluster is contiguous with the old one and
5135                  * adding this new cluster, we don't surpass the limit of
5136                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5137                  * initialized and used like other buckets in the previous
5138                  * cluster.
5139                  * So add it as a contiguous one. The caller will handle
5140                  * its init process.
5141                  */
5142                 v_start = prev_cpos + prev_clusters;
5143                 *num_clusters = prev_clusters + num_bits;
5144         } else {
5145                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5146                                                        handle,
5147                                                        first,
5148                                                        target,
5149                                                        block,
5150                                                        prev_clusters,
5151                                                        &v_start,
5152                                                        extend);
5153                 if (ret) {
5154                         mlog_errno(ret);
5155                         goto leave;
5156                 }
5157         }
5158
5159         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5160                                                  v_start, num_bits);
5161         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5162                                   num_bits, 0, ctxt->meta_ac);
5163         if (ret < 0) {
5164                 mlog_errno(ret);
5165                 goto leave;
5166         }
5167
5168         ocfs2_journal_dirty(handle, root_bh);
5169
5170 leave:
5171         return ret;
5172 }
5173
5174 /*
5175  * We are given an extent.  'first' is the bucket at the very front of
5176  * the extent.  The extent has space for an additional bucket past
5177  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5178  * of the target bucket.  We wish to shift every bucket past the target
5179  * down one, filling in that additional space.  When we get back to the
5180  * target, we split the target between itself and the now-empty bucket
5181  * at target+1 (aka, target_blkno + blks_per_bucket).
5182  */
5183 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5184                                      handle_t *handle,
5185                                      struct ocfs2_xattr_bucket *first,
5186                                      u64 target_blk,
5187                                      u32 num_clusters)
5188 {
5189         int ret, credits;
5190         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5191         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5192         u64 end_blk;
5193         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5194
5195         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5196                                         (unsigned long long)bucket_blkno(first),
5197                                         num_clusters, new_bucket);
5198
5199         /* The extent must have room for an additional bucket */
5200         BUG_ON(new_bucket >=
5201                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5202
5203         /* end_blk points to the last existing bucket */
5204         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5205
5206         /*
5207          * end_blk is the start of the last existing bucket.
5208          * Thus, (end_blk - target_blk) covers the target bucket and
5209          * every bucket after it up to, but not including, the last
5210          * existing bucket.  Then we add the last existing bucket, the
5211          * new bucket, and the first bucket (3 * blk_per_bucket).
5212          */
5213         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5214         ret = ocfs2_extend_trans(handle, credits);
5215         if (ret) {
5216                 mlog_errno(ret);
5217                 goto out;
5218         }
5219
5220         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5221                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5222         if (ret) {
5223                 mlog_errno(ret);
5224                 goto out;
5225         }
5226
5227         while (end_blk != target_blk) {
5228                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5229                                             end_blk + blk_per_bucket, 0);
5230                 if (ret)
5231                         goto out;
5232                 end_blk -= blk_per_bucket;
5233         }
5234
5235         /* Move half of the xattr in target_blkno to the next bucket. */
5236         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5237                                         target_blk + blk_per_bucket, NULL, 0);
5238
5239         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5240         ocfs2_xattr_bucket_journal_dirty(handle, first);
5241
5242 out:
5243         return ret;
5244 }
5245
5246 /*
5247  * Add new xattr bucket in an extent record and adjust the buckets
5248  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5249  * bucket we want to insert into.
5250  *
5251  * In the easy case, we will move all the buckets after target down by
5252  * one. Half of target's xattrs will be moved to the next bucket.
5253  *
5254  * If current cluster is full, we'll allocate a new one.  This may not
5255  * be contiguous.  The underlying calls will make sure that there is
5256  * space for the insert, shifting buckets around if necessary.
5257  * 'target' may be moved by those calls.
5258  */
5259 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5260                                       struct buffer_head *xb_bh,
5261                                       struct ocfs2_xattr_bucket *target,
5262                                       struct ocfs2_xattr_set_ctxt *ctxt)
5263 {
5264         struct ocfs2_xattr_block *xb =
5265                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5266         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5267         struct ocfs2_extent_list *el = &xb_root->xt_list;
5268         u32 name_hash =
5269                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5270         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5271         int ret, num_buckets, extend = 1;
5272         u64 p_blkno;
5273         u32 e_cpos, num_clusters;
5274         /* The bucket at the front of the extent */
5275         struct ocfs2_xattr_bucket *first;
5276
5277         trace_ocfs2_add_new_xattr_bucket(
5278                                 (unsigned long long)bucket_blkno(target));
5279
5280         /* The first bucket of the original extent */
5281         first = ocfs2_xattr_bucket_new(inode);
5282         if (!first) {
5283                 ret = -ENOMEM;
5284                 mlog_errno(ret);
5285                 goto out;
5286         }
5287
5288         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5289                                   &num_clusters, el);
5290         if (ret) {
5291                 mlog_errno(ret);
5292                 goto out;
5293         }
5294
5295         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5296         if (ret) {
5297                 mlog_errno(ret);
5298                 goto out;
5299         }
5300
5301         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5302         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5303                 /*
5304                  * This can move first+target if the target bucket moves
5305                  * to the new extent.
5306                  */
5307                 ret = ocfs2_add_new_xattr_cluster(inode,
5308                                                   xb_bh,
5309                                                   first,
5310                                                   target,
5311                                                   &num_clusters,
5312                                                   e_cpos,
5313                                                   &extend,
5314                                                   ctxt);
5315                 if (ret) {
5316                         mlog_errno(ret);
5317                         goto out;
5318                 }
5319         }
5320
5321         if (extend) {
5322                 ret = ocfs2_extend_xattr_bucket(inode,
5323                                                 ctxt->handle,
5324                                                 first,
5325                                                 bucket_blkno(target),
5326                                                 num_clusters);
5327                 if (ret)
5328                         mlog_errno(ret);
5329         }
5330
5331 out:
5332         ocfs2_xattr_bucket_free(first);
5333
5334         return ret;
5335 }
5336
5337 /*
5338  * Truncate the specified xe_off entry in xattr bucket.
5339  * bucket is indicated by header_bh and len is the new length.
5340  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5341  *
5342  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5343  */
5344 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5345                                              struct ocfs2_xattr_bucket *bucket,
5346                                              int xe_off,
5347                                              int len,
5348                                              struct ocfs2_xattr_set_ctxt *ctxt)
5349 {
5350         int ret, offset;
5351         u64 value_blk;
5352         struct ocfs2_xattr_entry *xe;
5353         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5354         size_t blocksize = inode->i_sb->s_blocksize;
5355         struct ocfs2_xattr_value_buf vb = {
5356                 .vb_access = ocfs2_journal_access,
5357         };
5358
5359         xe = &xh->xh_entries[xe_off];
5360
5361         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5362
5363         offset = le16_to_cpu(xe->xe_name_offset) +
5364                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5365
5366         value_blk = offset / blocksize;
5367
5368         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5369         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5370
5371         vb.vb_bh = bucket->bu_bhs[value_blk];
5372         BUG_ON(!vb.vb_bh);
5373
5374         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5375                 (vb.vb_bh->b_data + offset % blocksize);
5376
5377         /*
5378          * From here on out we have to dirty the bucket.  The generic
5379          * value calls only modify one of the bucket's bhs, but we need
5380          * to send the bucket at once.  So if they error, they *could* have
5381          * modified something.  We have to assume they did, and dirty
5382          * the whole bucket.  This leaves us in a consistent state.
5383          */
5384         trace_ocfs2_xattr_bucket_value_truncate(
5385                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5386         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5387         if (ret) {
5388                 mlog_errno(ret);
5389                 goto out;
5390         }
5391
5392         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5393                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5394         if (ret) {
5395                 mlog_errno(ret);
5396                 goto out;
5397         }
5398
5399         xe->xe_value_size = cpu_to_le64(len);
5400
5401         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5402
5403 out:
5404         return ret;
5405 }
5406
5407 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5408                                   struct buffer_head *root_bh,
5409                                   u64 blkno,
5410                                   u32 cpos,
5411                                   u32 len,
5412                                   void *para)
5413 {
5414         int ret;
5415         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5416         struct inode *tl_inode = osb->osb_tl_inode;
5417         handle_t *handle;
5418         struct ocfs2_xattr_block *xb =
5419                         (struct ocfs2_xattr_block *)root_bh->b_data;
5420         struct ocfs2_alloc_context *meta_ac = NULL;
5421         struct ocfs2_cached_dealloc_ctxt dealloc;
5422         struct ocfs2_extent_tree et;
5423
5424         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5425                                           ocfs2_delete_xattr_in_bucket, para);
5426         if (ret) {
5427                 mlog_errno(ret);
5428                 return ret;
5429         }
5430
5431         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5432
5433         ocfs2_init_dealloc_ctxt(&dealloc);
5434
5435         trace_ocfs2_rm_xattr_cluster(
5436                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5437                         (unsigned long long)blkno, cpos, len);
5438
5439         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5440                                                len);
5441
5442         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5443         if (ret) {
5444                 mlog_errno(ret);
5445                 return ret;
5446         }
5447
5448         inode_lock(tl_inode);
5449
5450         if (ocfs2_truncate_log_needs_flush(osb)) {
5451                 ret = __ocfs2_flush_truncate_log(osb);
5452                 if (ret < 0) {
5453                         mlog_errno(ret);
5454                         goto out;
5455                 }
5456         }
5457
5458         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5459         if (IS_ERR(handle)) {
5460                 ret = -ENOMEM;
5461                 mlog_errno(ret);
5462                 goto out;
5463         }
5464
5465         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5466                                       OCFS2_JOURNAL_ACCESS_WRITE);
5467         if (ret) {
5468                 mlog_errno(ret);
5469                 goto out_commit;
5470         }
5471
5472         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5473                                   &dealloc);
5474         if (ret) {
5475                 mlog_errno(ret);
5476                 goto out_commit;
5477         }
5478
5479         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5480         ocfs2_journal_dirty(handle, root_bh);
5481
5482         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5483         if (ret)
5484                 mlog_errno(ret);
5485         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5486
5487 out_commit:
5488         ocfs2_commit_trans(osb, handle);
5489 out:
5490         ocfs2_schedule_truncate_log_flush(osb, 1);
5491
5492         inode_unlock(tl_inode);
5493
5494         if (meta_ac)
5495                 ocfs2_free_alloc_context(meta_ac);
5496
5497         ocfs2_run_deallocs(osb, &dealloc);
5498
5499         return ret;
5500 }
5501
5502 /*
5503  * check whether the xattr bucket is filled up with the same hash value.
5504  * If we want to insert the xattr with the same hash, return -ENOSPC.
5505  * If we want to insert a xattr with different hash value, go ahead
5506  * and ocfs2_divide_xattr_bucket will handle this.
5507  */
5508 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5509                                               struct ocfs2_xattr_bucket *bucket,
5510                                               const char *name)
5511 {
5512         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5513         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5514
5515         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5516                 return 0;
5517
5518         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5519             xh->xh_entries[0].xe_name_hash) {
5520                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5521                      "hash = %u\n",
5522                      (unsigned long long)bucket_blkno(bucket),
5523                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5524                 return -ENOSPC;
5525         }
5526
5527         return 0;
5528 }
5529
5530 /*
5531  * Try to set the entry in the current bucket.  If we fail, the caller
5532  * will handle getting us another bucket.
5533  */
5534 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5535                                         struct ocfs2_xattr_info *xi,
5536                                         struct ocfs2_xattr_search *xs,
5537                                         struct ocfs2_xattr_set_ctxt *ctxt)
5538 {
5539         int ret;
5540         struct ocfs2_xa_loc loc;
5541
5542         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5543
5544         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5545                                        xs->not_found ? NULL : xs->here);
5546         ret = ocfs2_xa_set(&loc, xi, ctxt);
5547         if (!ret) {
5548                 xs->here = loc.xl_entry;
5549                 goto out;
5550         }
5551         if (ret != -ENOSPC) {
5552                 mlog_errno(ret);
5553                 goto out;
5554         }
5555
5556         /* Ok, we need space.  Let's try defragmenting the bucket. */
5557         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5558                                         xs->bucket);
5559         if (ret) {
5560                 mlog_errno(ret);
5561                 goto out;
5562         }
5563
5564         ret = ocfs2_xa_set(&loc, xi, ctxt);
5565         if (!ret) {
5566                 xs->here = loc.xl_entry;
5567                 goto out;
5568         }
5569         if (ret != -ENOSPC)
5570                 mlog_errno(ret);
5571
5572
5573 out:
5574         return ret;
5575 }
5576
5577 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5578                                              struct ocfs2_xattr_info *xi,
5579                                              struct ocfs2_xattr_search *xs,
5580                                              struct ocfs2_xattr_set_ctxt *ctxt)
5581 {
5582         int ret;
5583
5584         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5585
5586         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5587         if (!ret)
5588                 goto out;
5589         if (ret != -ENOSPC) {
5590                 mlog_errno(ret);
5591                 goto out;
5592         }
5593
5594         /* Ack, need more space.  Let's try to get another bucket! */
5595
5596         /*
5597          * We do not allow for overlapping ranges between buckets. And
5598          * the maximum number of collisions we will allow for then is
5599          * one bucket's worth, so check it here whether we need to
5600          * add a new bucket for the insert.
5601          */
5602         ret = ocfs2_check_xattr_bucket_collision(inode,
5603                                                  xs->bucket,
5604                                                  xi->xi_name);
5605         if (ret) {
5606                 mlog_errno(ret);
5607                 goto out;
5608         }
5609
5610         ret = ocfs2_add_new_xattr_bucket(inode,
5611                                          xs->xattr_bh,
5612                                          xs->bucket,
5613                                          ctxt);
5614         if (ret) {
5615                 mlog_errno(ret);
5616                 goto out;
5617         }
5618
5619         /*
5620          * ocfs2_add_new_xattr_bucket() will have updated
5621          * xs->bucket if it moved, but it will not have updated
5622          * any of the other search fields.  Thus, we drop it and
5623          * re-search.  Everything should be cached, so it'll be
5624          * quick.
5625          */
5626         ocfs2_xattr_bucket_relse(xs->bucket);
5627         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5628                                            xi->xi_name_index,
5629                                            xi->xi_name, xs);
5630         if (ret && ret != -ENODATA)
5631                 goto out;
5632         xs->not_found = ret;
5633
5634         /* Ok, we have a new bucket, let's try again */
5635         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5636         if (ret && (ret != -ENOSPC))
5637                 mlog_errno(ret);
5638
5639 out:
5640         return ret;
5641 }
5642
5643 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5644                                         struct ocfs2_xattr_bucket *bucket,
5645                                         void *para)
5646 {
5647         int ret = 0, ref_credits;
5648         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5649         u16 i;
5650         struct ocfs2_xattr_entry *xe;
5651         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5652         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5653         int credits = ocfs2_remove_extent_credits(osb->sb) +
5654                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5655         struct ocfs2_xattr_value_root *xv;
5656         struct ocfs2_rm_xattr_bucket_para *args =
5657                         (struct ocfs2_rm_xattr_bucket_para *)para;
5658
5659         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5660
5661         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5662                 xe = &xh->xh_entries[i];
5663                 if (ocfs2_xattr_is_local(xe))
5664                         continue;
5665
5666                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5667                                                       i, &xv, NULL);
5668                 if (ret) {
5669                         mlog_errno(ret);
5670                         break;
5671                 }
5672
5673                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5674                                                          args->ref_ci,
5675                                                          args->ref_root_bh,
5676                                                          &ctxt.meta_ac,
5677                                                          &ref_credits);
5678
5679                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5680                 if (IS_ERR(ctxt.handle)) {
5681                         ret = PTR_ERR(ctxt.handle);
5682                         mlog_errno(ret);
5683                         break;
5684                 }
5685
5686                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5687                                                         i, 0, &ctxt);
5688
5689                 ocfs2_commit_trans(osb, ctxt.handle);
5690                 if (ctxt.meta_ac) {
5691                         ocfs2_free_alloc_context(ctxt.meta_ac);
5692                         ctxt.meta_ac = NULL;
5693                 }
5694                 if (ret) {
5695                         mlog_errno(ret);
5696                         break;
5697                 }
5698         }
5699
5700         if (ctxt.meta_ac)
5701                 ocfs2_free_alloc_context(ctxt.meta_ac);
5702         ocfs2_schedule_truncate_log_flush(osb, 1);
5703         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5704         return ret;
5705 }
5706
5707 /*
5708  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5709  * or change the extent record flag), we need to recalculate
5710  * the metaecc for the whole bucket. So it is done here.
5711  *
5712  * Note:
5713  * We have to give the extra credits for the caller.
5714  */
5715 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5716                                             handle_t *handle,
5717                                             void *para)
5718 {
5719         int ret;
5720         struct ocfs2_xattr_bucket *bucket =
5721                         (struct ocfs2_xattr_bucket *)para;
5722
5723         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5724                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5725         if (ret) {
5726                 mlog_errno(ret);
5727                 return ret;
5728         }
5729
5730         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5731
5732         return 0;
5733 }
5734
5735 /*
5736  * Special action we need if the xattr value is refcounted.
5737  *
5738  * 1. If the xattr is refcounted, lock the tree.
5739  * 2. CoW the xattr if we are setting the new value and the value
5740  *    will be stored outside.
5741  * 3. In other case, decrease_refcount will work for us, so just
5742  *    lock the refcount tree, calculate the meta and credits is OK.
5743  *
5744  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5745  * currently CoW is a completed transaction, while this function
5746  * will also lock the allocators and let us deadlock. So we will
5747  * CoW the whole xattr value.
5748  */
5749 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5750                                         struct ocfs2_dinode *di,
5751                                         struct ocfs2_xattr_info *xi,
5752                                         struct ocfs2_xattr_search *xis,
5753                                         struct ocfs2_xattr_search *xbs,
5754                                         struct ocfs2_refcount_tree **ref_tree,
5755                                         int *meta_add,
5756                                         int *credits)
5757 {
5758         int ret = 0;
5759         struct ocfs2_xattr_block *xb;
5760         struct ocfs2_xattr_entry *xe;
5761         char *base;
5762         u32 p_cluster, num_clusters;
5763         unsigned int ext_flags;
5764         int name_offset, name_len;
5765         struct ocfs2_xattr_value_buf vb;
5766         struct ocfs2_xattr_bucket *bucket = NULL;
5767         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5768         struct ocfs2_post_refcount refcount;
5769         struct ocfs2_post_refcount *p = NULL;
5770         struct buffer_head *ref_root_bh = NULL;
5771
5772         if (!xis->not_found) {
5773                 xe = xis->here;
5774                 name_offset = le16_to_cpu(xe->xe_name_offset);
5775                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5776                 base = xis->base;
5777                 vb.vb_bh = xis->inode_bh;
5778                 vb.vb_access = ocfs2_journal_access_di;
5779         } else {
5780                 int i, block_off = 0;
5781                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5782                 xe = xbs->here;
5783                 name_offset = le16_to_cpu(xe->xe_name_offset);
5784                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5785                 i = xbs->here - xbs->header->xh_entries;
5786
5787                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5788                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5789                                                         bucket_xh(xbs->bucket),
5790                                                         i, &block_off,
5791                                                         &name_offset);
5792                         if (ret) {
5793                                 mlog_errno(ret);
5794                                 goto out;
5795                         }
5796                         base = bucket_block(xbs->bucket, block_off);
5797                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5798                         vb.vb_access = ocfs2_journal_access;
5799
5800                         if (ocfs2_meta_ecc(osb)) {
5801                                 /*create parameters for ocfs2_post_refcount. */
5802                                 bucket = xbs->bucket;
5803                                 refcount.credits = bucket->bu_blocks;
5804                                 refcount.para = bucket;
5805                                 refcount.func =
5806                                         ocfs2_xattr_bucket_post_refcount;
5807                                 p = &refcount;
5808                         }
5809                 } else {
5810                         base = xbs->base;
5811                         vb.vb_bh = xbs->xattr_bh;
5812                         vb.vb_access = ocfs2_journal_access_xb;
5813                 }
5814         }
5815
5816         if (ocfs2_xattr_is_local(xe))
5817                 goto out;
5818
5819         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5820                                 (base + name_offset + name_len);
5821
5822         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5823                                        &num_clusters, &vb.vb_xv->xr_list,
5824                                        &ext_flags);
5825         if (ret) {
5826                 mlog_errno(ret);
5827                 goto out;
5828         }
5829
5830         /*
5831          * We just need to check the 1st extent record, since we always
5832          * CoW the whole xattr. So there shouldn't be a xattr with
5833          * some REFCOUNT extent recs after the 1st one.
5834          */
5835         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5836                 goto out;
5837
5838         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5839                                        1, ref_tree, &ref_root_bh);
5840         if (ret) {
5841                 mlog_errno(ret);
5842                 goto out;
5843         }
5844
5845         /*
5846          * If we are deleting the xattr or the new size will be stored inside,
5847          * cool, leave it there, the xattr truncate process will remove them
5848          * for us(it still needs the refcount tree lock and the meta, credits).
5849          * And the worse case is that every cluster truncate will split the
5850          * refcount tree, and make the original extent become 3. So we will need
5851          * 2 * cluster more extent recs at most.
5852          */
5853         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5854
5855                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5856                                                          &(*ref_tree)->rf_ci,
5857                                                          ref_root_bh, vb.vb_xv,
5858                                                          meta_add, credits);
5859                 if (ret)
5860                         mlog_errno(ret);
5861                 goto out;
5862         }
5863
5864         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5865                                        *ref_tree, ref_root_bh, 0,
5866                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5867         if (ret)
5868                 mlog_errno(ret);
5869
5870 out:
5871         brelse(ref_root_bh);
5872         return ret;
5873 }
5874
5875 /*
5876  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5877  * The physical clusters will be added to refcount tree.
5878  */
5879 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5880                                 struct ocfs2_xattr_value_root *xv,
5881                                 struct ocfs2_extent_tree *value_et,
5882                                 struct ocfs2_caching_info *ref_ci,
5883                                 struct buffer_head *ref_root_bh,
5884                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5885                                 struct ocfs2_post_refcount *refcount)
5886 {
5887         int ret = 0;
5888         u32 clusters = le32_to_cpu(xv->xr_clusters);
5889         u32 cpos, p_cluster, num_clusters;
5890         struct ocfs2_extent_list *el = &xv->xr_list;
5891         unsigned int ext_flags;
5892
5893         cpos = 0;
5894         while (cpos < clusters) {
5895                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5896                                                &num_clusters, el, &ext_flags);
5897                 if (ret) {
5898                         mlog_errno(ret);
5899                         break;
5900                 }
5901
5902                 cpos += num_clusters;
5903                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5904                         continue;
5905
5906                 BUG_ON(!p_cluster);
5907
5908                 ret = ocfs2_add_refcount_flag(inode, value_et,
5909                                               ref_ci, ref_root_bh,
5910                                               cpos - num_clusters,
5911                                               p_cluster, num_clusters,
5912                                               dealloc, refcount);
5913                 if (ret) {
5914                         mlog_errno(ret);
5915                         break;
5916                 }
5917         }
5918
5919         return ret;
5920 }
5921
5922 /*
5923  * Given a normal ocfs2_xattr_header, refcount all the entries which
5924  * have value stored outside.
5925  * Used for xattrs stored in inode and ocfs2_xattr_block.
5926  */
5927 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5928                                 struct ocfs2_xattr_value_buf *vb,
5929                                 struct ocfs2_xattr_header *header,
5930                                 struct ocfs2_caching_info *ref_ci,
5931                                 struct buffer_head *ref_root_bh,
5932                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5933 {
5934
5935         struct ocfs2_xattr_entry *xe;
5936         struct ocfs2_xattr_value_root *xv;
5937         struct ocfs2_extent_tree et;
5938         int i, ret = 0;
5939
5940         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5941                 xe = &header->xh_entries[i];
5942
5943                 if (ocfs2_xattr_is_local(xe))
5944                         continue;
5945
5946                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5947                         le16_to_cpu(xe->xe_name_offset) +
5948                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5949
5950                 vb->vb_xv = xv;
5951                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5952
5953                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5954                                                         ref_ci, ref_root_bh,
5955                                                         dealloc, NULL);
5956                 if (ret) {
5957                         mlog_errno(ret);
5958                         break;
5959                 }
5960         }
5961
5962         return ret;
5963 }
5964
5965 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5966                                 struct buffer_head *fe_bh,
5967                                 struct ocfs2_caching_info *ref_ci,
5968                                 struct buffer_head *ref_root_bh,
5969                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5970 {
5971         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5972         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5973                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5974                                 le16_to_cpu(di->i_xattr_inline_size));
5975         struct ocfs2_xattr_value_buf vb = {
5976                 .vb_bh = fe_bh,
5977                 .vb_access = ocfs2_journal_access_di,
5978         };
5979
5980         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5981                                                   ref_ci, ref_root_bh, dealloc);
5982 }
5983
5984 struct ocfs2_xattr_tree_value_refcount_para {
5985         struct ocfs2_caching_info *ref_ci;
5986         struct buffer_head *ref_root_bh;
5987         struct ocfs2_cached_dealloc_ctxt *dealloc;
5988 };
5989
5990 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5991                                            struct ocfs2_xattr_bucket *bucket,
5992                                            int offset,
5993                                            struct ocfs2_xattr_value_root **xv,
5994                                            struct buffer_head **bh)
5995 {
5996         int ret, block_off, name_offset;
5997         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5998         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5999         void *base;
6000
6001         ret = ocfs2_xattr_bucket_get_name_value(sb,
6002                                                 bucket_xh(bucket),
6003                                                 offset,
6004                                                 &block_off,
6005                                                 &name_offset);
6006         if (ret) {
6007                 mlog_errno(ret);
6008                 goto out;
6009         }
6010
6011         base = bucket_block(bucket, block_off);
6012
6013         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6014                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6015
6016         if (bh)
6017                 *bh = bucket->bu_bhs[block_off];
6018 out:
6019         return ret;
6020 }
6021
6022 /*
6023  * For a given xattr bucket, refcount all the entries which
6024  * have value stored outside.
6025  */
6026 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6027                                              struct ocfs2_xattr_bucket *bucket,
6028                                              void *para)
6029 {
6030         int i, ret = 0;
6031         struct ocfs2_extent_tree et;
6032         struct ocfs2_xattr_tree_value_refcount_para *ref =
6033                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6034         struct ocfs2_xattr_header *xh =
6035                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6036         struct ocfs2_xattr_entry *xe;
6037         struct ocfs2_xattr_value_buf vb = {
6038                 .vb_access = ocfs2_journal_access,
6039         };
6040         struct ocfs2_post_refcount refcount = {
6041                 .credits = bucket->bu_blocks,
6042                 .para = bucket,
6043                 .func = ocfs2_xattr_bucket_post_refcount,
6044         };
6045         struct ocfs2_post_refcount *p = NULL;
6046
6047         /* We only need post_refcount if we support metaecc. */
6048         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6049                 p = &refcount;
6050
6051         trace_ocfs2_xattr_bucket_value_refcount(
6052                                 (unsigned long long)bucket_blkno(bucket),
6053                                 le16_to_cpu(xh->xh_count));
6054         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6055                 xe = &xh->xh_entries[i];
6056
6057                 if (ocfs2_xattr_is_local(xe))
6058                         continue;
6059
6060                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6061                                                       &vb.vb_xv, &vb.vb_bh);
6062                 if (ret) {
6063                         mlog_errno(ret);
6064                         break;
6065                 }
6066
6067                 ocfs2_init_xattr_value_extent_tree(&et,
6068                                                    INODE_CACHE(inode), &vb);
6069
6070                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6071                                                         &et, ref->ref_ci,
6072                                                         ref->ref_root_bh,
6073                                                         ref->dealloc, p);
6074                 if (ret) {
6075                         mlog_errno(ret);
6076                         break;
6077                 }
6078         }
6079
6080         return ret;
6081
6082 }
6083
6084 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6085                                      struct buffer_head *root_bh,
6086                                      u64 blkno, u32 cpos, u32 len, void *para)
6087 {
6088         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6089                                            ocfs2_xattr_bucket_value_refcount,
6090                                            para);
6091 }
6092
6093 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6094                                 struct buffer_head *blk_bh,
6095                                 struct ocfs2_caching_info *ref_ci,
6096                                 struct buffer_head *ref_root_bh,
6097                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6098 {
6099         int ret = 0;
6100         struct ocfs2_xattr_block *xb =
6101                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6102
6103         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6104                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6105                 struct ocfs2_xattr_value_buf vb = {
6106                         .vb_bh = blk_bh,
6107                         .vb_access = ocfs2_journal_access_xb,
6108                 };
6109
6110                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6111                                                          ref_ci, ref_root_bh,
6112                                                          dealloc);
6113         } else {
6114                 struct ocfs2_xattr_tree_value_refcount_para para = {
6115                         .ref_ci = ref_ci,
6116                         .ref_root_bh = ref_root_bh,
6117                         .dealloc = dealloc,
6118                 };
6119
6120                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6121                                                 ocfs2_refcount_xattr_tree_rec,
6122                                                 &para);
6123         }
6124
6125         return ret;
6126 }
6127
6128 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6129                                      struct buffer_head *fe_bh,
6130                                      struct ocfs2_caching_info *ref_ci,
6131                                      struct buffer_head *ref_root_bh,
6132                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6133 {
6134         int ret = 0;
6135         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6136         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6137         struct buffer_head *blk_bh = NULL;
6138
6139         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6140                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6141                                                          ref_ci, ref_root_bh,
6142                                                          dealloc);
6143                 if (ret) {
6144                         mlog_errno(ret);
6145                         goto out;
6146                 }
6147         }
6148
6149         if (!di->i_xattr_loc)
6150                 goto out;
6151
6152         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6153                                      &blk_bh);
6154         if (ret < 0) {
6155                 mlog_errno(ret);
6156                 goto out;
6157         }
6158
6159         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6160                                                 ref_root_bh, dealloc);
6161         if (ret)
6162                 mlog_errno(ret);
6163
6164         brelse(blk_bh);
6165 out:
6166
6167         return ret;
6168 }
6169
6170 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6171 /*
6172  * Store the information we need in xattr reflink.
6173  * old_bh and new_bh are inode bh for the old and new inode.
6174  */
6175 struct ocfs2_xattr_reflink {
6176         struct inode *old_inode;
6177         struct inode *new_inode;
6178         struct buffer_head *old_bh;
6179         struct buffer_head *new_bh;
6180         struct ocfs2_caching_info *ref_ci;
6181         struct buffer_head *ref_root_bh;
6182         struct ocfs2_cached_dealloc_ctxt *dealloc;
6183         should_xattr_reflinked *xattr_reflinked;
6184 };
6185
6186 /*
6187  * Given a xattr header and xe offset,
6188  * return the proper xv and the corresponding bh.
6189  * xattr in inode, block and xattr tree have different implementaions.
6190  */
6191 typedef int (get_xattr_value_root)(struct super_block *sb,
6192                                    struct buffer_head *bh,
6193                                    struct ocfs2_xattr_header *xh,
6194                                    int offset,
6195                                    struct ocfs2_xattr_value_root **xv,
6196                                    struct buffer_head **ret_bh,
6197                                    void *para);
6198
6199 /*
6200  * Calculate all the xattr value root metadata stored in this xattr header and
6201  * credits we need if we create them from the scratch.
6202  * We use get_xattr_value_root so that all types of xattr container can use it.
6203  */
6204 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6205                                              struct buffer_head *bh,
6206                                              struct ocfs2_xattr_header *xh,
6207                                              int *metas, int *credits,
6208                                              int *num_recs,
6209                                              get_xattr_value_root *func,
6210                                              void *para)
6211 {
6212         int i, ret = 0;
6213         struct ocfs2_xattr_value_root *xv;
6214         struct ocfs2_xattr_entry *xe;
6215
6216         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6217                 xe = &xh->xh_entries[i];
6218                 if (ocfs2_xattr_is_local(xe))
6219                         continue;
6220
6221                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6222                 if (ret) {
6223                         mlog_errno(ret);
6224                         break;
6225                 }
6226
6227                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6228                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6229
6230                 *credits += ocfs2_calc_extend_credits(sb,
6231                                                 &def_xv.xv.xr_list);
6232
6233                 /*
6234                  * If the value is a tree with depth > 1, We don't go deep
6235                  * to the extent block, so just calculate a maximum record num.
6236                  */
6237                 if (!xv->xr_list.l_tree_depth)
6238                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6239                 else
6240                         *num_recs += ocfs2_clusters_for_bytes(sb,
6241                                                               XATTR_SIZE_MAX);
6242         }
6243
6244         return ret;
6245 }
6246
6247 /* Used by xattr inode and block to return the right xv and buffer_head. */
6248 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6249                                       struct buffer_head *bh,
6250                                       struct ocfs2_xattr_header *xh,
6251                                       int offset,
6252                                       struct ocfs2_xattr_value_root **xv,
6253                                       struct buffer_head **ret_bh,
6254                                       void *para)
6255 {
6256         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6257
6258         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6259                 le16_to_cpu(xe->xe_name_offset) +
6260                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6261
6262         if (ret_bh)
6263                 *ret_bh = bh;
6264
6265         return 0;
6266 }
6267
6268 /*
6269  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6270  * It is only used for inline xattr and xattr block.
6271  */
6272 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6273                                         struct ocfs2_xattr_header *xh,
6274                                         struct buffer_head *ref_root_bh,
6275                                         int *credits,
6276                                         struct ocfs2_alloc_context **meta_ac)
6277 {
6278         int ret, meta_add = 0, num_recs = 0;
6279         struct ocfs2_refcount_block *rb =
6280                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6281
6282         *credits = 0;
6283
6284         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6285                                                 &meta_add, credits, &num_recs,
6286                                                 ocfs2_get_xattr_value_root,
6287                                                 NULL);
6288         if (ret) {
6289                 mlog_errno(ret);
6290                 goto out;
6291         }
6292
6293         /*
6294          * We need to add/modify num_recs in refcount tree, so just calculate
6295          * an approximate number we need for refcount tree change.
6296          * Sometimes we need to split the tree, and after split,  half recs
6297          * will be moved to the new block, and a new block can only provide
6298          * half number of recs. So we multiple new blocks by 2.
6299          */
6300         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6301         meta_add += num_recs;
6302         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6303         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6304                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6305                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6306         else
6307                 *credits += 1;
6308
6309         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6310         if (ret)
6311                 mlog_errno(ret);
6312
6313 out:
6314         return ret;
6315 }
6316
6317 /*
6318  * Given a xattr header, reflink all the xattrs in this container.
6319  * It can be used for inode, block and bucket.
6320  *
6321  * NOTE:
6322  * Before we call this function, the caller has memcpy the xattr in
6323  * old_xh to the new_xh.
6324  *
6325  * If args.xattr_reflinked is set, call it to decide whether the xe should
6326  * be reflinked or not. If not, remove it from the new xattr header.
6327  */
6328 static int ocfs2_reflink_xattr_header(handle_t *handle,
6329                                       struct ocfs2_xattr_reflink *args,
6330                                       struct buffer_head *old_bh,
6331                                       struct ocfs2_xattr_header *xh,
6332                                       struct buffer_head *new_bh,
6333                                       struct ocfs2_xattr_header *new_xh,
6334                                       struct ocfs2_xattr_value_buf *vb,
6335                                       struct ocfs2_alloc_context *meta_ac,
6336                                       get_xattr_value_root *func,
6337                                       void *para)
6338 {
6339         int ret = 0, i, j;
6340         struct super_block *sb = args->old_inode->i_sb;
6341         struct buffer_head *value_bh;
6342         struct ocfs2_xattr_entry *xe, *last;
6343         struct ocfs2_xattr_value_root *xv, *new_xv;
6344         struct ocfs2_extent_tree data_et;
6345         u32 clusters, cpos, p_cluster, num_clusters;
6346         unsigned int ext_flags = 0;
6347
6348         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6349                                          le16_to_cpu(xh->xh_count));
6350
6351         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6352         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6353                 xe = &xh->xh_entries[i];
6354
6355                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6356                         xe = &new_xh->xh_entries[j];
6357
6358                         le16_add_cpu(&new_xh->xh_count, -1);
6359                         if (new_xh->xh_count) {
6360                                 memmove(xe, xe + 1,
6361                                         (void *)last - (void *)xe);
6362                                 memset(last, 0,
6363                                        sizeof(struct ocfs2_xattr_entry));
6364                         }
6365
6366                         /*
6367                          * We don't want j to increase in the next round since
6368                          * it is already moved ahead.
6369                          */
6370                         j--;
6371                         continue;
6372                 }
6373
6374                 if (ocfs2_xattr_is_local(xe))
6375                         continue;
6376
6377                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6378                 if (ret) {
6379                         mlog_errno(ret);
6380                         break;
6381                 }
6382
6383                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6384                 if (ret) {
6385                         mlog_errno(ret);
6386                         break;
6387                 }
6388
6389                 /*
6390                  * For the xattr which has l_tree_depth = 0, all the extent
6391                  * recs have already be copied to the new xh with the
6392                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6393                  * increase the refount count int the refcount tree.
6394                  *
6395                  * For the xattr which has l_tree_depth > 0, we need
6396                  * to initialize it to the empty default value root,
6397                  * and then insert the extents one by one.
6398                  */
6399                 if (xv->xr_list.l_tree_depth) {
6400                         memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6401                         vb->vb_xv = new_xv;
6402                         vb->vb_bh = value_bh;
6403                         ocfs2_init_xattr_value_extent_tree(&data_et,
6404                                         INODE_CACHE(args->new_inode), vb);
6405                 }
6406
6407                 clusters = le32_to_cpu(xv->xr_clusters);
6408                 cpos = 0;
6409                 while (cpos < clusters) {
6410                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6411                                                        cpos,
6412                                                        &p_cluster,
6413                                                        &num_clusters,
6414                                                        &xv->xr_list,
6415                                                        &ext_flags);
6416                         if (ret) {
6417                                 mlog_errno(ret);
6418                                 goto out;
6419                         }
6420
6421                         BUG_ON(!p_cluster);
6422
6423                         if (xv->xr_list.l_tree_depth) {
6424                                 ret = ocfs2_insert_extent(handle,
6425                                                 &data_et, cpos,
6426                                                 ocfs2_clusters_to_blocks(
6427                                                         args->old_inode->i_sb,
6428                                                         p_cluster),
6429                                                 num_clusters, ext_flags,
6430                                                 meta_ac);
6431                                 if (ret) {
6432                                         mlog_errno(ret);
6433                                         goto out;
6434                                 }
6435                         }
6436
6437                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6438                                                       args->ref_root_bh,
6439                                                       p_cluster, num_clusters,
6440                                                       meta_ac, args->dealloc);
6441                         if (ret) {
6442                                 mlog_errno(ret);
6443                                 goto out;
6444                         }
6445
6446                         cpos += num_clusters;
6447                 }
6448         }
6449
6450 out:
6451         return ret;
6452 }
6453
6454 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6455 {
6456         int ret = 0, credits = 0;
6457         handle_t *handle;
6458         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6459         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6460         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6461         int header_off = osb->sb->s_blocksize - inline_size;
6462         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6463                                         (args->old_bh->b_data + header_off);
6464         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6465                                         (args->new_bh->b_data + header_off);
6466         struct ocfs2_alloc_context *meta_ac = NULL;
6467         struct ocfs2_inode_info *new_oi;
6468         struct ocfs2_dinode *new_di;
6469         struct ocfs2_xattr_value_buf vb = {
6470                 .vb_bh = args->new_bh,
6471                 .vb_access = ocfs2_journal_access_di,
6472         };
6473
6474         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6475                                                   &credits, &meta_ac);
6476         if (ret) {
6477                 mlog_errno(ret);
6478                 goto out;
6479         }
6480
6481         handle = ocfs2_start_trans(osb, credits);
6482         if (IS_ERR(handle)) {
6483                 ret = PTR_ERR(handle);
6484                 mlog_errno(ret);
6485                 goto out;
6486         }
6487
6488         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6489                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6490         if (ret) {
6491                 mlog_errno(ret);
6492                 goto out_commit;
6493         }
6494
6495         memcpy(args->new_bh->b_data + header_off,
6496                args->old_bh->b_data + header_off, inline_size);
6497
6498         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6499         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6500
6501         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6502                                          args->new_bh, new_xh, &vb, meta_ac,
6503                                          ocfs2_get_xattr_value_root, NULL);
6504         if (ret) {
6505                 mlog_errno(ret);
6506                 goto out_commit;
6507         }
6508
6509         new_oi = OCFS2_I(args->new_inode);
6510         /*
6511          * Adjust extent record count to reserve space for extended attribute.
6512          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6513          */
6514         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6515             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6516                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6517                 le16_add_cpu(&el->l_count, -(inline_size /
6518                                         sizeof(struct ocfs2_extent_rec)));
6519         }
6520         spin_lock(&new_oi->ip_lock);
6521         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6522         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6523         spin_unlock(&new_oi->ip_lock);
6524
6525         ocfs2_journal_dirty(handle, args->new_bh);
6526
6527 out_commit:
6528         ocfs2_commit_trans(osb, handle);
6529
6530 out:
6531         if (meta_ac)
6532                 ocfs2_free_alloc_context(meta_ac);
6533         return ret;
6534 }
6535
6536 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6537                                           struct buffer_head *fe_bh,
6538                                           struct buffer_head **ret_bh,
6539                                           int indexed)
6540 {
6541         int ret;
6542         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6543         struct ocfs2_xattr_set_ctxt ctxt;
6544
6545         memset(&ctxt, 0, sizeof(ctxt));
6546         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6547         if (ret < 0) {
6548                 mlog_errno(ret);
6549                 return ret;
6550         }
6551
6552         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6553         if (IS_ERR(ctxt.handle)) {
6554                 ret = PTR_ERR(ctxt.handle);
6555                 mlog_errno(ret);
6556                 goto out;
6557         }
6558
6559         trace_ocfs2_create_empty_xattr_block(
6560                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6561         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6562                                        ret_bh);
6563         if (ret)
6564                 mlog_errno(ret);
6565
6566         ocfs2_commit_trans(osb, ctxt.handle);
6567 out:
6568         ocfs2_free_alloc_context(ctxt.meta_ac);
6569         return ret;
6570 }
6571
6572 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6573                                      struct buffer_head *blk_bh,
6574                                      struct buffer_head *new_blk_bh)
6575 {
6576         int ret = 0, credits = 0;
6577         handle_t *handle;
6578         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6579         struct ocfs2_dinode *new_di;
6580         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6581         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6582         struct ocfs2_xattr_block *xb =
6583                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6584         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6585         struct ocfs2_xattr_block *new_xb =
6586                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6587         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6588         struct ocfs2_alloc_context *meta_ac;
6589         struct ocfs2_xattr_value_buf vb = {
6590                 .vb_bh = new_blk_bh,
6591                 .vb_access = ocfs2_journal_access_xb,
6592         };
6593
6594         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6595                                                   &credits, &meta_ac);
6596         if (ret) {
6597                 mlog_errno(ret);
6598                 return ret;
6599         }
6600
6601         /* One more credits in case we need to add xattr flags in new inode. */
6602         handle = ocfs2_start_trans(osb, credits + 1);
6603         if (IS_ERR(handle)) {
6604                 ret = PTR_ERR(handle);
6605                 mlog_errno(ret);
6606                 goto out;
6607         }
6608
6609         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6610                 ret = ocfs2_journal_access_di(handle,
6611                                               INODE_CACHE(args->new_inode),
6612                                               args->new_bh,
6613                                               OCFS2_JOURNAL_ACCESS_WRITE);
6614                 if (ret) {
6615                         mlog_errno(ret);
6616                         goto out_commit;
6617                 }
6618         }
6619
6620         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6621                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6622         if (ret) {
6623                 mlog_errno(ret);
6624                 goto out_commit;
6625         }
6626
6627         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6628                osb->sb->s_blocksize - header_off);
6629
6630         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6631                                          new_blk_bh, new_xh, &vb, meta_ac,
6632                                          ocfs2_get_xattr_value_root, NULL);
6633         if (ret) {
6634                 mlog_errno(ret);
6635                 goto out_commit;
6636         }
6637
6638         ocfs2_journal_dirty(handle, new_blk_bh);
6639
6640         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6641                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6642                 spin_lock(&new_oi->ip_lock);
6643                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6644                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6645                 spin_unlock(&new_oi->ip_lock);
6646
6647                 ocfs2_journal_dirty(handle, args->new_bh);
6648         }
6649
6650 out_commit:
6651         ocfs2_commit_trans(osb, handle);
6652
6653 out:
6654         ocfs2_free_alloc_context(meta_ac);
6655         return ret;
6656 }
6657
6658 struct ocfs2_reflink_xattr_tree_args {
6659         struct ocfs2_xattr_reflink *reflink;
6660         struct buffer_head *old_blk_bh;
6661         struct buffer_head *new_blk_bh;
6662         struct ocfs2_xattr_bucket *old_bucket;
6663         struct ocfs2_xattr_bucket *new_bucket;
6664 };
6665
6666 /*
6667  * NOTE:
6668  * We have to handle the case that both old bucket and new bucket
6669  * will call this function to get the right ret_bh.
6670  * So The caller must give us the right bh.
6671  */
6672 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6673                                         struct buffer_head *bh,
6674                                         struct ocfs2_xattr_header *xh,
6675                                         int offset,
6676                                         struct ocfs2_xattr_value_root **xv,
6677                                         struct buffer_head **ret_bh,
6678                                         void *para)
6679 {
6680         struct ocfs2_reflink_xattr_tree_args *args =
6681                         (struct ocfs2_reflink_xattr_tree_args *)para;
6682         struct ocfs2_xattr_bucket *bucket;
6683
6684         if (bh == args->old_bucket->bu_bhs[0])
6685                 bucket = args->old_bucket;
6686         else
6687                 bucket = args->new_bucket;
6688
6689         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6690                                                xv, ret_bh);
6691 }
6692
6693 struct ocfs2_value_tree_metas {
6694         int num_metas;
6695         int credits;
6696         int num_recs;
6697 };
6698
6699 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6700                                         struct buffer_head *bh,
6701                                         struct ocfs2_xattr_header *xh,
6702                                         int offset,
6703                                         struct ocfs2_xattr_value_root **xv,
6704                                         struct buffer_head **ret_bh,
6705                                         void *para)
6706 {
6707         struct ocfs2_xattr_bucket *bucket =
6708                                 (struct ocfs2_xattr_bucket *)para;
6709
6710         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6711                                                xv, ret_bh);
6712 }
6713
6714 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6715                                       struct ocfs2_xattr_bucket *bucket,
6716                                       void *para)
6717 {
6718         struct ocfs2_value_tree_metas *metas =
6719                         (struct ocfs2_value_tree_metas *)para;
6720         struct ocfs2_xattr_header *xh =
6721                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6722
6723         /* Add the credits for this bucket first. */
6724         metas->credits += bucket->bu_blocks;
6725         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6726                                         xh, &metas->num_metas,
6727                                         &metas->credits, &metas->num_recs,
6728                                         ocfs2_value_tree_metas_in_bucket,
6729                                         bucket);
6730 }
6731
6732 /*
6733  * Given a xattr extent rec starting from blkno and having len clusters,
6734  * iterate all the buckets calculate how much metadata we need for reflinking
6735  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6736  */
6737 static int ocfs2_lock_reflink_xattr_rec_allocators(
6738                                 struct ocfs2_reflink_xattr_tree_args *args,
6739                                 struct ocfs2_extent_tree *xt_et,
6740                                 u64 blkno, u32 len, int *credits,
6741                                 struct ocfs2_alloc_context **meta_ac,
6742                                 struct ocfs2_alloc_context **data_ac)
6743 {
6744         int ret, num_free_extents;
6745         struct ocfs2_value_tree_metas metas;
6746         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6747         struct ocfs2_refcount_block *rb;
6748
6749         memset(&metas, 0, sizeof(metas));
6750
6751         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6752                                           ocfs2_calc_value_tree_metas, &metas);
6753         if (ret) {
6754                 mlog_errno(ret);
6755                 goto out;
6756         }
6757
6758         *credits = metas.credits;
6759
6760         /*
6761          * Calculate we need for refcount tree change.
6762          *
6763          * We need to add/modify num_recs in refcount tree, so just calculate
6764          * an approximate number we need for refcount tree change.
6765          * Sometimes we need to split the tree, and after split,  half recs
6766          * will be moved to the new block, and a new block can only provide
6767          * half number of recs. So we multiple new blocks by 2.
6768          * In the end, we have to add credits for modifying the already
6769          * existed refcount block.
6770          */
6771         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6772         metas.num_recs =
6773                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6774                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6775         metas.num_metas += metas.num_recs;
6776         *credits += metas.num_recs +
6777                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6778         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6779                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6780                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6781         else
6782                 *credits += 1;
6783
6784         /* count in the xattr tree change. */
6785         num_free_extents = ocfs2_num_free_extents(xt_et);
6786         if (num_free_extents < 0) {
6787                 ret = num_free_extents;
6788                 mlog_errno(ret);
6789                 goto out;
6790         }
6791
6792         if (num_free_extents < len)
6793                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6794
6795         *credits += ocfs2_calc_extend_credits(osb->sb,
6796                                               xt_et->et_root_el);
6797
6798         if (metas.num_metas) {
6799                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6800                                                         meta_ac);
6801                 if (ret) {
6802                         mlog_errno(ret);
6803                         goto out;
6804                 }
6805         }
6806
6807         if (len) {
6808                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6809                 if (ret)
6810                         mlog_errno(ret);
6811         }
6812 out:
6813         if (ret) {
6814                 if (*meta_ac) {
6815                         ocfs2_free_alloc_context(*meta_ac);
6816                         *meta_ac = NULL;
6817                 }
6818         }
6819
6820         return ret;
6821 }
6822
6823 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6824                                 u64 blkno, u64 new_blkno, u32 clusters,
6825                                 u32 *cpos, int num_buckets,
6826                                 struct ocfs2_alloc_context *meta_ac,
6827                                 struct ocfs2_alloc_context *data_ac,
6828                                 struct ocfs2_reflink_xattr_tree_args *args)
6829 {
6830         int i, j, ret = 0;
6831         struct super_block *sb = args->reflink->old_inode->i_sb;
6832         int bpb = args->old_bucket->bu_blocks;
6833         struct ocfs2_xattr_value_buf vb = {
6834                 .vb_access = ocfs2_journal_access,
6835         };
6836
6837         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6838                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6839                 if (ret) {
6840                         mlog_errno(ret);
6841                         break;
6842                 }
6843
6844                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6845                 if (ret) {
6846                         mlog_errno(ret);
6847                         break;
6848                 }
6849
6850                 ret = ocfs2_xattr_bucket_journal_access(handle,
6851                                                 args->new_bucket,
6852                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6853                 if (ret) {
6854                         mlog_errno(ret);
6855                         break;
6856                 }
6857
6858                 for (j = 0; j < bpb; j++)
6859                         memcpy(bucket_block(args->new_bucket, j),
6860                                bucket_block(args->old_bucket, j),
6861                                sb->s_blocksize);
6862
6863                 /*
6864                  * Record the start cpos so that we can use it to initialize
6865                  * our xattr tree we also set the xh_num_bucket for the new
6866                  * bucket.
6867                  */
6868                 if (i == 0) {
6869                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6870                                             xh_entries[0].xe_name_hash);
6871                         bucket_xh(args->new_bucket)->xh_num_buckets =
6872                                 cpu_to_le16(num_buckets);
6873                 }
6874
6875                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6876
6877                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6878                                         args->old_bucket->bu_bhs[0],
6879                                         bucket_xh(args->old_bucket),
6880                                         args->new_bucket->bu_bhs[0],
6881                                         bucket_xh(args->new_bucket),
6882                                         &vb, meta_ac,
6883                                         ocfs2_get_reflink_xattr_value_root,
6884                                         args);
6885                 if (ret) {
6886                         mlog_errno(ret);
6887                         break;
6888                 }
6889
6890                 /*
6891                  * Re-access and dirty the bucket to calculate metaecc.
6892                  * Because we may extend the transaction in reflink_xattr_header
6893                  * which will let the already accessed block gone.
6894                  */
6895                 ret = ocfs2_xattr_bucket_journal_access(handle,
6896                                                 args->new_bucket,
6897                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6898                 if (ret) {
6899                         mlog_errno(ret);
6900                         break;
6901                 }
6902
6903                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6904
6905                 ocfs2_xattr_bucket_relse(args->old_bucket);
6906                 ocfs2_xattr_bucket_relse(args->new_bucket);
6907         }
6908
6909         ocfs2_xattr_bucket_relse(args->old_bucket);
6910         ocfs2_xattr_bucket_relse(args->new_bucket);
6911         return ret;
6912 }
6913
6914 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6915                                 struct inode *inode,
6916                                 struct ocfs2_reflink_xattr_tree_args *args,
6917                                 struct ocfs2_extent_tree *et,
6918                                 struct ocfs2_alloc_context *meta_ac,
6919                                 struct ocfs2_alloc_context *data_ac,
6920                                 u64 blkno, u32 cpos, u32 len)
6921 {
6922         int ret, first_inserted = 0;
6923         u32 p_cluster, num_clusters, reflink_cpos = 0;
6924         u64 new_blkno;
6925         unsigned int num_buckets, reflink_buckets;
6926         unsigned int bpc =
6927                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6928
6929         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6930         if (ret) {
6931                 mlog_errno(ret);
6932                 goto out;
6933         }
6934         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6935         ocfs2_xattr_bucket_relse(args->old_bucket);
6936
6937         while (len && num_buckets) {
6938                 ret = ocfs2_claim_clusters(handle, data_ac,
6939                                            1, &p_cluster, &num_clusters);
6940                 if (ret) {
6941                         mlog_errno(ret);
6942                         goto out;
6943                 }
6944
6945                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6946                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6947
6948                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6949                                                  new_blkno, num_clusters,
6950                                                  &reflink_cpos, reflink_buckets,
6951                                                  meta_ac, data_ac, args);
6952                 if (ret) {
6953                         mlog_errno(ret);
6954                         goto out;
6955                 }
6956
6957                 /*
6958                  * For the 1st allocated cluster, we make it use the same cpos
6959                  * so that the xattr tree looks the same as the original one
6960                  * in the most case.
6961                  */
6962                 if (!first_inserted) {
6963                         reflink_cpos = cpos;
6964                         first_inserted = 1;
6965                 }
6966                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6967                                           num_clusters, 0, meta_ac);
6968                 if (ret)
6969                         mlog_errno(ret);
6970
6971                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6972                                                   num_clusters, reflink_cpos);
6973
6974                 len -= num_clusters;
6975                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6976                 num_buckets -= reflink_buckets;
6977         }
6978 out:
6979         return ret;
6980 }
6981
6982 /*
6983  * Create the same xattr extent record in the new inode's xattr tree.
6984  */
6985 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6986                                    struct buffer_head *root_bh,
6987                                    u64 blkno,
6988                                    u32 cpos,
6989                                    u32 len,
6990                                    void *para)
6991 {
6992         int ret, credits = 0;
6993         handle_t *handle;
6994         struct ocfs2_reflink_xattr_tree_args *args =
6995                         (struct ocfs2_reflink_xattr_tree_args *)para;
6996         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6997         struct ocfs2_alloc_context *meta_ac = NULL;
6998         struct ocfs2_alloc_context *data_ac = NULL;
6999         struct ocfs2_extent_tree et;
7000
7001         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7002
7003         ocfs2_init_xattr_tree_extent_tree(&et,
7004                                           INODE_CACHE(args->reflink->new_inode),
7005                                           args->new_blk_bh);
7006
7007         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7008                                                       len, &credits,
7009                                                       &meta_ac, &data_ac);
7010         if (ret) {
7011                 mlog_errno(ret);
7012                 goto out;
7013         }
7014
7015         handle = ocfs2_start_trans(osb, credits);
7016         if (IS_ERR(handle)) {
7017                 ret = PTR_ERR(handle);
7018                 mlog_errno(ret);
7019                 goto out;
7020         }
7021
7022         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7023                                           meta_ac, data_ac,
7024                                           blkno, cpos, len);
7025         if (ret)
7026                 mlog_errno(ret);
7027
7028         ocfs2_commit_trans(osb, handle);
7029
7030 out:
7031         if (meta_ac)
7032                 ocfs2_free_alloc_context(meta_ac);
7033         if (data_ac)
7034                 ocfs2_free_alloc_context(data_ac);
7035         return ret;
7036 }
7037
7038 /*
7039  * Create reflinked xattr buckets.
7040  * We will add bucket one by one, and refcount all the xattrs in the bucket
7041  * if they are stored outside.
7042  */
7043 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7044                                     struct buffer_head *blk_bh,
7045                                     struct buffer_head *new_blk_bh)
7046 {
7047         int ret;
7048         struct ocfs2_reflink_xattr_tree_args para;
7049
7050         memset(&para, 0, sizeof(para));
7051         para.reflink = args;
7052         para.old_blk_bh = blk_bh;
7053         para.new_blk_bh = new_blk_bh;
7054
7055         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7056         if (!para.old_bucket) {
7057                 mlog_errno(-ENOMEM);
7058                 return -ENOMEM;
7059         }
7060
7061         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7062         if (!para.new_bucket) {
7063                 ret = -ENOMEM;
7064                 mlog_errno(ret);
7065                 goto out;
7066         }
7067
7068         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7069                                               ocfs2_reflink_xattr_rec,
7070                                               &para);
7071         if (ret)
7072                 mlog_errno(ret);
7073
7074 out:
7075         ocfs2_xattr_bucket_free(para.old_bucket);
7076         ocfs2_xattr_bucket_free(para.new_bucket);
7077         return ret;
7078 }
7079
7080 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7081                                         struct buffer_head *blk_bh)
7082 {
7083         int ret, indexed = 0;
7084         struct buffer_head *new_blk_bh = NULL;
7085         struct ocfs2_xattr_block *xb =
7086                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7087
7088
7089         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7090                 indexed = 1;
7091
7092         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7093                                              &new_blk_bh, indexed);
7094         if (ret) {
7095                 mlog_errno(ret);
7096                 goto out;
7097         }
7098
7099         if (!indexed)
7100                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7101         else
7102                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7103         if (ret)
7104                 mlog_errno(ret);
7105
7106 out:
7107         brelse(new_blk_bh);
7108         return ret;
7109 }
7110
7111 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7112 {
7113         int type = ocfs2_xattr_get_type(xe);
7114
7115         return type != OCFS2_XATTR_INDEX_SECURITY &&
7116                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7117                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7118 }
7119
7120 int ocfs2_reflink_xattrs(struct inode *old_inode,
7121                          struct buffer_head *old_bh,
7122                          struct inode *new_inode,
7123                          struct buffer_head *new_bh,
7124                          bool preserve_security)
7125 {
7126         int ret;
7127         struct ocfs2_xattr_reflink args;
7128         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7129         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7130         struct buffer_head *blk_bh = NULL;
7131         struct ocfs2_cached_dealloc_ctxt dealloc;
7132         struct ocfs2_refcount_tree *ref_tree;
7133         struct buffer_head *ref_root_bh = NULL;
7134
7135         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7136                                        le64_to_cpu(di->i_refcount_loc),
7137                                        1, &ref_tree, &ref_root_bh);
7138         if (ret) {
7139                 mlog_errno(ret);
7140                 goto out;
7141         }
7142
7143         ocfs2_init_dealloc_ctxt(&dealloc);
7144
7145         args.old_inode = old_inode;
7146         args.new_inode = new_inode;
7147         args.old_bh = old_bh;
7148         args.new_bh = new_bh;
7149         args.ref_ci = &ref_tree->rf_ci;
7150         args.ref_root_bh = ref_root_bh;
7151         args.dealloc = &dealloc;
7152         if (preserve_security)
7153                 args.xattr_reflinked = NULL;
7154         else
7155                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7156
7157         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7158                 ret = ocfs2_reflink_xattr_inline(&args);
7159                 if (ret) {
7160                         mlog_errno(ret);
7161                         goto out_unlock;
7162                 }
7163         }
7164
7165         if (!di->i_xattr_loc)
7166                 goto out_unlock;
7167
7168         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7169                                      &blk_bh);
7170         if (ret < 0) {
7171                 mlog_errno(ret);
7172                 goto out_unlock;
7173         }
7174
7175         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7176         if (ret)
7177                 mlog_errno(ret);
7178
7179         brelse(blk_bh);
7180
7181 out_unlock:
7182         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7183                                    ref_tree, 1);
7184         brelse(ref_root_bh);
7185
7186         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7187                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7188                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7189         }
7190
7191 out:
7192         return ret;
7193 }
7194
7195 /*
7196  * Initialize security and acl for a already created inode.
7197  * Used for reflink a non-preserve-security file.
7198  *
7199  * It uses common api like ocfs2_xattr_set, so the caller
7200  * must not hold any lock expect i_mutex.
7201  */
7202 int ocfs2_init_security_and_acl(struct inode *dir,
7203                                 struct inode *inode,
7204                                 const struct qstr *qstr)
7205 {
7206         int ret = 0;
7207         struct buffer_head *dir_bh = NULL;
7208
7209         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7210         if (ret) {
7211                 mlog_errno(ret);
7212                 goto leave;
7213         }
7214
7215         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7216         if (ret) {
7217                 mlog_errno(ret);
7218                 goto leave;
7219         }
7220         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7221         if (ret)
7222                 mlog_errno(ret);
7223
7224         ocfs2_inode_unlock(dir, 0);
7225         brelse(dir_bh);
7226 leave:
7227         return ret;
7228 }
7229
7230 /*
7231  * 'security' attributes support
7232  */
7233 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7234                                     struct dentry *unused, struct inode *inode,
7235                                     const char *name, void *buffer, size_t size)
7236 {
7237         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7238                                name, buffer, size);
7239 }
7240
7241 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7242                                     struct dentry *unused, struct inode *inode,
7243                                     const char *name, const void *value,
7244                                     size_t size, int flags)
7245 {
7246         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7247                                name, value, size, flags);
7248 }
7249
7250 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7251                      void *fs_info)
7252 {
7253         const struct xattr *xattr;
7254         int err = 0;
7255
7256         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7257                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7258                                       xattr->name, xattr->value,
7259                                       xattr->value_len, XATTR_CREATE);
7260                 if (err)
7261                         break;
7262         }
7263         return err;
7264 }
7265
7266 int ocfs2_init_security_get(struct inode *inode,
7267                             struct inode *dir,
7268                             const struct qstr *qstr,
7269                             struct ocfs2_security_xattr_info *si)
7270 {
7271         /* check whether ocfs2 support feature xattr */
7272         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7273                 return -EOPNOTSUPP;
7274         if (si)
7275                 return security_old_inode_init_security(inode, dir, qstr,
7276                                                         &si->name, &si->value,
7277                                                         &si->value_len);
7278
7279         return security_inode_init_security(inode, dir, qstr,
7280                                             &ocfs2_initxattrs, NULL);
7281 }
7282
7283 int ocfs2_init_security_set(handle_t *handle,
7284                             struct inode *inode,
7285                             struct buffer_head *di_bh,
7286                             struct ocfs2_security_xattr_info *si,
7287                             struct ocfs2_alloc_context *xattr_ac,
7288                             struct ocfs2_alloc_context *data_ac)
7289 {
7290         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7291                                      OCFS2_XATTR_INDEX_SECURITY,
7292                                      si->name, si->value, si->value_len, 0,
7293                                      xattr_ac, data_ac);
7294 }
7295
7296 const struct xattr_handler ocfs2_xattr_security_handler = {
7297         .prefix = XATTR_SECURITY_PREFIX,
7298         .get    = ocfs2_xattr_security_get,
7299         .set    = ocfs2_xattr_security_set,
7300 };
7301
7302 /*
7303  * 'trusted' attributes support
7304  */
7305 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7306                                    struct dentry *unused, struct inode *inode,
7307                                    const char *name, void *buffer, size_t size)
7308 {
7309         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7310                                name, buffer, size);
7311 }
7312
7313 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7314                                    struct dentry *unused, struct inode *inode,
7315                                    const char *name, const void *value,
7316                                    size_t size, int flags)
7317 {
7318         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7319                                name, value, size, flags);
7320 }
7321
7322 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7323         .prefix = XATTR_TRUSTED_PREFIX,
7324         .get    = ocfs2_xattr_trusted_get,
7325         .set    = ocfs2_xattr_trusted_set,
7326 };
7327
7328 /*
7329  * 'user' attributes support
7330  */
7331 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7332                                 struct dentry *unused, struct inode *inode,
7333                                 const char *name, void *buffer, size_t size)
7334 {
7335         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7336
7337         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7338                 return -EOPNOTSUPP;
7339         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7340                                buffer, size);
7341 }
7342
7343 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7344                                 struct dentry *unused, struct inode *inode,
7345                                 const char *name, const void *value,
7346                                 size_t size, int flags)
7347 {
7348         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7349
7350         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7351                 return -EOPNOTSUPP;
7352
7353         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7354                                name, value, size, flags);
7355 }
7356
7357 const struct xattr_handler ocfs2_xattr_user_handler = {
7358         .prefix = XATTR_USER_PREFIX,
7359         .get    = ocfs2_xattr_user_get,
7360         .set    = ocfs2_xattr_user_set,
7361 };