fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58 #include "refcounttree.h"
  59 #include "acl.h"
  60
  61 struct ocfs2_xattr_def_value_root {
  62         struct ocfs2_xattr_value_root   xv;
  63         struct ocfs2_extent_rec         er;
  64 };
  65
  66 struct ocfs2_xattr_bucket {
  67         /* The inode these xattrs are associated with */
  68         struct inode *bu_inode;
  69
  70         /* The actual buffers that make up the bucket */
  71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  72
  73         /* How many blocks make up one bucket for this filesystem */
  74         int bu_blocks;
  75 };
  76
  77 struct ocfs2_xattr_set_ctxt {
  78         handle_t *handle;
  79         struct ocfs2_alloc_context *meta_ac;
  80         struct ocfs2_alloc_context *data_ac;
  81         struct ocfs2_cached_dealloc_ctxt dealloc;
  82         int set_abort;
  83 };
  84
  85 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  86 #define OCFS2_XATTR_INLINE_SIZE 80
  87 #define OCFS2_XATTR_HEADER_GAP  4
  88 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  89                                          - sizeof(struct ocfs2_xattr_header) \
  90                                          - OCFS2_XATTR_HEADER_GAP)
  91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  92                                          - sizeof(struct ocfs2_xattr_block) \
  93                                          - sizeof(struct ocfs2_xattr_header) \
  94                                          - OCFS2_XATTR_HEADER_GAP)
  95
  96 static struct ocfs2_xattr_def_value_root def_xv = {
  97         .xv.xr_list.l_count = cpu_to_le16(1),
  98 };
  99
 100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
 101         &ocfs2_xattr_user_handler,
 102         &ocfs2_xattr_acl_access_handler,
 103         &ocfs2_xattr_acl_default_handler,
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 117 };
 118
 119 struct ocfs2_xattr_info {
 120         int             xi_name_index;
 121         const char      *xi_name;
 122         int             xi_name_len;
 123         const void      *xi_value;
 124         size_t          xi_value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 /* Operations on struct ocfs2_xa_entry */
 143 struct ocfs2_xa_loc;
 144 struct ocfs2_xa_loc_operations {
 145         /*
 146          * Journal functions
 147          */
 148         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 149                                   int type);
 150         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 151
 152         /*
 153          * Return a pointer to the appropriate buffer in loc->xl_storage
 154          * at the given offset from loc->xl_header.
 155          */
 156         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 157
 158         /* Can we reuse the existing entry for the new value? */
 159         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 160                              struct ocfs2_xattr_info *xi);
 161
 162         /* How much space is needed for the new value? */
 163         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 164                                struct ocfs2_xattr_info *xi);
 165
 166         /*
 167          * Return the offset of the first name+value pair.  This is
 168          * the start of our downward-filling free space.
 169          */
 170         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 171
 172         /*
 173          * Remove the name+value at this location.  Do whatever is
 174          * appropriate with the remaining name+value pairs.
 175          */
 176         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 177
 178         /* Fill xl_entry with a new entry */
 179         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 180
 181         /* Add name+value storage to an entry */
 182         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 183
 184         /*
 185          * Initialize the value buf's access and bh fields for this entry.
 186          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 187          */
 188         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 189                                    struct ocfs2_xattr_value_buf *vb);
 190 };
 191
 192 /*
 193  * Describes an xattr entry location.  This is a memory structure
 194  * tracking the on-disk structure.
 195  */
 196 struct ocfs2_xa_loc {
 197         /* This xattr belongs to this inode */
 198         struct inode *xl_inode;
 199
 200         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 201         struct ocfs2_xattr_header *xl_header;
 202
 203         /* Bytes from xl_header to the end of the storage */
 204         int xl_size;
 205
 206         /*
 207          * The ocfs2_xattr_entry this location describes.  If this is
 208          * NULL, this location describes the on-disk structure where it
 209          * would have been.
 210          */
 211         struct ocfs2_xattr_entry *xl_entry;
 212
 213         /*
 214          * Internal housekeeping
 215          */
 216
 217         /* Buffer(s) containing this entry */
 218         void *xl_storage;
 219
 220         /* Operations on the storage backing this location */
 221         const struct ocfs2_xa_loc_operations *xl_ops;
 222 };
 223
 224 /*
 225  * Convenience functions to calculate how much space is needed for a
 226  * given name+value pair
 227  */
 228 static int namevalue_size(int name_len, uint64_t value_len)
 229 {
 230         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 231                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 232         else
 233                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 234 }
 235
 236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237 {
 238         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239 }
 240
 241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 242 {
 243         u64 value_len = le64_to_cpu(xe->xe_value_size);
 244
 245         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 246                ocfs2_xattr_is_local(xe));
 247         return namevalue_size(xe->xe_name_len, value_len);
 248 }
 249
 250
 251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 252                                              struct ocfs2_xattr_header *xh,
 253                                              int index,
 254                                              int *block_off,
 255                                              int *new_offset);
 256
 257 static int ocfs2_xattr_block_find(struct inode *inode,
 258                                   int name_index,
 259                                   const char *name,
 260                                   struct ocfs2_xattr_search *xs);
 261 static int ocfs2_xattr_index_block_find(struct inode *inode,
 262                                         struct buffer_head *root_bh,
 263                                         int name_index,
 264                                         const char *name,
 265                                         struct ocfs2_xattr_search *xs);
 266
 267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 268                                         struct buffer_head *blk_bh,
 269                                         char *buffer,
 270                                         size_t buffer_size);
 271
 272 static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                           struct ocfs2_xattr_search *xs,
 274                                           struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 277                                              struct ocfs2_xattr_info *xi,
 278                                              struct ocfs2_xattr_search *xs,
 279                                              struct ocfs2_xattr_set_ctxt *ctxt);
 280
 281 typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno, u32 cpos, u32 len, void *para);
 284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 285                                            struct buffer_head *root_bh,
 286                                            xattr_tree_rec_func *rec_func,
 287                                            void *para);
 288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                         struct ocfs2_xattr_bucket *bucket,
 290                                         void *para);
 291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 292                                   struct buffer_head *root_bh,
 293                                   u64 blkno,
 294                                   u32 cpos,
 295                                   u32 len,
 296                                   void *para);
 297
 298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 299                                   u64 src_blk, u64 last_blk, u64 to_blk,
 300                                   unsigned int start_bucket,
 301                                   u32 *first_hash);
 302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 303                                         struct ocfs2_dinode *di,
 304                                         struct ocfs2_xattr_info *xi,
 305                                         struct ocfs2_xattr_search *xis,
 306                                         struct ocfs2_xattr_search *xbs,
 307                                         struct ocfs2_refcount_tree **ref_tree,
 308                                         int *meta_need,
 309                                         int *credits);
 310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 311                                            struct ocfs2_xattr_bucket *bucket,
 312                                            int offset,
 313                                            struct ocfs2_xattr_value_root **xv,
 314                                            struct buffer_head **bh);
 315
 316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 317 {
 318         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 319 }
 320
 321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 322 {
 323         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 324 }
 325
 326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 329
 330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 331 {
 332         struct ocfs2_xattr_bucket *bucket;
 333         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 334
 335         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 336
 337         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 338         if (bucket) {
 339                 bucket->bu_inode = inode;
 340                 bucket->bu_blocks = blks;
 341         }
 342
 343         return bucket;
 344 }
 345
 346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         int i;
 349
 350         for (i = 0; i < bucket->bu_blocks; i++) {
 351                 brelse(bucket->bu_bhs[i]);
 352                 bucket->bu_bhs[i] = NULL;
 353         }
 354 }
 355
 356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 357 {
 358         if (bucket) {
 359                 ocfs2_xattr_bucket_relse(bucket);
 360                 bucket->bu_inode = NULL;
 361                 kfree(bucket);
 362         }
 363 }
 364
 365 /*
 366  * A bucket that has never been written to disk doesn't need to be
 367  * read.  We just need the buffer_heads.  Don't call this for
 368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 369  * them fully.
 370  */
 371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                    u64 xb_blkno)
 373 {
 374         int i, rc = 0;
 375
 376         for (i = 0; i < bucket->bu_blocks; i++) {
 377                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                               xb_blkno + i);
 379                 if (!bucket->bu_bhs[i]) {
 380                         rc = -EIO;
 381                         mlog_errno(rc);
 382                         break;
 383                 }
 384
 385                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                            bucket->bu_bhs[i]))
 387                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 388                                                       bucket->bu_bhs[i]);
 389         }
 390
 391         if (rc)
 392                 ocfs2_xattr_bucket_relse(bucket);
 393         return rc;
 394 }
 395
 396 /* Read the xattr bucket at xb_blkno */
 397 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 398                                    u64 xb_blkno)
 399 {
 400         int rc;
 401
 402         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 403                                bucket->bu_blocks, bucket->bu_bhs, 0,
 404                                NULL);
 405         if (!rc) {
 406                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 407                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 408                                                  bucket->bu_bhs,
 409                                                  bucket->bu_blocks,
 410                                                  &bucket_xh(bucket)->xh_check);
 411                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 412                 if (rc)
 413                         mlog_errno(rc);
 414         }
 415
 416         if (rc)
 417                 ocfs2_xattr_bucket_relse(bucket);
 418         return rc;
 419 }
 420
 421 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 422                                              struct ocfs2_xattr_bucket *bucket,
 423                                              int type)
 424 {
 425         int i, rc = 0;
 426
 427         for (i = 0; i < bucket->bu_blocks; i++) {
 428                 rc = ocfs2_journal_access(handle,
 429                                           INODE_CACHE(bucket->bu_inode),
 430                                           bucket->bu_bhs[i], type);
 431                 if (rc) {
 432                         mlog_errno(rc);
 433                         break;
 434                 }
 435         }
 436
 437         return rc;
 438 }
 439
 440 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 441                                              struct ocfs2_xattr_bucket *bucket)
 442 {
 443         int i;
 444
 445         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 446         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 447                                    bucket->bu_bhs, bucket->bu_blocks,
 448                                    &bucket_xh(bucket)->xh_check);
 449         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 450
 451         for (i = 0; i < bucket->bu_blocks; i++)
 452                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 453 }
 454
 455 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 456                                          struct ocfs2_xattr_bucket *src)
 457 {
 458         int i;
 459         int blocksize = src->bu_inode->i_sb->s_blocksize;
 460
 461         BUG_ON(dest->bu_blocks != src->bu_blocks);
 462         BUG_ON(dest->bu_inode != src->bu_inode);
 463
 464         for (i = 0; i < src->bu_blocks; i++) {
 465                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 466                        blocksize);
 467         }
 468 }
 469
 470 static int ocfs2_validate_xattr_block(struct super_block *sb,
 471                                       struct buffer_head *bh)
 472 {
 473         int rc;
 474         struct ocfs2_xattr_block *xb =
 475                 (struct ocfs2_xattr_block *)bh->b_data;
 476
 477         mlog(0, "Validating xattr block %llu\n",
 478              (unsigned long long)bh->b_blocknr);
 479
 480         BUG_ON(!buffer_uptodate(bh));
 481
 482         /*
 483          * If the ecc fails, we return the error but otherwise
 484          * leave the filesystem running.  We know any error is
 485          * local to this block.
 486          */
 487         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 488         if (rc)
 489                 return rc;
 490
 491         /*
 492          * Errors after here are fatal
 493          */
 494
 495         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 496                 ocfs2_error(sb,
 497                             "Extended attribute block #%llu has bad "
 498                             "signature %.*s",
 499                             (unsigned long long)bh->b_blocknr, 7,
 500                             xb->xb_signature);
 501                 return -EINVAL;
 502         }
 503
 504         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 505                 ocfs2_error(sb,
 506                             "Extended attribute block #%llu has an "
 507                             "invalid xb_blkno of %llu",
 508                             (unsigned long long)bh->b_blocknr,
 509                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 510                 return -EINVAL;
 511         }
 512
 513         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 514                 ocfs2_error(sb,
 515                             "Extended attribute block #%llu has an invalid "
 516                             "xb_fs_generation of #%u",
 517                             (unsigned long long)bh->b_blocknr,
 518                             le32_to_cpu(xb->xb_fs_generation));
 519                 return -EINVAL;
 520         }
 521
 522         return 0;
 523 }
 524
 525 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 526                                   struct buffer_head **bh)
 527 {
 528         int rc;
 529         struct buffer_head *tmp = *bh;
 530
 531         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 532                               ocfs2_validate_xattr_block);
 533
 534         /* If ocfs2_read_block() got us a new bh, pass it up. */
 535         if (!rc && !*bh)
 536                 *bh = tmp;
 537
 538         return rc;
 539 }
 540
 541 static inline const char *ocfs2_xattr_prefix(int name_index)
 542 {
 543         const struct xattr_handler *handler = NULL;
 544
 545         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 546                 handler = ocfs2_xattr_handler_map[name_index];
 547
 548         return handler ? handler->prefix : NULL;
 549 }
 550
 551 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 552                                  const char *name,
 553                                  int name_len)
 554 {
 555         /* Get hash value of uuid from super block */
 556         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 557         int i;
 558
 559         /* hash extended attribute name */
 560         for (i = 0; i < name_len; i++) {
 561                 hash = (hash << OCFS2_HASH_SHIFT) ^
 562                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 563                        *name++;
 564         }
 565
 566         return hash;
 567 }
 568
 569 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 570 {
 571         return namevalue_size(name_len, value_len) +
 572                 sizeof(struct ocfs2_xattr_entry);
 573 }
 574
 575 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 576 {
 577         return namevalue_size_xi(xi) +
 578                 sizeof(struct ocfs2_xattr_entry);
 579 }
 580
 581 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 582 {
 583         return namevalue_size_xe(xe) +
 584                 sizeof(struct ocfs2_xattr_entry);
 585 }
 586
 587 int ocfs2_calc_security_init(struct inode *dir,
 588                              struct ocfs2_security_xattr_info *si,
 589                              int *want_clusters,
 590                              int *xattr_credits,
 591                              struct ocfs2_alloc_context **xattr_ac)
 592 {
 593         int ret = 0;
 594         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 595         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 596                                                  si->value_len);
 597
 598         /*
 599          * The max space of security xattr taken inline is
 600          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 601          * So reserve one metadata block for it is ok.
 602          */
 603         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 604             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 605                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 606                 if (ret) {
 607                         mlog_errno(ret);
 608                         return ret;
 609                 }
 610                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 611         }
 612
 613         /* reserve clusters for xattr value which will be set in B tree*/
 614         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 615                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 616                                                             si->value_len);
 617
 618                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 619                                                            new_clusters);
 620                 *want_clusters += new_clusters;
 621         }
 622         return ret;
 623 }
 624
 625 int ocfs2_calc_xattr_init(struct inode *dir,
 626                           struct buffer_head *dir_bh,
 627                           int mode,
 628                           struct ocfs2_security_xattr_info *si,
 629                           int *want_clusters,
 630                           int *xattr_credits,
 631                           int *want_meta)
 632 {
 633         int ret = 0;
 634         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 635         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 636
 637         if (si->enable)
 638                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 639                                                      si->value_len);
 640
 641         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 642                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 643                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 644                                         "", NULL, 0);
 645                 if (acl_len > 0) {
 646                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 647                         if (S_ISDIR(mode))
 648                                 a_size <<= 1;
 649                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 650                         mlog_errno(ret);
 651                         return ret;
 652                 }
 653         }
 654
 655         if (!(s_size + a_size))
 656                 return ret;
 657
 658         /*
 659          * The max space of security xattr taken inline is
 660          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 661          * The max space of acl xattr taken inline is
 662          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 663          * when blocksize = 512, may reserve one more cluser for
 664          * xattr bucket, otherwise reserve one metadata block
 665          * for them is ok.
 666          * If this is a new directory with inline data,
 667          * we choose to reserve the entire inline area for
 668          * directory contents and force an external xattr block.
 669          */
 670         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 671             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 672             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 673                 *want_meta = *want_meta + 1;
 674                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 675         }
 676
 677         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 678             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 679                 *want_clusters += 1;
 680                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 681         }
 682
 683         /*
 684          * reserve credits and clusters for xattrs which has large value
 685          * and have to be set outside
 686          */
 687         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 688                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 689                                                         si->value_len);
 690                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 691                                                            new_clusters);
 692                 *want_clusters += new_clusters;
 693         }
 694         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 695             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 696                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 697                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 698                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 699                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 700                                                            new_clusters);
 701                 *want_clusters += new_clusters;
 702         }
 703
 704         return ret;
 705 }
 706
 707 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 708                                          u32 clusters_to_add,
 709                                          struct ocfs2_xattr_value_buf *vb,
 710                                          struct ocfs2_xattr_set_ctxt *ctxt)
 711 {
 712         int status = 0, credits;
 713         handle_t *handle = ctxt->handle;
 714         enum ocfs2_alloc_restarted why;
 715         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 716         struct ocfs2_extent_tree et;
 717
 718         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 719
 720         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 721
 722         while (clusters_to_add) {
 723                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 724                                        OCFS2_JOURNAL_ACCESS_WRITE);
 725                 if (status < 0) {
 726                         mlog_errno(status);
 727                         break;
 728                 }
 729
 730                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 731                 status = ocfs2_add_clusters_in_btree(handle,
 732                                                      &et,
 733                                                      &logical_start,
 734                                                      clusters_to_add,
 735                                                      0,
 736                                                      ctxt->data_ac,
 737                                                      ctxt->meta_ac,
 738                                                      &why);
 739                 if ((status < 0) && (status != -EAGAIN)) {
 740                         if (status != -ENOSPC)
 741                                 mlog_errno(status);
 742                         break;
 743                 }
 744
 745                 ocfs2_journal_dirty(handle, vb->vb_bh);
 746
 747                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 748                                          prev_clusters;
 749
 750                 if (why != RESTART_NONE && clusters_to_add) {
 751                         /*
 752                          * We can only fail in case the alloc file doesn't give
 753                          * up enough clusters.
 754                          */
 755                         BUG_ON(why == RESTART_META);
 756
 757                         mlog(0, "restarting xattr value extension for %u"
 758                              " clusters,.\n", clusters_to_add);
 759                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 760                                                             &vb->vb_xv->xr_list,
 761                                                             clusters_to_add);
 762                         status = ocfs2_extend_trans(handle, credits);
 763                         if (status < 0) {
 764                                 status = -ENOMEM;
 765                                 mlog_errno(status);
 766                                 break;
 767                         }
 768                 }
 769         }
 770
 771         return status;
 772 }
 773
 774 static int __ocfs2_remove_xattr_range(struct inode *inode,
 775                                       struct ocfs2_xattr_value_buf *vb,
 776                                       u32 cpos, u32 phys_cpos, u32 len,
 777                                       unsigned int ext_flags,
 778                                       struct ocfs2_xattr_set_ctxt *ctxt)
 779 {
 780         int ret;
 781         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 782         handle_t *handle = ctxt->handle;
 783         struct ocfs2_extent_tree et;
 784
 785         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 786
 787         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 788                             OCFS2_JOURNAL_ACCESS_WRITE);
 789         if (ret) {
 790                 mlog_errno(ret);
 791                 goto out;
 792         }
 793
 794         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 795                                   &ctxt->dealloc);
 796         if (ret) {
 797                 mlog_errno(ret);
 798                 goto out;
 799         }
 800
 801         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 802         ocfs2_journal_dirty(handle, vb->vb_bh);
 803
 804         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 805                 ret = ocfs2_decrease_refcount(inode, handle,
 806                                         ocfs2_blocks_to_clusters(inode->i_sb,
 807                                                                  phys_blkno),
 808                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 809         else
 810                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 811                                                   phys_blkno, len);
 812         if (ret)
 813                 mlog_errno(ret);
 814
 815 out:
 816         return ret;
 817 }
 818
 819 static int ocfs2_xattr_shrink_size(struct inode *inode,
 820                                    u32 old_clusters,
 821                                    u32 new_clusters,
 822                                    struct ocfs2_xattr_value_buf *vb,
 823                                    struct ocfs2_xattr_set_ctxt *ctxt)
 824 {
 825         int ret = 0;
 826         unsigned int ext_flags;
 827         u32 trunc_len, cpos, phys_cpos, alloc_size;
 828         u64 block;
 829
 830         if (old_clusters <= new_clusters)
 831                 return 0;
 832
 833         cpos = new_clusters;
 834         trunc_len = old_clusters - new_clusters;
 835         while (trunc_len) {
 836                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 837                                                &alloc_size,
 838                                                &vb->vb_xv->xr_list, &ext_flags);
 839                 if (ret) {
 840                         mlog_errno(ret);
 841                         goto out;
 842                 }
 843
 844                 if (alloc_size > trunc_len)
 845                         alloc_size = trunc_len;
 846
 847                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 848                                                  phys_cpos, alloc_size,
 849                                                  ext_flags, ctxt);
 850                 if (ret) {
 851                         mlog_errno(ret);
 852                         goto out;
 853                 }
 854
 855                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 856                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 857                                                        block, alloc_size);
 858                 cpos += alloc_size;
 859                 trunc_len -= alloc_size;
 860         }
 861
 862 out:
 863         return ret;
 864 }
 865
 866 static int ocfs2_xattr_value_truncate(struct inode *inode,
 867                                       struct ocfs2_xattr_value_buf *vb,
 868                                       int len,
 869                                       struct ocfs2_xattr_set_ctxt *ctxt)
 870 {
 871         int ret;
 872         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 873         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 874
 875         if (new_clusters == old_clusters)
 876                 return 0;
 877
 878         if (new_clusters > old_clusters)
 879                 ret = ocfs2_xattr_extend_allocation(inode,
 880                                                     new_clusters - old_clusters,
 881                                                     vb, ctxt);
 882         else
 883                 ret = ocfs2_xattr_shrink_size(inode,
 884                                               old_clusters, new_clusters,
 885                                               vb, ctxt);
 886
 887         return ret;
 888 }
 889
 890 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 891                                   size_t *result, const char *prefix,
 892                                   const char *name, int name_len)
 893 {
 894         char *p = buffer + *result;
 895         int prefix_len = strlen(prefix);
 896         int total_len = prefix_len + name_len + 1;
 897
 898         *result += total_len;
 899
 900         /* we are just looking for how big our buffer needs to be */
 901         if (!size)
 902                 return 0;
 903
 904         if (*result > size)
 905                 return -ERANGE;
 906
 907         memcpy(p, prefix, prefix_len);
 908         memcpy(p + prefix_len, name, name_len);
 909         p[prefix_len + name_len] = '\0';
 910
 911         return 0;
 912 }
 913
 914 static int ocfs2_xattr_list_entries(struct inode *inode,
 915                                     struct ocfs2_xattr_header *header,
 916                                     char *buffer, size_t buffer_size)
 917 {
 918         size_t result = 0;
 919         int i, type, ret;
 920         const char *prefix, *name;
 921
 922         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 923                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 924                 type = ocfs2_xattr_get_type(entry);
 925                 prefix = ocfs2_xattr_prefix(type);
 926
 927                 if (prefix) {
 928                         name = (const char *)header +
 929                                 le16_to_cpu(entry->xe_name_offset);
 930
 931                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 932                                                      &result, prefix, name,
 933                                                      entry->xe_name_len);
 934                         if (ret)
 935                                 return ret;
 936                 }
 937         }
 938
 939         return result;
 940 }
 941
 942 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 943                                          struct ocfs2_dinode *di)
 944 {
 945         struct ocfs2_xattr_header *xh;
 946         int i;
 947
 948         xh = (struct ocfs2_xattr_header *)
 949                  ((void *)di + inode->i_sb->s_blocksize -
 950                  le16_to_cpu(di->i_xattr_inline_size));
 951
 952         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 953                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 954                         return 1;
 955
 956         return 0;
 957 }
 958
 959 static int ocfs2_xattr_ibody_list(struct inode *inode,
 960                                   struct ocfs2_dinode *di,
 961                                   char *buffer,
 962                                   size_t buffer_size)
 963 {
 964         struct ocfs2_xattr_header *header = NULL;
 965         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 966         int ret = 0;
 967
 968         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 969                 return ret;
 970
 971         header = (struct ocfs2_xattr_header *)
 972                  ((void *)di + inode->i_sb->s_blocksize -
 973                  le16_to_cpu(di->i_xattr_inline_size));
 974
 975         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 976
 977         return ret;
 978 }
 979
 980 static int ocfs2_xattr_block_list(struct inode *inode,
 981                                   struct ocfs2_dinode *di,
 982                                   char *buffer,
 983                                   size_t buffer_size)
 984 {
 985         struct buffer_head *blk_bh = NULL;
 986         struct ocfs2_xattr_block *xb;
 987         int ret = 0;
 988
 989         if (!di->i_xattr_loc)
 990                 return ret;
 991
 992         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 993                                      &blk_bh);
 994         if (ret < 0) {
 995                 mlog_errno(ret);
 996                 return ret;
 997         }
 998
 999         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1000         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1001                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1002                 ret = ocfs2_xattr_list_entries(inode, header,
1003                                                buffer, buffer_size);
1004         } else
1005                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1006                                                    buffer, buffer_size);
1007
1008         brelse(blk_bh);
1009
1010         return ret;
1011 }
1012
1013 ssize_t ocfs2_listxattr(struct dentry *dentry,
1014                         char *buffer,
1015                         size_t size)
1016 {
1017         int ret = 0, i_ret = 0, b_ret = 0;
1018         struct buffer_head *di_bh = NULL;
1019         struct ocfs2_dinode *di = NULL;
1020         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1021
1022         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1023                 return -EOPNOTSUPP;
1024
1025         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1026                 return ret;
1027
1028         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1029         if (ret < 0) {
1030                 mlog_errno(ret);
1031                 return ret;
1032         }
1033
1034         di = (struct ocfs2_dinode *)di_bh->b_data;
1035
1036         down_read(&oi->ip_xattr_sem);
1037         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1038         if (i_ret < 0)
1039                 b_ret = 0;
1040         else {
1041                 if (buffer) {
1042                         buffer += i_ret;
1043                         size -= i_ret;
1044                 }
1045                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1046                                                buffer, size);
1047                 if (b_ret < 0)
1048                         i_ret = 0;
1049         }
1050         up_read(&oi->ip_xattr_sem);
1051         ocfs2_inode_unlock(dentry->d_inode, 0);
1052
1053         brelse(di_bh);
1054
1055         return i_ret + b_ret;
1056 }
1057
1058 static int ocfs2_xattr_find_entry(int name_index,
1059                                   const char *name,
1060                                   struct ocfs2_xattr_search *xs)
1061 {
1062         struct ocfs2_xattr_entry *entry;
1063         size_t name_len;
1064         int i, cmp = 1;
1065
1066         if (name == NULL)
1067                 return -EINVAL;
1068
1069         name_len = strlen(name);
1070         entry = xs->here;
1071         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1072                 cmp = name_index - ocfs2_xattr_get_type(entry);
1073                 if (!cmp)
1074                         cmp = name_len - entry->xe_name_len;
1075                 if (!cmp)
1076                         cmp = memcmp(name, (xs->base +
1077                                      le16_to_cpu(entry->xe_name_offset)),
1078                                      name_len);
1079                 if (cmp == 0)
1080                         break;
1081                 entry += 1;
1082         }
1083         xs->here = entry;
1084
1085         return cmp ? -ENODATA : 0;
1086 }
1087
1088 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1089                                          struct ocfs2_xattr_value_root *xv,
1090                                          void *buffer,
1091                                          size_t len)
1092 {
1093         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1094         u64 blkno;
1095         int i, ret = 0;
1096         size_t cplen, blocksize;
1097         struct buffer_head *bh = NULL;
1098         struct ocfs2_extent_list *el;
1099
1100         el = &xv->xr_list;
1101         clusters = le32_to_cpu(xv->xr_clusters);
1102         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1103         blocksize = inode->i_sb->s_blocksize;
1104
1105         cpos = 0;
1106         while (cpos < clusters) {
1107                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1108                                                &num_clusters, el, NULL);
1109                 if (ret) {
1110                         mlog_errno(ret);
1111                         goto out;
1112                 }
1113
1114                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1115                 /* Copy ocfs2_xattr_value */
1116                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1117                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1118                                                &bh, NULL);
1119                         if (ret) {
1120                                 mlog_errno(ret);
1121                                 goto out;
1122                         }
1123
1124                         cplen = len >= blocksize ? blocksize : len;
1125                         memcpy(buffer, bh->b_data, cplen);
1126                         len -= cplen;
1127                         buffer += cplen;
1128
1129                         brelse(bh);
1130                         bh = NULL;
1131                         if (len == 0)
1132                                 break;
1133                 }
1134                 cpos += num_clusters;
1135         }
1136 out:
1137         return ret;
1138 }
1139
1140 static int ocfs2_xattr_ibody_get(struct inode *inode,
1141                                  int name_index,
1142                                  const char *name,
1143                                  void *buffer,
1144                                  size_t buffer_size,
1145                                  struct ocfs2_xattr_search *xs)
1146 {
1147         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1148         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1149         struct ocfs2_xattr_value_root *xv;
1150         size_t size;
1151         int ret = 0;
1152
1153         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1154                 return -ENODATA;
1155
1156         xs->end = (void *)di + inode->i_sb->s_blocksize;
1157         xs->header = (struct ocfs2_xattr_header *)
1158                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1159         xs->base = (void *)xs->header;
1160         xs->here = xs->header->xh_entries;
1161
1162         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1163         if (ret)
1164                 return ret;
1165         size = le64_to_cpu(xs->here->xe_value_size);
1166         if (buffer) {
1167                 if (size > buffer_size)
1168                         return -ERANGE;
1169                 if (ocfs2_xattr_is_local(xs->here)) {
1170                         memcpy(buffer, (void *)xs->base +
1171                                le16_to_cpu(xs->here->xe_name_offset) +
1172                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1173                 } else {
1174                         xv = (struct ocfs2_xattr_value_root *)
1175                                 (xs->base + le16_to_cpu(
1176                                  xs->here->xe_name_offset) +
1177                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1178                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1179                                                             buffer, size);
1180                         if (ret < 0) {
1181                                 mlog_errno(ret);
1182                                 return ret;
1183                         }
1184                 }
1185         }
1186
1187         return size;
1188 }
1189
1190 static int ocfs2_xattr_block_get(struct inode *inode,
1191                                  int name_index,
1192                                  const char *name,
1193                                  void *buffer,
1194                                  size_t buffer_size,
1195                                  struct ocfs2_xattr_search *xs)
1196 {
1197         struct ocfs2_xattr_block *xb;
1198         struct ocfs2_xattr_value_root *xv;
1199         size_t size;
1200         int ret = -ENODATA, name_offset, name_len, i;
1201         int uninitialized_var(block_off);
1202
1203         xs->bucket = ocfs2_xattr_bucket_new(inode);
1204         if (!xs->bucket) {
1205                 ret = -ENOMEM;
1206                 mlog_errno(ret);
1207                 goto cleanup;
1208         }
1209
1210         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1211         if (ret) {
1212                 mlog_errno(ret);
1213                 goto cleanup;
1214         }
1215
1216         if (xs->not_found) {
1217                 ret = -ENODATA;
1218                 goto cleanup;
1219         }
1220
1221         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1222         size = le64_to_cpu(xs->here->xe_value_size);
1223         if (buffer) {
1224                 ret = -ERANGE;
1225                 if (size > buffer_size)
1226                         goto cleanup;
1227
1228                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1229                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1230                 i = xs->here - xs->header->xh_entries;
1231
1232                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1233                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1234                                                                 bucket_xh(xs->bucket),
1235                                                                 i,
1236                                                                 &block_off,
1237                                                                 &name_offset);
1238                         xs->base = bucket_block(xs->bucket, block_off);
1239                 }
1240                 if (ocfs2_xattr_is_local(xs->here)) {
1241                         memcpy(buffer, (void *)xs->base +
1242                                name_offset + name_len, size);
1243                 } else {
1244                         xv = (struct ocfs2_xattr_value_root *)
1245                                 (xs->base + name_offset + name_len);
1246                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1247                                                             buffer, size);
1248                         if (ret < 0) {
1249                                 mlog_errno(ret);
1250                                 goto cleanup;
1251                         }
1252                 }
1253         }
1254         ret = size;
1255 cleanup:
1256         ocfs2_xattr_bucket_free(xs->bucket);
1257
1258         brelse(xs->xattr_bh);
1259         xs->xattr_bh = NULL;
1260         return ret;
1261 }
1262
1263 int ocfs2_xattr_get_nolock(struct inode *inode,
1264                            struct buffer_head *di_bh,
1265                            int name_index,
1266                            const char *name,
1267                            void *buffer,
1268                            size_t buffer_size)
1269 {
1270         int ret;
1271         struct ocfs2_dinode *di = NULL;
1272         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1273         struct ocfs2_xattr_search xis = {
1274                 .not_found = -ENODATA,
1275         };
1276         struct ocfs2_xattr_search xbs = {
1277                 .not_found = -ENODATA,
1278         };
1279
1280         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1281                 return -EOPNOTSUPP;
1282
1283         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1284                 ret = -ENODATA;
1285
1286         xis.inode_bh = xbs.inode_bh = di_bh;
1287         di = (struct ocfs2_dinode *)di_bh->b_data;
1288
1289         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1290                                     buffer_size, &xis);
1291         if (ret == -ENODATA && di->i_xattr_loc)
1292                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1293                                             buffer_size, &xbs);
1294
1295         return ret;
1296 }
1297
1298 /* ocfs2_xattr_get()
1299  *
1300  * Copy an extended attribute into the buffer provided.
1301  * Buffer is NULL to compute the size of buffer required.
1302  */
1303 static int ocfs2_xattr_get(struct inode *inode,
1304                            int name_index,
1305                            const char *name,
1306                            void *buffer,
1307                            size_t buffer_size)
1308 {
1309         int ret;
1310         struct buffer_head *di_bh = NULL;
1311
1312         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1313         if (ret < 0) {
1314                 mlog_errno(ret);
1315                 return ret;
1316         }
1317         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1318         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1319                                      name, buffer, buffer_size);
1320         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1321
1322         ocfs2_inode_unlock(inode, 0);
1323
1324         brelse(di_bh);
1325
1326         return ret;
1327 }
1328
1329 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1330                                            handle_t *handle,
1331                                            struct ocfs2_xattr_value_buf *vb,
1332                                            const void *value,
1333                                            int value_len)
1334 {
1335         int ret = 0, i, cp_len;
1336         u16 blocksize = inode->i_sb->s_blocksize;
1337         u32 p_cluster, num_clusters;
1338         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1339         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1340         u64 blkno;
1341         struct buffer_head *bh = NULL;
1342         unsigned int ext_flags;
1343         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1344
1345         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1346
1347         while (cpos < clusters) {
1348                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1349                                                &num_clusters, &xv->xr_list,
1350                                                &ext_flags);
1351                 if (ret) {
1352                         mlog_errno(ret);
1353                         goto out;
1354                 }
1355
1356                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1357
1358                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1359
1360                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1361                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1362                                                &bh, NULL);
1363                         if (ret) {
1364                                 mlog_errno(ret);
1365                                 goto out;
1366                         }
1367
1368                         ret = ocfs2_journal_access(handle,
1369                                                    INODE_CACHE(inode),
1370                                                    bh,
1371                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1372                         if (ret < 0) {
1373                                 mlog_errno(ret);
1374                                 goto out;
1375                         }
1376
1377                         cp_len = value_len > blocksize ? blocksize : value_len;
1378                         memcpy(bh->b_data, value, cp_len);
1379                         value_len -= cp_len;
1380                         value += cp_len;
1381                         if (cp_len < blocksize)
1382                                 memset(bh->b_data + cp_len, 0,
1383                                        blocksize - cp_len);
1384
1385                         ocfs2_journal_dirty(handle, bh);
1386                         brelse(bh);
1387                         bh = NULL;
1388
1389                         /*
1390                          * XXX: do we need to empty all the following
1391                          * blocks in this cluster?
1392                          */
1393                         if (!value_len)
1394                                 break;
1395                 }
1396                 cpos += num_clusters;
1397         }
1398 out:
1399         brelse(bh);
1400
1401         return ret;
1402 }
1403
1404 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1405                                        int num_entries)
1406 {
1407         int free_space;
1408
1409         if (!needed_space)
1410                 return 0;
1411
1412         free_space = free_start -
1413                 sizeof(struct ocfs2_xattr_header) -
1414                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1415                 OCFS2_XATTR_HEADER_GAP;
1416         if (free_space < 0)
1417                 return -EIO;
1418         if (free_space < needed_space)
1419                 return -ENOSPC;
1420
1421         return 0;
1422 }
1423
1424 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1425                                    int type)
1426 {
1427         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1428 }
1429
1430 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1431 {
1432         loc->xl_ops->xlo_journal_dirty(handle, loc);
1433 }
1434
1435 /* Give a pointer into the storage for the given offset */
1436 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1437 {
1438         BUG_ON(offset >= loc->xl_size);
1439         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1440 }
1441
1442 /*
1443  * Wipe the name+value pair and allow the storage to reclaim it.  This
1444  * must be followed by either removal of the entry or a call to
1445  * ocfs2_xa_add_namevalue().
1446  */
1447 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1448 {
1449         loc->xl_ops->xlo_wipe_namevalue(loc);
1450 }
1451
1452 /*
1453  * Find lowest offset to a name+value pair.  This is the start of our
1454  * downward-growing free space.
1455  */
1456 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1457 {
1458         return loc->xl_ops->xlo_get_free_start(loc);
1459 }
1460
1461 /* Can we reuse loc->xl_entry for xi? */
1462 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1463                                     struct ocfs2_xattr_info *xi)
1464 {
1465         return loc->xl_ops->xlo_can_reuse(loc, xi);
1466 }
1467
1468 /* How much free space is needed to set the new value */
1469 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1470                                 struct ocfs2_xattr_info *xi)
1471 {
1472         return loc->xl_ops->xlo_check_space(loc, xi);
1473 }
1474
1475 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1476 {
1477         loc->xl_ops->xlo_add_entry(loc, name_hash);
1478         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1479         /*
1480          * We can't leave the new entry's xe_name_offset at zero or
1481          * add_namevalue() will go nuts.  We set it to the size of our
1482          * storage so that it can never be less than any other entry.
1483          */
1484         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1485 }
1486
1487 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1488                                    struct ocfs2_xattr_info *xi)
1489 {
1490         int size = namevalue_size_xi(xi);
1491         int nameval_offset;
1492         char *nameval_buf;
1493
1494         loc->xl_ops->xlo_add_namevalue(loc, size);
1495         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1496         loc->xl_entry->xe_name_len = xi->xi_name_len;
1497         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1498         ocfs2_xattr_set_local(loc->xl_entry,
1499                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1500
1501         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1502         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1503         memset(nameval_buf, 0, size);
1504         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1505 }
1506
1507 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1508                                     struct ocfs2_xattr_value_buf *vb)
1509 {
1510         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1511         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1512
1513         /* Value bufs are for value trees */
1514         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1515         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1516                (name_size + OCFS2_XATTR_ROOT_SIZE));
1517
1518         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1519         vb->vb_xv =
1520                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1521                                                         nameval_offset +
1522                                                         name_size);
1523 }
1524
1525 static int ocfs2_xa_block_journal_access(handle_t *handle,
1526                                          struct ocfs2_xa_loc *loc, int type)
1527 {
1528         struct buffer_head *bh = loc->xl_storage;
1529         ocfs2_journal_access_func access;
1530
1531         if (loc->xl_size == (bh->b_size -
1532                              offsetof(struct ocfs2_xattr_block,
1533                                       xb_attrs.xb_header)))
1534                 access = ocfs2_journal_access_xb;
1535         else
1536                 access = ocfs2_journal_access_di;
1537         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1538 }
1539
1540 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1541                                          struct ocfs2_xa_loc *loc)
1542 {
1543         struct buffer_head *bh = loc->xl_storage;
1544
1545         ocfs2_journal_dirty(handle, bh);
1546 }
1547
1548 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1549                                            int offset)
1550 {
1551         return (char *)loc->xl_header + offset;
1552 }
1553
1554 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1555                                     struct ocfs2_xattr_info *xi)
1556 {
1557         /*
1558          * Block storage is strict.  If the sizes aren't exact, we will
1559          * remove the old one and reinsert the new.
1560          */
1561         return namevalue_size_xe(loc->xl_entry) ==
1562                 namevalue_size_xi(xi);
1563 }
1564
1565 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1566 {
1567         struct ocfs2_xattr_header *xh = loc->xl_header;
1568         int i, count = le16_to_cpu(xh->xh_count);
1569         int offset, free_start = loc->xl_size;
1570
1571         for (i = 0; i < count; i++) {
1572                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1573                 if (offset < free_start)
1574                         free_start = offset;
1575         }
1576
1577         return free_start;
1578 }
1579
1580 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1581                                       struct ocfs2_xattr_info *xi)
1582 {
1583         int count = le16_to_cpu(loc->xl_header->xh_count);
1584         int free_start = ocfs2_xa_get_free_start(loc);
1585         int needed_space = ocfs2_xi_entry_usage(xi);
1586
1587         /*
1588          * Block storage will reclaim the original entry before inserting
1589          * the new value, so we only need the difference.  If the new
1590          * entry is smaller than the old one, we don't need anything.
1591          */
1592         if (loc->xl_entry) {
1593                 /* Don't need space if we're reusing! */
1594                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1595                         needed_space = 0;
1596                 else
1597                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1598         }
1599         if (needed_space < 0)
1600                 needed_space = 0;
1601         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1602 }
1603
1604 /*
1605  * Block storage for xattrs keeps the name+value pairs compacted.  When
1606  * we remove one, we have to shift any that preceded it towards the end.
1607  */
1608 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1609 {
1610         int i, offset;
1611         int namevalue_offset, first_namevalue_offset, namevalue_size;
1612         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1613         struct ocfs2_xattr_header *xh = loc->xl_header;
1614         int count = le16_to_cpu(xh->xh_count);
1615
1616         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1617         namevalue_size = namevalue_size_xe(entry);
1618         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1619
1620         /* Shift the name+value pairs */
1621         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1622                 (char *)xh + first_namevalue_offset,
1623                 namevalue_offset - first_namevalue_offset);
1624         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1625
1626         /* Now tell xh->xh_entries about it */
1627         for (i = 0; i < count; i++) {
1628                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1629                 if (offset <= namevalue_offset)
1630                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1631                                      namevalue_size);
1632         }
1633
1634         /*
1635          * Note that we don't update xh_free_start or xh_name_value_len
1636          * because they're not used in block-stored xattrs.
1637          */
1638 }
1639
1640 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1641 {
1642         int count = le16_to_cpu(loc->xl_header->xh_count);
1643         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1644         le16_add_cpu(&loc->xl_header->xh_count, 1);
1645         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1646 }
1647
1648 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1649 {
1650         int free_start = ocfs2_xa_get_free_start(loc);
1651
1652         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1653 }
1654
1655 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1656                                           struct ocfs2_xattr_value_buf *vb)
1657 {
1658         struct buffer_head *bh = loc->xl_storage;
1659
1660         if (loc->xl_size == (bh->b_size -
1661                              offsetof(struct ocfs2_xattr_block,
1662                                       xb_attrs.xb_header)))
1663                 vb->vb_access = ocfs2_journal_access_xb;
1664         else
1665                 vb->vb_access = ocfs2_journal_access_di;
1666         vb->vb_bh = bh;
1667 }
1668
1669 /*
1670  * Operations for xattrs stored in blocks.  This includes inline inode
1671  * storage and unindexed ocfs2_xattr_blocks.
1672  */
1673 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1674         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1675         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1676         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1677         .xlo_check_space        = ocfs2_xa_block_check_space,
1678         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1679         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1680         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1681         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1682         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1683         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1684 };
1685
1686 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1687                                           struct ocfs2_xa_loc *loc, int type)
1688 {
1689         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1690
1691         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1692 }
1693
1694 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1695                                           struct ocfs2_xa_loc *loc)
1696 {
1697         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1698
1699         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1700 }
1701
1702 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1703                                             int offset)
1704 {
1705         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1706         int block, block_offset;
1707
1708         /* The header is at the front of the bucket */
1709         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1710         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1711
1712         return bucket_block(bucket, block) + block_offset;
1713 }
1714
1715 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1716                                      struct ocfs2_xattr_info *xi)
1717 {
1718         return namevalue_size_xe(loc->xl_entry) >=
1719                 namevalue_size_xi(xi);
1720 }
1721
1722 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1723 {
1724         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1725         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1726 }
1727
1728 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1729                                          int free_start, int size)
1730 {
1731         /*
1732          * We need to make sure that the name+value pair fits within
1733          * one block.
1734          */
1735         if (((free_start - size) >> sb->s_blocksize_bits) !=
1736             ((free_start - 1) >> sb->s_blocksize_bits))
1737                 free_start -= free_start % sb->s_blocksize;
1738
1739         return free_start;
1740 }
1741
1742 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1743                                        struct ocfs2_xattr_info *xi)
1744 {
1745         int rc;
1746         int count = le16_to_cpu(loc->xl_header->xh_count);
1747         int free_start = ocfs2_xa_get_free_start(loc);
1748         int needed_space = ocfs2_xi_entry_usage(xi);
1749         int size = namevalue_size_xi(xi);
1750         struct super_block *sb = loc->xl_inode->i_sb;
1751
1752         /*
1753          * Bucket storage does not reclaim name+value pairs it cannot
1754          * reuse.  They live as holes until the bucket fills, and then
1755          * the bucket is defragmented.  However, the bucket can reclaim
1756          * the ocfs2_xattr_entry.
1757          */
1758         if (loc->xl_entry) {
1759                 /* Don't need space if we're reusing! */
1760                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1761                         needed_space = 0;
1762                 else
1763                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1764         }
1765         BUG_ON(needed_space < 0);
1766
1767         if (free_start < size) {
1768                 if (needed_space)
1769                         return -ENOSPC;
1770         } else {
1771                 /*
1772                  * First we check if it would fit in the first place.
1773                  * Below, we align the free start to a block.  This may
1774                  * slide us below the minimum gap.  By checking unaligned
1775                  * first, we avoid that error.
1776                  */
1777                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1778                                                  count);
1779                 if (rc)
1780                         return rc;
1781                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1782                                                            size);
1783         }
1784         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1785 }
1786
1787 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1788 {
1789         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1790                      -namevalue_size_xe(loc->xl_entry));
1791 }
1792
1793 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1794 {
1795         struct ocfs2_xattr_header *xh = loc->xl_header;
1796         int count = le16_to_cpu(xh->xh_count);
1797         int low = 0, high = count - 1, tmp;
1798         struct ocfs2_xattr_entry *tmp_xe;
1799
1800         /*
1801          * We keep buckets sorted by name_hash, so we need to find
1802          * our insert place.
1803          */
1804         while (low <= high && count) {
1805                 tmp = (low + high) / 2;
1806                 tmp_xe = &xh->xh_entries[tmp];
1807
1808                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1809                         low = tmp + 1;
1810                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1811                         high = tmp - 1;
1812                 else {
1813                         low = tmp;
1814                         break;
1815                 }
1816         }
1817
1818         if (low != count)
1819                 memmove(&xh->xh_entries[low + 1],
1820                         &xh->xh_entries[low],
1821                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1822
1823         le16_add_cpu(&xh->xh_count, 1);
1824         loc->xl_entry = &xh->xh_entries[low];
1825         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1826 }
1827
1828 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1829 {
1830         int free_start = ocfs2_xa_get_free_start(loc);
1831         struct ocfs2_xattr_header *xh = loc->xl_header;
1832         struct super_block *sb = loc->xl_inode->i_sb;
1833         int nameval_offset;
1834
1835         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1836         nameval_offset = free_start - size;
1837         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1838         xh->xh_free_start = cpu_to_le16(nameval_offset);
1839         le16_add_cpu(&xh->xh_name_value_len, size);
1840
1841 }
1842
1843 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1844                                            struct ocfs2_xattr_value_buf *vb)
1845 {
1846         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1847         struct super_block *sb = loc->xl_inode->i_sb;
1848         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1849         int size = namevalue_size_xe(loc->xl_entry);
1850         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1851
1852         /* Values are not allowed to straddle block boundaries */
1853         BUG_ON(block_offset !=
1854                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1855         /* We expect the bucket to be filled in */
1856         BUG_ON(!bucket->bu_bhs[block_offset]);
1857
1858         vb->vb_access = ocfs2_journal_access;
1859         vb->vb_bh = bucket->bu_bhs[block_offset];
1860 }
1861
1862 /* Operations for xattrs stored in buckets. */
1863 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1864         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1865         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1866         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1867         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1868         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1869         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1870         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1871         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1872         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1873         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1874 };
1875
1876 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1877 {
1878         struct ocfs2_xattr_value_buf vb;
1879
1880         if (ocfs2_xattr_is_local(loc->xl_entry))
1881                 return 0;
1882
1883         ocfs2_xa_fill_value_buf(loc, &vb);
1884         return le32_to_cpu(vb.vb_xv->xr_clusters);
1885 }
1886
1887 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1888                                    struct ocfs2_xattr_set_ctxt *ctxt)
1889 {
1890         int trunc_rc, access_rc;
1891         struct ocfs2_xattr_value_buf vb;
1892
1893         ocfs2_xa_fill_value_buf(loc, &vb);
1894         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1895                                               ctxt);
1896
1897         /*
1898          * The caller of ocfs2_xa_value_truncate() has already called
1899          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1900          * calls ocfs2_extend_trans().  This may commit the previous
1901          * transaction and open a new one.  If this is a bucket, truncate
1902          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1903          * the caller is expecting to dirty the entire bucket.  So we must
1904          * reset the journal work.  We do this even if truncate has failed,
1905          * as it could have failed after committing the extend.
1906          */
1907         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1908                                             OCFS2_JOURNAL_ACCESS_WRITE);
1909
1910         /* Errors in truncate take precedence */
1911         return trunc_rc ? trunc_rc : access_rc;
1912 }
1913
1914 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1915 {
1916         int index, count;
1917         struct ocfs2_xattr_header *xh = loc->xl_header;
1918         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1919
1920         ocfs2_xa_wipe_namevalue(loc);
1921         loc->xl_entry = NULL;
1922
1923         le16_add_cpu(&xh->xh_count, -1);
1924         count = le16_to_cpu(xh->xh_count);
1925
1926         /*
1927          * Only zero out the entry if there are more remaining.  This is
1928          * important for an empty bucket, as it keeps track of the
1929          * bucket's hash value.  It doesn't hurt empty block storage.
1930          */
1931         if (count) {
1932                 index = ((char *)entry - (char *)&xh->xh_entries) /
1933                         sizeof(struct ocfs2_xattr_entry);
1934                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1935                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1936                 memset(&xh->xh_entries[count], 0,
1937                        sizeof(struct ocfs2_xattr_entry));
1938         }
1939 }
1940
1941 /*
1942  * If we have a problem adjusting the size of an external value during
1943  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1944  * in an intermediate state.  For example, the value may be partially
1945  * truncated.
1946  *
1947  * If the value tree hasn't changed, the extend/truncate went nowhere.
1948  * We have nothing to do.  The caller can treat it as a straight error.
1949  *
1950  * If the value tree got partially truncated, we now have a corrupted
1951  * extended attribute.  We're going to wipe its entry and leak the
1952  * clusters.  Better to leak some storage than leave a corrupt entry.
1953  *
1954  * If the value tree grew, it obviously didn't grow enough for the
1955  * new entry.  We're not going to try and reclaim those clusters either.
1956  * If there was already an external value there (orig_clusters != 0),
1957  * the new clusters are attached safely and we can just leave the old
1958  * value in place.  If there was no external value there, we remove
1959  * the entry.
1960  *
1961  * This way, the xattr block we store in the journal will be consistent.
1962  * If the size change broke because of the journal, no changes will hit
1963  * disk anyway.
1964  */
1965 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1966                                             const char *what,
1967                                             unsigned int orig_clusters)
1968 {
1969         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1970         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1971                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1972
1973         if (new_clusters < orig_clusters) {
1974                 mlog(ML_ERROR,
1975                      "Partial truncate while %s xattr %.*s.  Leaking "
1976                      "%u clusters and removing the entry\n",
1977                      what, loc->xl_entry->xe_name_len, nameval_buf,
1978                      orig_clusters - new_clusters);
1979                 ocfs2_xa_remove_entry(loc);
1980         } else if (!orig_clusters) {
1981                 mlog(ML_ERROR,
1982                      "Unable to allocate an external value for xattr "
1983                      "%.*s safely.  Leaking %u clusters and removing the "
1984                      "entry\n",
1985                      loc->xl_entry->xe_name_len, nameval_buf,
1986                      new_clusters - orig_clusters);
1987                 ocfs2_xa_remove_entry(loc);
1988         } else if (new_clusters > orig_clusters)
1989                 mlog(ML_ERROR,
1990                      "Unable to grow xattr %.*s safely.  %u new clusters "
1991                      "have been added, but the value will not be "
1992                      "modified\n",
1993                      loc->xl_entry->xe_name_len, nameval_buf,
1994                      new_clusters - orig_clusters);
1995 }
1996
1997 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1998                            struct ocfs2_xattr_set_ctxt *ctxt)
1999 {
2000         int rc = 0;
2001         unsigned int orig_clusters;
2002
2003         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2004                 orig_clusters = ocfs2_xa_value_clusters(loc);
2005                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2006                 if (rc) {
2007                         mlog_errno(rc);
2008                         /*
2009                          * Since this is remove, we can return 0 if
2010                          * ocfs2_xa_cleanup_value_truncate() is going to
2011                          * wipe the entry anyway.  So we check the
2012                          * cluster count as well.
2013                          */
2014                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2015                                 rc = 0;
2016                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2017                                                         orig_clusters);
2018                         if (rc)
2019                                 goto out;
2020                 }
2021         }
2022
2023         ocfs2_xa_remove_entry(loc);
2024
2025 out:
2026         return rc;
2027 }
2028
2029 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2030 {
2031         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2032         char *nameval_buf;
2033
2034         nameval_buf = ocfs2_xa_offset_pointer(loc,
2035                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2036         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2037 }
2038
2039 /*
2040  * Take an existing entry and make it ready for the new value.  This
2041  * won't allocate space, but it may free space.  It should be ready for
2042  * ocfs2_xa_prepare_entry() to finish the work.
2043  */
2044 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2045                                 struct ocfs2_xattr_info *xi,
2046                                 struct ocfs2_xattr_set_ctxt *ctxt)
2047 {
2048         int rc = 0;
2049         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2050         unsigned int orig_clusters;
2051         char *nameval_buf;
2052         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2053         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2054
2055         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2056                name_size);
2057
2058         nameval_buf = ocfs2_xa_offset_pointer(loc,
2059                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2060         if (xe_local) {
2061                 memset(nameval_buf + name_size, 0,
2062                        namevalue_size_xe(loc->xl_entry) - name_size);
2063                 if (!xi_local)
2064                         ocfs2_xa_install_value_root(loc);
2065         } else {
2066                 orig_clusters = ocfs2_xa_value_clusters(loc);
2067                 if (xi_local) {
2068                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2069                         if (rc < 0)
2070                                 mlog_errno(rc);
2071                         else
2072                                 memset(nameval_buf + name_size, 0,
2073                                        namevalue_size_xe(loc->xl_entry) -
2074                                        name_size);
2075                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2076                            xi->xi_value_len) {
2077                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2078                                                      ctxt);
2079                         if (rc < 0)
2080                                 mlog_errno(rc);
2081                 }
2082
2083                 if (rc) {
2084                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2085                                                         orig_clusters);
2086                         goto out;
2087                 }
2088         }
2089
2090         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2091         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2092
2093 out:
2094         return rc;
2095 }
2096
2097 /*
2098  * Prepares loc->xl_entry to receive the new xattr.  This includes
2099  * properly setting up the name+value pair region.  If loc->xl_entry
2100  * already exists, it will take care of modifying it appropriately.
2101  *
2102  * Note that this modifies the data.  You did journal_access already,
2103  * right?
2104  */
2105 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2106                                   struct ocfs2_xattr_info *xi,
2107                                   u32 name_hash,
2108                                   struct ocfs2_xattr_set_ctxt *ctxt)
2109 {
2110         int rc = 0;
2111         unsigned int orig_clusters;
2112         __le64 orig_value_size = 0;
2113
2114         rc = ocfs2_xa_check_space(loc, xi);
2115         if (rc)
2116                 goto out;
2117
2118         if (loc->xl_entry) {
2119                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2120                         orig_value_size = loc->xl_entry->xe_value_size;
2121                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2122                         if (rc)
2123                                 goto out;
2124                         goto alloc_value;
2125                 }
2126
2127                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2128                         orig_clusters = ocfs2_xa_value_clusters(loc);
2129                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2130                         if (rc) {
2131                                 mlog_errno(rc);
2132                                 ocfs2_xa_cleanup_value_truncate(loc,
2133                                                                 "overwriting",
2134                                                                 orig_clusters);
2135                                 goto out;
2136                         }
2137                 }
2138                 ocfs2_xa_wipe_namevalue(loc);
2139         } else
2140                 ocfs2_xa_add_entry(loc, name_hash);
2141
2142         /*
2143          * If we get here, we have a blank entry.  Fill it.  We grow our
2144          * name+value pair back from the end.
2145          */
2146         ocfs2_xa_add_namevalue(loc, xi);
2147         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2148                 ocfs2_xa_install_value_root(loc);
2149
2150 alloc_value:
2151         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2152                 orig_clusters = ocfs2_xa_value_clusters(loc);
2153                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2154                 if (rc < 0) {
2155                         ctxt->set_abort = 1;
2156                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2157                                                         orig_clusters);
2158                         /*
2159                          * If we were growing an existing value,
2160                          * ocfs2_xa_cleanup_value_truncate() won't remove
2161                          * the entry. We need to restore the original value
2162                          * size.
2163                          */
2164                         if (loc->xl_entry) {
2165                                 BUG_ON(!orig_value_size);
2166                                 loc->xl_entry->xe_value_size = orig_value_size;
2167                         }
2168                         mlog_errno(rc);
2169                 }
2170         }
2171
2172 out:
2173         return rc;
2174 }
2175
2176 /*
2177  * Store the value portion of the name+value pair.  This will skip
2178  * values that are stored externally.  Their tree roots were set up
2179  * by ocfs2_xa_prepare_entry().
2180  */
2181 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2182                                 struct ocfs2_xattr_info *xi,
2183                                 struct ocfs2_xattr_set_ctxt *ctxt)
2184 {
2185         int rc = 0;
2186         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2187         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2188         char *nameval_buf;
2189         struct ocfs2_xattr_value_buf vb;
2190
2191         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2192         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2193                 ocfs2_xa_fill_value_buf(loc, &vb);
2194                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2195                                                      ctxt->handle, &vb,
2196                                                      xi->xi_value,
2197                                                      xi->xi_value_len);
2198         } else
2199                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2200
2201         return rc;
2202 }
2203
2204 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2205                         struct ocfs2_xattr_info *xi,
2206                         struct ocfs2_xattr_set_ctxt *ctxt)
2207 {
2208         int ret;
2209         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2210                                               xi->xi_name_len);
2211
2212         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2213                                       OCFS2_JOURNAL_ACCESS_WRITE);
2214         if (ret) {
2215                 mlog_errno(ret);
2216                 goto out;
2217         }
2218
2219         /*
2220          * From here on out, everything is going to modify the buffer a
2221          * little.  Errors are going to leave the xattr header in a
2222          * sane state.  Thus, even with errors we dirty the sucker.
2223          */
2224
2225         /* Don't worry, we are never called with !xi_value and !xl_entry */
2226         if (!xi->xi_value) {
2227                 ret = ocfs2_xa_remove(loc, ctxt);
2228                 goto out_dirty;
2229         }
2230
2231         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2232         if (ret) {
2233                 if (ret != -ENOSPC)
2234                         mlog_errno(ret);
2235                 goto out_dirty;
2236         }
2237
2238         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2239         if (ret)
2240                 mlog_errno(ret);
2241
2242 out_dirty:
2243         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2244
2245 out:
2246         return ret;
2247 }
2248
2249 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2250                                      struct inode *inode,
2251                                      struct buffer_head *bh,
2252                                      struct ocfs2_xattr_entry *entry)
2253 {
2254         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2255
2256         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2257
2258         loc->xl_inode = inode;
2259         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2260         loc->xl_storage = bh;
2261         loc->xl_entry = entry;
2262         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2263         loc->xl_header =
2264                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2265                                               loc->xl_size);
2266 }
2267
2268 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2269                                           struct inode *inode,
2270                                           struct buffer_head *bh,
2271                                           struct ocfs2_xattr_entry *entry)
2272 {
2273         struct ocfs2_xattr_block *xb =
2274                 (struct ocfs2_xattr_block *)bh->b_data;
2275
2276         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2277
2278         loc->xl_inode = inode;
2279         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2280         loc->xl_storage = bh;
2281         loc->xl_header = &(xb->xb_attrs.xb_header);
2282         loc->xl_entry = entry;
2283         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2284                                              xb_attrs.xb_header);
2285 }
2286
2287 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2288                                            struct ocfs2_xattr_bucket *bucket,
2289                                            struct ocfs2_xattr_entry *entry)
2290 {
2291         loc->xl_inode = bucket->bu_inode;
2292         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2293         loc->xl_storage = bucket;
2294         loc->xl_header = bucket_xh(bucket);
2295         loc->xl_entry = entry;
2296         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2297 }
2298
2299 /*
2300  * In xattr remove, if it is stored outside and refcounted, we may have
2301  * the chance to split the refcount tree. So need the allocators.
2302  */
2303 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2304                                         struct ocfs2_xattr_value_root *xv,
2305                                         struct ocfs2_caching_info *ref_ci,
2306                                         struct buffer_head *ref_root_bh,
2307                                         struct ocfs2_alloc_context **meta_ac,
2308                                         int *ref_credits)
2309 {
2310         int ret, meta_add = 0;
2311         u32 p_cluster, num_clusters;
2312         unsigned int ext_flags;
2313
2314         *ref_credits = 0;
2315         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2316                                        &num_clusters,
2317                                        &xv->xr_list,
2318                                        &ext_flags);
2319         if (ret) {
2320                 mlog_errno(ret);
2321                 goto out;
2322         }
2323
2324         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2325                 goto out;
2326
2327         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2328                                                  ref_root_bh, xv,
2329                                                  &meta_add, ref_credits);
2330         if (ret) {
2331                 mlog_errno(ret);
2332                 goto out;
2333         }
2334
2335         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2336                                                 meta_add, meta_ac);
2337         if (ret)
2338                 mlog_errno(ret);
2339
2340 out:
2341         return ret;
2342 }
2343
2344 static int ocfs2_remove_value_outside(struct inode*inode,
2345                                       struct ocfs2_xattr_value_buf *vb,
2346                                       struct ocfs2_xattr_header *header,
2347                                       struct ocfs2_caching_info *ref_ci,
2348                                       struct buffer_head *ref_root_bh)
2349 {
2350         int ret = 0, i, ref_credits;
2351         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2352         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2353         void *val;
2354
2355         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2356
2357         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2358                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2359
2360                 if (ocfs2_xattr_is_local(entry))
2361                         continue;
2362
2363                 val = (void *)header +
2364                         le16_to_cpu(entry->xe_name_offset);
2365                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2366                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2367
2368                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2369                                                          ref_ci, ref_root_bh,
2370                                                          &ctxt.meta_ac,
2371                                                          &ref_credits);
2372
2373                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2374                                         ocfs2_remove_extent_credits(osb->sb));
2375                 if (IS_ERR(ctxt.handle)) {
2376                         ret = PTR_ERR(ctxt.handle);
2377                         mlog_errno(ret);
2378                         break;
2379                 }
2380
2381                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2382                 if (ret < 0) {
2383                         mlog_errno(ret);
2384                         break;
2385                 }
2386
2387                 ocfs2_commit_trans(osb, ctxt.handle);
2388                 if (ctxt.meta_ac) {
2389                         ocfs2_free_alloc_context(ctxt.meta_ac);
2390                         ctxt.meta_ac = NULL;
2391                 }
2392         }
2393
2394         if (ctxt.meta_ac)
2395                 ocfs2_free_alloc_context(ctxt.meta_ac);
2396         ocfs2_schedule_truncate_log_flush(osb, 1);
2397         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2398         return ret;
2399 }
2400
2401 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2402                                     struct buffer_head *di_bh,
2403                                     struct ocfs2_caching_info *ref_ci,
2404                                     struct buffer_head *ref_root_bh)
2405 {
2406
2407         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2408         struct ocfs2_xattr_header *header;
2409         int ret;
2410         struct ocfs2_xattr_value_buf vb = {
2411                 .vb_bh = di_bh,
2412                 .vb_access = ocfs2_journal_access_di,
2413         };
2414
2415         header = (struct ocfs2_xattr_header *)
2416                  ((void *)di + inode->i_sb->s_blocksize -
2417                  le16_to_cpu(di->i_xattr_inline_size));
2418
2419         ret = ocfs2_remove_value_outside(inode, &vb, header,
2420                                          ref_ci, ref_root_bh);
2421
2422         return ret;
2423 }
2424
2425 struct ocfs2_rm_xattr_bucket_para {
2426         struct ocfs2_caching_info *ref_ci;
2427         struct buffer_head *ref_root_bh;
2428 };
2429
2430 static int ocfs2_xattr_block_remove(struct inode *inode,
2431                                     struct buffer_head *blk_bh,
2432                                     struct ocfs2_caching_info *ref_ci,
2433                                     struct buffer_head *ref_root_bh)
2434 {
2435         struct ocfs2_xattr_block *xb;
2436         int ret = 0;
2437         struct ocfs2_xattr_value_buf vb = {
2438                 .vb_bh = blk_bh,
2439                 .vb_access = ocfs2_journal_access_xb,
2440         };
2441         struct ocfs2_rm_xattr_bucket_para args = {
2442                 .ref_ci = ref_ci,
2443                 .ref_root_bh = ref_root_bh,
2444         };
2445
2446         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2447         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2448                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2449                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2450                                                  ref_ci, ref_root_bh);
2451         } else
2452                 ret = ocfs2_iterate_xattr_index_block(inode,
2453                                                 blk_bh,
2454                                                 ocfs2_rm_xattr_cluster,
2455                                                 &args);
2456
2457         return ret;
2458 }
2459
2460 static int ocfs2_xattr_free_block(struct inode *inode,
2461                                   u64 block,
2462                                   struct ocfs2_caching_info *ref_ci,
2463                                   struct buffer_head *ref_root_bh)
2464 {
2465         struct inode *xb_alloc_inode;
2466         struct buffer_head *xb_alloc_bh = NULL;
2467         struct buffer_head *blk_bh = NULL;
2468         struct ocfs2_xattr_block *xb;
2469         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2470         handle_t *handle;
2471         int ret = 0;
2472         u64 blk, bg_blkno;
2473         u16 bit;
2474
2475         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2476         if (ret < 0) {
2477                 mlog_errno(ret);
2478                 goto out;
2479         }
2480
2481         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2482         if (ret < 0) {
2483                 mlog_errno(ret);
2484                 goto out;
2485         }
2486
2487         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2488         blk = le64_to_cpu(xb->xb_blkno);
2489         bit = le16_to_cpu(xb->xb_suballoc_bit);
2490         if (xb->xb_suballoc_loc)
2491                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2492         else
2493                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2494
2495         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2496                                 EXTENT_ALLOC_SYSTEM_INODE,
2497                                 le16_to_cpu(xb->xb_suballoc_slot));
2498         if (!xb_alloc_inode) {
2499                 ret = -ENOMEM;
2500                 mlog_errno(ret);
2501                 goto out;
2502         }
2503         mutex_lock(&xb_alloc_inode->i_mutex);
2504
2505         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2506         if (ret < 0) {
2507                 mlog_errno(ret);
2508                 goto out_mutex;
2509         }
2510
2511         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2512         if (IS_ERR(handle)) {
2513                 ret = PTR_ERR(handle);
2514                 mlog_errno(ret);
2515                 goto out_unlock;
2516         }
2517
2518         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2519                                        bit, bg_blkno, 1);
2520         if (ret < 0)
2521                 mlog_errno(ret);
2522
2523         ocfs2_commit_trans(osb, handle);
2524 out_unlock:
2525         ocfs2_inode_unlock(xb_alloc_inode, 1);
2526         brelse(xb_alloc_bh);
2527 out_mutex:
2528         mutex_unlock(&xb_alloc_inode->i_mutex);
2529         iput(xb_alloc_inode);
2530 out:
2531         brelse(blk_bh);
2532         return ret;
2533 }
2534
2535 /*
2536  * ocfs2_xattr_remove()
2537  *
2538  * Free extended attribute resources associated with this inode.
2539  */
2540 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2541 {
2542         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2543         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2544         struct ocfs2_refcount_tree *ref_tree = NULL;
2545         struct buffer_head *ref_root_bh = NULL;
2546         struct ocfs2_caching_info *ref_ci = NULL;
2547         handle_t *handle;
2548         int ret;
2549
2550         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2551                 return 0;
2552
2553         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2554                 return 0;
2555
2556         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2557                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2558                                                le64_to_cpu(di->i_refcount_loc),
2559                                                1, &ref_tree, &ref_root_bh);
2560                 if (ret) {
2561                         mlog_errno(ret);
2562                         goto out;
2563                 }
2564                 ref_ci = &ref_tree->rf_ci;
2565
2566         }
2567
2568         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2569                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2570                                                ref_ci, ref_root_bh);
2571                 if (ret < 0) {
2572                         mlog_errno(ret);
2573                         goto out;
2574                 }
2575         }
2576
2577         if (di->i_xattr_loc) {
2578                 ret = ocfs2_xattr_free_block(inode,
2579                                              le64_to_cpu(di->i_xattr_loc),
2580                                              ref_ci, ref_root_bh);
2581                 if (ret < 0) {
2582                         mlog_errno(ret);
2583                         goto out;
2584                 }
2585         }
2586
2587         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2588                                    OCFS2_INODE_UPDATE_CREDITS);
2589         if (IS_ERR(handle)) {
2590                 ret = PTR_ERR(handle);
2591                 mlog_errno(ret);
2592                 goto out;
2593         }
2594         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2595                                       OCFS2_JOURNAL_ACCESS_WRITE);
2596         if (ret) {
2597                 mlog_errno(ret);
2598                 goto out_commit;
2599         }
2600
2601         di->i_xattr_loc = 0;
2602
2603         spin_lock(&oi->ip_lock);
2604         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2605         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2606         spin_unlock(&oi->ip_lock);
2607
2608         ocfs2_journal_dirty(handle, di_bh);
2609 out_commit:
2610         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2611 out:
2612         if (ref_tree)
2613                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2614         brelse(ref_root_bh);
2615         return ret;
2616 }
2617
2618 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2619                                         struct ocfs2_dinode *di)
2620 {
2621         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2622         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2623         int free;
2624
2625         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2626                 return 0;
2627
2628         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2629                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2630                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2631         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2632                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2633                         le64_to_cpu(di->i_size);
2634         } else {
2635                 struct ocfs2_extent_list *el = &di->id2.i_list;
2636                 free = (le16_to_cpu(el->l_count) -
2637                         le16_to_cpu(el->l_next_free_rec)) *
2638                         sizeof(struct ocfs2_extent_rec);
2639         }
2640         if (free >= xattrsize)
2641                 return 1;
2642
2643         return 0;
2644 }
2645
2646 /*
2647  * ocfs2_xattr_ibody_find()
2648  *
2649  * Find extended attribute in inode block and
2650  * fill search info into struct ocfs2_xattr_search.
2651  */
2652 static int ocfs2_xattr_ibody_find(struct inode *inode,
2653                                   int name_index,
2654                                   const char *name,
2655                                   struct ocfs2_xattr_search *xs)
2656 {
2657         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2658         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2659         int ret;
2660         int has_space = 0;
2661
2662         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2663                 return 0;
2664
2665         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2666                 down_read(&oi->ip_alloc_sem);
2667                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2668                 up_read(&oi->ip_alloc_sem);
2669                 if (!has_space)
2670                         return 0;
2671         }
2672
2673         xs->xattr_bh = xs->inode_bh;
2674         xs->end = (void *)di + inode->i_sb->s_blocksize;
2675         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2676                 xs->header = (struct ocfs2_xattr_header *)
2677                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2678         else
2679                 xs->header = (struct ocfs2_xattr_header *)
2680                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2681         xs->base = (void *)xs->header;
2682         xs->here = xs->header->xh_entries;
2683
2684         /* Find the named attribute. */
2685         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2686                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2687                 if (ret && ret != -ENODATA)
2688                         return ret;
2689                 xs->not_found = ret;
2690         }
2691
2692         return 0;
2693 }
2694
2695 static int ocfs2_xattr_ibody_init(struct inode *inode,
2696                                   struct buffer_head *di_bh,
2697                                   struct ocfs2_xattr_set_ctxt *ctxt)
2698 {
2699         int ret;
2700         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2701         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2702         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2703         unsigned int xattrsize = osb->s_xattr_inline_size;
2704
2705         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2706                 ret = -ENOSPC;
2707                 goto out;
2708         }
2709
2710         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2711                                       OCFS2_JOURNAL_ACCESS_WRITE);
2712         if (ret) {
2713                 mlog_errno(ret);
2714                 goto out;
2715         }
2716
2717         /*
2718          * Adjust extent record count or inline data size
2719          * to reserve space for extended attribute.
2720          */
2721         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2722                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2723                 le16_add_cpu(&idata->id_count, -xattrsize);
2724         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2725                 struct ocfs2_extent_list *el = &di->id2.i_list;
2726                 le16_add_cpu(&el->l_count, -(xattrsize /
2727                                              sizeof(struct ocfs2_extent_rec)));
2728         }
2729         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2730
2731         spin_lock(&oi->ip_lock);
2732         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2733         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2734         spin_unlock(&oi->ip_lock);
2735
2736         ocfs2_journal_dirty(ctxt->handle, di_bh);
2737
2738 out:
2739         return ret;
2740 }
2741
2742 /*
2743  * ocfs2_xattr_ibody_set()
2744  *
2745  * Set, replace or remove an extended attribute into inode block.
2746  *
2747  */
2748 static int ocfs2_xattr_ibody_set(struct inode *inode,
2749                                  struct ocfs2_xattr_info *xi,
2750                                  struct ocfs2_xattr_search *xs,
2751                                  struct ocfs2_xattr_set_ctxt *ctxt)
2752 {
2753         int ret;
2754         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2755         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2756         struct ocfs2_xa_loc loc;
2757
2758         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2759                 return -ENOSPC;
2760
2761         down_write(&oi->ip_alloc_sem);
2762         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2763                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2764                         ret = -ENOSPC;
2765                         goto out;
2766                 }
2767         }
2768
2769         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2770                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2771                 if (ret) {
2772                         if (ret != -ENOSPC)
2773                                 mlog_errno(ret);
2774                         goto out;
2775                 }
2776         }
2777
2778         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2779                                  xs->not_found ? NULL : xs->here);
2780         ret = ocfs2_xa_set(&loc, xi, ctxt);
2781         if (ret) {
2782                 if (ret != -ENOSPC)
2783                         mlog_errno(ret);
2784                 goto out;
2785         }
2786         xs->here = loc.xl_entry;
2787
2788 out:
2789         up_write(&oi->ip_alloc_sem);
2790
2791         return ret;
2792 }
2793
2794 /*
2795  * ocfs2_xattr_block_find()
2796  *
2797  * Find extended attribute in external block and
2798  * fill search info into struct ocfs2_xattr_search.
2799  */
2800 static int ocfs2_xattr_block_find(struct inode *inode,
2801                                   int name_index,
2802                                   const char *name,
2803                                   struct ocfs2_xattr_search *xs)
2804 {
2805         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2806         struct buffer_head *blk_bh = NULL;
2807         struct ocfs2_xattr_block *xb;
2808         int ret = 0;
2809
2810         if (!di->i_xattr_loc)
2811                 return ret;
2812
2813         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2814                                      &blk_bh);
2815         if (ret < 0) {
2816                 mlog_errno(ret);
2817                 return ret;
2818         }
2819
2820         xs->xattr_bh = blk_bh;
2821         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2822
2823         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2824                 xs->header = &xb->xb_attrs.xb_header;
2825                 xs->base = (void *)xs->header;
2826                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2827                 xs->here = xs->header->xh_entries;
2828
2829                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2830         } else
2831                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2832                                                    name_index,
2833                                                    name, xs);
2834
2835         if (ret && ret != -ENODATA) {
2836                 xs->xattr_bh = NULL;
2837                 goto cleanup;
2838         }
2839         xs->not_found = ret;
2840         return 0;
2841 cleanup:
2842         brelse(blk_bh);
2843
2844         return ret;
2845 }
2846
2847 static int ocfs2_create_xattr_block(struct inode *inode,
2848                                     struct buffer_head *inode_bh,
2849                                     struct ocfs2_xattr_set_ctxt *ctxt,
2850                                     int indexed,
2851                                     struct buffer_head **ret_bh)
2852 {
2853         int ret;
2854         u16 suballoc_bit_start;
2855         u32 num_got;
2856         u64 suballoc_loc, first_blkno;
2857         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2858         struct buffer_head *new_bh = NULL;
2859         struct ocfs2_xattr_block *xblk;
2860
2861         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2862                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2863         if (ret < 0) {
2864                 mlog_errno(ret);
2865                 goto end;
2866         }
2867
2868         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2869                                    &suballoc_loc, &suballoc_bit_start,
2870                                    &num_got, &first_blkno);
2871         if (ret < 0) {
2872                 mlog_errno(ret);
2873                 goto end;
2874         }
2875
2876         new_bh = sb_getblk(inode->i_sb, first_blkno);
2877         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2878
2879         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2880                                       new_bh,
2881                                       OCFS2_JOURNAL_ACCESS_CREATE);
2882         if (ret < 0) {
2883                 mlog_errno(ret);
2884                 goto end;
2885         }
2886
2887         /* Initialize ocfs2_xattr_block */
2888         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2889         memset(xblk, 0, inode->i_sb->s_blocksize);
2890         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2891         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2892         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2893         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2894         xblk->xb_fs_generation =
2895                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2896         xblk->xb_blkno = cpu_to_le64(first_blkno);
2897         if (indexed) {
2898                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2899                 xr->xt_clusters = cpu_to_le32(1);
2900                 xr->xt_last_eb_blk = 0;
2901                 xr->xt_list.l_tree_depth = 0;
2902                 xr->xt_list.l_count = cpu_to_le16(
2903                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2904                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2905                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2906         }
2907         ocfs2_journal_dirty(ctxt->handle, new_bh);
2908
2909         /* Add it to the inode */
2910         di->i_xattr_loc = cpu_to_le64(first_blkno);
2911
2912         spin_lock(&OCFS2_I(inode)->ip_lock);
2913         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2914         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2915         spin_unlock(&OCFS2_I(inode)->ip_lock);
2916
2917         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2918
2919         *ret_bh = new_bh;
2920         new_bh = NULL;
2921
2922 end:
2923         brelse(new_bh);
2924         return ret;
2925 }
2926
2927 /*
2928  * ocfs2_xattr_block_set()
2929  *
2930  * Set, replace or remove an extended attribute into external block.
2931  *
2932  */
2933 static int ocfs2_xattr_block_set(struct inode *inode,
2934                                  struct ocfs2_xattr_info *xi,
2935                                  struct ocfs2_xattr_search *xs,
2936                                  struct ocfs2_xattr_set_ctxt *ctxt)
2937 {
2938         struct buffer_head *new_bh = NULL;
2939         struct ocfs2_xattr_block *xblk = NULL;
2940         int ret;
2941         struct ocfs2_xa_loc loc;
2942
2943         if (!xs->xattr_bh) {
2944                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2945                                                0, &new_bh);
2946                 if (ret) {
2947                         mlog_errno(ret);
2948                         goto end;
2949                 }
2950
2951                 xs->xattr_bh = new_bh;
2952                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2953                 xs->header = &xblk->xb_attrs.xb_header;
2954                 xs->base = (void *)xs->header;
2955                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2956                 xs->here = xs->header->xh_entries;
2957         } else
2958                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2959
2960         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2961                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2962                                               xs->not_found ? NULL : xs->here);
2963
2964                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2965                 if (!ret)
2966                         xs->here = loc.xl_entry;
2967                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2968                         goto end;
2969                 else {
2970                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2971                         if (ret)
2972                                 goto end;
2973                 }
2974         }
2975
2976         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2977                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2978
2979 end:
2980         return ret;
2981 }
2982
2983 /* Check whether the new xattr can be inserted into the inode. */
2984 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2985                                        struct ocfs2_xattr_info *xi,
2986                                        struct ocfs2_xattr_search *xs)
2987 {
2988         struct ocfs2_xattr_entry *last;
2989         int free, i;
2990         size_t min_offs = xs->end - xs->base;
2991
2992         if (!xs->header)
2993                 return 0;
2994
2995         last = xs->header->xh_entries;
2996
2997         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2998                 size_t offs = le16_to_cpu(last->xe_name_offset);
2999                 if (offs < min_offs)
3000                         min_offs = offs;
3001                 last += 1;
3002         }
3003
3004         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3005         if (free < 0)
3006                 return 0;
3007
3008         BUG_ON(!xs->not_found);
3009
3010         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3011                 return 1;
3012
3013         return 0;
3014 }
3015
3016 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3017                                      struct ocfs2_dinode *di,
3018                                      struct ocfs2_xattr_info *xi,
3019                                      struct ocfs2_xattr_search *xis,
3020                                      struct ocfs2_xattr_search *xbs,
3021                                      int *clusters_need,
3022                                      int *meta_need,
3023                                      int *credits_need)
3024 {
3025         int ret = 0, old_in_xb = 0;
3026         int clusters_add = 0, meta_add = 0, credits = 0;
3027         struct buffer_head *bh = NULL;
3028         struct ocfs2_xattr_block *xb = NULL;
3029         struct ocfs2_xattr_entry *xe = NULL;
3030         struct ocfs2_xattr_value_root *xv = NULL;
3031         char *base = NULL;
3032         int name_offset, name_len = 0;
3033         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3034                                                     xi->xi_value_len);
3035         u64 value_size;
3036
3037         /*
3038          * Calculate the clusters we need to write.
3039          * No matter whether we replace an old one or add a new one,
3040          * we need this for writing.
3041          */
3042         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3043                 credits += new_clusters *
3044                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3045
3046         if (xis->not_found && xbs->not_found) {
3047                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3048
3049                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3050                         clusters_add += new_clusters;
3051                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3052                                                         &def_xv.xv.xr_list,
3053                                                         new_clusters);
3054                 }
3055
3056                 goto meta_guess;
3057         }
3058
3059         if (!xis->not_found) {
3060                 xe = xis->here;
3061                 name_offset = le16_to_cpu(xe->xe_name_offset);
3062                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3063                 base = xis->base;
3064                 credits += OCFS2_INODE_UPDATE_CREDITS;
3065         } else {
3066                 int i, block_off = 0;
3067                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3068                 xe = xbs->here;
3069                 name_offset = le16_to_cpu(xe->xe_name_offset);
3070                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3071                 i = xbs->here - xbs->header->xh_entries;
3072                 old_in_xb = 1;
3073
3074                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3075                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3076                                                         bucket_xh(xbs->bucket),
3077                                                         i, &block_off,
3078                                                         &name_offset);
3079                         base = bucket_block(xbs->bucket, block_off);
3080                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3081                 } else {
3082                         base = xbs->base;
3083                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3084                 }
3085         }
3086
3087         /*
3088          * delete a xattr doesn't need metadata and cluster allocation.
3089          * so just calculate the credits and return.
3090          *
3091          * The credits for removing the value tree will be extended
3092          * by ocfs2_remove_extent itself.
3093          */
3094         if (!xi->xi_value) {
3095                 if (!ocfs2_xattr_is_local(xe))
3096                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3097
3098                 goto out;
3099         }
3100
3101         /* do cluster allocation guess first. */
3102         value_size = le64_to_cpu(xe->xe_value_size);
3103
3104         if (old_in_xb) {
3105                 /*
3106                  * In xattr set, we always try to set the xe in inode first,
3107                  * so if it can be inserted into inode successfully, the old
3108                  * one will be removed from the xattr block, and this xattr
3109                  * will be inserted into inode as a new xattr in inode.
3110                  */
3111                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3112                         clusters_add += new_clusters;
3113                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3114                                     OCFS2_INODE_UPDATE_CREDITS;
3115                         if (!ocfs2_xattr_is_local(xe))
3116                                 credits += ocfs2_calc_extend_credits(
3117                                                         inode->i_sb,
3118                                                         &def_xv.xv.xr_list,
3119                                                         new_clusters);
3120                         goto out;
3121                 }
3122         }
3123
3124         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3125                 /* the new values will be stored outside. */
3126                 u32 old_clusters = 0;
3127
3128                 if (!ocfs2_xattr_is_local(xe)) {
3129                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3130                                                                  value_size);
3131                         xv = (struct ocfs2_xattr_value_root *)
3132                              (base + name_offset + name_len);
3133                         value_size = OCFS2_XATTR_ROOT_SIZE;
3134                 } else
3135                         xv = &def_xv.xv;
3136
3137                 if (old_clusters >= new_clusters) {
3138                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3139                         goto out;
3140                 } else {
3141                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3142                         clusters_add += new_clusters - old_clusters;
3143                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3144                                                              &xv->xr_list,
3145                                                              new_clusters -
3146                                                              old_clusters);
3147                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3148                                 goto out;
3149                 }
3150         } else {
3151                 /*
3152                  * Now the new value will be stored inside. So if the new
3153                  * value is smaller than the size of value root or the old
3154                  * value, we don't need any allocation, otherwise we have
3155                  * to guess metadata allocation.
3156                  */
3157                 if ((ocfs2_xattr_is_local(xe) &&
3158                      (value_size >= xi->xi_value_len)) ||
3159                     (!ocfs2_xattr_is_local(xe) &&
3160                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3161                         goto out;
3162         }
3163
3164 meta_guess:
3165         /* calculate metadata allocation. */
3166         if (di->i_xattr_loc) {
3167                 if (!xbs->xattr_bh) {
3168                         ret = ocfs2_read_xattr_block(inode,
3169                                                      le64_to_cpu(di->i_xattr_loc),
3170                                                      &bh);
3171                         if (ret) {
3172                                 mlog_errno(ret);
3173                                 goto out;
3174                         }
3175
3176                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3177                 } else
3178                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3179
3180                 /*
3181                  * If there is already an xattr tree, good, we can calculate
3182                  * like other b-trees. Otherwise we may have the chance of
3183                  * create a tree, the credit calculation is borrowed from
3184                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3185                  * new tree will be cluster based, so no meta is needed.
3186                  */
3187                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3188                         struct ocfs2_extent_list *el =
3189                                  &xb->xb_attrs.xb_root.xt_list;
3190                         meta_add += ocfs2_extend_meta_needed(el);
3191                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3192                                                              el, 1);
3193                 } else
3194                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3195
3196                 /*
3197                  * This cluster will be used either for new bucket or for
3198                  * new xattr block.
3199                  * If the cluster size is the same as the bucket size, one
3200                  * more is needed since we may need to extend the bucket
3201                  * also.
3202                  */
3203                 clusters_add += 1;
3204                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3205                 if (OCFS2_XATTR_BUCKET_SIZE ==
3206                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3207                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3208                         clusters_add += 1;
3209                 }
3210         } else {
3211                 meta_add += 1;
3212                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3213         }
3214 out:
3215         if (clusters_need)
3216                 *clusters_need = clusters_add;
3217         if (meta_need)
3218                 *meta_need = meta_add;
3219         if (credits_need)
3220                 *credits_need = credits;
3221         brelse(bh);
3222         return ret;
3223 }
3224
3225 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3226                                      struct ocfs2_dinode *di,
3227                                      struct ocfs2_xattr_info *xi,
3228                                      struct ocfs2_xattr_search *xis,
3229                                      struct ocfs2_xattr_search *xbs,
3230                                      struct ocfs2_xattr_set_ctxt *ctxt,
3231                                      int extra_meta,
3232                                      int *credits)
3233 {
3234         int clusters_add, meta_add, ret;
3235         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3236
3237         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3238
3239         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3240
3241         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3242                                         &clusters_add, &meta_add, credits);
3243         if (ret) {
3244                 mlog_errno(ret);
3245                 return ret;
3246         }
3247
3248         meta_add += extra_meta;
3249         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
3250              "credits = %d\n", xi->xi_name, meta_add, clusters_add, *credits);
3251
3252         if (meta_add) {
3253                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3254                                                         &ctxt->meta_ac);
3255                 if (ret) {
3256                         mlog_errno(ret);
3257                         goto out;
3258                 }
3259         }
3260
3261         if (clusters_add) {
3262                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3263                 if (ret)
3264                         mlog_errno(ret);
3265         }
3266 out:
3267         if (ret) {
3268                 if (ctxt->meta_ac) {
3269                         ocfs2_free_alloc_context(ctxt->meta_ac);
3270                         ctxt->meta_ac = NULL;
3271                 }
3272
3273                 /*
3274                  * We cannot have an error and a non null ctxt->data_ac.
3275                  */
3276         }
3277
3278         return ret;
3279 }
3280
3281 static int __ocfs2_xattr_set_handle(struct inode *inode,
3282                                     struct ocfs2_dinode *di,
3283                                     struct ocfs2_xattr_info *xi,
3284                                     struct ocfs2_xattr_search *xis,
3285                                     struct ocfs2_xattr_search *xbs,
3286                                     struct ocfs2_xattr_set_ctxt *ctxt)
3287 {
3288         int ret = 0, credits, old_found;
3289
3290         if (!xi->xi_value) {
3291                 /* Remove existing extended attribute */
3292                 if (!xis->not_found)
3293                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3294                 else if (!xbs->not_found)
3295                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3296         } else {
3297                 /* We always try to set extended attribute into inode first*/
3298                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3299                 if (!ret && !xbs->not_found) {
3300                         /*
3301                          * If succeed and that extended attribute existing in
3302                          * external block, then we will remove it.
3303                          */
3304                         xi->xi_value = NULL;
3305                         xi->xi_value_len = 0;
3306
3307                         old_found = xis->not_found;
3308                         xis->not_found = -ENODATA;
3309                         ret = ocfs2_calc_xattr_set_need(inode,
3310                                                         di,
3311                                                         xi,
3312                                                         xis,
3313                                                         xbs,
3314                                                         NULL,
3315                                                         NULL,
3316                                                         &credits);
3317                         xis->not_found = old_found;
3318                         if (ret) {
3319                                 mlog_errno(ret);
3320                                 goto out;
3321                         }
3322
3323                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3324                         if (ret) {
3325                                 mlog_errno(ret);
3326                                 goto out;
3327                         }
3328                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3329                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3330                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3331                                 ret = ocfs2_xattr_block_find(inode,
3332                                                              xi->xi_name_index,
3333                                                              xi->xi_name, xbs);
3334                                 if (ret)
3335                                         goto out;
3336
3337                                 old_found = xis->not_found;
3338                                 xis->not_found = -ENODATA;
3339                                 ret = ocfs2_calc_xattr_set_need(inode,
3340                                                                 di,
3341                                                                 xi,
3342                                                                 xis,
3343                                                                 xbs,
3344                                                                 NULL,
3345                                                                 NULL,
3346                                                                 &credits);
3347                                 xis->not_found = old_found;
3348                                 if (ret) {
3349                                         mlog_errno(ret);
3350                                         goto out;
3351                                 }
3352
3353                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3354                                 if (ret) {
3355                                         mlog_errno(ret);
3356                                         goto out;
3357                                 }
3358                         }
3359                         /*
3360                          * If no space in inode, we will set extended attribute
3361                          * into external block.
3362                          */
3363                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3364                         if (ret)
3365                                 goto out;
3366                         if (!xis->not_found) {
3367                                 /*
3368                                  * If succeed and that extended attribute
3369                                  * existing in inode, we will remove it.
3370                                  */
3371                                 xi->xi_value = NULL;
3372                                 xi->xi_value_len = 0;
3373                                 xbs->not_found = -ENODATA;
3374                                 ret = ocfs2_calc_xattr_set_need(inode,
3375                                                                 di,
3376                                                                 xi,
3377                                                                 xis,
3378                                                                 xbs,
3379                                                                 NULL,
3380                                                                 NULL,
3381                                                                 &credits);
3382                                 if (ret) {
3383                                         mlog_errno(ret);
3384                                         goto out;
3385                                 }
3386
3387                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3388                                 if (ret) {
3389                                         mlog_errno(ret);
3390                                         goto out;
3391                                 }
3392                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3393                                                             xis, ctxt);
3394                         }
3395                 }
3396         }
3397
3398         if (!ret) {
3399                 /* Update inode ctime. */
3400                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3401                                               xis->inode_bh,
3402                                               OCFS2_JOURNAL_ACCESS_WRITE);
3403                 if (ret) {
3404                         mlog_errno(ret);
3405                         goto out;
3406                 }
3407
3408                 inode->i_ctime = CURRENT_TIME;
3409                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3410                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3411                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3412         }
3413 out:
3414         return ret;
3415 }
3416
3417 /*
3418  * This function only called duing creating inode
3419  * for init security/acl xattrs of the new inode.
3420  * All transanction credits have been reserved in mknod.
3421  */
3422 int ocfs2_xattr_set_handle(handle_t *handle,
3423                            struct inode *inode,
3424                            struct buffer_head *di_bh,
3425                            int name_index,
3426                            const char *name,
3427                            const void *value,
3428                            size_t value_len,
3429                            int flags,
3430                            struct ocfs2_alloc_context *meta_ac,
3431                            struct ocfs2_alloc_context *data_ac)
3432 {
3433         struct ocfs2_dinode *di;
3434         int ret;
3435
3436         struct ocfs2_xattr_info xi = {
3437                 .xi_name_index = name_index,
3438                 .xi_name = name,
3439                 .xi_name_len = strlen(name),
3440                 .xi_value = value,
3441                 .xi_value_len = value_len,
3442         };
3443
3444         struct ocfs2_xattr_search xis = {
3445                 .not_found = -ENODATA,
3446         };
3447
3448         struct ocfs2_xattr_search xbs = {
3449                 .not_found = -ENODATA,
3450         };
3451
3452         struct ocfs2_xattr_set_ctxt ctxt = {
3453                 .handle = handle,
3454                 .meta_ac = meta_ac,
3455                 .data_ac = data_ac,
3456         };
3457
3458         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3459                 return -EOPNOTSUPP;
3460
3461         /*
3462          * In extreme situation, may need xattr bucket when
3463          * block size is too small. And we have already reserved
3464          * the credits for bucket in mknod.
3465          */
3466         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3467                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3468                 if (!xbs.bucket) {
3469                         mlog_errno(-ENOMEM);
3470                         return -ENOMEM;
3471                 }
3472         }
3473
3474         xis.inode_bh = xbs.inode_bh = di_bh;
3475         di = (struct ocfs2_dinode *)di_bh->b_data;
3476
3477         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3478
3479         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3480         if (ret)
3481                 goto cleanup;
3482         if (xis.not_found) {
3483                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3484                 if (ret)
3485                         goto cleanup;
3486         }
3487
3488         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3489
3490 cleanup:
3491         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3492         brelse(xbs.xattr_bh);
3493         ocfs2_xattr_bucket_free(xbs.bucket);
3494
3495         return ret;
3496 }
3497
3498 /*
3499  * ocfs2_xattr_set()
3500  *
3501  * Set, replace or remove an extended attribute for this inode.
3502  * value is NULL to remove an existing extended attribute, else either
3503  * create or replace an extended attribute.
3504  */
3505 int ocfs2_xattr_set(struct inode *inode,
3506                     int name_index,
3507                     const char *name,
3508                     const void *value,
3509                     size_t value_len,
3510                     int flags)
3511 {
3512         struct buffer_head *di_bh = NULL;
3513         struct ocfs2_dinode *di;
3514         int ret, credits, ref_meta = 0, ref_credits = 0;
3515         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3516         struct inode *tl_inode = osb->osb_tl_inode;
3517         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3518         struct ocfs2_refcount_tree *ref_tree = NULL;
3519
3520         struct ocfs2_xattr_info xi = {
3521                 .xi_name_index = name_index,
3522                 .xi_name = name,
3523                 .xi_name_len = strlen(name),
3524                 .xi_value = value,
3525                 .xi_value_len = value_len,
3526         };
3527
3528         struct ocfs2_xattr_search xis = {
3529                 .not_found = -ENODATA,
3530         };
3531
3532         struct ocfs2_xattr_search xbs = {
3533                 .not_found = -ENODATA,
3534         };
3535
3536         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3537                 return -EOPNOTSUPP;
3538
3539         /*
3540          * Only xbs will be used on indexed trees.  xis doesn't need a
3541          * bucket.
3542          */
3543         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3544         if (!xbs.bucket) {
3545                 mlog_errno(-ENOMEM);
3546                 return -ENOMEM;
3547         }
3548
3549         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3550         if (ret < 0) {
3551                 mlog_errno(ret);
3552                 goto cleanup_nolock;
3553         }
3554         xis.inode_bh = xbs.inode_bh = di_bh;
3555         di = (struct ocfs2_dinode *)di_bh->b_data;
3556
3557         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3558         /*
3559          * Scan inode and external block to find the same name
3560          * extended attribute and collect search infomation.
3561          */
3562         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3563         if (ret)
3564                 goto cleanup;
3565         if (xis.not_found) {
3566                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3567                 if (ret)
3568                         goto cleanup;
3569         }
3570
3571         if (xis.not_found && xbs.not_found) {
3572                 ret = -ENODATA;
3573                 if (flags & XATTR_REPLACE)
3574                         goto cleanup;
3575                 ret = 0;
3576                 if (!value)
3577                         goto cleanup;
3578         } else {
3579                 ret = -EEXIST;
3580                 if (flags & XATTR_CREATE)
3581                         goto cleanup;
3582         }
3583
3584         /* Check whether the value is refcounted and do some prepartion. */
3585         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3586             (!xis.not_found || !xbs.not_found)) {
3587                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3588                                                    &xis, &xbs, &ref_tree,
3589                                                    &ref_meta, &ref_credits);
3590                 if (ret) {
3591                         mlog_errno(ret);
3592                         goto cleanup;
3593                 }
3594         }
3595
3596         mutex_lock(&tl_inode->i_mutex);
3597
3598         if (ocfs2_truncate_log_needs_flush(osb)) {
3599                 ret = __ocfs2_flush_truncate_log(osb);
3600                 if (ret < 0) {
3601                         mutex_unlock(&tl_inode->i_mutex);
3602                         mlog_errno(ret);
3603                         goto cleanup;
3604                 }
3605         }
3606         mutex_unlock(&tl_inode->i_mutex);
3607
3608         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3609                                         &xbs, &ctxt, ref_meta, &credits);
3610         if (ret) {
3611                 mlog_errno(ret);
3612                 goto cleanup;
3613         }
3614
3615         /* we need to update inode's ctime field, so add credit for it. */
3616         credits += OCFS2_INODE_UPDATE_CREDITS;
3617         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3618         if (IS_ERR(ctxt.handle)) {
3619                 ret = PTR_ERR(ctxt.handle);
3620                 mlog_errno(ret);
3621                 goto cleanup;
3622         }
3623
3624         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3625
3626         ocfs2_commit_trans(osb, ctxt.handle);
3627
3628         if (ctxt.data_ac)
3629                 ocfs2_free_alloc_context(ctxt.data_ac);
3630         if (ctxt.meta_ac)
3631                 ocfs2_free_alloc_context(ctxt.meta_ac);
3632         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3633                 ocfs2_schedule_truncate_log_flush(osb, 1);
3634         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3635
3636 cleanup:
3637         if (ref_tree)
3638                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3639         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3640         if (!value && !ret) {
3641                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3642                 if (ret)
3643                         mlog_errno(ret);
3644         }
3645         ocfs2_inode_unlock(inode, 1);
3646 cleanup_nolock:
3647         brelse(di_bh);
3648         brelse(xbs.xattr_bh);
3649         ocfs2_xattr_bucket_free(xbs.bucket);
3650
3651         return ret;
3652 }
3653
3654 /*
3655  * Find the xattr extent rec which may contains name_hash.
3656  * e_cpos will be the first name hash of the xattr rec.
3657  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3658  */
3659 static int ocfs2_xattr_get_rec(struct inode *inode,
3660                                u32 name_hash,
3661                                u64 *p_blkno,
3662                                u32 *e_cpos,
3663                                u32 *num_clusters,
3664                                struct ocfs2_extent_list *el)
3665 {
3666         int ret = 0, i;
3667         struct buffer_head *eb_bh = NULL;
3668         struct ocfs2_extent_block *eb;
3669         struct ocfs2_extent_rec *rec = NULL;
3670         u64 e_blkno = 0;
3671
3672         if (el->l_tree_depth) {
3673                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3674                                       &eb_bh);
3675                 if (ret) {
3676                         mlog_errno(ret);
3677                         goto out;
3678                 }
3679
3680                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3681                 el = &eb->h_list;
3682
3683                 if (el->l_tree_depth) {
3684                         ocfs2_error(inode->i_sb,
3685                                     "Inode %lu has non zero tree depth in "
3686                                     "xattr tree block %llu\n", inode->i_ino,
3687                                     (unsigned long long)eb_bh->b_blocknr);
3688                         ret = -EROFS;
3689                         goto out;
3690                 }
3691         }
3692
3693         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3694                 rec = &el->l_recs[i];
3695
3696                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3697                         e_blkno = le64_to_cpu(rec->e_blkno);
3698                         break;
3699                 }
3700         }
3701
3702         if (!e_blkno) {
3703                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3704                             "record (%u, %u, 0) in xattr", inode->i_ino,
3705                             le32_to_cpu(rec->e_cpos),
3706                             ocfs2_rec_clusters(el, rec));
3707                 ret = -EROFS;
3708                 goto out;
3709         }
3710
3711         *p_blkno = le64_to_cpu(rec->e_blkno);
3712         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3713         if (e_cpos)
3714                 *e_cpos = le32_to_cpu(rec->e_cpos);
3715 out:
3716         brelse(eb_bh);
3717         return ret;
3718 }
3719
3720 typedef int (xattr_bucket_func)(struct inode *inode,
3721                                 struct ocfs2_xattr_bucket *bucket,
3722                                 void *para);
3723
3724 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3725                                    struct ocfs2_xattr_bucket *bucket,
3726                                    int name_index,
3727                                    const char *name,
3728                                    u32 name_hash,
3729                                    u16 *xe_index,
3730                                    int *found)
3731 {
3732         int i, ret = 0, cmp = 1, block_off, new_offset;
3733         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3734         size_t name_len = strlen(name);
3735         struct ocfs2_xattr_entry *xe = NULL;
3736         char *xe_name;
3737
3738         /*
3739          * We don't use binary search in the bucket because there
3740          * may be multiple entries with the same name hash.
3741          */
3742         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3743                 xe = &xh->xh_entries[i];
3744
3745                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3746                         continue;
3747                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3748                         break;
3749
3750                 cmp = name_index - ocfs2_xattr_get_type(xe);
3751                 if (!cmp)
3752                         cmp = name_len - xe->xe_name_len;
3753                 if (cmp)
3754                         continue;
3755
3756                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3757                                                         xh,
3758                                                         i,
3759                                                         &block_off,
3760                                                         &new_offset);
3761                 if (ret) {
3762                         mlog_errno(ret);
3763                         break;
3764                 }
3765
3766
3767                 xe_name = bucket_block(bucket, block_off) + new_offset;
3768                 if (!memcmp(name, xe_name, name_len)) {
3769                         *xe_index = i;
3770                         *found = 1;
3771                         ret = 0;
3772                         break;
3773                 }
3774         }
3775
3776         return ret;
3777 }
3778
3779 /*
3780  * Find the specified xattr entry in a series of buckets.
3781  * This series start from p_blkno and last for num_clusters.
3782  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3783  * the num of the valid buckets.
3784  *
3785  * Return the buffer_head this xattr should reside in. And if the xattr's
3786  * hash is in the gap of 2 buckets, return the lower bucket.
3787  */
3788 static int ocfs2_xattr_bucket_find(struct inode *inode,
3789                                    int name_index,
3790                                    const char *name,
3791                                    u32 name_hash,
3792                                    u64 p_blkno,
3793                                    u32 first_hash,
3794                                    u32 num_clusters,
3795                                    struct ocfs2_xattr_search *xs)
3796 {
3797         int ret, found = 0;
3798         struct ocfs2_xattr_header *xh = NULL;
3799         struct ocfs2_xattr_entry *xe = NULL;
3800         u16 index = 0;
3801         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3802         int low_bucket = 0, bucket, high_bucket;
3803         struct ocfs2_xattr_bucket *search;
3804         u32 last_hash;
3805         u64 blkno, lower_blkno = 0;
3806
3807         search = ocfs2_xattr_bucket_new(inode);
3808         if (!search) {
3809                 ret = -ENOMEM;
3810                 mlog_errno(ret);
3811                 goto out;
3812         }
3813
3814         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3815         if (ret) {
3816                 mlog_errno(ret);
3817                 goto out;
3818         }
3819
3820         xh = bucket_xh(search);
3821         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3822         while (low_bucket <= high_bucket) {
3823                 ocfs2_xattr_bucket_relse(search);
3824
3825                 bucket = (low_bucket + high_bucket) / 2;
3826                 blkno = p_blkno + bucket * blk_per_bucket;
3827                 ret = ocfs2_read_xattr_bucket(search, blkno);
3828                 if (ret) {
3829                         mlog_errno(ret);
3830                         goto out;
3831                 }
3832
3833                 xh = bucket_xh(search);
3834                 xe = &xh->xh_entries[0];
3835                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3836                         high_bucket = bucket - 1;
3837                         continue;
3838                 }
3839
3840                 /*
3841                  * Check whether the hash of the last entry in our
3842                  * bucket is larger than the search one. for an empty
3843                  * bucket, the last one is also the first one.
3844                  */
3845                 if (xh->xh_count)
3846                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3847
3848                 last_hash = le32_to_cpu(xe->xe_name_hash);
3849
3850                 /* record lower_blkno which may be the insert place. */
3851                 lower_blkno = blkno;
3852
3853                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3854                         low_bucket = bucket + 1;
3855                         continue;
3856                 }
3857
3858                 /* the searched xattr should reside in this bucket if exists. */
3859                 ret = ocfs2_find_xe_in_bucket(inode, search,
3860                                               name_index, name, name_hash,
3861                                               &index, &found);
3862                 if (ret) {
3863                         mlog_errno(ret);
3864                         goto out;
3865                 }
3866                 break;
3867         }
3868
3869         /*
3870          * Record the bucket we have found.
3871          * When the xattr's hash value is in the gap of 2 buckets, we will
3872          * always set it to the previous bucket.
3873          */
3874         if (!lower_blkno)
3875                 lower_blkno = p_blkno;
3876
3877         /* This should be in cache - we just read it during the search */
3878         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3879         if (ret) {
3880                 mlog_errno(ret);
3881                 goto out;
3882         }
3883
3884         xs->header = bucket_xh(xs->bucket);
3885         xs->base = bucket_block(xs->bucket, 0);
3886         xs->end = xs->base + inode->i_sb->s_blocksize;
3887
3888         if (found) {
3889                 xs->here = &xs->header->xh_entries[index];
3890                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3891                      (unsigned long long)bucket_blkno(xs->bucket), index);
3892         } else
3893                 ret = -ENODATA;
3894
3895 out:
3896         ocfs2_xattr_bucket_free(search);
3897         return ret;
3898 }
3899
3900 static int ocfs2_xattr_index_block_find(struct inode *inode,
3901                                         struct buffer_head *root_bh,
3902                                         int name_index,
3903                                         const char *name,
3904                                         struct ocfs2_xattr_search *xs)
3905 {
3906         int ret;
3907         struct ocfs2_xattr_block *xb =
3908                         (struct ocfs2_xattr_block *)root_bh->b_data;
3909         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3910         struct ocfs2_extent_list *el = &xb_root->xt_list;
3911         u64 p_blkno = 0;
3912         u32 first_hash, num_clusters = 0;
3913         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3914
3915         if (le16_to_cpu(el->l_next_free_rec) == 0)
3916                 return -ENODATA;
3917
3918         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3919              name, name_hash, name_index);
3920
3921         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3922                                   &num_clusters, el);
3923         if (ret) {
3924                 mlog_errno(ret);
3925                 goto out;
3926         }
3927
3928         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3929
3930         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3931              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3932              first_hash);
3933
3934         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3935                                       p_blkno, first_hash, num_clusters, xs);
3936
3937 out:
3938         return ret;
3939 }
3940
3941 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3942                                        u64 blkno,
3943                                        u32 clusters,
3944                                        xattr_bucket_func *func,
3945                                        void *para)
3946 {
3947         int i, ret = 0;
3948         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3949         u32 num_buckets = clusters * bpc;
3950         struct ocfs2_xattr_bucket *bucket;
3951
3952         bucket = ocfs2_xattr_bucket_new(inode);
3953         if (!bucket) {
3954                 mlog_errno(-ENOMEM);
3955                 return -ENOMEM;
3956         }
3957
3958         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3959              clusters, (unsigned long long)blkno);
3960
3961         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3962                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3963                 if (ret) {
3964                         mlog_errno(ret);
3965                         break;
3966                 }
3967
3968                 /*
3969                  * The real bucket num in this series of blocks is stored
3970                  * in the 1st bucket.
3971                  */
3972                 if (i == 0)
3973                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3974
3975                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3976                      (unsigned long long)blkno,
3977                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3978                 if (func) {
3979                         ret = func(inode, bucket, para);
3980                         if (ret && ret != -ERANGE)
3981                                 mlog_errno(ret);
3982                         /* Fall through to bucket_relse() */
3983                 }
3984
3985                 ocfs2_xattr_bucket_relse(bucket);
3986                 if (ret)
3987                         break;
3988         }
3989
3990         ocfs2_xattr_bucket_free(bucket);
3991         return ret;
3992 }
3993
3994 struct ocfs2_xattr_tree_list {
3995         char *buffer;
3996         size_t buffer_size;
3997         size_t result;
3998 };
3999
4000 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4001                                              struct ocfs2_xattr_header *xh,
4002                                              int index,
4003                                              int *block_off,
4004                                              int *new_offset)
4005 {
4006         u16 name_offset;
4007
4008         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4009                 return -EINVAL;
4010
4011         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4012
4013         *block_off = name_offset >> sb->s_blocksize_bits;
4014         *new_offset = name_offset % sb->s_blocksize;
4015
4016         return 0;
4017 }
4018
4019 static int ocfs2_list_xattr_bucket(struct inode *inode,
4020                                    struct ocfs2_xattr_bucket *bucket,
4021                                    void *para)
4022 {
4023         int ret = 0, type;
4024         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4025         int i, block_off, new_offset;
4026         const char *prefix, *name;
4027
4028         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4029                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4030                 type = ocfs2_xattr_get_type(entry);
4031                 prefix = ocfs2_xattr_prefix(type);
4032
4033                 if (prefix) {
4034                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4035                                                                 bucket_xh(bucket),
4036                                                                 i,
4037                                                                 &block_off,
4038                                                                 &new_offset);
4039                         if (ret)
4040                                 break;
4041
4042                         name = (const char *)bucket_block(bucket, block_off) +
4043                                 new_offset;
4044                         ret = ocfs2_xattr_list_entry(xl->buffer,
4045                                                      xl->buffer_size,
4046                                                      &xl->result,
4047                                                      prefix, name,
4048                                                      entry->xe_name_len);
4049                         if (ret)
4050                                 break;
4051                 }
4052         }
4053
4054         return ret;
4055 }
4056
4057 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4058                                            struct buffer_head *blk_bh,
4059                                            xattr_tree_rec_func *rec_func,
4060                                            void *para)
4061 {
4062         struct ocfs2_xattr_block *xb =
4063                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4064         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4065         int ret = 0;
4066         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4067         u64 p_blkno = 0;
4068
4069         if (!el->l_next_free_rec || !rec_func)
4070                 return 0;
4071
4072         while (name_hash > 0) {
4073                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4074                                           &e_cpos, &num_clusters, el);
4075                 if (ret) {
4076                         mlog_errno(ret);
4077                         break;
4078                 }
4079
4080                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4081                                num_clusters, para);
4082                 if (ret) {
4083                         if (ret != -ERANGE)
4084                                 mlog_errno(ret);
4085                         break;
4086                 }
4087
4088                 if (e_cpos == 0)
4089                         break;
4090
4091                 name_hash = e_cpos - 1;
4092         }
4093
4094         return ret;
4095
4096 }
4097
4098 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4099                                      struct buffer_head *root_bh,
4100                                      u64 blkno, u32 cpos, u32 len, void *para)
4101 {
4102         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4103                                            ocfs2_list_xattr_bucket, para);
4104 }
4105
4106 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4107                                              struct buffer_head *blk_bh,
4108                                              char *buffer,
4109                                              size_t buffer_size)
4110 {
4111         int ret;
4112         struct ocfs2_xattr_tree_list xl = {
4113                 .buffer = buffer,
4114                 .buffer_size = buffer_size,
4115                 .result = 0,
4116         };
4117
4118         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4119                                               ocfs2_list_xattr_tree_rec, &xl);
4120         if (ret) {
4121                 mlog_errno(ret);
4122                 goto out;
4123         }
4124
4125         ret = xl.result;
4126 out:
4127         return ret;
4128 }
4129
4130 static int cmp_xe(const void *a, const void *b)
4131 {
4132         const struct ocfs2_xattr_entry *l = a, *r = b;
4133         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4134         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4135
4136         if (l_hash > r_hash)
4137                 return 1;
4138         if (l_hash < r_hash)
4139                 return -1;
4140         return 0;
4141 }
4142
4143 static void swap_xe(void *a, void *b, int size)
4144 {
4145         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4146
4147         tmp = *l;
4148         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4149         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4150 }
4151
4152 /*
4153  * When the ocfs2_xattr_block is filled up, new bucket will be created
4154  * and all the xattr entries will be moved to the new bucket.
4155  * The header goes at the start of the bucket, and the names+values are
4156  * filled from the end.  This is why *target starts as the last buffer.
4157  * Note: we need to sort the entries since they are not saved in order
4158  * in the ocfs2_xattr_block.
4159  */
4160 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4161                                            struct buffer_head *xb_bh,
4162                                            struct ocfs2_xattr_bucket *bucket)
4163 {
4164         int i, blocksize = inode->i_sb->s_blocksize;
4165         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4166         u16 offset, size, off_change;
4167         struct ocfs2_xattr_entry *xe;
4168         struct ocfs2_xattr_block *xb =
4169                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4170         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4171         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4172         u16 count = le16_to_cpu(xb_xh->xh_count);
4173         char *src = xb_bh->b_data;
4174         char *target = bucket_block(bucket, blks - 1);
4175
4176         mlog(0, "cp xattr from block %llu to bucket %llu\n",
4177              (unsigned long long)xb_bh->b_blocknr,
4178              (unsigned long long)bucket_blkno(bucket));
4179
4180         for (i = 0; i < blks; i++)
4181                 memset(bucket_block(bucket, i), 0, blocksize);
4182
4183         /*
4184          * Since the xe_name_offset is based on ocfs2_xattr_header,
4185          * there is a offset change corresponding to the change of
4186          * ocfs2_xattr_header's position.
4187          */
4188         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4189         xe = &xb_xh->xh_entries[count - 1];
4190         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4191         size = blocksize - offset;
4192
4193         /* copy all the names and values. */
4194         memcpy(target + offset, src + offset, size);
4195
4196         /* Init new header now. */
4197         xh->xh_count = xb_xh->xh_count;
4198         xh->xh_num_buckets = cpu_to_le16(1);
4199         xh->xh_name_value_len = cpu_to_le16(size);
4200         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4201
4202         /* copy all the entries. */
4203         target = bucket_block(bucket, 0);
4204         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4205         size = count * sizeof(struct ocfs2_xattr_entry);
4206         memcpy(target + offset, (char *)xb_xh + offset, size);
4207
4208         /* Change the xe offset for all the xe because of the move. */
4209         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4210                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4211         for (i = 0; i < count; i++)
4212                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4213
4214         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
4215              offset, size, off_change);
4216
4217         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4218              cmp_xe, swap_xe);
4219 }
4220
4221 /*
4222  * After we move xattr from block to index btree, we have to
4223  * update ocfs2_xattr_search to the new xe and base.
4224  *
4225  * When the entry is in xattr block, xattr_bh indicates the storage place.
4226  * While if the entry is in index b-tree, "bucket" indicates the
4227  * real place of the xattr.
4228  */
4229 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4230                                             struct ocfs2_xattr_search *xs,
4231                                             struct buffer_head *old_bh)
4232 {
4233         char *buf = old_bh->b_data;
4234         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4235         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4236         int i;
4237
4238         xs->header = bucket_xh(xs->bucket);
4239         xs->base = bucket_block(xs->bucket, 0);
4240         xs->end = xs->base + inode->i_sb->s_blocksize;
4241
4242         if (xs->not_found)
4243                 return;
4244
4245         i = xs->here - old_xh->xh_entries;
4246         xs->here = &xs->header->xh_entries[i];
4247 }
4248
4249 static int ocfs2_xattr_create_index_block(struct inode *inode,
4250                                           struct ocfs2_xattr_search *xs,
4251                                           struct ocfs2_xattr_set_ctxt *ctxt)
4252 {
4253         int ret;
4254         u32 bit_off, len;
4255         u64 blkno;
4256         handle_t *handle = ctxt->handle;
4257         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4258         struct buffer_head *xb_bh = xs->xattr_bh;
4259         struct ocfs2_xattr_block *xb =
4260                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4261         struct ocfs2_xattr_tree_root *xr;
4262         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4263
4264         mlog(0, "create xattr index block for %llu\n",
4265              (unsigned long long)xb_bh->b_blocknr);
4266
4267         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4268         BUG_ON(!xs->bucket);
4269
4270         /*
4271          * XXX:
4272          * We can use this lock for now, and maybe move to a dedicated mutex
4273          * if performance becomes a problem later.
4274          */
4275         down_write(&oi->ip_alloc_sem);
4276
4277         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4278                                       OCFS2_JOURNAL_ACCESS_WRITE);
4279         if (ret) {
4280                 mlog_errno(ret);
4281                 goto out;
4282         }
4283
4284         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4285                                      1, 1, &bit_off, &len);
4286         if (ret) {
4287                 mlog_errno(ret);
4288                 goto out;
4289         }
4290
4291         /*
4292          * The bucket may spread in many blocks, and
4293          * we will only touch the 1st block and the last block
4294          * in the whole bucket(one for entry and one for data).
4295          */
4296         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4297
4298         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
4299              (unsigned long long)blkno);
4300
4301         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4302         if (ret) {
4303                 mlog_errno(ret);
4304                 goto out;
4305         }
4306
4307         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4308                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4309         if (ret) {
4310                 mlog_errno(ret);
4311                 goto out;
4312         }
4313
4314         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4315         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4316
4317         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4318
4319         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4320         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4321                offsetof(struct ocfs2_xattr_block, xb_attrs));
4322
4323         xr = &xb->xb_attrs.xb_root;
4324         xr->xt_clusters = cpu_to_le32(1);
4325         xr->xt_last_eb_blk = 0;
4326         xr->xt_list.l_tree_depth = 0;
4327         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4328         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4329
4330         xr->xt_list.l_recs[0].e_cpos = 0;
4331         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4332         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4333
4334         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4335
4336         ocfs2_journal_dirty(handle, xb_bh);
4337
4338 out:
4339         up_write(&oi->ip_alloc_sem);
4340
4341         return ret;
4342 }
4343
4344 static int cmp_xe_offset(const void *a, const void *b)
4345 {
4346         const struct ocfs2_xattr_entry *l = a, *r = b;
4347         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4348         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4349
4350         if (l_name_offset < r_name_offset)
4351                 return 1;
4352         if (l_name_offset > r_name_offset)
4353                 return -1;
4354         return 0;
4355 }
4356
4357 /*
4358  * defrag a xattr bucket if we find that the bucket has some
4359  * holes beteen name/value pairs.
4360  * We will move all the name/value pairs to the end of the bucket
4361  * so that we can spare some space for insertion.
4362  */
4363 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4364                                      handle_t *handle,
4365                                      struct ocfs2_xattr_bucket *bucket)
4366 {
4367         int ret, i;
4368         size_t end, offset, len;
4369         struct ocfs2_xattr_header *xh;
4370         char *entries, *buf, *bucket_buf = NULL;
4371         u64 blkno = bucket_blkno(bucket);
4372         u16 xh_free_start;
4373         size_t blocksize = inode->i_sb->s_blocksize;
4374         struct ocfs2_xattr_entry *xe;
4375
4376         /*
4377          * In order to make the operation more efficient and generic,
4378          * we copy all the blocks into a contiguous memory and do the
4379          * defragment there, so if anything is error, we will not touch
4380          * the real block.
4381          */
4382         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4383         if (!bucket_buf) {
4384                 ret = -EIO;
4385                 goto out;
4386         }
4387
4388         buf = bucket_buf;
4389         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4390                 memcpy(buf, bucket_block(bucket, i), blocksize);
4391
4392         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4393                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4394         if (ret < 0) {
4395                 mlog_errno(ret);
4396                 goto out;
4397         }
4398
4399         xh = (struct ocfs2_xattr_header *)bucket_buf;
4400         entries = (char *)xh->xh_entries;
4401         xh_free_start = le16_to_cpu(xh->xh_free_start);
4402
4403         mlog(0, "adjust xattr bucket in %llu, count = %u, "
4404              "xh_free_start = %u, xh_name_value_len = %u.\n",
4405              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4406              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4407
4408         /*
4409          * sort all the entries by their offset.
4410          * the largest will be the first, so that we can
4411          * move them to the end one by one.
4412          */
4413         sort(entries, le16_to_cpu(xh->xh_count),
4414              sizeof(struct ocfs2_xattr_entry),
4415              cmp_xe_offset, swap_xe);
4416
4417         /* Move all name/values to the end of the bucket. */
4418         xe = xh->xh_entries;
4419         end = OCFS2_XATTR_BUCKET_SIZE;
4420         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4421                 offset = le16_to_cpu(xe->xe_name_offset);
4422                 len = namevalue_size_xe(xe);
4423
4424                 /*
4425                  * We must make sure that the name/value pair
4426                  * exist in the same block. So adjust end to
4427                  * the previous block end if needed.
4428                  */
4429                 if (((end - len) / blocksize !=
4430                         (end - 1) / blocksize))
4431                         end = end - end % blocksize;
4432
4433                 if (end > offset + len) {
4434                         memmove(bucket_buf + end - len,
4435                                 bucket_buf + offset, len);
4436                         xe->xe_name_offset = cpu_to_le16(end - len);
4437                 }
4438
4439                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4440                                 "bucket %llu\n", (unsigned long long)blkno);
4441
4442                 end -= len;
4443         }
4444
4445         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4446                         "bucket %llu\n", (unsigned long long)blkno);
4447
4448         if (xh_free_start == end)
4449                 goto out;
4450
4451         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4452         xh->xh_free_start = cpu_to_le16(end);
4453
4454         /* sort the entries by their name_hash. */
4455         sort(entries, le16_to_cpu(xh->xh_count),
4456              sizeof(struct ocfs2_xattr_entry),
4457              cmp_xe, swap_xe);
4458
4459         buf = bucket_buf;
4460         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4461                 memcpy(bucket_block(bucket, i), buf, blocksize);
4462         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4463
4464 out:
4465         kfree(bucket_buf);
4466         return ret;
4467 }
4468
4469 /*
4470  * prev_blkno points to the start of an existing extent.  new_blkno
4471  * points to a newly allocated extent.  Because we know each of our
4472  * clusters contains more than bucket, we can easily split one cluster
4473  * at a bucket boundary.  So we take the last cluster of the existing
4474  * extent and split it down the middle.  We move the last half of the
4475  * buckets in the last cluster of the existing extent over to the new
4476  * extent.
4477  *
4478  * first_bh is the buffer at prev_blkno so we can update the existing
4479  * extent's bucket count.  header_bh is the bucket were we were hoping
4480  * to insert our xattr.  If the bucket move places the target in the new
4481  * extent, we'll update first_bh and header_bh after modifying the old
4482  * extent.
4483  *
4484  * first_hash will be set as the 1st xe's name_hash in the new extent.
4485  */
4486 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4487                                                handle_t *handle,
4488                                                struct ocfs2_xattr_bucket *first,
4489                                                struct ocfs2_xattr_bucket *target,
4490                                                u64 new_blkno,
4491                                                u32 num_clusters,
4492                                                u32 *first_hash)
4493 {
4494         int ret;
4495         struct super_block *sb = inode->i_sb;
4496         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4497         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4498         int to_move = num_buckets / 2;
4499         u64 src_blkno;
4500         u64 last_cluster_blkno = bucket_blkno(first) +
4501                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4502
4503         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4504         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4505
4506         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4507              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4508
4509         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4510                                      last_cluster_blkno, new_blkno,
4511                                      to_move, first_hash);
4512         if (ret) {
4513                 mlog_errno(ret);
4514                 goto out;
4515         }
4516
4517         /* This is the first bucket that got moved */
4518         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4519
4520         /*
4521          * If the target bucket was part of the moved buckets, we need to
4522          * update first and target.
4523          */
4524         if (bucket_blkno(target) >= src_blkno) {
4525                 /* Find the block for the new target bucket */
4526                 src_blkno = new_blkno +
4527                         (bucket_blkno(target) - src_blkno);
4528
4529                 ocfs2_xattr_bucket_relse(first);
4530                 ocfs2_xattr_bucket_relse(target);
4531
4532                 /*
4533                  * These shouldn't fail - the buffers are in the
4534                  * journal from ocfs2_cp_xattr_bucket().
4535                  */
4536                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4537                 if (ret) {
4538                         mlog_errno(ret);
4539                         goto out;
4540                 }
4541                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4542                 if (ret)
4543                         mlog_errno(ret);
4544
4545         }
4546
4547 out:
4548         return ret;
4549 }
4550
4551 /*
4552  * Find the suitable pos when we divide a bucket into 2.
4553  * We have to make sure the xattrs with the same hash value exist
4554  * in the same bucket.
4555  *
4556  * If this ocfs2_xattr_header covers more than one hash value, find a
4557  * place where the hash value changes.  Try to find the most even split.
4558  * The most common case is that all entries have different hash values,
4559  * and the first check we make will find a place to split.
4560  */
4561 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4562 {
4563         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4564         int count = le16_to_cpu(xh->xh_count);
4565         int delta, middle = count / 2;
4566
4567         /*
4568          * We start at the middle.  Each step gets farther away in both
4569          * directions.  We therefore hit the change in hash value
4570          * nearest to the middle.  Note that this loop does not execute for
4571          * count < 2.
4572          */
4573         for (delta = 0; delta < middle; delta++) {
4574                 /* Let's check delta earlier than middle */
4575                 if (cmp_xe(&entries[middle - delta - 1],
4576                            &entries[middle - delta]))
4577                         return middle - delta;
4578
4579                 /* For even counts, don't walk off the end */
4580                 if ((middle + delta + 1) == count)
4581                         continue;
4582
4583                 /* Now try delta past middle */
4584                 if (cmp_xe(&entries[middle + delta],
4585                            &entries[middle + delta + 1]))
4586                         return middle + delta + 1;
4587         }
4588
4589         /* Every entry had the same hash */
4590         return count;
4591 }
4592
4593 /*
4594  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4595  * first_hash will record the 1st hash of the new bucket.
4596  *
4597  * Normally half of the xattrs will be moved.  But we have to make
4598  * sure that the xattrs with the same hash value are stored in the
4599  * same bucket. If all the xattrs in this bucket have the same hash
4600  * value, the new bucket will be initialized as an empty one and the
4601  * first_hash will be initialized as (hash_value+1).
4602  */
4603 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4604                                     handle_t *handle,
4605                                     u64 blk,
4606                                     u64 new_blk,
4607                                     u32 *first_hash,
4608                                     int new_bucket_head)
4609 {
4610         int ret, i;
4611         int count, start, len, name_value_len = 0, name_offset = 0;
4612         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4613         struct ocfs2_xattr_header *xh;
4614         struct ocfs2_xattr_entry *xe;
4615         int blocksize = inode->i_sb->s_blocksize;
4616
4617         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4618              (unsigned long long)blk, (unsigned long long)new_blk);
4619
4620         s_bucket = ocfs2_xattr_bucket_new(inode);
4621         t_bucket = ocfs2_xattr_bucket_new(inode);
4622         if (!s_bucket || !t_bucket) {
4623                 ret = -ENOMEM;
4624                 mlog_errno(ret);
4625                 goto out;
4626         }
4627
4628         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4629         if (ret) {
4630                 mlog_errno(ret);
4631                 goto out;
4632         }
4633
4634         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4635                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4636         if (ret) {
4637                 mlog_errno(ret);
4638                 goto out;
4639         }
4640
4641         /*
4642          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4643          * there's no need to read it.
4644          */
4645         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4646         if (ret) {
4647                 mlog_errno(ret);
4648                 goto out;
4649         }
4650
4651         /*
4652          * Hey, if we're overwriting t_bucket, what difference does
4653          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4654          * same part of ocfs2_cp_xattr_bucket().
4655          */
4656         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4657                                                 new_bucket_head ?
4658                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4659                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4660         if (ret) {
4661                 mlog_errno(ret);
4662                 goto out;
4663         }
4664
4665         xh = bucket_xh(s_bucket);
4666         count = le16_to_cpu(xh->xh_count);
4667         start = ocfs2_xattr_find_divide_pos(xh);
4668
4669         if (start == count) {
4670                 xe = &xh->xh_entries[start-1];
4671
4672                 /*
4673                  * initialized a new empty bucket here.
4674                  * The hash value is set as one larger than
4675                  * that of the last entry in the previous bucket.
4676                  */
4677                 for (i = 0; i < t_bucket->bu_blocks; i++)
4678                         memset(bucket_block(t_bucket, i), 0, blocksize);
4679
4680                 xh = bucket_xh(t_bucket);
4681                 xh->xh_free_start = cpu_to_le16(blocksize);
4682                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4683                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4684
4685                 goto set_num_buckets;
4686         }
4687
4688         /* copy the whole bucket to the new first. */
4689         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4690
4691         /* update the new bucket. */
4692         xh = bucket_xh(t_bucket);
4693
4694         /*
4695          * Calculate the total name/value len and xh_free_start for
4696          * the old bucket first.
4697          */
4698         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4699         name_value_len = 0;
4700         for (i = 0; i < start; i++) {
4701                 xe = &xh->xh_entries[i];
4702                 name_value_len += namevalue_size_xe(xe);
4703                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4704                         name_offset = le16_to_cpu(xe->xe_name_offset);
4705         }
4706
4707         /*
4708          * Now begin the modification to the new bucket.
4709          *
4710          * In the new bucket, We just move the xattr entry to the beginning
4711          * and don't touch the name/value. So there will be some holes in the
4712          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4713          * called.
4714          */
4715         xe = &xh->xh_entries[start];
4716         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4717         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4718              (int)((char *)xe - (char *)xh),
4719              (int)((char *)xh->xh_entries - (char *)xh));
4720         memmove((char *)xh->xh_entries, (char *)xe, len);
4721         xe = &xh->xh_entries[count - start];
4722         len = sizeof(struct ocfs2_xattr_entry) * start;
4723         memset((char *)xe, 0, len);
4724
4725         le16_add_cpu(&xh->xh_count, -start);
4726         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4727
4728         /* Calculate xh_free_start for the new bucket. */
4729         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4730         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4731                 xe = &xh->xh_entries[i];
4732                 if (le16_to_cpu(xe->xe_name_offset) <
4733                     le16_to_cpu(xh->xh_free_start))
4734                         xh->xh_free_start = xe->xe_name_offset;
4735         }
4736
4737 set_num_buckets:
4738         /* set xh->xh_num_buckets for the new xh. */
4739         if (new_bucket_head)
4740                 xh->xh_num_buckets = cpu_to_le16(1);
4741         else
4742                 xh->xh_num_buckets = 0;
4743
4744         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4745
4746         /* store the first_hash of the new bucket. */
4747         if (first_hash)
4748                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4749
4750         /*
4751          * Now only update the 1st block of the old bucket.  If we
4752          * just added a new empty bucket, there is no need to modify
4753          * it.
4754          */
4755         if (start == count)
4756                 goto out;
4757
4758         xh = bucket_xh(s_bucket);
4759         memset(&xh->xh_entries[start], 0,
4760                sizeof(struct ocfs2_xattr_entry) * (count - start));
4761         xh->xh_count = cpu_to_le16(start);
4762         xh->xh_free_start = cpu_to_le16(name_offset);
4763         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4764
4765         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4766
4767 out:
4768         ocfs2_xattr_bucket_free(s_bucket);
4769         ocfs2_xattr_bucket_free(t_bucket);
4770
4771         return ret;
4772 }
4773
4774 /*
4775  * Copy xattr from one bucket to another bucket.
4776  *
4777  * The caller must make sure that the journal transaction
4778  * has enough space for journaling.
4779  */
4780 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4781                                  handle_t *handle,
4782                                  u64 s_blkno,
4783                                  u64 t_blkno,
4784                                  int t_is_new)
4785 {
4786         int ret;
4787         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4788
4789         BUG_ON(s_blkno == t_blkno);
4790
4791         mlog(0, "cp bucket %llu to %llu, target is %d\n",
4792              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4793              t_is_new);
4794
4795         s_bucket = ocfs2_xattr_bucket_new(inode);
4796         t_bucket = ocfs2_xattr_bucket_new(inode);
4797         if (!s_bucket || !t_bucket) {
4798                 ret = -ENOMEM;
4799                 mlog_errno(ret);
4800                 goto out;
4801         }
4802
4803         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4804         if (ret)
4805                 goto out;
4806
4807         /*
4808          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4809          * there's no need to read it.
4810          */
4811         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4812         if (ret)
4813                 goto out;
4814
4815         /*
4816          * Hey, if we're overwriting t_bucket, what difference does
4817          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4818          * cluster to fill, we came here from
4819          * ocfs2_mv_xattr_buckets(), and it is really new -
4820          * ACCESS_CREATE is required.  But we also might have moved data
4821          * out of t_bucket before extending back into it.
4822          * ocfs2_add_new_xattr_bucket() can do this - its call to
4823          * ocfs2_add_new_xattr_cluster() may have created a new extent
4824          * and copied out the end of the old extent.  Then it re-extends
4825          * the old extent back to create space for new xattrs.  That's
4826          * how we get here, and the bucket isn't really new.
4827          */
4828         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4829                                                 t_is_new ?
4830                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4831                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4832         if (ret)
4833                 goto out;
4834
4835         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4836         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4837
4838 out:
4839         ocfs2_xattr_bucket_free(t_bucket);
4840         ocfs2_xattr_bucket_free(s_bucket);
4841
4842         return ret;
4843 }
4844
4845 /*
4846  * src_blk points to the start of an existing extent.  last_blk points to
4847  * last cluster in that extent.  to_blk points to a newly allocated
4848  * extent.  We copy the buckets from the cluster at last_blk to the new
4849  * extent.  If start_bucket is non-zero, we skip that many buckets before
4850  * we start copying.  The new extent's xh_num_buckets gets set to the
4851  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4852  * by the same amount.
4853  */
4854 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4855                                   u64 src_blk, u64 last_blk, u64 to_blk,
4856                                   unsigned int start_bucket,
4857                                   u32 *first_hash)
4858 {
4859         int i, ret, credits;
4860         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4861         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4862         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4863         struct ocfs2_xattr_bucket *old_first, *new_first;
4864
4865         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4866              (unsigned long long)last_blk, (unsigned long long)to_blk);
4867
4868         BUG_ON(start_bucket >= num_buckets);
4869         if (start_bucket) {
4870                 num_buckets -= start_bucket;
4871                 last_blk += (start_bucket * blks_per_bucket);
4872         }
4873
4874         /* The first bucket of the original extent */
4875         old_first = ocfs2_xattr_bucket_new(inode);
4876         /* The first bucket of the new extent */
4877         new_first = ocfs2_xattr_bucket_new(inode);
4878         if (!old_first || !new_first) {
4879                 ret = -ENOMEM;
4880                 mlog_errno(ret);
4881                 goto out;
4882         }
4883
4884         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4885         if (ret) {
4886                 mlog_errno(ret);
4887                 goto out;
4888         }
4889
4890         /*
4891          * We need to update the first bucket of the old extent and all
4892          * the buckets going to the new extent.
4893          */
4894         credits = ((num_buckets + 1) * blks_per_bucket);
4895         ret = ocfs2_extend_trans(handle, credits);
4896         if (ret) {
4897                 mlog_errno(ret);
4898                 goto out;
4899         }
4900
4901         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4902                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4903         if (ret) {
4904                 mlog_errno(ret);
4905                 goto out;
4906         }
4907
4908         for (i = 0; i < num_buckets; i++) {
4909                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4910                                             last_blk + (i * blks_per_bucket),
4911                                             to_blk + (i * blks_per_bucket),
4912                                             1);
4913                 if (ret) {
4914                         mlog_errno(ret);
4915                         goto out;
4916                 }
4917         }
4918
4919         /*
4920          * Get the new bucket ready before we dirty anything
4921          * (This actually shouldn't fail, because we already dirtied
4922          * it once in ocfs2_cp_xattr_bucket()).
4923          */
4924         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4925         if (ret) {
4926                 mlog_errno(ret);
4927                 goto out;
4928         }
4929         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4930                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4931         if (ret) {
4932                 mlog_errno(ret);
4933                 goto out;
4934         }
4935
4936         /* Now update the headers */
4937         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4938         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4939
4940         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4941         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4942
4943         if (first_hash)
4944                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4945
4946 out:
4947         ocfs2_xattr_bucket_free(new_first);
4948         ocfs2_xattr_bucket_free(old_first);
4949         return ret;
4950 }
4951
4952 /*
4953  * Move some xattrs in this cluster to the new cluster.
4954  * This function should only be called when bucket size == cluster size.
4955  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4956  */
4957 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4958                                       handle_t *handle,
4959                                       u64 prev_blk,
4960                                       u64 new_blk,
4961                                       u32 *first_hash)
4962 {
4963         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4964         int ret, credits = 2 * blk_per_bucket;
4965
4966         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4967
4968         ret = ocfs2_extend_trans(handle, credits);
4969         if (ret) {
4970                 mlog_errno(ret);
4971                 return ret;
4972         }
4973
4974         /* Move half of the xattr in start_blk to the next bucket. */
4975         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4976                                           new_blk, first_hash, 1);
4977 }
4978
4979 /*
4980  * Move some xattrs from the old cluster to the new one since they are not
4981  * contiguous in ocfs2 xattr tree.
4982  *
4983  * new_blk starts a new separate cluster, and we will move some xattrs from
4984  * prev_blk to it. v_start will be set as the first name hash value in this
4985  * new cluster so that it can be used as e_cpos during tree insertion and
4986  * don't collide with our original b-tree operations. first_bh and header_bh
4987  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4988  * to extend the insert bucket.
4989  *
4990  * The problem is how much xattr should we move to the new one and when should
4991  * we update first_bh and header_bh?
4992  * 1. If cluster size > bucket size, that means the previous cluster has more
4993  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4994  *    update the first_bh and header_bh if the insert bucket has been moved
4995  *    to the new cluster.
4996  * 2. If cluster_size == bucket_size:
4997  *    a) If the previous extent rec has more than one cluster and the insert
4998  *       place isn't in the last cluster, copy the entire last cluster to the
4999  *       new one. This time, we don't need to upate the first_bh and header_bh
5000  *       since they will not be moved into the new cluster.
5001  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5002  *       the new one. And we set the extend flag to zero if the insert place is
5003  *       moved into the new allocated cluster since no extend is needed.
5004  */
5005 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5006                                             handle_t *handle,
5007                                             struct ocfs2_xattr_bucket *first,
5008                                             struct ocfs2_xattr_bucket *target,
5009                                             u64 new_blk,
5010                                             u32 prev_clusters,
5011                                             u32 *v_start,
5012                                             int *extend)
5013 {
5014         int ret;
5015
5016         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
5017              (unsigned long long)bucket_blkno(first), prev_clusters,
5018              (unsigned long long)new_blk);
5019
5020         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5021                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5022                                                           handle,
5023                                                           first, target,
5024                                                           new_blk,
5025                                                           prev_clusters,
5026                                                           v_start);
5027                 if (ret)
5028                         mlog_errno(ret);
5029         } else {
5030                 /* The start of the last cluster in the first extent */
5031                 u64 last_blk = bucket_blkno(first) +
5032                         ((prev_clusters - 1) *
5033                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5034
5035                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5036                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5037                                                      bucket_blkno(first),
5038                                                      last_blk, new_blk, 0,
5039                                                      v_start);
5040                         if (ret)
5041                                 mlog_errno(ret);
5042                 } else {
5043                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5044                                                          last_blk, new_blk,
5045                                                          v_start);
5046                         if (ret)
5047                                 mlog_errno(ret);
5048
5049                         if ((bucket_blkno(target) == last_blk) && extend)
5050                                 *extend = 0;
5051                 }
5052         }
5053
5054         return ret;
5055 }
5056
5057 /*
5058  * Add a new cluster for xattr storage.
5059  *
5060  * If the new cluster is contiguous with the previous one, it will be
5061  * appended to the same extent record, and num_clusters will be updated.
5062  * If not, we will insert a new extent for it and move some xattrs in
5063  * the last cluster into the new allocated one.
5064  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5065  * lose the benefits of hashing because we'll have to search large leaves.
5066  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5067  * if it's bigger).
5068  *
5069  * first_bh is the first block of the previous extent rec and header_bh
5070  * indicates the bucket we will insert the new xattrs. They will be updated
5071  * when the header_bh is moved into the new cluster.
5072  */
5073 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5074                                        struct buffer_head *root_bh,
5075                                        struct ocfs2_xattr_bucket *first,
5076                                        struct ocfs2_xattr_bucket *target,
5077                                        u32 *num_clusters,
5078                                        u32 prev_cpos,
5079                                        int *extend,
5080                                        struct ocfs2_xattr_set_ctxt *ctxt)
5081 {
5082         int ret;
5083         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5084         u32 prev_clusters = *num_clusters;
5085         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5086         u64 block;
5087         handle_t *handle = ctxt->handle;
5088         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5089         struct ocfs2_extent_tree et;
5090
5091         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
5092              "previous xattr blkno = %llu\n",
5093              (unsigned long long)OCFS2_I(inode)->ip_blkno,
5094              prev_cpos, (unsigned long long)bucket_blkno(first));
5095
5096         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5097
5098         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5099                                       OCFS2_JOURNAL_ACCESS_WRITE);
5100         if (ret < 0) {
5101                 mlog_errno(ret);
5102                 goto leave;
5103         }
5104
5105         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5106                                      clusters_to_add, &bit_off, &num_bits);
5107         if (ret < 0) {
5108                 if (ret != -ENOSPC)
5109                         mlog_errno(ret);
5110                 goto leave;
5111         }
5112
5113         BUG_ON(num_bits > clusters_to_add);
5114
5115         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5116         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
5117              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
5118
5119         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5120             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5121              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5122                 /*
5123                  * If this cluster is contiguous with the old one and
5124                  * adding this new cluster, we don't surpass the limit of
5125                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5126                  * initialized and used like other buckets in the previous
5127                  * cluster.
5128                  * So add it as a contiguous one. The caller will handle
5129                  * its init process.
5130                  */
5131                 v_start = prev_cpos + prev_clusters;
5132                 *num_clusters = prev_clusters + num_bits;
5133                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
5134                      num_bits);
5135         } else {
5136                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5137                                                        handle,
5138                                                        first,
5139                                                        target,
5140                                                        block,
5141                                                        prev_clusters,
5142                                                        &v_start,
5143                                                        extend);
5144                 if (ret) {
5145                         mlog_errno(ret);
5146                         goto leave;
5147                 }
5148         }
5149
5150         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
5151              num_bits, (unsigned long long)block, v_start);
5152         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5153                                   num_bits, 0, ctxt->meta_ac);
5154         if (ret < 0) {
5155                 mlog_errno(ret);
5156                 goto leave;
5157         }
5158
5159         ocfs2_journal_dirty(handle, root_bh);
5160
5161 leave:
5162         return ret;
5163 }
5164
5165 /*
5166  * We are given an extent.  'first' is the bucket at the very front of
5167  * the extent.  The extent has space for an additional bucket past
5168  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5169  * of the target bucket.  We wish to shift every bucket past the target
5170  * down one, filling in that additional space.  When we get back to the
5171  * target, we split the target between itself and the now-empty bucket
5172  * at target+1 (aka, target_blkno + blks_per_bucket).
5173  */
5174 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5175                                      handle_t *handle,
5176                                      struct ocfs2_xattr_bucket *first,
5177                                      u64 target_blk,
5178                                      u32 num_clusters)
5179 {
5180         int ret, credits;
5181         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5182         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5183         u64 end_blk;
5184         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5185
5186         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
5187              "from %llu, len = %u\n", (unsigned long long)target_blk,
5188              (unsigned long long)bucket_blkno(first), num_clusters);
5189
5190         /* The extent must have room for an additional bucket */
5191         BUG_ON(new_bucket >=
5192                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5193
5194         /* end_blk points to the last existing bucket */
5195         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5196
5197         /*
5198          * end_blk is the start of the last existing bucket.
5199          * Thus, (end_blk - target_blk) covers the target bucket and
5200          * every bucket after it up to, but not including, the last
5201          * existing bucket.  Then we add the last existing bucket, the
5202          * new bucket, and the first bucket (3 * blk_per_bucket).
5203          */
5204         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5205         ret = ocfs2_extend_trans(handle, credits);
5206         if (ret) {
5207                 mlog_errno(ret);
5208                 goto out;
5209         }
5210
5211         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5212                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5213         if (ret) {
5214                 mlog_errno(ret);
5215                 goto out;
5216         }
5217
5218         while (end_blk != target_blk) {
5219                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5220                                             end_blk + blk_per_bucket, 0);
5221                 if (ret)
5222                         goto out;
5223                 end_blk -= blk_per_bucket;
5224         }
5225
5226         /* Move half of the xattr in target_blkno to the next bucket. */
5227         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5228                                         target_blk + blk_per_bucket, NULL, 0);
5229
5230         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5231         ocfs2_xattr_bucket_journal_dirty(handle, first);
5232
5233 out:
5234         return ret;
5235 }
5236
5237 /*
5238  * Add new xattr bucket in an extent record and adjust the buckets
5239  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5240  * bucket we want to insert into.
5241  *
5242  * In the easy case, we will move all the buckets after target down by
5243  * one. Half of target's xattrs will be moved to the next bucket.
5244  *
5245  * If current cluster is full, we'll allocate a new one.  This may not
5246  * be contiguous.  The underlying calls will make sure that there is
5247  * space for the insert, shifting buckets around if necessary.
5248  * 'target' may be moved by those calls.
5249  */
5250 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5251                                       struct buffer_head *xb_bh,
5252                                       struct ocfs2_xattr_bucket *target,
5253                                       struct ocfs2_xattr_set_ctxt *ctxt)
5254 {
5255         struct ocfs2_xattr_block *xb =
5256                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5257         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5258         struct ocfs2_extent_list *el = &xb_root->xt_list;
5259         u32 name_hash =
5260                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5261         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5262         int ret, num_buckets, extend = 1;
5263         u64 p_blkno;
5264         u32 e_cpos, num_clusters;
5265         /* The bucket at the front of the extent */
5266         struct ocfs2_xattr_bucket *first;
5267
5268         mlog(0, "Add new xattr bucket starting from %llu\n",
5269              (unsigned long long)bucket_blkno(target));
5270
5271         /* The first bucket of the original extent */
5272         first = ocfs2_xattr_bucket_new(inode);
5273         if (!first) {
5274                 ret = -ENOMEM;
5275                 mlog_errno(ret);
5276                 goto out;
5277         }
5278
5279         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5280                                   &num_clusters, el);
5281         if (ret) {
5282                 mlog_errno(ret);
5283                 goto out;
5284         }
5285
5286         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5287         if (ret) {
5288                 mlog_errno(ret);
5289                 goto out;
5290         }
5291
5292         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5293         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5294                 /*
5295                  * This can move first+target if the target bucket moves
5296                  * to the new extent.
5297                  */
5298                 ret = ocfs2_add_new_xattr_cluster(inode,
5299                                                   xb_bh,
5300                                                   first,
5301                                                   target,
5302                                                   &num_clusters,
5303                                                   e_cpos,
5304                                                   &extend,
5305                                                   ctxt);
5306                 if (ret) {
5307                         mlog_errno(ret);
5308                         goto out;
5309                 }
5310         }
5311
5312         if (extend) {
5313                 ret = ocfs2_extend_xattr_bucket(inode,
5314                                                 ctxt->handle,
5315                                                 first,
5316                                                 bucket_blkno(target),
5317                                                 num_clusters);
5318                 if (ret)
5319                         mlog_errno(ret);
5320         }
5321
5322 out:
5323         ocfs2_xattr_bucket_free(first);
5324
5325         return ret;
5326 }
5327
5328 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5329                                         struct ocfs2_xattr_bucket *bucket,
5330                                         int offs)
5331 {
5332         int block_off = offs >> inode->i_sb->s_blocksize_bits;
5333
5334         offs = offs % inode->i_sb->s_blocksize;
5335         return bucket_block(bucket, block_off) + offs;
5336 }
5337
5338 /*
5339  * Truncate the specified xe_off entry in xattr bucket.
5340  * bucket is indicated by header_bh and len is the new length.
5341  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5342  *
5343  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5344  */
5345 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5346                                              struct ocfs2_xattr_bucket *bucket,
5347                                              int xe_off,
5348                                              int len,
5349                                              struct ocfs2_xattr_set_ctxt *ctxt)
5350 {
5351         int ret, offset;
5352         u64 value_blk;
5353         struct ocfs2_xattr_entry *xe;
5354         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5355         size_t blocksize = inode->i_sb->s_blocksize;
5356         struct ocfs2_xattr_value_buf vb = {
5357                 .vb_access = ocfs2_journal_access,
5358         };
5359
5360         xe = &xh->xh_entries[xe_off];
5361
5362         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5363
5364         offset = le16_to_cpu(xe->xe_name_offset) +
5365                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5366
5367         value_blk = offset / blocksize;
5368
5369         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5370         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5371
5372         vb.vb_bh = bucket->bu_bhs[value_blk];
5373         BUG_ON(!vb.vb_bh);
5374
5375         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5376                 (vb.vb_bh->b_data + offset % blocksize);
5377
5378         /*
5379          * From here on out we have to dirty the bucket.  The generic
5380          * value calls only modify one of the bucket's bhs, but we need
5381          * to send the bucket at once.  So if they error, they *could* have
5382          * modified something.  We have to assume they did, and dirty
5383          * the whole bucket.  This leaves us in a consistent state.
5384          */
5385         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5386              xe_off, (unsigned long long)bucket_blkno(bucket), len);
5387         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5388         if (ret) {
5389                 mlog_errno(ret);
5390                 goto out;
5391         }
5392
5393         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5394                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5395         if (ret) {
5396                 mlog_errno(ret);
5397                 goto out;
5398         }
5399
5400         xe->xe_value_size = cpu_to_le64(len);
5401
5402         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5403
5404 out:
5405         return ret;
5406 }
5407
5408 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5409                                   struct buffer_head *root_bh,
5410                                   u64 blkno,
5411                                   u32 cpos,
5412                                   u32 len,
5413                                   void *para)
5414 {
5415         int ret;
5416         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5417         struct inode *tl_inode = osb->osb_tl_inode;
5418         handle_t *handle;
5419         struct ocfs2_xattr_block *xb =
5420                         (struct ocfs2_xattr_block *)root_bh->b_data;
5421         struct ocfs2_alloc_context *meta_ac = NULL;
5422         struct ocfs2_cached_dealloc_ctxt dealloc;
5423         struct ocfs2_extent_tree et;
5424
5425         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5426                                           ocfs2_delete_xattr_in_bucket, para);
5427         if (ret) {
5428                 mlog_errno(ret);
5429                 return ret;
5430         }
5431
5432         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5433
5434         ocfs2_init_dealloc_ctxt(&dealloc);
5435
5436         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5437              cpos, len, (unsigned long long)blkno);
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         mutex_lock(&tl_inode->i_mutex);
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
5486 out_commit:
5487         ocfs2_commit_trans(osb, handle);
5488 out:
5489         ocfs2_schedule_truncate_log_flush(osb, 1);
5490
5491         mutex_unlock(&tl_inode->i_mutex);
5492
5493         if (meta_ac)
5494                 ocfs2_free_alloc_context(meta_ac);
5495
5496         ocfs2_run_deallocs(osb, &dealloc);
5497
5498         return ret;
5499 }
5500
5501 /*
5502  * check whether the xattr bucket is filled up with the same hash value.
5503  * If we want to insert the xattr with the same hash, return -ENOSPC.
5504  * If we want to insert a xattr with different hash value, go ahead
5505  * and ocfs2_divide_xattr_bucket will handle this.
5506  */
5507 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5508                                               struct ocfs2_xattr_bucket *bucket,
5509                                               const char *name)
5510 {
5511         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5512         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5513
5514         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5515                 return 0;
5516
5517         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5518             xh->xh_entries[0].xe_name_hash) {
5519                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5520                      "hash = %u\n",
5521                      (unsigned long long)bucket_blkno(bucket),
5522                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5523                 return -ENOSPC;
5524         }
5525
5526         return 0;
5527 }
5528
5529 /*
5530  * Try to set the entry in the current bucket.  If we fail, the caller
5531  * will handle getting us another bucket.
5532  */
5533 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5534                                         struct ocfs2_xattr_info *xi,
5535                                         struct ocfs2_xattr_search *xs,
5536                                         struct ocfs2_xattr_set_ctxt *ctxt)
5537 {
5538         int ret;
5539         struct ocfs2_xa_loc loc;
5540
5541         mlog_entry("Set xattr %s in xattr bucket\n", xi->xi_name);
5542
5543         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5544                                        xs->not_found ? NULL : xs->here);
5545         ret = ocfs2_xa_set(&loc, xi, ctxt);
5546         if (!ret) {
5547                 xs->here = loc.xl_entry;
5548                 goto out;
5549         }
5550         if (ret != -ENOSPC) {
5551                 mlog_errno(ret);
5552                 goto out;
5553         }
5554
5555         /* Ok, we need space.  Let's try defragmenting the bucket. */
5556         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5557                                         xs->bucket);
5558         if (ret) {
5559                 mlog_errno(ret);
5560                 goto out;
5561         }
5562
5563         ret = ocfs2_xa_set(&loc, xi, ctxt);
5564         if (!ret) {
5565                 xs->here = loc.xl_entry;
5566                 goto out;
5567         }
5568         if (ret != -ENOSPC)
5569                 mlog_errno(ret);
5570
5571
5572 out:
5573         mlog_exit(ret);
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         mlog_entry("Set xattr %s in xattr index block\n", 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         mlog_exit(ret);
5641         return ret;
5642 }
5643
5644 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5645                                         struct ocfs2_xattr_bucket *bucket,
5646                                         void *para)
5647 {
5648         int ret = 0, ref_credits;
5649         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5650         u16 i;
5651         struct ocfs2_xattr_entry *xe;
5652         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5653         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5654         int credits = ocfs2_remove_extent_credits(osb->sb) +
5655                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5656         struct ocfs2_xattr_value_root *xv;
5657         struct ocfs2_rm_xattr_bucket_para *args =
5658                         (struct ocfs2_rm_xattr_bucket_para *)para;
5659
5660         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5661
5662         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5663                 xe = &xh->xh_entries[i];
5664                 if (ocfs2_xattr_is_local(xe))
5665                         continue;
5666
5667                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5668                                                       i, &xv, NULL);
5669
5670                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5671                                                          args->ref_ci,
5672                                                          args->ref_root_bh,
5673                                                          &ctxt.meta_ac,
5674                                                          &ref_credits);
5675
5676                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5677                 if (IS_ERR(ctxt.handle)) {
5678                         ret = PTR_ERR(ctxt.handle);
5679                         mlog_errno(ret);
5680                         break;
5681                 }
5682
5683                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5684                                                         i, 0, &ctxt);
5685
5686                 ocfs2_commit_trans(osb, ctxt.handle);
5687                 if (ctxt.meta_ac) {
5688                         ocfs2_free_alloc_context(ctxt.meta_ac);
5689                         ctxt.meta_ac = NULL;
5690                 }
5691                 if (ret) {
5692                         mlog_errno(ret);
5693                         break;
5694                 }
5695         }
5696
5697         if (ctxt.meta_ac)
5698                 ocfs2_free_alloc_context(ctxt.meta_ac);
5699         ocfs2_schedule_truncate_log_flush(osb, 1);
5700         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5701         return ret;
5702 }
5703
5704 /*
5705  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5706  * or change the extent record flag), we need to recalculate
5707  * the metaecc for the whole bucket. So it is done here.
5708  *
5709  * Note:
5710  * We have to give the extra credits for the caller.
5711  */
5712 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5713                                             handle_t *handle,
5714                                             void *para)
5715 {
5716         int ret;
5717         struct ocfs2_xattr_bucket *bucket =
5718                         (struct ocfs2_xattr_bucket *)para;
5719
5720         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5721                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5722         if (ret) {
5723                 mlog_errno(ret);
5724                 return ret;
5725         }
5726
5727         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5728
5729         return 0;
5730 }
5731
5732 /*
5733  * Special action we need if the xattr value is refcounted.
5734  *
5735  * 1. If the xattr is refcounted, lock the tree.
5736  * 2. CoW the xattr if we are setting the new value and the value
5737  *    will be stored outside.
5738  * 3. In other case, decrease_refcount will work for us, so just
5739  *    lock the refcount tree, calculate the meta and credits is OK.
5740  *
5741  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5742  * currently CoW is a completed transaction, while this function
5743  * will also lock the allocators and let us deadlock. So we will
5744  * CoW the whole xattr value.
5745  */
5746 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5747                                         struct ocfs2_dinode *di,
5748                                         struct ocfs2_xattr_info *xi,
5749                                         struct ocfs2_xattr_search *xis,
5750                                         struct ocfs2_xattr_search *xbs,
5751                                         struct ocfs2_refcount_tree **ref_tree,
5752                                         int *meta_add,
5753                                         int *credits)
5754 {
5755         int ret = 0;
5756         struct ocfs2_xattr_block *xb;
5757         struct ocfs2_xattr_entry *xe;
5758         char *base;
5759         u32 p_cluster, num_clusters;
5760         unsigned int ext_flags;
5761         int name_offset, name_len;
5762         struct ocfs2_xattr_value_buf vb;
5763         struct ocfs2_xattr_bucket *bucket = NULL;
5764         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5765         struct ocfs2_post_refcount refcount;
5766         struct ocfs2_post_refcount *p = NULL;
5767         struct buffer_head *ref_root_bh = NULL;
5768
5769         if (!xis->not_found) {
5770                 xe = xis->here;
5771                 name_offset = le16_to_cpu(xe->xe_name_offset);
5772                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5773                 base = xis->base;
5774                 vb.vb_bh = xis->inode_bh;
5775                 vb.vb_access = ocfs2_journal_access_di;
5776         } else {
5777                 int i, block_off = 0;
5778                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5779                 xe = xbs->here;
5780                 name_offset = le16_to_cpu(xe->xe_name_offset);
5781                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5782                 i = xbs->here - xbs->header->xh_entries;
5783
5784                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5785                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5786                                                         bucket_xh(xbs->bucket),
5787                                                         i, &block_off,
5788                                                         &name_offset);
5789                         if (ret) {
5790                                 mlog_errno(ret);
5791                                 goto out;
5792                         }
5793                         base = bucket_block(xbs->bucket, block_off);
5794                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5795                         vb.vb_access = ocfs2_journal_access;
5796
5797                         if (ocfs2_meta_ecc(osb)) {
5798                                 /*create parameters for ocfs2_post_refcount. */
5799                                 bucket = xbs->bucket;
5800                                 refcount.credits = bucket->bu_blocks;
5801                                 refcount.para = bucket;
5802                                 refcount.func =
5803                                         ocfs2_xattr_bucket_post_refcount;
5804                                 p = &refcount;
5805                         }
5806                 } else {
5807                         base = xbs->base;
5808                         vb.vb_bh = xbs->xattr_bh;
5809                         vb.vb_access = ocfs2_journal_access_xb;
5810                 }
5811         }
5812
5813         if (ocfs2_xattr_is_local(xe))
5814                 goto out;
5815
5816         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5817                                 (base + name_offset + name_len);
5818
5819         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5820                                        &num_clusters, &vb.vb_xv->xr_list,
5821                                        &ext_flags);
5822         if (ret) {
5823                 mlog_errno(ret);
5824                 goto out;
5825         }
5826
5827         /*
5828          * We just need to check the 1st extent record, since we always
5829          * CoW the whole xattr. So there shouldn't be a xattr with
5830          * some REFCOUNT extent recs after the 1st one.
5831          */
5832         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5833                 goto out;
5834
5835         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5836                                        1, ref_tree, &ref_root_bh);
5837         if (ret) {
5838                 mlog_errno(ret);
5839                 goto out;
5840         }
5841
5842         /*
5843          * If we are deleting the xattr or the new size will be stored inside,
5844          * cool, leave it there, the xattr truncate process will remove them
5845          * for us(it still needs the refcount tree lock and the meta, credits).
5846          * And the worse case is that every cluster truncate will split the
5847          * refcount tree, and make the original extent become 3. So we will need
5848          * 2 * cluster more extent recs at most.
5849          */
5850         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5851
5852                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5853                                                          &(*ref_tree)->rf_ci,
5854                                                          ref_root_bh, vb.vb_xv,
5855                                                          meta_add, credits);
5856                 if (ret)
5857                         mlog_errno(ret);
5858                 goto out;
5859         }
5860
5861         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5862                                        *ref_tree, ref_root_bh, 0,
5863                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5864         if (ret)
5865                 mlog_errno(ret);
5866
5867 out:
5868         brelse(ref_root_bh);
5869         return ret;
5870 }
5871
5872 /*
5873  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5874  * The physical clusters will be added to refcount tree.
5875  */
5876 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5877                                 struct ocfs2_xattr_value_root *xv,
5878                                 struct ocfs2_extent_tree *value_et,
5879                                 struct ocfs2_caching_info *ref_ci,
5880                                 struct buffer_head *ref_root_bh,
5881                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5882                                 struct ocfs2_post_refcount *refcount)
5883 {
5884         int ret = 0;
5885         u32 clusters = le32_to_cpu(xv->xr_clusters);
5886         u32 cpos, p_cluster, num_clusters;
5887         struct ocfs2_extent_list *el = &xv->xr_list;
5888         unsigned int ext_flags;
5889
5890         cpos = 0;
5891         while (cpos < clusters) {
5892                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5893                                                &num_clusters, el, &ext_flags);
5894
5895                 cpos += num_clusters;
5896                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5897                         continue;
5898
5899                 BUG_ON(!p_cluster);
5900
5901                 ret = ocfs2_add_refcount_flag(inode, value_et,
5902                                               ref_ci, ref_root_bh,
5903                                               cpos - num_clusters,
5904                                               p_cluster, num_clusters,
5905                                               dealloc, refcount);
5906                 if (ret) {
5907                         mlog_errno(ret);
5908                         break;
5909                 }
5910         }
5911
5912         return ret;
5913 }
5914
5915 /*
5916  * Given a normal ocfs2_xattr_header, refcount all the entries which
5917  * have value stored outside.
5918  * Used for xattrs stored in inode and ocfs2_xattr_block.
5919  */
5920 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5921                                 struct ocfs2_xattr_value_buf *vb,
5922                                 struct ocfs2_xattr_header *header,
5923                                 struct ocfs2_caching_info *ref_ci,
5924                                 struct buffer_head *ref_root_bh,
5925                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5926 {
5927
5928         struct ocfs2_xattr_entry *xe;
5929         struct ocfs2_xattr_value_root *xv;
5930         struct ocfs2_extent_tree et;
5931         int i, ret = 0;
5932
5933         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5934                 xe = &header->xh_entries[i];
5935
5936                 if (ocfs2_xattr_is_local(xe))
5937                         continue;
5938
5939                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5940                         le16_to_cpu(xe->xe_name_offset) +
5941                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5942
5943                 vb->vb_xv = xv;
5944                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5945
5946                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5947                                                         ref_ci, ref_root_bh,
5948                                                         dealloc, NULL);
5949                 if (ret) {
5950                         mlog_errno(ret);
5951                         break;
5952                 }
5953         }
5954
5955         return ret;
5956 }
5957
5958 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5959                                 struct buffer_head *fe_bh,
5960                                 struct ocfs2_caching_info *ref_ci,
5961                                 struct buffer_head *ref_root_bh,
5962                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5963 {
5964         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5965         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5966                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5967                                 le16_to_cpu(di->i_xattr_inline_size));
5968         struct ocfs2_xattr_value_buf vb = {
5969                 .vb_bh = fe_bh,
5970                 .vb_access = ocfs2_journal_access_di,
5971         };
5972
5973         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5974                                                   ref_ci, ref_root_bh, dealloc);
5975 }
5976
5977 struct ocfs2_xattr_tree_value_refcount_para {
5978         struct ocfs2_caching_info *ref_ci;
5979         struct buffer_head *ref_root_bh;
5980         struct ocfs2_cached_dealloc_ctxt *dealloc;
5981 };
5982
5983 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5984                                            struct ocfs2_xattr_bucket *bucket,
5985                                            int offset,
5986                                            struct ocfs2_xattr_value_root **xv,
5987                                            struct buffer_head **bh)
5988 {
5989         int ret, block_off, name_offset;
5990         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5991         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5992         void *base;
5993
5994         ret = ocfs2_xattr_bucket_get_name_value(sb,
5995                                                 bucket_xh(bucket),
5996                                                 offset,
5997                                                 &block_off,
5998                                                 &name_offset);
5999         if (ret) {
6000                 mlog_errno(ret);
6001                 goto out;
6002         }
6003
6004         base = bucket_block(bucket, block_off);
6005
6006         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6007                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6008
6009         if (bh)
6010                 *bh = bucket->bu_bhs[block_off];
6011 out:
6012         return ret;
6013 }
6014
6015 /*
6016  * For a given xattr bucket, refcount all the entries which
6017  * have value stored outside.
6018  */
6019 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6020                                              struct ocfs2_xattr_bucket *bucket,
6021                                              void *para)
6022 {
6023         int i, ret = 0;
6024         struct ocfs2_extent_tree et;
6025         struct ocfs2_xattr_tree_value_refcount_para *ref =
6026                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6027         struct ocfs2_xattr_header *xh =
6028                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6029         struct ocfs2_xattr_entry *xe;
6030         struct ocfs2_xattr_value_buf vb = {
6031                 .vb_access = ocfs2_journal_access,
6032         };
6033         struct ocfs2_post_refcount refcount = {
6034                 .credits = bucket->bu_blocks,
6035                 .para = bucket,
6036                 .func = ocfs2_xattr_bucket_post_refcount,
6037         };
6038         struct ocfs2_post_refcount *p = NULL;
6039
6040         /* We only need post_refcount if we support metaecc. */
6041         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6042                 p = &refcount;
6043
6044         mlog(0, "refcount bucket %llu, count = %u\n",
6045              (unsigned long long)bucket_blkno(bucket),
6046              le16_to_cpu(xh->xh_count));
6047         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6048                 xe = &xh->xh_entries[i];
6049
6050                 if (ocfs2_xattr_is_local(xe))
6051                         continue;
6052
6053                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6054                                                       &vb.vb_xv, &vb.vb_bh);
6055                 if (ret) {
6056                         mlog_errno(ret);
6057                         break;
6058                 }
6059
6060                 ocfs2_init_xattr_value_extent_tree(&et,
6061                                                    INODE_CACHE(inode), &vb);
6062
6063                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6064                                                         &et, ref->ref_ci,
6065                                                         ref->ref_root_bh,
6066                                                         ref->dealloc, p);
6067                 if (ret) {
6068                         mlog_errno(ret);
6069                         break;
6070                 }
6071         }
6072
6073         return ret;
6074
6075 }
6076
6077 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6078                                      struct buffer_head *root_bh,
6079                                      u64 blkno, u32 cpos, u32 len, void *para)
6080 {
6081         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6082                                            ocfs2_xattr_bucket_value_refcount,
6083                                            para);
6084 }
6085
6086 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6087                                 struct buffer_head *blk_bh,
6088                                 struct ocfs2_caching_info *ref_ci,
6089                                 struct buffer_head *ref_root_bh,
6090                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6091 {
6092         int ret = 0;
6093         struct ocfs2_xattr_block *xb =
6094                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6095
6096         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6097                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6098                 struct ocfs2_xattr_value_buf vb = {
6099                         .vb_bh = blk_bh,
6100                         .vb_access = ocfs2_journal_access_xb,
6101                 };
6102
6103                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6104                                                          ref_ci, ref_root_bh,
6105                                                          dealloc);
6106         } else {
6107                 struct ocfs2_xattr_tree_value_refcount_para para = {
6108                         .ref_ci = ref_ci,
6109                         .ref_root_bh = ref_root_bh,
6110                         .dealloc = dealloc,
6111                 };
6112
6113                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6114                                                 ocfs2_refcount_xattr_tree_rec,
6115                                                 &para);
6116         }
6117
6118         return ret;
6119 }
6120
6121 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6122                                      struct buffer_head *fe_bh,
6123                                      struct ocfs2_caching_info *ref_ci,
6124                                      struct buffer_head *ref_root_bh,
6125                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6126 {
6127         int ret = 0;
6128         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6129         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6130         struct buffer_head *blk_bh = NULL;
6131
6132         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6133                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6134                                                          ref_ci, ref_root_bh,
6135                                                          dealloc);
6136                 if (ret) {
6137                         mlog_errno(ret);
6138                         goto out;
6139                 }
6140         }
6141
6142         if (!di->i_xattr_loc)
6143                 goto out;
6144
6145         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6146                                      &blk_bh);
6147         if (ret < 0) {
6148                 mlog_errno(ret);
6149                 goto out;
6150         }
6151
6152         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6153                                                 ref_root_bh, dealloc);
6154         if (ret)
6155                 mlog_errno(ret);
6156
6157         brelse(blk_bh);
6158 out:
6159
6160         return ret;
6161 }
6162
6163 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6164 /*
6165  * Store the information we need in xattr reflink.
6166  * old_bh and new_bh are inode bh for the old and new inode.
6167  */
6168 struct ocfs2_xattr_reflink {
6169         struct inode *old_inode;
6170         struct inode *new_inode;
6171         struct buffer_head *old_bh;
6172         struct buffer_head *new_bh;
6173         struct ocfs2_caching_info *ref_ci;
6174         struct buffer_head *ref_root_bh;
6175         struct ocfs2_cached_dealloc_ctxt *dealloc;
6176         should_xattr_reflinked *xattr_reflinked;
6177 };
6178
6179 /*
6180  * Given a xattr header and xe offset,
6181  * return the proper xv and the corresponding bh.
6182  * xattr in inode, block and xattr tree have different implementaions.
6183  */
6184 typedef int (get_xattr_value_root)(struct super_block *sb,
6185                                    struct buffer_head *bh,
6186                                    struct ocfs2_xattr_header *xh,
6187                                    int offset,
6188                                    struct ocfs2_xattr_value_root **xv,
6189                                    struct buffer_head **ret_bh,
6190                                    void *para);
6191
6192 /*
6193  * Calculate all the xattr value root metadata stored in this xattr header and
6194  * credits we need if we create them from the scratch.
6195  * We use get_xattr_value_root so that all types of xattr container can use it.
6196  */
6197 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6198                                              struct buffer_head *bh,
6199                                              struct ocfs2_xattr_header *xh,
6200                                              int *metas, int *credits,
6201                                              int *num_recs,
6202                                              get_xattr_value_root *func,
6203                                              void *para)
6204 {
6205         int i, ret = 0;
6206         struct ocfs2_xattr_value_root *xv;
6207         struct ocfs2_xattr_entry *xe;
6208
6209         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6210                 xe = &xh->xh_entries[i];
6211                 if (ocfs2_xattr_is_local(xe))
6212                         continue;
6213
6214                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6215                 if (ret) {
6216                         mlog_errno(ret);
6217                         break;
6218                 }
6219
6220                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6221                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6222
6223                 *credits += ocfs2_calc_extend_credits(sb,
6224                                                 &def_xv.xv.xr_list,
6225                                                 le32_to_cpu(xv->xr_clusters));
6226
6227                 /*
6228                  * If the value is a tree with depth > 1, We don't go deep
6229                  * to the extent block, so just calculate a maximum record num.
6230                  */
6231                 if (!xv->xr_list.l_tree_depth)
6232                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6233                 else
6234                         *num_recs += ocfs2_clusters_for_bytes(sb,
6235                                                               XATTR_SIZE_MAX);
6236         }
6237
6238         return ret;
6239 }
6240
6241 /* Used by xattr inode and block to return the right xv and buffer_head. */
6242 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6243                                       struct buffer_head *bh,
6244                                       struct ocfs2_xattr_header *xh,
6245                                       int offset,
6246                                       struct ocfs2_xattr_value_root **xv,
6247                                       struct buffer_head **ret_bh,
6248                                       void *para)
6249 {
6250         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6251
6252         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6253                 le16_to_cpu(xe->xe_name_offset) +
6254                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6255
6256         if (ret_bh)
6257                 *ret_bh = bh;
6258
6259         return 0;
6260 }
6261
6262 /*
6263  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6264  * It is only used for inline xattr and xattr block.
6265  */
6266 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6267                                         struct ocfs2_xattr_header *xh,
6268                                         struct buffer_head *ref_root_bh,
6269                                         int *credits,
6270                                         struct ocfs2_alloc_context **meta_ac)
6271 {
6272         int ret, meta_add = 0, num_recs = 0;
6273         struct ocfs2_refcount_block *rb =
6274                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6275
6276         *credits = 0;
6277
6278         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6279                                                 &meta_add, credits, &num_recs,
6280                                                 ocfs2_get_xattr_value_root,
6281                                                 NULL);
6282         if (ret) {
6283                 mlog_errno(ret);
6284                 goto out;
6285         }
6286
6287         /*
6288          * We need to add/modify num_recs in refcount tree, so just calculate
6289          * an approximate number we need for refcount tree change.
6290          * Sometimes we need to split the tree, and after split,  half recs
6291          * will be moved to the new block, and a new block can only provide
6292          * half number of recs. So we multiple new blocks by 2.
6293          */
6294         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6295         meta_add += num_recs;
6296         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6297         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6298                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6299                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6300         else
6301                 *credits += 1;
6302
6303         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6304         if (ret)
6305                 mlog_errno(ret);
6306
6307 out:
6308         return ret;
6309 }
6310
6311 /*
6312  * Given a xattr header, reflink all the xattrs in this container.
6313  * It can be used for inode, block and bucket.
6314  *
6315  * NOTE:
6316  * Before we call this function, the caller has memcpy the xattr in
6317  * old_xh to the new_xh.
6318  *
6319  * If args.xattr_reflinked is set, call it to decide whether the xe should
6320  * be reflinked or not. If not, remove it from the new xattr header.
6321  */
6322 static int ocfs2_reflink_xattr_header(handle_t *handle,
6323                                       struct ocfs2_xattr_reflink *args,
6324                                       struct buffer_head *old_bh,
6325                                       struct ocfs2_xattr_header *xh,
6326                                       struct buffer_head *new_bh,
6327                                       struct ocfs2_xattr_header *new_xh,
6328                                       struct ocfs2_xattr_value_buf *vb,
6329                                       struct ocfs2_alloc_context *meta_ac,
6330                                       get_xattr_value_root *func,
6331                                       void *para)
6332 {
6333         int ret = 0, i, j;
6334         struct super_block *sb = args->old_inode->i_sb;
6335         struct buffer_head *value_bh;
6336         struct ocfs2_xattr_entry *xe, *last;
6337         struct ocfs2_xattr_value_root *xv, *new_xv;
6338         struct ocfs2_extent_tree data_et;
6339         u32 clusters, cpos, p_cluster, num_clusters;
6340         unsigned int ext_flags = 0;
6341
6342         mlog(0, "reflink xattr in container %llu, count = %u\n",
6343              (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
6344
6345         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6346         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6347                 xe = &xh->xh_entries[i];
6348
6349                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6350                         xe = &new_xh->xh_entries[j];
6351
6352                         le16_add_cpu(&new_xh->xh_count, -1);
6353                         if (new_xh->xh_count) {
6354                                 memmove(xe, xe + 1,
6355                                         (void *)last - (void *)xe);
6356                                 memset(last, 0,
6357                                        sizeof(struct ocfs2_xattr_entry));
6358                         }
6359
6360                         /*
6361                          * We don't want j to increase in the next round since
6362                          * it is already moved ahead.
6363                          */
6364                         j--;
6365                         continue;
6366                 }
6367
6368                 if (ocfs2_xattr_is_local(xe))
6369                         continue;
6370
6371                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6372                 if (ret) {
6373                         mlog_errno(ret);
6374                         break;
6375                 }
6376
6377                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6378                 if (ret) {
6379                         mlog_errno(ret);
6380                         break;
6381                 }
6382
6383                 /*
6384                  * For the xattr which has l_tree_depth = 0, all the extent
6385                  * recs have already be copied to the new xh with the
6386                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6387                  * increase the refount count int the refcount tree.
6388                  *
6389                  * For the xattr which has l_tree_depth > 0, we need
6390                  * to initialize it to the empty default value root,
6391                  * and then insert the extents one by one.
6392                  */
6393                 if (xv->xr_list.l_tree_depth) {
6394                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6395                         vb->vb_xv = new_xv;
6396                         vb->vb_bh = value_bh;
6397                         ocfs2_init_xattr_value_extent_tree(&data_et,
6398                                         INODE_CACHE(args->new_inode), vb);
6399                 }
6400
6401                 clusters = le32_to_cpu(xv->xr_clusters);
6402                 cpos = 0;
6403                 while (cpos < clusters) {
6404                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6405                                                        cpos,
6406                                                        &p_cluster,
6407                                                        &num_clusters,
6408                                                        &xv->xr_list,
6409                                                        &ext_flags);
6410                         if (ret) {
6411                                 mlog_errno(ret);
6412                                 goto out;
6413                         }
6414
6415                         BUG_ON(!p_cluster);
6416
6417                         if (xv->xr_list.l_tree_depth) {
6418                                 ret = ocfs2_insert_extent(handle,
6419                                                 &data_et, cpos,
6420                                                 ocfs2_clusters_to_blocks(
6421                                                         args->old_inode->i_sb,
6422                                                         p_cluster),
6423                                                 num_clusters, ext_flags,
6424                                                 meta_ac);
6425                                 if (ret) {
6426                                         mlog_errno(ret);
6427                                         goto out;
6428                                 }
6429                         }
6430
6431                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6432                                                       args->ref_root_bh,
6433                                                       p_cluster, num_clusters,
6434                                                       meta_ac, args->dealloc);
6435                         if (ret) {
6436                                 mlog_errno(ret);
6437                                 goto out;
6438                         }
6439
6440                         cpos += num_clusters;
6441                 }
6442         }
6443
6444 out:
6445         return ret;
6446 }
6447
6448 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6449 {
6450         int ret = 0, credits = 0;
6451         handle_t *handle;
6452         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6453         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6454         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6455         int header_off = osb->sb->s_blocksize - inline_size;
6456         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6457                                         (args->old_bh->b_data + header_off);
6458         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6459                                         (args->new_bh->b_data + header_off);
6460         struct ocfs2_alloc_context *meta_ac = NULL;
6461         struct ocfs2_inode_info *new_oi;
6462         struct ocfs2_dinode *new_di;
6463         struct ocfs2_xattr_value_buf vb = {
6464                 .vb_bh = args->new_bh,
6465                 .vb_access = ocfs2_journal_access_di,
6466         };
6467
6468         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6469                                                   &credits, &meta_ac);
6470         if (ret) {
6471                 mlog_errno(ret);
6472                 goto out;
6473         }
6474
6475         handle = ocfs2_start_trans(osb, credits);
6476         if (IS_ERR(handle)) {
6477                 ret = PTR_ERR(handle);
6478                 mlog_errno(ret);
6479                 goto out;
6480         }
6481
6482         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6483                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6484         if (ret) {
6485                 mlog_errno(ret);
6486                 goto out_commit;
6487         }
6488
6489         memcpy(args->new_bh->b_data + header_off,
6490                args->old_bh->b_data + header_off, inline_size);
6491
6492         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6493         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6494
6495         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6496                                          args->new_bh, new_xh, &vb, meta_ac,
6497                                          ocfs2_get_xattr_value_root, NULL);
6498         if (ret) {
6499                 mlog_errno(ret);
6500                 goto out_commit;
6501         }
6502
6503         new_oi = OCFS2_I(args->new_inode);
6504         spin_lock(&new_oi->ip_lock);
6505         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6506         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6507         spin_unlock(&new_oi->ip_lock);
6508
6509         ocfs2_journal_dirty(handle, args->new_bh);
6510
6511 out_commit:
6512         ocfs2_commit_trans(osb, handle);
6513
6514 out:
6515         if (meta_ac)
6516                 ocfs2_free_alloc_context(meta_ac);
6517         return ret;
6518 }
6519
6520 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6521                                           struct buffer_head *fe_bh,
6522                                           struct buffer_head **ret_bh,
6523                                           int indexed)
6524 {
6525         int ret;
6526         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6527         struct ocfs2_xattr_set_ctxt ctxt;
6528
6529         memset(&ctxt, 0, sizeof(ctxt));
6530         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6531         if (ret < 0) {
6532                 mlog_errno(ret);
6533                 return ret;
6534         }
6535
6536         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6537         if (IS_ERR(ctxt.handle)) {
6538                 ret = PTR_ERR(ctxt.handle);
6539                 mlog_errno(ret);
6540                 goto out;
6541         }
6542
6543         mlog(0, "create new xattr block for inode %llu, index = %d\n",
6544              (unsigned long long)fe_bh->b_blocknr, indexed);
6545         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6546                                        ret_bh);
6547         if (ret)
6548                 mlog_errno(ret);
6549
6550         ocfs2_commit_trans(osb, ctxt.handle);
6551 out:
6552         ocfs2_free_alloc_context(ctxt.meta_ac);
6553         return ret;
6554 }
6555
6556 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6557                                      struct buffer_head *blk_bh,
6558                                      struct buffer_head *new_blk_bh)
6559 {
6560         int ret = 0, credits = 0;
6561         handle_t *handle;
6562         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6563         struct ocfs2_dinode *new_di;
6564         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6565         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6566         struct ocfs2_xattr_block *xb =
6567                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6568         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6569         struct ocfs2_xattr_block *new_xb =
6570                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6571         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6572         struct ocfs2_alloc_context *meta_ac;
6573         struct ocfs2_xattr_value_buf vb = {
6574                 .vb_bh = new_blk_bh,
6575                 .vb_access = ocfs2_journal_access_xb,
6576         };
6577
6578         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6579                                                   &credits, &meta_ac);
6580         if (ret) {
6581                 mlog_errno(ret);
6582                 return ret;
6583         }
6584
6585         /* One more credits in case we need to add xattr flags in new inode. */
6586         handle = ocfs2_start_trans(osb, credits + 1);
6587         if (IS_ERR(handle)) {
6588                 ret = PTR_ERR(handle);
6589                 mlog_errno(ret);
6590                 goto out;
6591         }
6592
6593         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6594                 ret = ocfs2_journal_access_di(handle,
6595                                               INODE_CACHE(args->new_inode),
6596                                               args->new_bh,
6597                                               OCFS2_JOURNAL_ACCESS_WRITE);
6598                 if (ret) {
6599                         mlog_errno(ret);
6600                         goto out_commit;
6601                 }
6602         }
6603
6604         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6605                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6606         if (ret) {
6607                 mlog_errno(ret);
6608                 goto out_commit;
6609         }
6610
6611         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6612                osb->sb->s_blocksize - header_off);
6613
6614         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6615                                          new_blk_bh, new_xh, &vb, meta_ac,
6616                                          ocfs2_get_xattr_value_root, NULL);
6617         if (ret) {
6618                 mlog_errno(ret);
6619                 goto out_commit;
6620         }
6621
6622         ocfs2_journal_dirty(handle, new_blk_bh);
6623
6624         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6625                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6626                 spin_lock(&new_oi->ip_lock);
6627                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6628                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6629                 spin_unlock(&new_oi->ip_lock);
6630
6631                 ocfs2_journal_dirty(handle, args->new_bh);
6632         }
6633
6634 out_commit:
6635         ocfs2_commit_trans(osb, handle);
6636
6637 out:
6638         ocfs2_free_alloc_context(meta_ac);
6639         return ret;
6640 }
6641
6642 struct ocfs2_reflink_xattr_tree_args {
6643         struct ocfs2_xattr_reflink *reflink;
6644         struct buffer_head *old_blk_bh;
6645         struct buffer_head *new_blk_bh;
6646         struct ocfs2_xattr_bucket *old_bucket;
6647         struct ocfs2_xattr_bucket *new_bucket;
6648 };
6649
6650 /*
6651  * NOTE:
6652  * We have to handle the case that both old bucket and new bucket
6653  * will call this function to get the right ret_bh.
6654  * So The caller must give us the right bh.
6655  */
6656 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6657                                         struct buffer_head *bh,
6658                                         struct ocfs2_xattr_header *xh,
6659                                         int offset,
6660                                         struct ocfs2_xattr_value_root **xv,
6661                                         struct buffer_head **ret_bh,
6662                                         void *para)
6663 {
6664         struct ocfs2_reflink_xattr_tree_args *args =
6665                         (struct ocfs2_reflink_xattr_tree_args *)para;
6666         struct ocfs2_xattr_bucket *bucket;
6667
6668         if (bh == args->old_bucket->bu_bhs[0])
6669                 bucket = args->old_bucket;
6670         else
6671                 bucket = args->new_bucket;
6672
6673         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6674                                                xv, ret_bh);
6675 }
6676
6677 struct ocfs2_value_tree_metas {
6678         int num_metas;
6679         int credits;
6680         int num_recs;
6681 };
6682
6683 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6684                                         struct buffer_head *bh,
6685                                         struct ocfs2_xattr_header *xh,
6686                                         int offset,
6687                                         struct ocfs2_xattr_value_root **xv,
6688                                         struct buffer_head **ret_bh,
6689                                         void *para)
6690 {
6691         struct ocfs2_xattr_bucket *bucket =
6692                                 (struct ocfs2_xattr_bucket *)para;
6693
6694         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6695                                                xv, ret_bh);
6696 }
6697
6698 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6699                                       struct ocfs2_xattr_bucket *bucket,
6700                                       void *para)
6701 {
6702         struct ocfs2_value_tree_metas *metas =
6703                         (struct ocfs2_value_tree_metas *)para;
6704         struct ocfs2_xattr_header *xh =
6705                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6706
6707         /* Add the credits for this bucket first. */
6708         metas->credits += bucket->bu_blocks;
6709         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6710                                         xh, &metas->num_metas,
6711                                         &metas->credits, &metas->num_recs,
6712                                         ocfs2_value_tree_metas_in_bucket,
6713                                         bucket);
6714 }
6715
6716 /*
6717  * Given a xattr extent rec starting from blkno and having len clusters,
6718  * iterate all the buckets calculate how much metadata we need for reflinking
6719  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6720  */
6721 static int ocfs2_lock_reflink_xattr_rec_allocators(
6722                                 struct ocfs2_reflink_xattr_tree_args *args,
6723                                 struct ocfs2_extent_tree *xt_et,
6724                                 u64 blkno, u32 len, int *credits,
6725                                 struct ocfs2_alloc_context **meta_ac,
6726                                 struct ocfs2_alloc_context **data_ac)
6727 {
6728         int ret, num_free_extents;
6729         struct ocfs2_value_tree_metas metas;
6730         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6731         struct ocfs2_refcount_block *rb;
6732
6733         memset(&metas, 0, sizeof(metas));
6734
6735         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6736                                           ocfs2_calc_value_tree_metas, &metas);
6737         if (ret) {
6738                 mlog_errno(ret);
6739                 goto out;
6740         }
6741
6742         *credits = metas.credits;
6743
6744         /*
6745          * Calculate we need for refcount tree change.
6746          *
6747          * We need to add/modify num_recs in refcount tree, so just calculate
6748          * an approximate number we need for refcount tree change.
6749          * Sometimes we need to split the tree, and after split,  half recs
6750          * will be moved to the new block, and a new block can only provide
6751          * half number of recs. So we multiple new blocks by 2.
6752          * In the end, we have to add credits for modifying the already
6753          * existed refcount block.
6754          */
6755         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6756         metas.num_recs =
6757                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6758                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6759         metas.num_metas += metas.num_recs;
6760         *credits += metas.num_recs +
6761                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6762         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6763                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6764                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6765         else
6766                 *credits += 1;
6767
6768         /* count in the xattr tree change. */
6769         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6770         if (num_free_extents < 0) {
6771                 ret = num_free_extents;
6772                 mlog_errno(ret);
6773                 goto out;
6774         }
6775
6776         if (num_free_extents < len)
6777                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6778
6779         *credits += ocfs2_calc_extend_credits(osb->sb,
6780                                               xt_et->et_root_el, len);
6781
6782         if (metas.num_metas) {
6783                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6784                                                         meta_ac);
6785                 if (ret) {
6786                         mlog_errno(ret);
6787                         goto out;
6788                 }
6789         }
6790
6791         if (len) {
6792                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6793                 if (ret)
6794                         mlog_errno(ret);
6795         }
6796 out:
6797         if (ret) {
6798                 if (*meta_ac) {
6799                         ocfs2_free_alloc_context(*meta_ac);
6800                         meta_ac = NULL;
6801                 }
6802         }
6803
6804         return ret;
6805 }
6806
6807 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6808                                 u64 blkno, u64 new_blkno, u32 clusters,
6809                                 u32 *cpos, int num_buckets,
6810                                 struct ocfs2_alloc_context *meta_ac,
6811                                 struct ocfs2_alloc_context *data_ac,
6812                                 struct ocfs2_reflink_xattr_tree_args *args)
6813 {
6814         int i, j, ret = 0;
6815         struct super_block *sb = args->reflink->old_inode->i_sb;
6816         int bpb = args->old_bucket->bu_blocks;
6817         struct ocfs2_xattr_value_buf vb = {
6818                 .vb_access = ocfs2_journal_access,
6819         };
6820
6821         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6822                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6823                 if (ret) {
6824                         mlog_errno(ret);
6825                         break;
6826                 }
6827
6828                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6829                 if (ret) {
6830                         mlog_errno(ret);
6831                         break;
6832                 }
6833
6834                 ret = ocfs2_xattr_bucket_journal_access(handle,
6835                                                 args->new_bucket,
6836                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6837                 if (ret) {
6838                         mlog_errno(ret);
6839                         break;
6840                 }
6841
6842                 for (j = 0; j < bpb; j++)
6843                         memcpy(bucket_block(args->new_bucket, j),
6844                                bucket_block(args->old_bucket, j),
6845                                sb->s_blocksize);
6846
6847                 /*
6848                  * Record the start cpos so that we can use it to initialize
6849                  * our xattr tree we also set the xh_num_bucket for the new
6850                  * bucket.
6851                  */
6852                 if (i == 0) {
6853                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6854                                             xh_entries[0].xe_name_hash);
6855                         bucket_xh(args->new_bucket)->xh_num_buckets =
6856                                 cpu_to_le16(num_buckets);
6857                 }
6858
6859                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6860
6861                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6862                                         args->old_bucket->bu_bhs[0],
6863                                         bucket_xh(args->old_bucket),
6864                                         args->new_bucket->bu_bhs[0],
6865                                         bucket_xh(args->new_bucket),
6866                                         &vb, meta_ac,
6867                                         ocfs2_get_reflink_xattr_value_root,
6868                                         args);
6869                 if (ret) {
6870                         mlog_errno(ret);
6871                         break;
6872                 }
6873
6874                 /*
6875                  * Re-access and dirty the bucket to calculate metaecc.
6876                  * Because we may extend the transaction in reflink_xattr_header
6877                  * which will let the already accessed block gone.
6878                  */
6879                 ret = ocfs2_xattr_bucket_journal_access(handle,
6880                                                 args->new_bucket,
6881                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6882                 if (ret) {
6883                         mlog_errno(ret);
6884                         break;
6885                 }
6886
6887                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6888
6889                 ocfs2_xattr_bucket_relse(args->old_bucket);
6890                 ocfs2_xattr_bucket_relse(args->new_bucket);
6891         }
6892
6893         ocfs2_xattr_bucket_relse(args->old_bucket);
6894         ocfs2_xattr_bucket_relse(args->new_bucket);
6895         return ret;
6896 }
6897
6898 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6899                                 struct inode *inode,
6900                                 struct ocfs2_reflink_xattr_tree_args *args,
6901                                 struct ocfs2_extent_tree *et,
6902                                 struct ocfs2_alloc_context *meta_ac,
6903                                 struct ocfs2_alloc_context *data_ac,
6904                                 u64 blkno, u32 cpos, u32 len)
6905 {
6906         int ret, first_inserted = 0;
6907         u32 p_cluster, num_clusters, reflink_cpos = 0;
6908         u64 new_blkno;
6909         unsigned int num_buckets, reflink_buckets;
6910         unsigned int bpc =
6911                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6912
6913         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6914         if (ret) {
6915                 mlog_errno(ret);
6916                 goto out;
6917         }
6918         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6919         ocfs2_xattr_bucket_relse(args->old_bucket);
6920
6921         while (len && num_buckets) {
6922                 ret = ocfs2_claim_clusters(handle, data_ac,
6923                                            1, &p_cluster, &num_clusters);
6924                 if (ret) {
6925                         mlog_errno(ret);
6926                         goto out;
6927                 }
6928
6929                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6930                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6931
6932                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6933                                                  new_blkno, num_clusters,
6934                                                  &reflink_cpos, reflink_buckets,
6935                                                  meta_ac, data_ac, args);
6936                 if (ret) {
6937                         mlog_errno(ret);
6938                         goto out;
6939                 }
6940
6941                 /*
6942                  * For the 1st allocated cluster, we make it use the same cpos
6943                  * so that the xattr tree looks the same as the original one
6944                  * in the most case.
6945                  */
6946                 if (!first_inserted) {
6947                         reflink_cpos = cpos;
6948                         first_inserted = 1;
6949                 }
6950                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6951                                           num_clusters, 0, meta_ac);
6952                 if (ret)
6953                         mlog_errno(ret);
6954
6955                 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6956                      (unsigned long long)new_blkno, num_clusters, reflink_cpos);
6957
6958                 len -= num_clusters;
6959                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6960                 num_buckets -= reflink_buckets;
6961         }
6962 out:
6963         return ret;
6964 }
6965
6966 /*
6967  * Create the same xattr extent record in the new inode's xattr tree.
6968  */
6969 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6970                                    struct buffer_head *root_bh,
6971                                    u64 blkno,
6972                                    u32 cpos,
6973                                    u32 len,
6974                                    void *para)
6975 {
6976         int ret, credits = 0;
6977         handle_t *handle;
6978         struct ocfs2_reflink_xattr_tree_args *args =
6979                         (struct ocfs2_reflink_xattr_tree_args *)para;
6980         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6981         struct ocfs2_alloc_context *meta_ac = NULL;
6982         struct ocfs2_alloc_context *data_ac = NULL;
6983         struct ocfs2_extent_tree et;
6984
6985         mlog(0, "reflink xattr buckets %llu len %u\n",
6986              (unsigned long long)blkno, len);
6987
6988         ocfs2_init_xattr_tree_extent_tree(&et,
6989                                           INODE_CACHE(args->reflink->new_inode),
6990                                           args->new_blk_bh);
6991
6992         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6993                                                       len, &credits,
6994                                                       &meta_ac, &data_ac);
6995         if (ret) {
6996                 mlog_errno(ret);
6997                 goto out;
6998         }
6999
7000         handle = ocfs2_start_trans(osb, credits);
7001         if (IS_ERR(handle)) {
7002                 ret = PTR_ERR(handle);
7003                 mlog_errno(ret);
7004                 goto out;
7005         }
7006
7007         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7008                                           meta_ac, data_ac,
7009                                           blkno, cpos, len);
7010         if (ret)
7011                 mlog_errno(ret);
7012
7013         ocfs2_commit_trans(osb, handle);
7014
7015 out:
7016         if (meta_ac)
7017                 ocfs2_free_alloc_context(meta_ac);
7018         if (data_ac)
7019                 ocfs2_free_alloc_context(data_ac);
7020         return ret;
7021 }
7022
7023 /*
7024  * Create reflinked xattr buckets.
7025  * We will add bucket one by one, and refcount all the xattrs in the bucket
7026  * if they are stored outside.
7027  */
7028 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7029                                     struct buffer_head *blk_bh,
7030                                     struct buffer_head *new_blk_bh)
7031 {
7032         int ret;
7033         struct ocfs2_reflink_xattr_tree_args para;
7034
7035         memset(&para, 0, sizeof(para));
7036         para.reflink = args;
7037         para.old_blk_bh = blk_bh;
7038         para.new_blk_bh = new_blk_bh;
7039
7040         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7041         if (!para.old_bucket) {
7042                 mlog_errno(-ENOMEM);
7043                 return -ENOMEM;
7044         }
7045
7046         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7047         if (!para.new_bucket) {
7048                 ret = -ENOMEM;
7049                 mlog_errno(ret);
7050                 goto out;
7051         }
7052
7053         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7054                                               ocfs2_reflink_xattr_rec,
7055                                               &para);
7056         if (ret)
7057                 mlog_errno(ret);
7058
7059 out:
7060         ocfs2_xattr_bucket_free(para.old_bucket);
7061         ocfs2_xattr_bucket_free(para.new_bucket);
7062         return ret;
7063 }
7064
7065 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7066                                         struct buffer_head *blk_bh)
7067 {
7068         int ret, indexed = 0;
7069         struct buffer_head *new_blk_bh = NULL;
7070         struct ocfs2_xattr_block *xb =
7071                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7072
7073
7074         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7075                 indexed = 1;
7076
7077         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7078                                              &new_blk_bh, indexed);
7079         if (ret) {
7080                 mlog_errno(ret);
7081                 goto out;
7082         }
7083
7084         if (!indexed)
7085                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7086         else
7087                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7088         if (ret)
7089                 mlog_errno(ret);
7090
7091 out:
7092         brelse(new_blk_bh);
7093         return ret;
7094 }
7095
7096 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7097 {
7098         int type = ocfs2_xattr_get_type(xe);
7099
7100         return type != OCFS2_XATTR_INDEX_SECURITY &&
7101                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7102                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7103 }
7104
7105 int ocfs2_reflink_xattrs(struct inode *old_inode,
7106                          struct buffer_head *old_bh,
7107                          struct inode *new_inode,
7108                          struct buffer_head *new_bh,
7109                          bool preserve_security)
7110 {
7111         int ret;
7112         struct ocfs2_xattr_reflink args;
7113         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7114         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7115         struct buffer_head *blk_bh = NULL;
7116         struct ocfs2_cached_dealloc_ctxt dealloc;
7117         struct ocfs2_refcount_tree *ref_tree;
7118         struct buffer_head *ref_root_bh = NULL;
7119
7120         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7121                                        le64_to_cpu(di->i_refcount_loc),
7122                                        1, &ref_tree, &ref_root_bh);
7123         if (ret) {
7124                 mlog_errno(ret);
7125                 goto out;
7126         }
7127
7128         ocfs2_init_dealloc_ctxt(&dealloc);
7129
7130         args.old_inode = old_inode;
7131         args.new_inode = new_inode;
7132         args.old_bh = old_bh;
7133         args.new_bh = new_bh;
7134         args.ref_ci = &ref_tree->rf_ci;
7135         args.ref_root_bh = ref_root_bh;
7136         args.dealloc = &dealloc;
7137         if (preserve_security)
7138                 args.xattr_reflinked = NULL;
7139         else
7140                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7141
7142         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7143                 ret = ocfs2_reflink_xattr_inline(&args);
7144                 if (ret) {
7145                         mlog_errno(ret);
7146                         goto out_unlock;
7147                 }
7148         }
7149
7150         if (!di->i_xattr_loc)
7151                 goto out_unlock;
7152
7153         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7154                                      &blk_bh);
7155         if (ret < 0) {
7156                 mlog_errno(ret);
7157                 goto out_unlock;
7158         }
7159
7160         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7161         if (ret)
7162                 mlog_errno(ret);
7163
7164         brelse(blk_bh);
7165
7166 out_unlock:
7167         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7168                                    ref_tree, 1);
7169         brelse(ref_root_bh);
7170
7171         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7172                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7173                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7174         }
7175
7176 out:
7177         return ret;
7178 }
7179
7180 /*
7181  * Initialize security and acl for a already created inode.
7182  * Used for reflink a non-preserve-security file.
7183  *
7184  * It uses common api like ocfs2_xattr_set, so the caller
7185  * must not hold any lock expect i_mutex.
7186  */
7187 int ocfs2_init_security_and_acl(struct inode *dir,
7188                                 struct inode *inode)
7189 {
7190         int ret = 0;
7191         struct buffer_head *dir_bh = NULL;
7192         struct ocfs2_security_xattr_info si = {
7193                 .enable = 1,
7194         };
7195
7196         ret = ocfs2_init_security_get(inode, dir, &si);
7197         if (!ret) {
7198                 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7199                                       si.name, si.value, si.value_len,
7200                                       XATTR_CREATE);
7201                 if (ret) {
7202                         mlog_errno(ret);
7203                         goto leave;
7204                 }
7205         } else if (ret != -EOPNOTSUPP) {
7206                 mlog_errno(ret);
7207                 goto leave;
7208         }
7209
7210         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7211         if (ret) {
7212                 mlog_errno(ret);
7213                 goto leave;
7214         }
7215
7216         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7217         if (ret)
7218                 mlog_errno(ret);
7219
7220         ocfs2_inode_unlock(dir, 0);
7221         brelse(dir_bh);
7222 leave:
7223         return ret;
7224 }
7225 /*
7226  * 'security' attributes support
7227  */
7228 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7229                                         size_t list_size, const char *name,
7230                                         size_t name_len, int type)
7231 {
7232         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7233         const size_t total_len = prefix_len + name_len + 1;
7234
7235         if (list && total_len <= list_size) {
7236                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7237                 memcpy(list + prefix_len, name, name_len);
7238                 list[prefix_len + name_len] = '\0';
7239         }
7240         return total_len;
7241 }
7242
7243 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7244                                     void *buffer, size_t size, int type)
7245 {
7246         if (strcmp(name, "") == 0)
7247                 return -EINVAL;
7248         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7249                                name, buffer, size);
7250 }
7251
7252 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7253                 const void *value, size_t size, int flags, int type)
7254 {
7255         if (strcmp(name, "") == 0)
7256                 return -EINVAL;
7257
7258         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7259                                name, value, size, flags);
7260 }
7261
7262 int ocfs2_init_security_get(struct inode *inode,
7263                             struct inode *dir,
7264                             struct ocfs2_security_xattr_info *si)
7265 {
7266         /* check whether ocfs2 support feature xattr */
7267         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7268                 return -EOPNOTSUPP;
7269         return security_inode_init_security(inode, dir, &si->name, &si->value,
7270                                             &si->value_len);
7271 }
7272
7273 int ocfs2_init_security_set(handle_t *handle,
7274                             struct inode *inode,
7275                             struct buffer_head *di_bh,
7276                             struct ocfs2_security_xattr_info *si,
7277                             struct ocfs2_alloc_context *xattr_ac,
7278                             struct ocfs2_alloc_context *data_ac)
7279 {
7280         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7281                                      OCFS2_XATTR_INDEX_SECURITY,
7282                                      si->name, si->value, si->value_len, 0,
7283                                      xattr_ac, data_ac);
7284 }
7285
7286 const struct xattr_handler ocfs2_xattr_security_handler = {
7287         .prefix = XATTR_SECURITY_PREFIX,
7288         .list   = ocfs2_xattr_security_list,
7289         .get    = ocfs2_xattr_security_get,
7290         .set    = ocfs2_xattr_security_set,
7291 };
7292
7293 /*
7294  * 'trusted' attributes support
7295  */
7296 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7297                                        size_t list_size, const char *name,
7298                                        size_t name_len, int type)
7299 {
7300         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7301         const size_t total_len = prefix_len + name_len + 1;
7302
7303         if (list && total_len <= list_size) {
7304                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7305                 memcpy(list + prefix_len, name, name_len);
7306                 list[prefix_len + name_len] = '\0';
7307         }
7308         return total_len;
7309 }
7310
7311 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7312                 void *buffer, size_t size, int type)
7313 {
7314         if (strcmp(name, "") == 0)
7315                 return -EINVAL;
7316         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7317                                name, buffer, size);
7318 }
7319
7320 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7321                 const void *value, size_t size, int flags, int type)
7322 {
7323         if (strcmp(name, "") == 0)
7324                 return -EINVAL;
7325
7326         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7327                                name, value, size, flags);
7328 }
7329
7330 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7331         .prefix = XATTR_TRUSTED_PREFIX,
7332         .list   = ocfs2_xattr_trusted_list,
7333         .get    = ocfs2_xattr_trusted_get,
7334         .set    = ocfs2_xattr_trusted_set,
7335 };
7336
7337 /*
7338  * 'user' attributes support
7339  */
7340 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7341                                     size_t list_size, const char *name,
7342                                     size_t name_len, int type)
7343 {
7344         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7345         const size_t total_len = prefix_len + name_len + 1;
7346         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7347
7348         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7349                 return 0;
7350
7351         if (list && total_len <= list_size) {
7352                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7353                 memcpy(list + prefix_len, name, name_len);
7354                 list[prefix_len + name_len] = '\0';
7355         }
7356         return total_len;
7357 }
7358
7359 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7360                 void *buffer, size_t size, int type)
7361 {
7362         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7363
7364         if (strcmp(name, "") == 0)
7365                 return -EINVAL;
7366         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7367                 return -EOPNOTSUPP;
7368         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7369                                buffer, size);
7370 }
7371
7372 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7373                 const void *value, size_t size, int flags, int type)
7374 {
7375         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7376
7377         if (strcmp(name, "") == 0)
7378                 return -EINVAL;
7379         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7380                 return -EOPNOTSUPP;
7381
7382         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7383                                name, value, size, flags);
7384 }
7385
7386 const struct xattr_handler ocfs2_xattr_user_handler = {
7387         .prefix = XATTR_USER_PREFIX,
7388         .list   = ocfs2_xattr_user_list,
7389         .get    = ocfs2_xattr_user_get,
7390         .set    = ocfs2_xattr_user_set,
7391 };