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 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43 #include "alloc.h"
  44 #include "blockcheck.h"
  45 #include "dlmglue.h"
  46 #include "file.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "inode.h"
  50 #include "journal.h"
  51 #include "ocfs2_fs.h"
  52 #include "suballoc.h"
  53 #include "uptodate.h"
  54 #include "buffer_head_io.h"
  55 #include "super.h"
  56 #include "xattr.h"
  57 #include "refcounttree.h"
  58 #include "acl.h"
  59 #include "ocfs2_trace.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         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 478
 479         BUG_ON(!buffer_uptodate(bh));
 480
 481         /*
 482          * If the ecc fails, we return the error but otherwise
 483          * leave the filesystem running.  We know any error is
 484          * local to this block.
 485          */
 486         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 487         if (rc)
 488                 return rc;
 489
 490         /*
 491          * Errors after here are fatal
 492          */
 493
 494         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 495                 ocfs2_error(sb,
 496                             "Extended attribute block #%llu has bad "
 497                             "signature %.*s",
 498                             (unsigned long long)bh->b_blocknr, 7,
 499                             xb->xb_signature);
 500                 return -EINVAL;
 501         }
 502
 503         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 504                 ocfs2_error(sb,
 505                             "Extended attribute block #%llu has an "
 506                             "invalid xb_blkno of %llu",
 507                             (unsigned long long)bh->b_blocknr,
 508                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 509                 return -EINVAL;
 510         }
 511
 512         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 513                 ocfs2_error(sb,
 514                             "Extended attribute block #%llu has an invalid "
 515                             "xb_fs_generation of #%u",
 516                             (unsigned long long)bh->b_blocknr,
 517                             le32_to_cpu(xb->xb_fs_generation));
 518                 return -EINVAL;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 525                                   struct buffer_head **bh)
 526 {
 527         int rc;
 528         struct buffer_head *tmp = *bh;
 529
 530         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 531                               ocfs2_validate_xattr_block);
 532
 533         /* If ocfs2_read_block() got us a new bh, pass it up. */
 534         if (!rc && !*bh)
 535                 *bh = tmp;
 536
 537         return rc;
 538 }
 539
 540 static inline const char *ocfs2_xattr_prefix(int name_index)
 541 {
 542         const struct xattr_handler *handler = NULL;
 543
 544         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 545                 handler = ocfs2_xattr_handler_map[name_index];
 546
 547         return handler ? handler->prefix : NULL;
 548 }
 549
 550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 551                                  const char *name,
 552                                  int name_len)
 553 {
 554         /* Get hash value of uuid from super block */
 555         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 556         int i;
 557
 558         /* hash extended attribute name */
 559         for (i = 0; i < name_len; i++) {
 560                 hash = (hash << OCFS2_HASH_SHIFT) ^
 561                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 562                        *name++;
 563         }
 564
 565         return hash;
 566 }
 567
 568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 569 {
 570         return namevalue_size(name_len, value_len) +
 571                 sizeof(struct ocfs2_xattr_entry);
 572 }
 573
 574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 575 {
 576         return namevalue_size_xi(xi) +
 577                 sizeof(struct ocfs2_xattr_entry);
 578 }
 579
 580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 581 {
 582         return namevalue_size_xe(xe) +
 583                 sizeof(struct ocfs2_xattr_entry);
 584 }
 585
 586 int ocfs2_calc_security_init(struct inode *dir,
 587                              struct ocfs2_security_xattr_info *si,
 588                              int *want_clusters,
 589                              int *xattr_credits,
 590                              struct ocfs2_alloc_context **xattr_ac)
 591 {
 592         int ret = 0;
 593         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 594         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 595                                                  si->value_len);
 596
 597         /*
 598          * The max space of security xattr taken inline is
 599          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 600          * So reserve one metadata block for it is ok.
 601          */
 602         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 603             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 604                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 605                 if (ret) {
 606                         mlog_errno(ret);
 607                         return ret;
 608                 }
 609                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 610         }
 611
 612         /* reserve clusters for xattr value which will be set in B tree*/
 613         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 614                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 615                                                             si->value_len);
 616
 617                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 618                                                            new_clusters);
 619                 *want_clusters += new_clusters;
 620         }
 621         return ret;
 622 }
 623
 624 int ocfs2_calc_xattr_init(struct inode *dir,
 625                           struct buffer_head *dir_bh,
 626                           umode_t mode,
 627                           struct ocfs2_security_xattr_info *si,
 628                           int *want_clusters,
 629                           int *xattr_credits,
 630                           int *want_meta)
 631 {
 632         int ret = 0;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 635
 636         if (si->enable)
 637                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 638                                                      si->value_len);
 639
 640         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 641                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 642                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 643                                         "", NULL, 0);
 644                 if (acl_len > 0) {
 645                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 646                         if (S_ISDIR(mode))
 647                                 a_size <<= 1;
 648                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 649                         mlog_errno(ret);
 650                         return ret;
 651                 }
 652         }
 653
 654         if (!(s_size + a_size))
 655                 return ret;
 656
 657         /*
 658          * The max space of security xattr taken inline is
 659          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 660          * The max space of acl xattr taken inline is
 661          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 662          * when blocksize = 512, may reserve one more cluser for
 663          * xattr bucket, otherwise reserve one metadata block
 664          * for them is ok.
 665          * If this is a new directory with inline data,
 666          * we choose to reserve the entire inline area for
 667          * directory contents and force an external xattr block.
 668          */
 669         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 670             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 671             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 672                 *want_meta = *want_meta + 1;
 673                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 674         }
 675
 676         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 677             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 678                 *want_clusters += 1;
 679                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 680         }
 681
 682         /*
 683          * reserve credits and clusters for xattrs which has large value
 684          * and have to be set outside
 685          */
 686         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 687                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 688                                                         si->value_len);
 689                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 690                                                            new_clusters);
 691                 *want_clusters += new_clusters;
 692         }
 693         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 694             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 695                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 696                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 697                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 698                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 699                                                            new_clusters);
 700                 *want_clusters += new_clusters;
 701         }
 702
 703         return ret;
 704 }
 705
 706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 707                                          u32 clusters_to_add,
 708                                          struct ocfs2_xattr_value_buf *vb,
 709                                          struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int status = 0, credits;
 712         handle_t *handle = ctxt->handle;
 713         enum ocfs2_alloc_restarted why;
 714         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 715         struct ocfs2_extent_tree et;
 716
 717         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 718
 719         while (clusters_to_add) {
 720                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 721
 722                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 723                                        OCFS2_JOURNAL_ACCESS_WRITE);
 724                 if (status < 0) {
 725                         mlog_errno(status);
 726                         break;
 727                 }
 728
 729                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 730                 status = ocfs2_add_clusters_in_btree(handle,
 731                                                      &et,
 732                                                      &logical_start,
 733                                                      clusters_to_add,
 734                                                      0,
 735                                                      ctxt->data_ac,
 736                                                      ctxt->meta_ac,
 737                                                      &why);
 738                 if ((status < 0) && (status != -EAGAIN)) {
 739                         if (status != -ENOSPC)
 740                                 mlog_errno(status);
 741                         break;
 742                 }
 743
 744                 ocfs2_journal_dirty(handle, vb->vb_bh);
 745
 746                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 747                                          prev_clusters;
 748
 749                 if (why != RESTART_NONE && clusters_to_add) {
 750                         /*
 751                          * We can only fail in case the alloc file doesn't give
 752                          * up enough clusters.
 753                          */
 754                         BUG_ON(why == RESTART_META);
 755
 756                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 757                                                             &vb->vb_xv->xr_list,
 758                                                             clusters_to_add);
 759                         status = ocfs2_extend_trans(handle, credits);
 760                         if (status < 0) {
 761                                 status = -ENOMEM;
 762                                 mlog_errno(status);
 763                                 break;
 764                         }
 765                 }
 766         }
 767
 768         return status;
 769 }
 770
 771 static int __ocfs2_remove_xattr_range(struct inode *inode,
 772                                       struct ocfs2_xattr_value_buf *vb,
 773                                       u32 cpos, u32 phys_cpos, u32 len,
 774                                       unsigned int ext_flags,
 775                                       struct ocfs2_xattr_set_ctxt *ctxt)
 776 {
 777         int ret;
 778         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 779         handle_t *handle = ctxt->handle;
 780         struct ocfs2_extent_tree et;
 781
 782         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 783
 784         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 785                             OCFS2_JOURNAL_ACCESS_WRITE);
 786         if (ret) {
 787                 mlog_errno(ret);
 788                 goto out;
 789         }
 790
 791         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 792                                   &ctxt->dealloc);
 793         if (ret) {
 794                 mlog_errno(ret);
 795                 goto out;
 796         }
 797
 798         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 799         ocfs2_journal_dirty(handle, vb->vb_bh);
 800
 801         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 802                 ret = ocfs2_decrease_refcount(inode, handle,
 803                                         ocfs2_blocks_to_clusters(inode->i_sb,
 804                                                                  phys_blkno),
 805                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 806         else
 807                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 808                                                   phys_blkno, len);
 809         if (ret)
 810                 mlog_errno(ret);
 811
 812 out:
 813         return ret;
 814 }
 815
 816 static int ocfs2_xattr_shrink_size(struct inode *inode,
 817                                    u32 old_clusters,
 818                                    u32 new_clusters,
 819                                    struct ocfs2_xattr_value_buf *vb,
 820                                    struct ocfs2_xattr_set_ctxt *ctxt)
 821 {
 822         int ret = 0;
 823         unsigned int ext_flags;
 824         u32 trunc_len, cpos, phys_cpos, alloc_size;
 825         u64 block;
 826
 827         if (old_clusters <= new_clusters)
 828                 return 0;
 829
 830         cpos = new_clusters;
 831         trunc_len = old_clusters - new_clusters;
 832         while (trunc_len) {
 833                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 834                                                &alloc_size,
 835                                                &vb->vb_xv->xr_list, &ext_flags);
 836                 if (ret) {
 837                         mlog_errno(ret);
 838                         goto out;
 839                 }
 840
 841                 if (alloc_size > trunc_len)
 842                         alloc_size = trunc_len;
 843
 844                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 845                                                  phys_cpos, alloc_size,
 846                                                  ext_flags, ctxt);
 847                 if (ret) {
 848                         mlog_errno(ret);
 849                         goto out;
 850                 }
 851
 852                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 853                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 854                                                        block, alloc_size);
 855                 cpos += alloc_size;
 856                 trunc_len -= alloc_size;
 857         }
 858
 859 out:
 860         return ret;
 861 }
 862
 863 static int ocfs2_xattr_value_truncate(struct inode *inode,
 864                                       struct ocfs2_xattr_value_buf *vb,
 865                                       int len,
 866                                       struct ocfs2_xattr_set_ctxt *ctxt)
 867 {
 868         int ret;
 869         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 870         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 871
 872         if (new_clusters == old_clusters)
 873                 return 0;
 874
 875         if (new_clusters > old_clusters)
 876                 ret = ocfs2_xattr_extend_allocation(inode,
 877                                                     new_clusters - old_clusters,
 878                                                     vb, ctxt);
 879         else
 880                 ret = ocfs2_xattr_shrink_size(inode,
 881                                               old_clusters, new_clusters,
 882                                               vb, ctxt);
 883
 884         return ret;
 885 }
 886
 887 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 888                                   size_t *result, const char *prefix,
 889                                   const char *name, int name_len)
 890 {
 891         char *p = buffer + *result;
 892         int prefix_len = strlen(prefix);
 893         int total_len = prefix_len + name_len + 1;
 894
 895         *result += total_len;
 896
 897         /* we are just looking for how big our buffer needs to be */
 898         if (!size)
 899                 return 0;
 900
 901         if (*result > size)
 902                 return -ERANGE;
 903
 904         memcpy(p, prefix, prefix_len);
 905         memcpy(p + prefix_len, name, name_len);
 906         p[prefix_len + name_len] = '\0';
 907
 908         return 0;
 909 }
 910
 911 static int ocfs2_xattr_list_entries(struct inode *inode,
 912                                     struct ocfs2_xattr_header *header,
 913                                     char *buffer, size_t buffer_size)
 914 {
 915         size_t result = 0;
 916         int i, type, ret;
 917         const char *prefix, *name;
 918
 919         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 920                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 921                 type = ocfs2_xattr_get_type(entry);
 922                 prefix = ocfs2_xattr_prefix(type);
 923
 924                 if (prefix) {
 925                         name = (const char *)header +
 926                                 le16_to_cpu(entry->xe_name_offset);
 927
 928                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 929                                                      &result, prefix, name,
 930                                                      entry->xe_name_len);
 931                         if (ret)
 932                                 return ret;
 933                 }
 934         }
 935
 936         return result;
 937 }
 938
 939 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 940                                          struct ocfs2_dinode *di)
 941 {
 942         struct ocfs2_xattr_header *xh;
 943         int i;
 944
 945         xh = (struct ocfs2_xattr_header *)
 946                  ((void *)di + inode->i_sb->s_blocksize -
 947                  le16_to_cpu(di->i_xattr_inline_size));
 948
 949         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 950                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 951                         return 1;
 952
 953         return 0;
 954 }
 955
 956 static int ocfs2_xattr_ibody_list(struct inode *inode,
 957                                   struct ocfs2_dinode *di,
 958                                   char *buffer,
 959                                   size_t buffer_size)
 960 {
 961         struct ocfs2_xattr_header *header = NULL;
 962         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 963         int ret = 0;
 964
 965         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 966                 return ret;
 967
 968         header = (struct ocfs2_xattr_header *)
 969                  ((void *)di + inode->i_sb->s_blocksize -
 970                  le16_to_cpu(di->i_xattr_inline_size));
 971
 972         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 973
 974         return ret;
 975 }
 976
 977 static int ocfs2_xattr_block_list(struct inode *inode,
 978                                   struct ocfs2_dinode *di,
 979                                   char *buffer,
 980                                   size_t buffer_size)
 981 {
 982         struct buffer_head *blk_bh = NULL;
 983         struct ocfs2_xattr_block *xb;
 984         int ret = 0;
 985
 986         if (!di->i_xattr_loc)
 987                 return ret;
 988
 989         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 990                                      &blk_bh);
 991         if (ret < 0) {
 992                 mlog_errno(ret);
 993                 return ret;
 994         }
 995
 996         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 997         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 998                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 999                 ret = ocfs2_xattr_list_entries(inode, header,
1000                                                buffer, buffer_size);
1001         } else
1002                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1003                                                    buffer, buffer_size);
1004
1005         brelse(blk_bh);
1006
1007         return ret;
1008 }
1009
1010 ssize_t ocfs2_listxattr(struct dentry *dentry,
1011                         char *buffer,
1012                         size_t size)
1013 {
1014         int ret = 0, i_ret = 0, b_ret = 0;
1015         struct buffer_head *di_bh = NULL;
1016         struct ocfs2_dinode *di = NULL;
1017         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1018
1019         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1020                 return -EOPNOTSUPP;
1021
1022         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1023                 return ret;
1024
1025         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1026         if (ret < 0) {
1027                 mlog_errno(ret);
1028                 return ret;
1029         }
1030
1031         di = (struct ocfs2_dinode *)di_bh->b_data;
1032
1033         down_read(&oi->ip_xattr_sem);
1034         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1035         if (i_ret < 0)
1036                 b_ret = 0;
1037         else {
1038                 if (buffer) {
1039                         buffer += i_ret;
1040                         size -= i_ret;
1041                 }
1042                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1043                                                buffer, size);
1044                 if (b_ret < 0)
1045                         i_ret = 0;
1046         }
1047         up_read(&oi->ip_xattr_sem);
1048         ocfs2_inode_unlock(dentry->d_inode, 0);
1049
1050         brelse(di_bh);
1051
1052         return i_ret + b_ret;
1053 }
1054
1055 static int ocfs2_xattr_find_entry(int name_index,
1056                                   const char *name,
1057                                   struct ocfs2_xattr_search *xs)
1058 {
1059         struct ocfs2_xattr_entry *entry;
1060         size_t name_len;
1061         int i, cmp = 1;
1062
1063         if (name == NULL)
1064                 return -EINVAL;
1065
1066         name_len = strlen(name);
1067         entry = xs->here;
1068         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1069                 cmp = name_index - ocfs2_xattr_get_type(entry);
1070                 if (!cmp)
1071                         cmp = name_len - entry->xe_name_len;
1072                 if (!cmp)
1073                         cmp = memcmp(name, (xs->base +
1074                                      le16_to_cpu(entry->xe_name_offset)),
1075                                      name_len);
1076                 if (cmp == 0)
1077                         break;
1078                 entry += 1;
1079         }
1080         xs->here = entry;
1081
1082         return cmp ? -ENODATA : 0;
1083 }
1084
1085 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1086                                          struct ocfs2_xattr_value_root *xv,
1087                                          void *buffer,
1088                                          size_t len)
1089 {
1090         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1091         u64 blkno;
1092         int i, ret = 0;
1093         size_t cplen, blocksize;
1094         struct buffer_head *bh = NULL;
1095         struct ocfs2_extent_list *el;
1096
1097         el = &xv->xr_list;
1098         clusters = le32_to_cpu(xv->xr_clusters);
1099         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1100         blocksize = inode->i_sb->s_blocksize;
1101
1102         cpos = 0;
1103         while (cpos < clusters) {
1104                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1105                                                &num_clusters, el, NULL);
1106                 if (ret) {
1107                         mlog_errno(ret);
1108                         goto out;
1109                 }
1110
1111                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1112                 /* Copy ocfs2_xattr_value */
1113                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1114                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1115                                                &bh, NULL);
1116                         if (ret) {
1117                                 mlog_errno(ret);
1118                                 goto out;
1119                         }
1120
1121                         cplen = len >= blocksize ? blocksize : len;
1122                         memcpy(buffer, bh->b_data, cplen);
1123                         len -= cplen;
1124                         buffer += cplen;
1125
1126                         brelse(bh);
1127                         bh = NULL;
1128                         if (len == 0)
1129                                 break;
1130                 }
1131                 cpos += num_clusters;
1132         }
1133 out:
1134         return ret;
1135 }
1136
1137 static int ocfs2_xattr_ibody_get(struct inode *inode,
1138                                  int name_index,
1139                                  const char *name,
1140                                  void *buffer,
1141                                  size_t buffer_size,
1142                                  struct ocfs2_xattr_search *xs)
1143 {
1144         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1145         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1146         struct ocfs2_xattr_value_root *xv;
1147         size_t size;
1148         int ret = 0;
1149
1150         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1151                 return -ENODATA;
1152
1153         xs->end = (void *)di + inode->i_sb->s_blocksize;
1154         xs->header = (struct ocfs2_xattr_header *)
1155                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1156         xs->base = (void *)xs->header;
1157         xs->here = xs->header->xh_entries;
1158
1159         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1160         if (ret)
1161                 return ret;
1162         size = le64_to_cpu(xs->here->xe_value_size);
1163         if (buffer) {
1164                 if (size > buffer_size)
1165                         return -ERANGE;
1166                 if (ocfs2_xattr_is_local(xs->here)) {
1167                         memcpy(buffer, (void *)xs->base +
1168                                le16_to_cpu(xs->here->xe_name_offset) +
1169                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1170                 } else {
1171                         xv = (struct ocfs2_xattr_value_root *)
1172                                 (xs->base + le16_to_cpu(
1173                                  xs->here->xe_name_offset) +
1174                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1175                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1176                                                             buffer, size);
1177                         if (ret < 0) {
1178                                 mlog_errno(ret);
1179                                 return ret;
1180                         }
1181                 }
1182         }
1183
1184         return size;
1185 }
1186
1187 static int ocfs2_xattr_block_get(struct inode *inode,
1188                                  int name_index,
1189                                  const char *name,
1190                                  void *buffer,
1191                                  size_t buffer_size,
1192                                  struct ocfs2_xattr_search *xs)
1193 {
1194         struct ocfs2_xattr_block *xb;
1195         struct ocfs2_xattr_value_root *xv;
1196         size_t size;
1197         int ret = -ENODATA, name_offset, name_len, i;
1198         int uninitialized_var(block_off);
1199
1200         xs->bucket = ocfs2_xattr_bucket_new(inode);
1201         if (!xs->bucket) {
1202                 ret = -ENOMEM;
1203                 mlog_errno(ret);
1204                 goto cleanup;
1205         }
1206
1207         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1208         if (ret) {
1209                 mlog_errno(ret);
1210                 goto cleanup;
1211         }
1212
1213         if (xs->not_found) {
1214                 ret = -ENODATA;
1215                 goto cleanup;
1216         }
1217
1218         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1219         size = le64_to_cpu(xs->here->xe_value_size);
1220         if (buffer) {
1221                 ret = -ERANGE;
1222                 if (size > buffer_size)
1223                         goto cleanup;
1224
1225                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1226                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1227                 i = xs->here - xs->header->xh_entries;
1228
1229                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1230                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1231                                                                 bucket_xh(xs->bucket),
1232                                                                 i,
1233                                                                 &block_off,
1234                                                                 &name_offset);
1235                         xs->base = bucket_block(xs->bucket, block_off);
1236                 }
1237                 if (ocfs2_xattr_is_local(xs->here)) {
1238                         memcpy(buffer, (void *)xs->base +
1239                                name_offset + name_len, size);
1240                 } else {
1241                         xv = (struct ocfs2_xattr_value_root *)
1242                                 (xs->base + name_offset + name_len);
1243                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1244                                                             buffer, size);
1245                         if (ret < 0) {
1246                                 mlog_errno(ret);
1247                                 goto cleanup;
1248                         }
1249                 }
1250         }
1251         ret = size;
1252 cleanup:
1253         ocfs2_xattr_bucket_free(xs->bucket);
1254
1255         brelse(xs->xattr_bh);
1256         xs->xattr_bh = NULL;
1257         return ret;
1258 }
1259
1260 int ocfs2_xattr_get_nolock(struct inode *inode,
1261                            struct buffer_head *di_bh,
1262                            int name_index,
1263                            const char *name,
1264                            void *buffer,
1265                            size_t buffer_size)
1266 {
1267         int ret;
1268         struct ocfs2_dinode *di = NULL;
1269         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1270         struct ocfs2_xattr_search xis = {
1271                 .not_found = -ENODATA,
1272         };
1273         struct ocfs2_xattr_search xbs = {
1274                 .not_found = -ENODATA,
1275         };
1276
1277         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1278                 return -EOPNOTSUPP;
1279
1280         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1281                 ret = -ENODATA;
1282
1283         xis.inode_bh = xbs.inode_bh = di_bh;
1284         di = (struct ocfs2_dinode *)di_bh->b_data;
1285
1286         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1287                                     buffer_size, &xis);
1288         if (ret == -ENODATA && di->i_xattr_loc)
1289                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1290                                             buffer_size, &xbs);
1291
1292         return ret;
1293 }
1294
1295 /* ocfs2_xattr_get()
1296  *
1297  * Copy an extended attribute into the buffer provided.
1298  * Buffer is NULL to compute the size of buffer required.
1299  */
1300 static int ocfs2_xattr_get(struct inode *inode,
1301                            int name_index,
1302                            const char *name,
1303                            void *buffer,
1304                            size_t buffer_size)
1305 {
1306         int ret;
1307         struct buffer_head *di_bh = NULL;
1308
1309         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1310         if (ret < 0) {
1311                 mlog_errno(ret);
1312                 return ret;
1313         }
1314         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1315         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1316                                      name, buffer, buffer_size);
1317         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1318
1319         ocfs2_inode_unlock(inode, 0);
1320
1321         brelse(di_bh);
1322
1323         return ret;
1324 }
1325
1326 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1327                                            handle_t *handle,
1328                                            struct ocfs2_xattr_value_buf *vb,
1329                                            const void *value,
1330                                            int value_len)
1331 {
1332         int ret = 0, i, cp_len;
1333         u16 blocksize = inode->i_sb->s_blocksize;
1334         u32 p_cluster, num_clusters;
1335         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1336         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1337         u64 blkno;
1338         struct buffer_head *bh = NULL;
1339         unsigned int ext_flags;
1340         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1341
1342         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1343
1344         while (cpos < clusters) {
1345                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1346                                                &num_clusters, &xv->xr_list,
1347                                                &ext_flags);
1348                 if (ret) {
1349                         mlog_errno(ret);
1350                         goto out;
1351                 }
1352
1353                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1354
1355                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1356
1357                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1358                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1359                                                &bh, NULL);
1360                         if (ret) {
1361                                 mlog_errno(ret);
1362                                 goto out;
1363                         }
1364
1365                         ret = ocfs2_journal_access(handle,
1366                                                    INODE_CACHE(inode),
1367                                                    bh,
1368                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1369                         if (ret < 0) {
1370                                 mlog_errno(ret);
1371                                 goto out;
1372                         }
1373
1374                         cp_len = value_len > blocksize ? blocksize : value_len;
1375                         memcpy(bh->b_data, value, cp_len);
1376                         value_len -= cp_len;
1377                         value += cp_len;
1378                         if (cp_len < blocksize)
1379                                 memset(bh->b_data + cp_len, 0,
1380                                        blocksize - cp_len);
1381
1382                         ocfs2_journal_dirty(handle, bh);
1383                         brelse(bh);
1384                         bh = NULL;
1385
1386                         /*
1387                          * XXX: do we need to empty all the following
1388                          * blocks in this cluster?
1389                          */
1390                         if (!value_len)
1391                                 break;
1392                 }
1393                 cpos += num_clusters;
1394         }
1395 out:
1396         brelse(bh);
1397
1398         return ret;
1399 }
1400
1401 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1402                                        int num_entries)
1403 {
1404         int free_space;
1405
1406         if (!needed_space)
1407                 return 0;
1408
1409         free_space = free_start -
1410                 sizeof(struct ocfs2_xattr_header) -
1411                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1412                 OCFS2_XATTR_HEADER_GAP;
1413         if (free_space < 0)
1414                 return -EIO;
1415         if (free_space < needed_space)
1416                 return -ENOSPC;
1417
1418         return 0;
1419 }
1420
1421 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1422                                    int type)
1423 {
1424         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1425 }
1426
1427 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1428 {
1429         loc->xl_ops->xlo_journal_dirty(handle, loc);
1430 }
1431
1432 /* Give a pointer into the storage for the given offset */
1433 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1434 {
1435         BUG_ON(offset >= loc->xl_size);
1436         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1437 }
1438
1439 /*
1440  * Wipe the name+value pair and allow the storage to reclaim it.  This
1441  * must be followed by either removal of the entry or a call to
1442  * ocfs2_xa_add_namevalue().
1443  */
1444 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1445 {
1446         loc->xl_ops->xlo_wipe_namevalue(loc);
1447 }
1448
1449 /*
1450  * Find lowest offset to a name+value pair.  This is the start of our
1451  * downward-growing free space.
1452  */
1453 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1454 {
1455         return loc->xl_ops->xlo_get_free_start(loc);
1456 }
1457
1458 /* Can we reuse loc->xl_entry for xi? */
1459 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1460                                     struct ocfs2_xattr_info *xi)
1461 {
1462         return loc->xl_ops->xlo_can_reuse(loc, xi);
1463 }
1464
1465 /* How much free space is needed to set the new value */
1466 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1467                                 struct ocfs2_xattr_info *xi)
1468 {
1469         return loc->xl_ops->xlo_check_space(loc, xi);
1470 }
1471
1472 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1473 {
1474         loc->xl_ops->xlo_add_entry(loc, name_hash);
1475         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1476         /*
1477          * We can't leave the new entry's xe_name_offset at zero or
1478          * add_namevalue() will go nuts.  We set it to the size of our
1479          * storage so that it can never be less than any other entry.
1480          */
1481         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1482 }
1483
1484 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1485                                    struct ocfs2_xattr_info *xi)
1486 {
1487         int size = namevalue_size_xi(xi);
1488         int nameval_offset;
1489         char *nameval_buf;
1490
1491         loc->xl_ops->xlo_add_namevalue(loc, size);
1492         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1493         loc->xl_entry->xe_name_len = xi->xi_name_len;
1494         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1495         ocfs2_xattr_set_local(loc->xl_entry,
1496                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1497
1498         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1499         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1500         memset(nameval_buf, 0, size);
1501         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1502 }
1503
1504 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1505                                     struct ocfs2_xattr_value_buf *vb)
1506 {
1507         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1508         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1509
1510         /* Value bufs are for value trees */
1511         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1512         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1513                (name_size + OCFS2_XATTR_ROOT_SIZE));
1514
1515         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1516         vb->vb_xv =
1517                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1518                                                         nameval_offset +
1519                                                         name_size);
1520 }
1521
1522 static int ocfs2_xa_block_journal_access(handle_t *handle,
1523                                          struct ocfs2_xa_loc *loc, int type)
1524 {
1525         struct buffer_head *bh = loc->xl_storage;
1526         ocfs2_journal_access_func access;
1527
1528         if (loc->xl_size == (bh->b_size -
1529                              offsetof(struct ocfs2_xattr_block,
1530                                       xb_attrs.xb_header)))
1531                 access = ocfs2_journal_access_xb;
1532         else
1533                 access = ocfs2_journal_access_di;
1534         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1535 }
1536
1537 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1538                                          struct ocfs2_xa_loc *loc)
1539 {
1540         struct buffer_head *bh = loc->xl_storage;
1541
1542         ocfs2_journal_dirty(handle, bh);
1543 }
1544
1545 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1546                                            int offset)
1547 {
1548         return (char *)loc->xl_header + offset;
1549 }
1550
1551 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1552                                     struct ocfs2_xattr_info *xi)
1553 {
1554         /*
1555          * Block storage is strict.  If the sizes aren't exact, we will
1556          * remove the old one and reinsert the new.
1557          */
1558         return namevalue_size_xe(loc->xl_entry) ==
1559                 namevalue_size_xi(xi);
1560 }
1561
1562 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1563 {
1564         struct ocfs2_xattr_header *xh = loc->xl_header;
1565         int i, count = le16_to_cpu(xh->xh_count);
1566         int offset, free_start = loc->xl_size;
1567
1568         for (i = 0; i < count; i++) {
1569                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1570                 if (offset < free_start)
1571                         free_start = offset;
1572         }
1573
1574         return free_start;
1575 }
1576
1577 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1578                                       struct ocfs2_xattr_info *xi)
1579 {
1580         int count = le16_to_cpu(loc->xl_header->xh_count);
1581         int free_start = ocfs2_xa_get_free_start(loc);
1582         int needed_space = ocfs2_xi_entry_usage(xi);
1583
1584         /*
1585          * Block storage will reclaim the original entry before inserting
1586          * the new value, so we only need the difference.  If the new
1587          * entry is smaller than the old one, we don't need anything.
1588          */
1589         if (loc->xl_entry) {
1590                 /* Don't need space if we're reusing! */
1591                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1592                         needed_space = 0;
1593                 else
1594                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1595         }
1596         if (needed_space < 0)
1597                 needed_space = 0;
1598         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1599 }
1600
1601 /*
1602  * Block storage for xattrs keeps the name+value pairs compacted.  When
1603  * we remove one, we have to shift any that preceded it towards the end.
1604  */
1605 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1606 {
1607         int i, offset;
1608         int namevalue_offset, first_namevalue_offset, namevalue_size;
1609         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1610         struct ocfs2_xattr_header *xh = loc->xl_header;
1611         int count = le16_to_cpu(xh->xh_count);
1612
1613         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1614         namevalue_size = namevalue_size_xe(entry);
1615         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1616
1617         /* Shift the name+value pairs */
1618         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1619                 (char *)xh + first_namevalue_offset,
1620                 namevalue_offset - first_namevalue_offset);
1621         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1622
1623         /* Now tell xh->xh_entries about it */
1624         for (i = 0; i < count; i++) {
1625                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1626                 if (offset <= namevalue_offset)
1627                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1628                                      namevalue_size);
1629         }
1630
1631         /*
1632          * Note that we don't update xh_free_start or xh_name_value_len
1633          * because they're not used in block-stored xattrs.
1634          */
1635 }
1636
1637 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1638 {
1639         int count = le16_to_cpu(loc->xl_header->xh_count);
1640         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1641         le16_add_cpu(&loc->xl_header->xh_count, 1);
1642         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1643 }
1644
1645 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1646 {
1647         int free_start = ocfs2_xa_get_free_start(loc);
1648
1649         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1650 }
1651
1652 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1653                                           struct ocfs2_xattr_value_buf *vb)
1654 {
1655         struct buffer_head *bh = loc->xl_storage;
1656
1657         if (loc->xl_size == (bh->b_size -
1658                              offsetof(struct ocfs2_xattr_block,
1659                                       xb_attrs.xb_header)))
1660                 vb->vb_access = ocfs2_journal_access_xb;
1661         else
1662                 vb->vb_access = ocfs2_journal_access_di;
1663         vb->vb_bh = bh;
1664 }
1665
1666 /*
1667  * Operations for xattrs stored in blocks.  This includes inline inode
1668  * storage and unindexed ocfs2_xattr_blocks.
1669  */
1670 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1671         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1672         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1673         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1674         .xlo_check_space        = ocfs2_xa_block_check_space,
1675         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1676         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1677         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1678         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1679         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1680         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1681 };
1682
1683 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1684                                           struct ocfs2_xa_loc *loc, int type)
1685 {
1686         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1687
1688         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1689 }
1690
1691 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1692                                           struct ocfs2_xa_loc *loc)
1693 {
1694         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1695
1696         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1697 }
1698
1699 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1700                                             int offset)
1701 {
1702         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1703         int block, block_offset;
1704
1705         /* The header is at the front of the bucket */
1706         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1707         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1708
1709         return bucket_block(bucket, block) + block_offset;
1710 }
1711
1712 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1713                                      struct ocfs2_xattr_info *xi)
1714 {
1715         return namevalue_size_xe(loc->xl_entry) >=
1716                 namevalue_size_xi(xi);
1717 }
1718
1719 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1720 {
1721         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1722         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1723 }
1724
1725 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1726                                          int free_start, int size)
1727 {
1728         /*
1729          * We need to make sure that the name+value pair fits within
1730          * one block.
1731          */
1732         if (((free_start - size) >> sb->s_blocksize_bits) !=
1733             ((free_start - 1) >> sb->s_blocksize_bits))
1734                 free_start -= free_start % sb->s_blocksize;
1735
1736         return free_start;
1737 }
1738
1739 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1740                                        struct ocfs2_xattr_info *xi)
1741 {
1742         int rc;
1743         int count = le16_to_cpu(loc->xl_header->xh_count);
1744         int free_start = ocfs2_xa_get_free_start(loc);
1745         int needed_space = ocfs2_xi_entry_usage(xi);
1746         int size = namevalue_size_xi(xi);
1747         struct super_block *sb = loc->xl_inode->i_sb;
1748
1749         /*
1750          * Bucket storage does not reclaim name+value pairs it cannot
1751          * reuse.  They live as holes until the bucket fills, and then
1752          * the bucket is defragmented.  However, the bucket can reclaim
1753          * the ocfs2_xattr_entry.
1754          */
1755         if (loc->xl_entry) {
1756                 /* Don't need space if we're reusing! */
1757                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1758                         needed_space = 0;
1759                 else
1760                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1761         }
1762         BUG_ON(needed_space < 0);
1763
1764         if (free_start < size) {
1765                 if (needed_space)
1766                         return -ENOSPC;
1767         } else {
1768                 /*
1769                  * First we check if it would fit in the first place.
1770                  * Below, we align the free start to a block.  This may
1771                  * slide us below the minimum gap.  By checking unaligned
1772                  * first, we avoid that error.
1773                  */
1774                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1775                                                  count);
1776                 if (rc)
1777                         return rc;
1778                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1779                                                            size);
1780         }
1781         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1782 }
1783
1784 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1785 {
1786         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1787                      -namevalue_size_xe(loc->xl_entry));
1788 }
1789
1790 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1791 {
1792         struct ocfs2_xattr_header *xh = loc->xl_header;
1793         int count = le16_to_cpu(xh->xh_count);
1794         int low = 0, high = count - 1, tmp;
1795         struct ocfs2_xattr_entry *tmp_xe;
1796
1797         /*
1798          * We keep buckets sorted by name_hash, so we need to find
1799          * our insert place.
1800          */
1801         while (low <= high && count) {
1802                 tmp = (low + high) / 2;
1803                 tmp_xe = &xh->xh_entries[tmp];
1804
1805                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1806                         low = tmp + 1;
1807                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1808                         high = tmp - 1;
1809                 else {
1810                         low = tmp;
1811                         break;
1812                 }
1813         }
1814
1815         if (low != count)
1816                 memmove(&xh->xh_entries[low + 1],
1817                         &xh->xh_entries[low],
1818                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1819
1820         le16_add_cpu(&xh->xh_count, 1);
1821         loc->xl_entry = &xh->xh_entries[low];
1822         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1823 }
1824
1825 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1826 {
1827         int free_start = ocfs2_xa_get_free_start(loc);
1828         struct ocfs2_xattr_header *xh = loc->xl_header;
1829         struct super_block *sb = loc->xl_inode->i_sb;
1830         int nameval_offset;
1831
1832         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1833         nameval_offset = free_start - size;
1834         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1835         xh->xh_free_start = cpu_to_le16(nameval_offset);
1836         le16_add_cpu(&xh->xh_name_value_len, size);
1837
1838 }
1839
1840 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1841                                            struct ocfs2_xattr_value_buf *vb)
1842 {
1843         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1844         struct super_block *sb = loc->xl_inode->i_sb;
1845         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1846         int size = namevalue_size_xe(loc->xl_entry);
1847         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1848
1849         /* Values are not allowed to straddle block boundaries */
1850         BUG_ON(block_offset !=
1851                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1852         /* We expect the bucket to be filled in */
1853         BUG_ON(!bucket->bu_bhs[block_offset]);
1854
1855         vb->vb_access = ocfs2_journal_access;
1856         vb->vb_bh = bucket->bu_bhs[block_offset];
1857 }
1858
1859 /* Operations for xattrs stored in buckets. */
1860 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1861         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1862         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1863         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1864         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1865         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1866         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1867         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1868         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1869         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1870         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1871 };
1872
1873 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1874 {
1875         struct ocfs2_xattr_value_buf vb;
1876
1877         if (ocfs2_xattr_is_local(loc->xl_entry))
1878                 return 0;
1879
1880         ocfs2_xa_fill_value_buf(loc, &vb);
1881         return le32_to_cpu(vb.vb_xv->xr_clusters);
1882 }
1883
1884 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1885                                    struct ocfs2_xattr_set_ctxt *ctxt)
1886 {
1887         int trunc_rc, access_rc;
1888         struct ocfs2_xattr_value_buf vb;
1889
1890         ocfs2_xa_fill_value_buf(loc, &vb);
1891         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1892                                               ctxt);
1893
1894         /*
1895          * The caller of ocfs2_xa_value_truncate() has already called
1896          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1897          * calls ocfs2_extend_trans().  This may commit the previous
1898          * transaction and open a new one.  If this is a bucket, truncate
1899          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1900          * the caller is expecting to dirty the entire bucket.  So we must
1901          * reset the journal work.  We do this even if truncate has failed,
1902          * as it could have failed after committing the extend.
1903          */
1904         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1905                                             OCFS2_JOURNAL_ACCESS_WRITE);
1906
1907         /* Errors in truncate take precedence */
1908         return trunc_rc ? trunc_rc : access_rc;
1909 }
1910
1911 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1912 {
1913         int index, count;
1914         struct ocfs2_xattr_header *xh = loc->xl_header;
1915         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1916
1917         ocfs2_xa_wipe_namevalue(loc);
1918         loc->xl_entry = NULL;
1919
1920         le16_add_cpu(&xh->xh_count, -1);
1921         count = le16_to_cpu(xh->xh_count);
1922
1923         /*
1924          * Only zero out the entry if there are more remaining.  This is
1925          * important for an empty bucket, as it keeps track of the
1926          * bucket's hash value.  It doesn't hurt empty block storage.
1927          */
1928         if (count) {
1929                 index = ((char *)entry - (char *)&xh->xh_entries) /
1930                         sizeof(struct ocfs2_xattr_entry);
1931                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1932                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1933                 memset(&xh->xh_entries[count], 0,
1934                        sizeof(struct ocfs2_xattr_entry));
1935         }
1936 }
1937
1938 /*
1939  * If we have a problem adjusting the size of an external value during
1940  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1941  * in an intermediate state.  For example, the value may be partially
1942  * truncated.
1943  *
1944  * If the value tree hasn't changed, the extend/truncate went nowhere.
1945  * We have nothing to do.  The caller can treat it as a straight error.
1946  *
1947  * If the value tree got partially truncated, we now have a corrupted
1948  * extended attribute.  We're going to wipe its entry and leak the
1949  * clusters.  Better to leak some storage than leave a corrupt entry.
1950  *
1951  * If the value tree grew, it obviously didn't grow enough for the
1952  * new entry.  We're not going to try and reclaim those clusters either.
1953  * If there was already an external value there (orig_clusters != 0),
1954  * the new clusters are attached safely and we can just leave the old
1955  * value in place.  If there was no external value there, we remove
1956  * the entry.
1957  *
1958  * This way, the xattr block we store in the journal will be consistent.
1959  * If the size change broke because of the journal, no changes will hit
1960  * disk anyway.
1961  */
1962 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1963                                             const char *what,
1964                                             unsigned int orig_clusters)
1965 {
1966         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1967         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1968                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1969
1970         if (new_clusters < orig_clusters) {
1971                 mlog(ML_ERROR,
1972                      "Partial truncate while %s xattr %.*s.  Leaking "
1973                      "%u clusters and removing the entry\n",
1974                      what, loc->xl_entry->xe_name_len, nameval_buf,
1975                      orig_clusters - new_clusters);
1976                 ocfs2_xa_remove_entry(loc);
1977         } else if (!orig_clusters) {
1978                 mlog(ML_ERROR,
1979                      "Unable to allocate an external value for xattr "
1980                      "%.*s safely.  Leaking %u clusters and removing the "
1981                      "entry\n",
1982                      loc->xl_entry->xe_name_len, nameval_buf,
1983                      new_clusters - orig_clusters);
1984                 ocfs2_xa_remove_entry(loc);
1985         } else if (new_clusters > orig_clusters)
1986                 mlog(ML_ERROR,
1987                      "Unable to grow xattr %.*s safely.  %u new clusters "
1988                      "have been added, but the value will not be "
1989                      "modified\n",
1990                      loc->xl_entry->xe_name_len, nameval_buf,
1991                      new_clusters - orig_clusters);
1992 }
1993
1994 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1995                            struct ocfs2_xattr_set_ctxt *ctxt)
1996 {
1997         int rc = 0;
1998         unsigned int orig_clusters;
1999
2000         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2001                 orig_clusters = ocfs2_xa_value_clusters(loc);
2002                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2003                 if (rc) {
2004                         mlog_errno(rc);
2005                         /*
2006                          * Since this is remove, we can return 0 if
2007                          * ocfs2_xa_cleanup_value_truncate() is going to
2008                          * wipe the entry anyway.  So we check the
2009                          * cluster count as well.
2010                          */
2011                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2012                                 rc = 0;
2013                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2014                                                         orig_clusters);
2015                         if (rc)
2016                                 goto out;
2017                 }
2018         }
2019
2020         ocfs2_xa_remove_entry(loc);
2021
2022 out:
2023         return rc;
2024 }
2025
2026 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2027 {
2028         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2029         char *nameval_buf;
2030
2031         nameval_buf = ocfs2_xa_offset_pointer(loc,
2032                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2033         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2034 }
2035
2036 /*
2037  * Take an existing entry and make it ready for the new value.  This
2038  * won't allocate space, but it may free space.  It should be ready for
2039  * ocfs2_xa_prepare_entry() to finish the work.
2040  */
2041 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2042                                 struct ocfs2_xattr_info *xi,
2043                                 struct ocfs2_xattr_set_ctxt *ctxt)
2044 {
2045         int rc = 0;
2046         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2047         unsigned int orig_clusters;
2048         char *nameval_buf;
2049         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2050         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2051
2052         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2053                name_size);
2054
2055         nameval_buf = ocfs2_xa_offset_pointer(loc,
2056                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2057         if (xe_local) {
2058                 memset(nameval_buf + name_size, 0,
2059                        namevalue_size_xe(loc->xl_entry) - name_size);
2060                 if (!xi_local)
2061                         ocfs2_xa_install_value_root(loc);
2062         } else {
2063                 orig_clusters = ocfs2_xa_value_clusters(loc);
2064                 if (xi_local) {
2065                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2066                         if (rc < 0)
2067                                 mlog_errno(rc);
2068                         else
2069                                 memset(nameval_buf + name_size, 0,
2070                                        namevalue_size_xe(loc->xl_entry) -
2071                                        name_size);
2072                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2073                            xi->xi_value_len) {
2074                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2075                                                      ctxt);
2076                         if (rc < 0)
2077                                 mlog_errno(rc);
2078                 }
2079
2080                 if (rc) {
2081                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2082                                                         orig_clusters);
2083                         goto out;
2084                 }
2085         }
2086
2087         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2088         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2089
2090 out:
2091         return rc;
2092 }
2093
2094 /*
2095  * Prepares loc->xl_entry to receive the new xattr.  This includes
2096  * properly setting up the name+value pair region.  If loc->xl_entry
2097  * already exists, it will take care of modifying it appropriately.
2098  *
2099  * Note that this modifies the data.  You did journal_access already,
2100  * right?
2101  */
2102 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2103                                   struct ocfs2_xattr_info *xi,
2104                                   u32 name_hash,
2105                                   struct ocfs2_xattr_set_ctxt *ctxt)
2106 {
2107         int rc = 0;
2108         unsigned int orig_clusters;
2109         __le64 orig_value_size = 0;
2110
2111         rc = ocfs2_xa_check_space(loc, xi);
2112         if (rc)
2113                 goto out;
2114
2115         if (loc->xl_entry) {
2116                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2117                         orig_value_size = loc->xl_entry->xe_value_size;
2118                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2119                         if (rc)
2120                                 goto out;
2121                         goto alloc_value;
2122                 }
2123
2124                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2125                         orig_clusters = ocfs2_xa_value_clusters(loc);
2126                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2127                         if (rc) {
2128                                 mlog_errno(rc);
2129                                 ocfs2_xa_cleanup_value_truncate(loc,
2130                                                                 "overwriting",
2131                                                                 orig_clusters);
2132                                 goto out;
2133                         }
2134                 }
2135                 ocfs2_xa_wipe_namevalue(loc);
2136         } else
2137                 ocfs2_xa_add_entry(loc, name_hash);
2138
2139         /*
2140          * If we get here, we have a blank entry.  Fill it.  We grow our
2141          * name+value pair back from the end.
2142          */
2143         ocfs2_xa_add_namevalue(loc, xi);
2144         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2145                 ocfs2_xa_install_value_root(loc);
2146
2147 alloc_value:
2148         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2149                 orig_clusters = ocfs2_xa_value_clusters(loc);
2150                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2151                 if (rc < 0) {
2152                         ctxt->set_abort = 1;
2153                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2154                                                         orig_clusters);
2155                         /*
2156                          * If we were growing an existing value,
2157                          * ocfs2_xa_cleanup_value_truncate() won't remove
2158                          * the entry. We need to restore the original value
2159                          * size.
2160                          */
2161                         if (loc->xl_entry) {
2162                                 BUG_ON(!orig_value_size);
2163                                 loc->xl_entry->xe_value_size = orig_value_size;
2164                         }
2165                         mlog_errno(rc);
2166                 }
2167         }
2168
2169 out:
2170         return rc;
2171 }
2172
2173 /*
2174  * Store the value portion of the name+value pair.  This will skip
2175  * values that are stored externally.  Their tree roots were set up
2176  * by ocfs2_xa_prepare_entry().
2177  */
2178 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2179                                 struct ocfs2_xattr_info *xi,
2180                                 struct ocfs2_xattr_set_ctxt *ctxt)
2181 {
2182         int rc = 0;
2183         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2184         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2185         char *nameval_buf;
2186         struct ocfs2_xattr_value_buf vb;
2187
2188         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2189         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2190                 ocfs2_xa_fill_value_buf(loc, &vb);
2191                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2192                                                      ctxt->handle, &vb,
2193                                                      xi->xi_value,
2194                                                      xi->xi_value_len);
2195         } else
2196                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2197
2198         return rc;
2199 }
2200
2201 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2202                         struct ocfs2_xattr_info *xi,
2203                         struct ocfs2_xattr_set_ctxt *ctxt)
2204 {
2205         int ret;
2206         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2207                                               xi->xi_name_len);
2208
2209         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2210                                       OCFS2_JOURNAL_ACCESS_WRITE);
2211         if (ret) {
2212                 mlog_errno(ret);
2213                 goto out;
2214         }
2215
2216         /*
2217          * From here on out, everything is going to modify the buffer a
2218          * little.  Errors are going to leave the xattr header in a
2219          * sane state.  Thus, even with errors we dirty the sucker.
2220          */
2221
2222         /* Don't worry, we are never called with !xi_value and !xl_entry */
2223         if (!xi->xi_value) {
2224                 ret = ocfs2_xa_remove(loc, ctxt);
2225                 goto out_dirty;
2226         }
2227
2228         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2229         if (ret) {
2230                 if (ret != -ENOSPC)
2231                         mlog_errno(ret);
2232                 goto out_dirty;
2233         }
2234
2235         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2236         if (ret)
2237                 mlog_errno(ret);
2238
2239 out_dirty:
2240         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2241
2242 out:
2243         return ret;
2244 }
2245
2246 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2247                                      struct inode *inode,
2248                                      struct buffer_head *bh,
2249                                      struct ocfs2_xattr_entry *entry)
2250 {
2251         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2252
2253         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2254
2255         loc->xl_inode = inode;
2256         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2257         loc->xl_storage = bh;
2258         loc->xl_entry = entry;
2259         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2260         loc->xl_header =
2261                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2262                                               loc->xl_size);
2263 }
2264
2265 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2266                                           struct inode *inode,
2267                                           struct buffer_head *bh,
2268                                           struct ocfs2_xattr_entry *entry)
2269 {
2270         struct ocfs2_xattr_block *xb =
2271                 (struct ocfs2_xattr_block *)bh->b_data;
2272
2273         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2274
2275         loc->xl_inode = inode;
2276         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2277         loc->xl_storage = bh;
2278         loc->xl_header = &(xb->xb_attrs.xb_header);
2279         loc->xl_entry = entry;
2280         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2281                                              xb_attrs.xb_header);
2282 }
2283
2284 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2285                                            struct ocfs2_xattr_bucket *bucket,
2286                                            struct ocfs2_xattr_entry *entry)
2287 {
2288         loc->xl_inode = bucket->bu_inode;
2289         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2290         loc->xl_storage = bucket;
2291         loc->xl_header = bucket_xh(bucket);
2292         loc->xl_entry = entry;
2293         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2294 }
2295
2296 /*
2297  * In xattr remove, if it is stored outside and refcounted, we may have
2298  * the chance to split the refcount tree. So need the allocators.
2299  */
2300 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2301                                         struct ocfs2_xattr_value_root *xv,
2302                                         struct ocfs2_caching_info *ref_ci,
2303                                         struct buffer_head *ref_root_bh,
2304                                         struct ocfs2_alloc_context **meta_ac,
2305                                         int *ref_credits)
2306 {
2307         int ret, meta_add = 0;
2308         u32 p_cluster, num_clusters;
2309         unsigned int ext_flags;
2310
2311         *ref_credits = 0;
2312         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2313                                        &num_clusters,
2314                                        &xv->xr_list,
2315                                        &ext_flags);
2316         if (ret) {
2317                 mlog_errno(ret);
2318                 goto out;
2319         }
2320
2321         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2322                 goto out;
2323
2324         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2325                                                  ref_root_bh, xv,
2326                                                  &meta_add, ref_credits);
2327         if (ret) {
2328                 mlog_errno(ret);
2329                 goto out;
2330         }
2331
2332         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2333                                                 meta_add, meta_ac);
2334         if (ret)
2335                 mlog_errno(ret);
2336
2337 out:
2338         return ret;
2339 }
2340
2341 static int ocfs2_remove_value_outside(struct inode*inode,
2342                                       struct ocfs2_xattr_value_buf *vb,
2343                                       struct ocfs2_xattr_header *header,
2344                                       struct ocfs2_caching_info *ref_ci,
2345                                       struct buffer_head *ref_root_bh)
2346 {
2347         int ret = 0, i, ref_credits;
2348         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2349         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2350         void *val;
2351
2352         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2353
2354         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2355                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2356
2357                 if (ocfs2_xattr_is_local(entry))
2358                         continue;
2359
2360                 val = (void *)header +
2361                         le16_to_cpu(entry->xe_name_offset);
2362                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2363                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2364
2365                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2366                                                          ref_ci, ref_root_bh,
2367                                                          &ctxt.meta_ac,
2368                                                          &ref_credits);
2369
2370                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2371                                         ocfs2_remove_extent_credits(osb->sb));
2372                 if (IS_ERR(ctxt.handle)) {
2373                         ret = PTR_ERR(ctxt.handle);
2374                         mlog_errno(ret);
2375                         break;
2376                 }
2377
2378                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2379
2380                 ocfs2_commit_trans(osb, ctxt.handle);
2381                 if (ctxt.meta_ac) {
2382                         ocfs2_free_alloc_context(ctxt.meta_ac);
2383                         ctxt.meta_ac = NULL;
2384                 }
2385
2386                 if (ret < 0) {
2387                         mlog_errno(ret);
2388                         break;
2389                 }
2390
2391         }
2392
2393         if (ctxt.meta_ac)
2394                 ocfs2_free_alloc_context(ctxt.meta_ac);
2395         ocfs2_schedule_truncate_log_flush(osb, 1);
2396         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2397         return ret;
2398 }
2399
2400 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2401                                     struct buffer_head *di_bh,
2402                                     struct ocfs2_caching_info *ref_ci,
2403                                     struct buffer_head *ref_root_bh)
2404 {
2405
2406         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2407         struct ocfs2_xattr_header *header;
2408         int ret;
2409         struct ocfs2_xattr_value_buf vb = {
2410                 .vb_bh = di_bh,
2411                 .vb_access = ocfs2_journal_access_di,
2412         };
2413
2414         header = (struct ocfs2_xattr_header *)
2415                  ((void *)di + inode->i_sb->s_blocksize -
2416                  le16_to_cpu(di->i_xattr_inline_size));
2417
2418         ret = ocfs2_remove_value_outside(inode, &vb, header,
2419                                          ref_ci, ref_root_bh);
2420
2421         return ret;
2422 }
2423
2424 struct ocfs2_rm_xattr_bucket_para {
2425         struct ocfs2_caching_info *ref_ci;
2426         struct buffer_head *ref_root_bh;
2427 };
2428
2429 static int ocfs2_xattr_block_remove(struct inode *inode,
2430                                     struct buffer_head *blk_bh,
2431                                     struct ocfs2_caching_info *ref_ci,
2432                                     struct buffer_head *ref_root_bh)
2433 {
2434         struct ocfs2_xattr_block *xb;
2435         int ret = 0;
2436         struct ocfs2_xattr_value_buf vb = {
2437                 .vb_bh = blk_bh,
2438                 .vb_access = ocfs2_journal_access_xb,
2439         };
2440         struct ocfs2_rm_xattr_bucket_para args = {
2441                 .ref_ci = ref_ci,
2442                 .ref_root_bh = ref_root_bh,
2443         };
2444
2445         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2446         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2447                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2448                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2449                                                  ref_ci, ref_root_bh);
2450         } else
2451                 ret = ocfs2_iterate_xattr_index_block(inode,
2452                                                 blk_bh,
2453                                                 ocfs2_rm_xattr_cluster,
2454                                                 &args);
2455
2456         return ret;
2457 }
2458
2459 static int ocfs2_xattr_free_block(struct inode *inode,
2460                                   u64 block,
2461                                   struct ocfs2_caching_info *ref_ci,
2462                                   struct buffer_head *ref_root_bh)
2463 {
2464         struct inode *xb_alloc_inode;
2465         struct buffer_head *xb_alloc_bh = NULL;
2466         struct buffer_head *blk_bh = NULL;
2467         struct ocfs2_xattr_block *xb;
2468         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2469         handle_t *handle;
2470         int ret = 0;
2471         u64 blk, bg_blkno;
2472         u16 bit;
2473
2474         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2475         if (ret < 0) {
2476                 mlog_errno(ret);
2477                 goto out;
2478         }
2479
2480         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2481         if (ret < 0) {
2482                 mlog_errno(ret);
2483                 goto out;
2484         }
2485
2486         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2487         blk = le64_to_cpu(xb->xb_blkno);
2488         bit = le16_to_cpu(xb->xb_suballoc_bit);
2489         if (xb->xb_suballoc_loc)
2490                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2491         else
2492                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2493
2494         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2495                                 EXTENT_ALLOC_SYSTEM_INODE,
2496                                 le16_to_cpu(xb->xb_suballoc_slot));
2497         if (!xb_alloc_inode) {
2498                 ret = -ENOMEM;
2499                 mlog_errno(ret);
2500                 goto out;
2501         }
2502         mutex_lock(&xb_alloc_inode->i_mutex);
2503
2504         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2505         if (ret < 0) {
2506                 mlog_errno(ret);
2507                 goto out_mutex;
2508         }
2509
2510         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2511         if (IS_ERR(handle)) {
2512                 ret = PTR_ERR(handle);
2513                 mlog_errno(ret);
2514                 goto out_unlock;
2515         }
2516
2517         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2518                                        bit, bg_blkno, 1);
2519         if (ret < 0)
2520                 mlog_errno(ret);
2521
2522         ocfs2_commit_trans(osb, handle);
2523 out_unlock:
2524         ocfs2_inode_unlock(xb_alloc_inode, 1);
2525         brelse(xb_alloc_bh);
2526 out_mutex:
2527         mutex_unlock(&xb_alloc_inode->i_mutex);
2528         iput(xb_alloc_inode);
2529 out:
2530         brelse(blk_bh);
2531         return ret;
2532 }
2533
2534 /*
2535  * ocfs2_xattr_remove()
2536  *
2537  * Free extended attribute resources associated with this inode.
2538  */
2539 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2540 {
2541         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2542         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2543         struct ocfs2_refcount_tree *ref_tree = NULL;
2544         struct buffer_head *ref_root_bh = NULL;
2545         struct ocfs2_caching_info *ref_ci = NULL;
2546         handle_t *handle;
2547         int ret;
2548
2549         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2550                 return 0;
2551
2552         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2553                 return 0;
2554
2555         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2556                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2557                                                le64_to_cpu(di->i_refcount_loc),
2558                                                1, &ref_tree, &ref_root_bh);
2559                 if (ret) {
2560                         mlog_errno(ret);
2561                         goto out;
2562                 }
2563                 ref_ci = &ref_tree->rf_ci;
2564
2565         }
2566
2567         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2568                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2569                                                ref_ci, ref_root_bh);
2570                 if (ret < 0) {
2571                         mlog_errno(ret);
2572                         goto out;
2573                 }
2574         }
2575
2576         if (di->i_xattr_loc) {
2577                 ret = ocfs2_xattr_free_block(inode,
2578                                              le64_to_cpu(di->i_xattr_loc),
2579                                              ref_ci, ref_root_bh);
2580                 if (ret < 0) {
2581                         mlog_errno(ret);
2582                         goto out;
2583                 }
2584         }
2585
2586         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2587                                    OCFS2_INODE_UPDATE_CREDITS);
2588         if (IS_ERR(handle)) {
2589                 ret = PTR_ERR(handle);
2590                 mlog_errno(ret);
2591                 goto out;
2592         }
2593         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2594                                       OCFS2_JOURNAL_ACCESS_WRITE);
2595         if (ret) {
2596                 mlog_errno(ret);
2597                 goto out_commit;
2598         }
2599
2600         di->i_xattr_loc = 0;
2601
2602         spin_lock(&oi->ip_lock);
2603         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2604         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2605         spin_unlock(&oi->ip_lock);
2606
2607         ocfs2_journal_dirty(handle, di_bh);
2608 out_commit:
2609         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2610 out:
2611         if (ref_tree)
2612                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2613         brelse(ref_root_bh);
2614         return ret;
2615 }
2616
2617 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2618                                         struct ocfs2_dinode *di)
2619 {
2620         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2621         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2622         int free;
2623
2624         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2625                 return 0;
2626
2627         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2628                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2629                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2630         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2631                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2632                         le64_to_cpu(di->i_size);
2633         } else {
2634                 struct ocfs2_extent_list *el = &di->id2.i_list;
2635                 free = (le16_to_cpu(el->l_count) -
2636                         le16_to_cpu(el->l_next_free_rec)) *
2637                         sizeof(struct ocfs2_extent_rec);
2638         }
2639         if (free >= xattrsize)
2640                 return 1;
2641
2642         return 0;
2643 }
2644
2645 /*
2646  * ocfs2_xattr_ibody_find()
2647  *
2648  * Find extended attribute in inode block and
2649  * fill search info into struct ocfs2_xattr_search.
2650  */
2651 static int ocfs2_xattr_ibody_find(struct inode *inode,
2652                                   int name_index,
2653                                   const char *name,
2654                                   struct ocfs2_xattr_search *xs)
2655 {
2656         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2657         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2658         int ret;
2659         int has_space = 0;
2660
2661         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2662                 return 0;
2663
2664         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2665                 down_read(&oi->ip_alloc_sem);
2666                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2667                 up_read(&oi->ip_alloc_sem);
2668                 if (!has_space)
2669                         return 0;
2670         }
2671
2672         xs->xattr_bh = xs->inode_bh;
2673         xs->end = (void *)di + inode->i_sb->s_blocksize;
2674         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2675                 xs->header = (struct ocfs2_xattr_header *)
2676                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2677         else
2678                 xs->header = (struct ocfs2_xattr_header *)
2679                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2680         xs->base = (void *)xs->header;
2681         xs->here = xs->header->xh_entries;
2682
2683         /* Find the named attribute. */
2684         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2685                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2686                 if (ret && ret != -ENODATA)
2687                         return ret;
2688                 xs->not_found = ret;
2689         }
2690
2691         return 0;
2692 }
2693
2694 static int ocfs2_xattr_ibody_init(struct inode *inode,
2695                                   struct buffer_head *di_bh,
2696                                   struct ocfs2_xattr_set_ctxt *ctxt)
2697 {
2698         int ret;
2699         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2700         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2701         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2702         unsigned int xattrsize = osb->s_xattr_inline_size;
2703
2704         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2705                 ret = -ENOSPC;
2706                 goto out;
2707         }
2708
2709         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2710                                       OCFS2_JOURNAL_ACCESS_WRITE);
2711         if (ret) {
2712                 mlog_errno(ret);
2713                 goto out;
2714         }
2715
2716         /*
2717          * Adjust extent record count or inline data size
2718          * to reserve space for extended attribute.
2719          */
2720         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2721                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2722                 le16_add_cpu(&idata->id_count, -xattrsize);
2723         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2724                 struct ocfs2_extent_list *el = &di->id2.i_list;
2725                 le16_add_cpu(&el->l_count, -(xattrsize /
2726                                              sizeof(struct ocfs2_extent_rec)));
2727         }
2728         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2729
2730         spin_lock(&oi->ip_lock);
2731         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2732         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2733         spin_unlock(&oi->ip_lock);
2734
2735         ocfs2_journal_dirty(ctxt->handle, di_bh);
2736
2737 out:
2738         return ret;
2739 }
2740
2741 /*
2742  * ocfs2_xattr_ibody_set()
2743  *
2744  * Set, replace or remove an extended attribute into inode block.
2745  *
2746  */
2747 static int ocfs2_xattr_ibody_set(struct inode *inode,
2748                                  struct ocfs2_xattr_info *xi,
2749                                  struct ocfs2_xattr_search *xs,
2750                                  struct ocfs2_xattr_set_ctxt *ctxt)
2751 {
2752         int ret;
2753         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2754         struct ocfs2_xa_loc loc;
2755
2756         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2757                 return -ENOSPC;
2758
2759         down_write(&oi->ip_alloc_sem);
2760         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2761                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2762                 if (ret) {
2763                         if (ret != -ENOSPC)
2764                                 mlog_errno(ret);
2765                         goto out;
2766                 }
2767         }
2768
2769         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2770                                  xs->not_found ? NULL : xs->here);
2771         ret = ocfs2_xa_set(&loc, xi, ctxt);
2772         if (ret) {
2773                 if (ret != -ENOSPC)
2774                         mlog_errno(ret);
2775                 goto out;
2776         }
2777         xs->here = loc.xl_entry;
2778
2779 out:
2780         up_write(&oi->ip_alloc_sem);
2781
2782         return ret;
2783 }
2784
2785 /*
2786  * ocfs2_xattr_block_find()
2787  *
2788  * Find extended attribute in external block and
2789  * fill search info into struct ocfs2_xattr_search.
2790  */
2791 static int ocfs2_xattr_block_find(struct inode *inode,
2792                                   int name_index,
2793                                   const char *name,
2794                                   struct ocfs2_xattr_search *xs)
2795 {
2796         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2797         struct buffer_head *blk_bh = NULL;
2798         struct ocfs2_xattr_block *xb;
2799         int ret = 0;
2800
2801         if (!di->i_xattr_loc)
2802                 return ret;
2803
2804         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2805                                      &blk_bh);
2806         if (ret < 0) {
2807                 mlog_errno(ret);
2808                 return ret;
2809         }
2810
2811         xs->xattr_bh = blk_bh;
2812         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2813
2814         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2815                 xs->header = &xb->xb_attrs.xb_header;
2816                 xs->base = (void *)xs->header;
2817                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2818                 xs->here = xs->header->xh_entries;
2819
2820                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2821         } else
2822                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2823                                                    name_index,
2824                                                    name, xs);
2825
2826         if (ret && ret != -ENODATA) {
2827                 xs->xattr_bh = NULL;
2828                 goto cleanup;
2829         }
2830         xs->not_found = ret;
2831         return 0;
2832 cleanup:
2833         brelse(blk_bh);
2834
2835         return ret;
2836 }
2837
2838 static int ocfs2_create_xattr_block(struct inode *inode,
2839                                     struct buffer_head *inode_bh,
2840                                     struct ocfs2_xattr_set_ctxt *ctxt,
2841                                     int indexed,
2842                                     struct buffer_head **ret_bh)
2843 {
2844         int ret;
2845         u16 suballoc_bit_start;
2846         u32 num_got;
2847         u64 suballoc_loc, first_blkno;
2848         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2849         struct buffer_head *new_bh = NULL;
2850         struct ocfs2_xattr_block *xblk;
2851
2852         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2853                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2854         if (ret < 0) {
2855                 mlog_errno(ret);
2856                 goto end;
2857         }
2858
2859         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2860                                    &suballoc_loc, &suballoc_bit_start,
2861                                    &num_got, &first_blkno);
2862         if (ret < 0) {
2863                 mlog_errno(ret);
2864                 goto end;
2865         }
2866
2867         new_bh = sb_getblk(inode->i_sb, first_blkno);
2868         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2869
2870         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2871                                       new_bh,
2872                                       OCFS2_JOURNAL_ACCESS_CREATE);
2873         if (ret < 0) {
2874                 mlog_errno(ret);
2875                 goto end;
2876         }
2877
2878         /* Initialize ocfs2_xattr_block */
2879         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2880         memset(xblk, 0, inode->i_sb->s_blocksize);
2881         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2882         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2883         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2884         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2885         xblk->xb_fs_generation =
2886                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2887         xblk->xb_blkno = cpu_to_le64(first_blkno);
2888         if (indexed) {
2889                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2890                 xr->xt_clusters = cpu_to_le32(1);
2891                 xr->xt_last_eb_blk = 0;
2892                 xr->xt_list.l_tree_depth = 0;
2893                 xr->xt_list.l_count = cpu_to_le16(
2894                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2895                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2896                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2897         }
2898         ocfs2_journal_dirty(ctxt->handle, new_bh);
2899
2900         /* Add it to the inode */
2901         di->i_xattr_loc = cpu_to_le64(first_blkno);
2902
2903         spin_lock(&OCFS2_I(inode)->ip_lock);
2904         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2905         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2906         spin_unlock(&OCFS2_I(inode)->ip_lock);
2907
2908         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2909
2910         *ret_bh = new_bh;
2911         new_bh = NULL;
2912
2913 end:
2914         brelse(new_bh);
2915         return ret;
2916 }
2917
2918 /*
2919  * ocfs2_xattr_block_set()
2920  *
2921  * Set, replace or remove an extended attribute into external block.
2922  *
2923  */
2924 static int ocfs2_xattr_block_set(struct inode *inode,
2925                                  struct ocfs2_xattr_info *xi,
2926                                  struct ocfs2_xattr_search *xs,
2927                                  struct ocfs2_xattr_set_ctxt *ctxt)
2928 {
2929         struct buffer_head *new_bh = NULL;
2930         struct ocfs2_xattr_block *xblk = NULL;
2931         int ret;
2932         struct ocfs2_xa_loc loc;
2933
2934         if (!xs->xattr_bh) {
2935                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2936                                                0, &new_bh);
2937                 if (ret) {
2938                         mlog_errno(ret);
2939                         goto end;
2940                 }
2941
2942                 xs->xattr_bh = new_bh;
2943                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2944                 xs->header = &xblk->xb_attrs.xb_header;
2945                 xs->base = (void *)xs->header;
2946                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2947                 xs->here = xs->header->xh_entries;
2948         } else
2949                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2950
2951         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2952                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2953                                               xs->not_found ? NULL : xs->here);
2954
2955                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2956                 if (!ret)
2957                         xs->here = loc.xl_entry;
2958                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2959                         goto end;
2960                 else {
2961                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2962                         if (ret)
2963                                 goto end;
2964                 }
2965         }
2966
2967         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2968                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2969
2970 end:
2971         return ret;
2972 }
2973
2974 /* Check whether the new xattr can be inserted into the inode. */
2975 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2976                                        struct ocfs2_xattr_info *xi,
2977                                        struct ocfs2_xattr_search *xs)
2978 {
2979         struct ocfs2_xattr_entry *last;
2980         int free, i;
2981         size_t min_offs = xs->end - xs->base;
2982
2983         if (!xs->header)
2984                 return 0;
2985
2986         last = xs->header->xh_entries;
2987
2988         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2989                 size_t offs = le16_to_cpu(last->xe_name_offset);
2990                 if (offs < min_offs)
2991                         min_offs = offs;
2992                 last += 1;
2993         }
2994
2995         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2996         if (free < 0)
2997                 return 0;
2998
2999         BUG_ON(!xs->not_found);
3000
3001         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3002                 return 1;
3003
3004         return 0;
3005 }
3006
3007 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3008                                      struct ocfs2_dinode *di,
3009                                      struct ocfs2_xattr_info *xi,
3010                                      struct ocfs2_xattr_search *xis,
3011                                      struct ocfs2_xattr_search *xbs,
3012                                      int *clusters_need,
3013                                      int *meta_need,
3014                                      int *credits_need)
3015 {
3016         int ret = 0, old_in_xb = 0;
3017         int clusters_add = 0, meta_add = 0, credits = 0;
3018         struct buffer_head *bh = NULL;
3019         struct ocfs2_xattr_block *xb = NULL;
3020         struct ocfs2_xattr_entry *xe = NULL;
3021         struct ocfs2_xattr_value_root *xv = NULL;
3022         char *base = NULL;
3023         int name_offset, name_len = 0;
3024         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3025                                                     xi->xi_value_len);
3026         u64 value_size;
3027
3028         /*
3029          * Calculate the clusters we need to write.
3030          * No matter whether we replace an old one or add a new one,
3031          * we need this for writing.
3032          */
3033         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3034                 credits += new_clusters *
3035                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3036
3037         if (xis->not_found && xbs->not_found) {
3038                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3039
3040                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3041                         clusters_add += new_clusters;
3042                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3043                                                         &def_xv.xv.xr_list,
3044                                                         new_clusters);
3045                 }
3046
3047                 goto meta_guess;
3048         }
3049
3050         if (!xis->not_found) {
3051                 xe = xis->here;
3052                 name_offset = le16_to_cpu(xe->xe_name_offset);
3053                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3054                 base = xis->base;
3055                 credits += OCFS2_INODE_UPDATE_CREDITS;
3056         } else {
3057                 int i, block_off = 0;
3058                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3059                 xe = xbs->here;
3060                 name_offset = le16_to_cpu(xe->xe_name_offset);
3061                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3062                 i = xbs->here - xbs->header->xh_entries;
3063                 old_in_xb = 1;
3064
3065                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3066                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3067                                                         bucket_xh(xbs->bucket),
3068                                                         i, &block_off,
3069                                                         &name_offset);
3070                         base = bucket_block(xbs->bucket, block_off);
3071                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3072                 } else {
3073                         base = xbs->base;
3074                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3075                 }
3076         }
3077
3078         /*
3079          * delete a xattr doesn't need metadata and cluster allocation.
3080          * so just calculate the credits and return.
3081          *
3082          * The credits for removing the value tree will be extended
3083          * by ocfs2_remove_extent itself.
3084          */
3085         if (!xi->xi_value) {
3086                 if (!ocfs2_xattr_is_local(xe))
3087                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3088
3089                 goto out;
3090         }
3091
3092         /* do cluster allocation guess first. */
3093         value_size = le64_to_cpu(xe->xe_value_size);
3094
3095         if (old_in_xb) {
3096                 /*
3097                  * In xattr set, we always try to set the xe in inode first,
3098                  * so if it can be inserted into inode successfully, the old
3099                  * one will be removed from the xattr block, and this xattr
3100                  * will be inserted into inode as a new xattr in inode.
3101                  */
3102                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3103                         clusters_add += new_clusters;
3104                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3105                                     OCFS2_INODE_UPDATE_CREDITS;
3106                         if (!ocfs2_xattr_is_local(xe))
3107                                 credits += ocfs2_calc_extend_credits(
3108                                                         inode->i_sb,
3109                                                         &def_xv.xv.xr_list,
3110                                                         new_clusters);
3111                         goto out;
3112                 }
3113         }
3114
3115         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3116                 /* the new values will be stored outside. */
3117                 u32 old_clusters = 0;
3118
3119                 if (!ocfs2_xattr_is_local(xe)) {
3120                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3121                                                                  value_size);
3122                         xv = (struct ocfs2_xattr_value_root *)
3123                              (base + name_offset + name_len);
3124                         value_size = OCFS2_XATTR_ROOT_SIZE;
3125                 } else
3126                         xv = &def_xv.xv;
3127
3128                 if (old_clusters >= new_clusters) {
3129                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3130                         goto out;
3131                 } else {
3132                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3133                         clusters_add += new_clusters - old_clusters;
3134                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3135                                                              &xv->xr_list,
3136                                                              new_clusters -
3137                                                              old_clusters);
3138                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3139                                 goto out;
3140                 }
3141         } else {
3142                 /*
3143                  * Now the new value will be stored inside. So if the new
3144                  * value is smaller than the size of value root or the old
3145                  * value, we don't need any allocation, otherwise we have
3146                  * to guess metadata allocation.
3147                  */
3148                 if ((ocfs2_xattr_is_local(xe) &&
3149                      (value_size >= xi->xi_value_len)) ||
3150                     (!ocfs2_xattr_is_local(xe) &&
3151                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3152                         goto out;
3153         }
3154
3155 meta_guess:
3156         /* calculate metadata allocation. */
3157         if (di->i_xattr_loc) {
3158                 if (!xbs->xattr_bh) {
3159                         ret = ocfs2_read_xattr_block(inode,
3160                                                      le64_to_cpu(di->i_xattr_loc),
3161                                                      &bh);
3162                         if (ret) {
3163                                 mlog_errno(ret);
3164                                 goto out;
3165                         }
3166
3167                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3168                 } else
3169                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3170
3171                 /*
3172                  * If there is already an xattr tree, good, we can calculate
3173                  * like other b-trees. Otherwise we may have the chance of
3174                  * create a tree, the credit calculation is borrowed from
3175                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3176                  * new tree will be cluster based, so no meta is needed.
3177                  */
3178                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3179                         struct ocfs2_extent_list *el =
3180                                  &xb->xb_attrs.xb_root.xt_list;
3181                         meta_add += ocfs2_extend_meta_needed(el);
3182                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3183                                                              el, 1);
3184                 } else
3185                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3186
3187                 /*
3188                  * This cluster will be used either for new bucket or for
3189                  * new xattr block.
3190                  * If the cluster size is the same as the bucket size, one
3191                  * more is needed since we may need to extend the bucket
3192                  * also.
3193                  */
3194                 clusters_add += 1;
3195                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3196                 if (OCFS2_XATTR_BUCKET_SIZE ==
3197                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3198                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3199                         clusters_add += 1;
3200                 }
3201         } else {
3202                 meta_add += 1;
3203                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3204         }
3205 out:
3206         if (clusters_need)
3207                 *clusters_need = clusters_add;
3208         if (meta_need)
3209                 *meta_need = meta_add;
3210         if (credits_need)
3211                 *credits_need = credits;
3212         brelse(bh);
3213         return ret;
3214 }
3215
3216 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3217                                      struct ocfs2_dinode *di,
3218                                      struct ocfs2_xattr_info *xi,
3219                                      struct ocfs2_xattr_search *xis,
3220                                      struct ocfs2_xattr_search *xbs,
3221                                      struct ocfs2_xattr_set_ctxt *ctxt,
3222                                      int extra_meta,
3223                                      int *credits)
3224 {
3225         int clusters_add, meta_add, ret;
3226         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3227
3228         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3229
3230         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3231
3232         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3233                                         &clusters_add, &meta_add, credits);
3234         if (ret) {
3235                 mlog_errno(ret);
3236                 return ret;
3237         }
3238
3239         meta_add += extra_meta;
3240         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3241                                         clusters_add, *credits);
3242
3243         if (meta_add) {
3244                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3245                                                         &ctxt->meta_ac);
3246                 if (ret) {
3247                         mlog_errno(ret);
3248                         goto out;
3249                 }
3250         }
3251
3252         if (clusters_add) {
3253                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3254                 if (ret)
3255                         mlog_errno(ret);
3256         }
3257 out:
3258         if (ret) {
3259                 if (ctxt->meta_ac) {
3260                         ocfs2_free_alloc_context(ctxt->meta_ac);
3261                         ctxt->meta_ac = NULL;
3262                 }
3263
3264                 /*
3265                  * We cannot have an error and a non null ctxt->data_ac.
3266                  */
3267         }
3268
3269         return ret;
3270 }
3271
3272 static int __ocfs2_xattr_set_handle(struct inode *inode,
3273                                     struct ocfs2_dinode *di,
3274                                     struct ocfs2_xattr_info *xi,
3275                                     struct ocfs2_xattr_search *xis,
3276                                     struct ocfs2_xattr_search *xbs,
3277                                     struct ocfs2_xattr_set_ctxt *ctxt)
3278 {
3279         int ret = 0, credits, old_found;
3280
3281         if (!xi->xi_value) {
3282                 /* Remove existing extended attribute */
3283                 if (!xis->not_found)
3284                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3285                 else if (!xbs->not_found)
3286                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3287         } else {
3288                 /* We always try to set extended attribute into inode first*/
3289                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3290                 if (!ret && !xbs->not_found) {
3291                         /*
3292                          * If succeed and that extended attribute existing in
3293                          * external block, then we will remove it.
3294                          */
3295                         xi->xi_value = NULL;
3296                         xi->xi_value_len = 0;
3297
3298                         old_found = xis->not_found;
3299                         xis->not_found = -ENODATA;
3300                         ret = ocfs2_calc_xattr_set_need(inode,
3301                                                         di,
3302                                                         xi,
3303                                                         xis,
3304                                                         xbs,
3305                                                         NULL,
3306                                                         NULL,
3307                                                         &credits);
3308                         xis->not_found = old_found;
3309                         if (ret) {
3310                                 mlog_errno(ret);
3311                                 goto out;
3312                         }
3313
3314                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3315                         if (ret) {
3316                                 mlog_errno(ret);
3317                                 goto out;
3318                         }
3319                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3320                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3321                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3322                                 ret = ocfs2_xattr_block_find(inode,
3323                                                              xi->xi_name_index,
3324                                                              xi->xi_name, xbs);
3325                                 if (ret)
3326                                         goto out;
3327
3328                                 old_found = xis->not_found;
3329                                 xis->not_found = -ENODATA;
3330                                 ret = ocfs2_calc_xattr_set_need(inode,
3331                                                                 di,
3332                                                                 xi,
3333                                                                 xis,
3334                                                                 xbs,
3335                                                                 NULL,
3336                                                                 NULL,
3337                                                                 &credits);
3338                                 xis->not_found = old_found;
3339                                 if (ret) {
3340                                         mlog_errno(ret);
3341                                         goto out;
3342                                 }
3343
3344                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3345                                 if (ret) {
3346                                         mlog_errno(ret);
3347                                         goto out;
3348                                 }
3349                         }
3350                         /*
3351                          * If no space in inode, we will set extended attribute
3352                          * into external block.
3353                          */
3354                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3355                         if (ret)
3356                                 goto out;
3357                         if (!xis->not_found) {
3358                                 /*
3359                                  * If succeed and that extended attribute
3360                                  * existing in inode, we will remove it.
3361                                  */
3362                                 xi->xi_value = NULL;
3363                                 xi->xi_value_len = 0;
3364                                 xbs->not_found = -ENODATA;
3365                                 ret = ocfs2_calc_xattr_set_need(inode,
3366                                                                 di,
3367                                                                 xi,
3368                                                                 xis,
3369                                                                 xbs,
3370                                                                 NULL,
3371                                                                 NULL,
3372                                                                 &credits);
3373                                 if (ret) {
3374                                         mlog_errno(ret);
3375                                         goto out;
3376                                 }
3377
3378                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3379                                 if (ret) {
3380                                         mlog_errno(ret);
3381                                         goto out;
3382                                 }
3383                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3384                                                             xis, ctxt);
3385                         }
3386                 }
3387         }
3388
3389         if (!ret) {
3390                 /* Update inode ctime. */
3391                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3392                                               xis->inode_bh,
3393                                               OCFS2_JOURNAL_ACCESS_WRITE);
3394                 if (ret) {
3395                         mlog_errno(ret);
3396                         goto out;
3397                 }
3398
3399                 inode->i_ctime = CURRENT_TIME;
3400                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3401                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3402                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3403         }
3404 out:
3405         return ret;
3406 }
3407
3408 /*
3409  * This function only called duing creating inode
3410  * for init security/acl xattrs of the new inode.
3411  * All transanction credits have been reserved in mknod.
3412  */
3413 int ocfs2_xattr_set_handle(handle_t *handle,
3414                            struct inode *inode,
3415                            struct buffer_head *di_bh,
3416                            int name_index,
3417                            const char *name,
3418                            const void *value,
3419                            size_t value_len,
3420                            int flags,
3421                            struct ocfs2_alloc_context *meta_ac,
3422                            struct ocfs2_alloc_context *data_ac)
3423 {
3424         struct ocfs2_dinode *di;
3425         int ret;
3426
3427         struct ocfs2_xattr_info xi = {
3428                 .xi_name_index = name_index,
3429                 .xi_name = name,
3430                 .xi_name_len = strlen(name),
3431                 .xi_value = value,
3432                 .xi_value_len = value_len,
3433         };
3434
3435         struct ocfs2_xattr_search xis = {
3436                 .not_found = -ENODATA,
3437         };
3438
3439         struct ocfs2_xattr_search xbs = {
3440                 .not_found = -ENODATA,
3441         };
3442
3443         struct ocfs2_xattr_set_ctxt ctxt = {
3444                 .handle = handle,
3445                 .meta_ac = meta_ac,
3446                 .data_ac = data_ac,
3447         };
3448
3449         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3450                 return -EOPNOTSUPP;
3451
3452         /*
3453          * In extreme situation, may need xattr bucket when
3454          * block size is too small. And we have already reserved
3455          * the credits for bucket in mknod.
3456          */
3457         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3458                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3459                 if (!xbs.bucket) {
3460                         mlog_errno(-ENOMEM);
3461                         return -ENOMEM;
3462                 }
3463         }
3464
3465         xis.inode_bh = xbs.inode_bh = di_bh;
3466         di = (struct ocfs2_dinode *)di_bh->b_data;
3467
3468         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3469
3470         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3471         if (ret)
3472                 goto cleanup;
3473         if (xis.not_found) {
3474                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3475                 if (ret)
3476                         goto cleanup;
3477         }
3478
3479         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3480
3481 cleanup:
3482         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3483         brelse(xbs.xattr_bh);
3484         ocfs2_xattr_bucket_free(xbs.bucket);
3485
3486         return ret;
3487 }
3488
3489 /*
3490  * ocfs2_xattr_set()
3491  *
3492  * Set, replace or remove an extended attribute for this inode.
3493  * value is NULL to remove an existing extended attribute, else either
3494  * create or replace an extended attribute.
3495  */
3496 int ocfs2_xattr_set(struct inode *inode,
3497                     int name_index,
3498                     const char *name,
3499                     const void *value,
3500                     size_t value_len,
3501                     int flags)
3502 {
3503         struct buffer_head *di_bh = NULL;
3504         struct ocfs2_dinode *di;
3505         int ret, credits, ref_meta = 0, ref_credits = 0;
3506         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3507         struct inode *tl_inode = osb->osb_tl_inode;
3508         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3509         struct ocfs2_refcount_tree *ref_tree = NULL;
3510
3511         struct ocfs2_xattr_info xi = {
3512                 .xi_name_index = name_index,
3513                 .xi_name = name,
3514                 .xi_name_len = strlen(name),
3515                 .xi_value = value,
3516                 .xi_value_len = value_len,
3517         };
3518
3519         struct ocfs2_xattr_search xis = {
3520                 .not_found = -ENODATA,
3521         };
3522
3523         struct ocfs2_xattr_search xbs = {
3524                 .not_found = -ENODATA,
3525         };
3526
3527         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3528                 return -EOPNOTSUPP;
3529
3530         /*
3531          * Only xbs will be used on indexed trees.  xis doesn't need a
3532          * bucket.
3533          */
3534         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3535         if (!xbs.bucket) {
3536                 mlog_errno(-ENOMEM);
3537                 return -ENOMEM;
3538         }
3539
3540         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3541         if (ret < 0) {
3542                 mlog_errno(ret);
3543                 goto cleanup_nolock;
3544         }
3545         xis.inode_bh = xbs.inode_bh = di_bh;
3546         di = (struct ocfs2_dinode *)di_bh->b_data;
3547
3548         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3549         /*
3550          * Scan inode and external block to find the same name
3551          * extended attribute and collect search information.
3552          */
3553         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3554         if (ret)
3555                 goto cleanup;
3556         if (xis.not_found) {
3557                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3558                 if (ret)
3559                         goto cleanup;
3560         }
3561
3562         if (xis.not_found && xbs.not_found) {
3563                 ret = -ENODATA;
3564                 if (flags & XATTR_REPLACE)
3565                         goto cleanup;
3566                 ret = 0;
3567                 if (!value)
3568                         goto cleanup;
3569         } else {
3570                 ret = -EEXIST;
3571                 if (flags & XATTR_CREATE)
3572                         goto cleanup;
3573         }
3574
3575         /* Check whether the value is refcounted and do some preparation. */
3576         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3577             (!xis.not_found || !xbs.not_found)) {
3578                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3579                                                    &xis, &xbs, &ref_tree,
3580                                                    &ref_meta, &ref_credits);
3581                 if (ret) {
3582                         mlog_errno(ret);
3583                         goto cleanup;
3584                 }
3585         }
3586
3587         mutex_lock(&tl_inode->i_mutex);
3588
3589         if (ocfs2_truncate_log_needs_flush(osb)) {
3590                 ret = __ocfs2_flush_truncate_log(osb);
3591                 if (ret < 0) {
3592                         mutex_unlock(&tl_inode->i_mutex);
3593                         mlog_errno(ret);
3594                         goto cleanup;
3595                 }
3596         }
3597         mutex_unlock(&tl_inode->i_mutex);
3598
3599         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3600                                         &xbs, &ctxt, ref_meta, &credits);
3601         if (ret) {
3602                 mlog_errno(ret);
3603                 goto cleanup;
3604         }
3605
3606         /* we need to update inode's ctime field, so add credit for it. */
3607         credits += OCFS2_INODE_UPDATE_CREDITS;
3608         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3609         if (IS_ERR(ctxt.handle)) {
3610                 ret = PTR_ERR(ctxt.handle);
3611                 mlog_errno(ret);
3612                 goto out_free_ac;
3613         }
3614
3615         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3616
3617         ocfs2_commit_trans(osb, ctxt.handle);
3618
3619 out_free_ac:
3620         if (ctxt.data_ac)
3621                 ocfs2_free_alloc_context(ctxt.data_ac);
3622         if (ctxt.meta_ac)
3623                 ocfs2_free_alloc_context(ctxt.meta_ac);
3624         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3625                 ocfs2_schedule_truncate_log_flush(osb, 1);
3626         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3627
3628 cleanup:
3629         if (ref_tree)
3630                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3631         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3632         if (!value && !ret) {
3633                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3634                 if (ret)
3635                         mlog_errno(ret);
3636         }
3637         ocfs2_inode_unlock(inode, 1);
3638 cleanup_nolock:
3639         brelse(di_bh);
3640         brelse(xbs.xattr_bh);
3641         ocfs2_xattr_bucket_free(xbs.bucket);
3642
3643         return ret;
3644 }
3645
3646 /*
3647  * Find the xattr extent rec which may contains name_hash.
3648  * e_cpos will be the first name hash of the xattr rec.
3649  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3650  */
3651 static int ocfs2_xattr_get_rec(struct inode *inode,
3652                                u32 name_hash,
3653                                u64 *p_blkno,
3654                                u32 *e_cpos,
3655                                u32 *num_clusters,
3656                                struct ocfs2_extent_list *el)
3657 {
3658         int ret = 0, i;
3659         struct buffer_head *eb_bh = NULL;
3660         struct ocfs2_extent_block *eb;
3661         struct ocfs2_extent_rec *rec = NULL;
3662         u64 e_blkno = 0;
3663
3664         if (el->l_tree_depth) {
3665                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3666                                       &eb_bh);
3667                 if (ret) {
3668                         mlog_errno(ret);
3669                         goto out;
3670                 }
3671
3672                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3673                 el = &eb->h_list;
3674
3675                 if (el->l_tree_depth) {
3676                         ocfs2_error(inode->i_sb,
3677                                     "Inode %lu has non zero tree depth in "
3678                                     "xattr tree block %llu\n", inode->i_ino,
3679                                     (unsigned long long)eb_bh->b_blocknr);
3680                         ret = -EROFS;
3681                         goto out;
3682                 }
3683         }
3684
3685         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3686                 rec = &el->l_recs[i];
3687
3688                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3689                         e_blkno = le64_to_cpu(rec->e_blkno);
3690                         break;
3691                 }
3692         }
3693
3694         if (!e_blkno) {
3695                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3696                             "record (%u, %u, 0) in xattr", inode->i_ino,
3697                             le32_to_cpu(rec->e_cpos),
3698                             ocfs2_rec_clusters(el, rec));
3699                 ret = -EROFS;
3700                 goto out;
3701         }
3702
3703         *p_blkno = le64_to_cpu(rec->e_blkno);
3704         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3705         if (e_cpos)
3706                 *e_cpos = le32_to_cpu(rec->e_cpos);
3707 out:
3708         brelse(eb_bh);
3709         return ret;
3710 }
3711
3712 typedef int (xattr_bucket_func)(struct inode *inode,
3713                                 struct ocfs2_xattr_bucket *bucket,
3714                                 void *para);
3715
3716 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3717                                    struct ocfs2_xattr_bucket *bucket,
3718                                    int name_index,
3719                                    const char *name,
3720                                    u32 name_hash,
3721                                    u16 *xe_index,
3722                                    int *found)
3723 {
3724         int i, ret = 0, cmp = 1, block_off, new_offset;
3725         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3726         size_t name_len = strlen(name);
3727         struct ocfs2_xattr_entry *xe = NULL;
3728         char *xe_name;
3729
3730         /*
3731          * We don't use binary search in the bucket because there
3732          * may be multiple entries with the same name hash.
3733          */
3734         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3735                 xe = &xh->xh_entries[i];
3736
3737                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3738                         continue;
3739                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3740                         break;
3741
3742                 cmp = name_index - ocfs2_xattr_get_type(xe);
3743                 if (!cmp)
3744                         cmp = name_len - xe->xe_name_len;
3745                 if (cmp)
3746                         continue;
3747
3748                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3749                                                         xh,
3750                                                         i,
3751                                                         &block_off,
3752                                                         &new_offset);
3753                 if (ret) {
3754                         mlog_errno(ret);
3755                         break;
3756                 }
3757
3758
3759                 xe_name = bucket_block(bucket, block_off) + new_offset;
3760                 if (!memcmp(name, xe_name, name_len)) {
3761                         *xe_index = i;
3762                         *found = 1;
3763                         ret = 0;
3764                         break;
3765                 }
3766         }
3767
3768         return ret;
3769 }
3770
3771 /*
3772  * Find the specified xattr entry in a series of buckets.
3773  * This series start from p_blkno and last for num_clusters.
3774  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3775  * the num of the valid buckets.
3776  *
3777  * Return the buffer_head this xattr should reside in. And if the xattr's
3778  * hash is in the gap of 2 buckets, return the lower bucket.
3779  */
3780 static int ocfs2_xattr_bucket_find(struct inode *inode,
3781                                    int name_index,
3782                                    const char *name,
3783                                    u32 name_hash,
3784                                    u64 p_blkno,
3785                                    u32 first_hash,
3786                                    u32 num_clusters,
3787                                    struct ocfs2_xattr_search *xs)
3788 {
3789         int ret, found = 0;
3790         struct ocfs2_xattr_header *xh = NULL;
3791         struct ocfs2_xattr_entry *xe = NULL;
3792         u16 index = 0;
3793         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3794         int low_bucket = 0, bucket, high_bucket;
3795         struct ocfs2_xattr_bucket *search;
3796         u32 last_hash;
3797         u64 blkno, lower_blkno = 0;
3798
3799         search = ocfs2_xattr_bucket_new(inode);
3800         if (!search) {
3801                 ret = -ENOMEM;
3802                 mlog_errno(ret);
3803                 goto out;
3804         }
3805
3806         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3807         if (ret) {
3808                 mlog_errno(ret);
3809                 goto out;
3810         }
3811
3812         xh = bucket_xh(search);
3813         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3814         while (low_bucket <= high_bucket) {
3815                 ocfs2_xattr_bucket_relse(search);
3816
3817                 bucket = (low_bucket + high_bucket) / 2;
3818                 blkno = p_blkno + bucket * blk_per_bucket;
3819                 ret = ocfs2_read_xattr_bucket(search, blkno);
3820                 if (ret) {
3821                         mlog_errno(ret);
3822                         goto out;
3823                 }
3824
3825                 xh = bucket_xh(search);
3826                 xe = &xh->xh_entries[0];
3827                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3828                         high_bucket = bucket - 1;
3829                         continue;
3830                 }
3831
3832                 /*
3833                  * Check whether the hash of the last entry in our
3834                  * bucket is larger than the search one. for an empty
3835                  * bucket, the last one is also the first one.
3836                  */
3837                 if (xh->xh_count)
3838                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3839
3840                 last_hash = le32_to_cpu(xe->xe_name_hash);
3841
3842                 /* record lower_blkno which may be the insert place. */
3843                 lower_blkno = blkno;
3844
3845                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3846                         low_bucket = bucket + 1;
3847                         continue;
3848                 }
3849
3850                 /* the searched xattr should reside in this bucket if exists. */
3851                 ret = ocfs2_find_xe_in_bucket(inode, search,
3852                                               name_index, name, name_hash,
3853                                               &index, &found);
3854                 if (ret) {
3855                         mlog_errno(ret);
3856                         goto out;
3857                 }
3858                 break;
3859         }
3860
3861         /*
3862          * Record the bucket we have found.
3863          * When the xattr's hash value is in the gap of 2 buckets, we will
3864          * always set it to the previous bucket.
3865          */
3866         if (!lower_blkno)
3867                 lower_blkno = p_blkno;
3868
3869         /* This should be in cache - we just read it during the search */
3870         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3871         if (ret) {
3872                 mlog_errno(ret);
3873                 goto out;
3874         }
3875
3876         xs->header = bucket_xh(xs->bucket);
3877         xs->base = bucket_block(xs->bucket, 0);
3878         xs->end = xs->base + inode->i_sb->s_blocksize;
3879
3880         if (found) {
3881                 xs->here = &xs->header->xh_entries[index];
3882                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3883                         name, name_index, name_hash,
3884                         (unsigned long long)bucket_blkno(xs->bucket),
3885                         index);
3886         } else
3887                 ret = -ENODATA;
3888
3889 out:
3890         ocfs2_xattr_bucket_free(search);
3891         return ret;
3892 }
3893
3894 static int ocfs2_xattr_index_block_find(struct inode *inode,
3895                                         struct buffer_head *root_bh,
3896                                         int name_index,
3897                                         const char *name,
3898                                         struct ocfs2_xattr_search *xs)
3899 {
3900         int ret;
3901         struct ocfs2_xattr_block *xb =
3902                         (struct ocfs2_xattr_block *)root_bh->b_data;
3903         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3904         struct ocfs2_extent_list *el = &xb_root->xt_list;
3905         u64 p_blkno = 0;
3906         u32 first_hash, num_clusters = 0;
3907         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3908
3909         if (le16_to_cpu(el->l_next_free_rec) == 0)
3910                 return -ENODATA;
3911
3912         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3913                                         name, name_index, name_hash,
3914                                         (unsigned long long)root_bh->b_blocknr,
3915                                         -1);
3916
3917         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3918                                   &num_clusters, el);
3919         if (ret) {
3920                 mlog_errno(ret);
3921                 goto out;
3922         }
3923
3924         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3925
3926         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3927                                         name, name_index, first_hash,
3928                                         (unsigned long long)p_blkno,
3929                                         num_clusters);
3930
3931         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3932                                       p_blkno, first_hash, num_clusters, xs);
3933
3934 out:
3935         return ret;
3936 }
3937
3938 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3939                                        u64 blkno,
3940                                        u32 clusters,
3941                                        xattr_bucket_func *func,
3942                                        void *para)
3943 {
3944         int i, ret = 0;
3945         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3946         u32 num_buckets = clusters * bpc;
3947         struct ocfs2_xattr_bucket *bucket;
3948
3949         bucket = ocfs2_xattr_bucket_new(inode);
3950         if (!bucket) {
3951                 mlog_errno(-ENOMEM);
3952                 return -ENOMEM;
3953         }
3954
3955         trace_ocfs2_iterate_xattr_buckets(
3956                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3957                 (unsigned long long)blkno, clusters);
3958
3959         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3960                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3961                 if (ret) {
3962                         mlog_errno(ret);
3963                         break;
3964                 }
3965
3966                 /*
3967                  * The real bucket num in this series of blocks is stored
3968                  * in the 1st bucket.
3969                  */
3970                 if (i == 0)
3971                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3972
3973                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3974                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3975                 if (func) {
3976                         ret = func(inode, bucket, para);
3977                         if (ret && ret != -ERANGE)
3978                                 mlog_errno(ret);
3979                         /* Fall through to bucket_relse() */
3980                 }
3981
3982                 ocfs2_xattr_bucket_relse(bucket);
3983                 if (ret)
3984                         break;
3985         }
3986
3987         ocfs2_xattr_bucket_free(bucket);
3988         return ret;
3989 }
3990
3991 struct ocfs2_xattr_tree_list {
3992         char *buffer;
3993         size_t buffer_size;
3994         size_t result;
3995 };
3996
3997 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3998                                              struct ocfs2_xattr_header *xh,
3999                                              int index,
4000                                              int *block_off,
4001                                              int *new_offset)
4002 {
4003         u16 name_offset;
4004
4005         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4006                 return -EINVAL;
4007
4008         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4009
4010         *block_off = name_offset >> sb->s_blocksize_bits;
4011         *new_offset = name_offset % sb->s_blocksize;
4012
4013         return 0;
4014 }
4015
4016 static int ocfs2_list_xattr_bucket(struct inode *inode,
4017                                    struct ocfs2_xattr_bucket *bucket,
4018                                    void *para)
4019 {
4020         int ret = 0, type;
4021         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4022         int i, block_off, new_offset;
4023         const char *prefix, *name;
4024
4025         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4026                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4027                 type = ocfs2_xattr_get_type(entry);
4028                 prefix = ocfs2_xattr_prefix(type);
4029
4030                 if (prefix) {
4031                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4032                                                                 bucket_xh(bucket),
4033                                                                 i,
4034                                                                 &block_off,
4035                                                                 &new_offset);
4036                         if (ret)
4037                                 break;
4038
4039                         name = (const char *)bucket_block(bucket, block_off) +
4040                                 new_offset;
4041                         ret = ocfs2_xattr_list_entry(xl->buffer,
4042                                                      xl->buffer_size,
4043                                                      &xl->result,
4044                                                      prefix, name,
4045                                                      entry->xe_name_len);
4046                         if (ret)
4047                                 break;
4048                 }
4049         }
4050
4051         return ret;
4052 }
4053
4054 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4055                                            struct buffer_head *blk_bh,
4056                                            xattr_tree_rec_func *rec_func,
4057                                            void *para)
4058 {
4059         struct ocfs2_xattr_block *xb =
4060                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4061         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4062         int ret = 0;
4063         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4064         u64 p_blkno = 0;
4065
4066         if (!el->l_next_free_rec || !rec_func)
4067                 return 0;
4068
4069         while (name_hash > 0) {
4070                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4071                                           &e_cpos, &num_clusters, el);
4072                 if (ret) {
4073                         mlog_errno(ret);
4074                         break;
4075                 }
4076
4077                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4078                                num_clusters, para);
4079                 if (ret) {
4080                         if (ret != -ERANGE)
4081                                 mlog_errno(ret);
4082                         break;
4083                 }
4084
4085                 if (e_cpos == 0)
4086                         break;
4087
4088                 name_hash = e_cpos - 1;
4089         }
4090
4091         return ret;
4092
4093 }
4094
4095 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4096                                      struct buffer_head *root_bh,
4097                                      u64 blkno, u32 cpos, u32 len, void *para)
4098 {
4099         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4100                                            ocfs2_list_xattr_bucket, para);
4101 }
4102
4103 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4104                                              struct buffer_head *blk_bh,
4105                                              char *buffer,
4106                                              size_t buffer_size)
4107 {
4108         int ret;
4109         struct ocfs2_xattr_tree_list xl = {
4110                 .buffer = buffer,
4111                 .buffer_size = buffer_size,
4112                 .result = 0,
4113         };
4114
4115         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4116                                               ocfs2_list_xattr_tree_rec, &xl);
4117         if (ret) {
4118                 mlog_errno(ret);
4119                 goto out;
4120         }
4121
4122         ret = xl.result;
4123 out:
4124         return ret;
4125 }
4126
4127 static int cmp_xe(const void *a, const void *b)
4128 {
4129         const struct ocfs2_xattr_entry *l = a, *r = b;
4130         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4131         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4132
4133         if (l_hash > r_hash)
4134                 return 1;
4135         if (l_hash < r_hash)
4136                 return -1;
4137         return 0;
4138 }
4139
4140 static void swap_xe(void *a, void *b, int size)
4141 {
4142         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4143
4144         tmp = *l;
4145         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4146         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4147 }
4148
4149 /*
4150  * When the ocfs2_xattr_block is filled up, new bucket will be created
4151  * and all the xattr entries will be moved to the new bucket.
4152  * The header goes at the start of the bucket, and the names+values are
4153  * filled from the end.  This is why *target starts as the last buffer.
4154  * Note: we need to sort the entries since they are not saved in order
4155  * in the ocfs2_xattr_block.
4156  */
4157 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4158                                            struct buffer_head *xb_bh,
4159                                            struct ocfs2_xattr_bucket *bucket)
4160 {
4161         int i, blocksize = inode->i_sb->s_blocksize;
4162         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4163         u16 offset, size, off_change;
4164         struct ocfs2_xattr_entry *xe;
4165         struct ocfs2_xattr_block *xb =
4166                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4167         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4168         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4169         u16 count = le16_to_cpu(xb_xh->xh_count);
4170         char *src = xb_bh->b_data;
4171         char *target = bucket_block(bucket, blks - 1);
4172
4173         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4174                                 (unsigned long long)xb_bh->b_blocknr,
4175                                 (unsigned long long)bucket_blkno(bucket));
4176
4177         for (i = 0; i < blks; i++)
4178                 memset(bucket_block(bucket, i), 0, blocksize);
4179
4180         /*
4181          * Since the xe_name_offset is based on ocfs2_xattr_header,
4182          * there is a offset change corresponding to the change of
4183          * ocfs2_xattr_header's position.
4184          */
4185         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4186         xe = &xb_xh->xh_entries[count - 1];
4187         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4188         size = blocksize - offset;
4189
4190         /* copy all the names and values. */
4191         memcpy(target + offset, src + offset, size);
4192
4193         /* Init new header now. */
4194         xh->xh_count = xb_xh->xh_count;
4195         xh->xh_num_buckets = cpu_to_le16(1);
4196         xh->xh_name_value_len = cpu_to_le16(size);
4197         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4198
4199         /* copy all the entries. */
4200         target = bucket_block(bucket, 0);
4201         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4202         size = count * sizeof(struct ocfs2_xattr_entry);
4203         memcpy(target + offset, (char *)xb_xh + offset, size);
4204
4205         /* Change the xe offset for all the xe because of the move. */
4206         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4207                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4208         for (i = 0; i < count; i++)
4209                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4210
4211         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4212
4213         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4214              cmp_xe, swap_xe);
4215 }
4216
4217 /*
4218  * After we move xattr from block to index btree, we have to
4219  * update ocfs2_xattr_search to the new xe and base.
4220  *
4221  * When the entry is in xattr block, xattr_bh indicates the storage place.
4222  * While if the entry is in index b-tree, "bucket" indicates the
4223  * real place of the xattr.
4224  */
4225 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4226                                             struct ocfs2_xattr_search *xs,
4227                                             struct buffer_head *old_bh)
4228 {
4229         char *buf = old_bh->b_data;
4230         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4231         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4232         int i;
4233
4234         xs->header = bucket_xh(xs->bucket);
4235         xs->base = bucket_block(xs->bucket, 0);
4236         xs->end = xs->base + inode->i_sb->s_blocksize;
4237
4238         if (xs->not_found)
4239                 return;
4240
4241         i = xs->here - old_xh->xh_entries;
4242         xs->here = &xs->header->xh_entries[i];
4243 }
4244
4245 static int ocfs2_xattr_create_index_block(struct inode *inode,
4246                                           struct ocfs2_xattr_search *xs,
4247                                           struct ocfs2_xattr_set_ctxt *ctxt)
4248 {
4249         int ret;
4250         u32 bit_off, len;
4251         u64 blkno;
4252         handle_t *handle = ctxt->handle;
4253         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4254         struct buffer_head *xb_bh = xs->xattr_bh;
4255         struct ocfs2_xattr_block *xb =
4256                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4257         struct ocfs2_xattr_tree_root *xr;
4258         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4259
4260         trace_ocfs2_xattr_create_index_block_begin(
4261                                 (unsigned long long)xb_bh->b_blocknr);
4262
4263         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4264         BUG_ON(!xs->bucket);
4265
4266         /*
4267          * XXX:
4268          * We can use this lock for now, and maybe move to a dedicated mutex
4269          * if performance becomes a problem later.
4270          */
4271         down_write(&oi->ip_alloc_sem);
4272
4273         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4274                                       OCFS2_JOURNAL_ACCESS_WRITE);
4275         if (ret) {
4276                 mlog_errno(ret);
4277                 goto out;
4278         }
4279
4280         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4281                                      1, 1, &bit_off, &len);
4282         if (ret) {
4283                 mlog_errno(ret);
4284                 goto out;
4285         }
4286
4287         /*
4288          * The bucket may spread in many blocks, and
4289          * we will only touch the 1st block and the last block
4290          * in the whole bucket(one for entry and one for data).
4291          */
4292         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4293
4294         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4295
4296         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4297         if (ret) {
4298                 mlog_errno(ret);
4299                 goto out;
4300         }
4301
4302         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4303                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4304         if (ret) {
4305                 mlog_errno(ret);
4306                 goto out;
4307         }
4308
4309         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4310         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4311
4312         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4313
4314         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4315         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4316                offsetof(struct ocfs2_xattr_block, xb_attrs));
4317
4318         xr = &xb->xb_attrs.xb_root;
4319         xr->xt_clusters = cpu_to_le32(1);
4320         xr->xt_last_eb_blk = 0;
4321         xr->xt_list.l_tree_depth = 0;
4322         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4323         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4324
4325         xr->xt_list.l_recs[0].e_cpos = 0;
4326         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4327         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4328
4329         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4330
4331         ocfs2_journal_dirty(handle, xb_bh);
4332
4333 out:
4334         up_write(&oi->ip_alloc_sem);
4335
4336         return ret;
4337 }
4338
4339 static int cmp_xe_offset(const void *a, const void *b)
4340 {
4341         const struct ocfs2_xattr_entry *l = a, *r = b;
4342         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4343         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4344
4345         if (l_name_offset < r_name_offset)
4346                 return 1;
4347         if (l_name_offset > r_name_offset)
4348                 return -1;
4349         return 0;
4350 }
4351
4352 /*
4353  * defrag a xattr bucket if we find that the bucket has some
4354  * holes beteen name/value pairs.
4355  * We will move all the name/value pairs to the end of the bucket
4356  * so that we can spare some space for insertion.
4357  */
4358 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4359                                      handle_t *handle,
4360                                      struct ocfs2_xattr_bucket *bucket)
4361 {
4362         int ret, i;
4363         size_t end, offset, len;
4364         struct ocfs2_xattr_header *xh;
4365         char *entries, *buf, *bucket_buf = NULL;
4366         u64 blkno = bucket_blkno(bucket);
4367         u16 xh_free_start;
4368         size_t blocksize = inode->i_sb->s_blocksize;
4369         struct ocfs2_xattr_entry *xe;
4370
4371         /*
4372          * In order to make the operation more efficient and generic,
4373          * we copy all the blocks into a contiguous memory and do the
4374          * defragment there, so if anything is error, we will not touch
4375          * the real block.
4376          */
4377         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4378         if (!bucket_buf) {
4379                 ret = -EIO;
4380                 goto out;
4381         }
4382
4383         buf = bucket_buf;
4384         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4385                 memcpy(buf, bucket_block(bucket, i), blocksize);
4386
4387         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4388                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4389         if (ret < 0) {
4390                 mlog_errno(ret);
4391                 goto out;
4392         }
4393
4394         xh = (struct ocfs2_xattr_header *)bucket_buf;
4395         entries = (char *)xh->xh_entries;
4396         xh_free_start = le16_to_cpu(xh->xh_free_start);
4397
4398         trace_ocfs2_defrag_xattr_bucket(
4399              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4400              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4401
4402         /*
4403          * sort all the entries by their offset.
4404          * the largest will be the first, so that we can
4405          * move them to the end one by one.
4406          */
4407         sort(entries, le16_to_cpu(xh->xh_count),
4408              sizeof(struct ocfs2_xattr_entry),
4409              cmp_xe_offset, swap_xe);
4410
4411         /* Move all name/values to the end of the bucket. */
4412         xe = xh->xh_entries;
4413         end = OCFS2_XATTR_BUCKET_SIZE;
4414         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4415                 offset = le16_to_cpu(xe->xe_name_offset);
4416                 len = namevalue_size_xe(xe);
4417
4418                 /*
4419                  * We must make sure that the name/value pair
4420                  * exist in the same block. So adjust end to
4421                  * the previous block end if needed.
4422                  */
4423                 if (((end - len) / blocksize !=
4424                         (end - 1) / blocksize))
4425                         end = end - end % blocksize;
4426
4427                 if (end > offset + len) {
4428                         memmove(bucket_buf + end - len,
4429                                 bucket_buf + offset, len);
4430                         xe->xe_name_offset = cpu_to_le16(end - len);
4431                 }
4432
4433                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4434                                 "bucket %llu\n", (unsigned long long)blkno);
4435
4436                 end -= len;
4437         }
4438
4439         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4440                         "bucket %llu\n", (unsigned long long)blkno);
4441
4442         if (xh_free_start == end)
4443                 goto out;
4444
4445         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4446         xh->xh_free_start = cpu_to_le16(end);
4447
4448         /* sort the entries by their name_hash. */
4449         sort(entries, le16_to_cpu(xh->xh_count),
4450              sizeof(struct ocfs2_xattr_entry),
4451              cmp_xe, swap_xe);
4452
4453         buf = bucket_buf;
4454         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4455                 memcpy(bucket_block(bucket, i), buf, blocksize);
4456         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4457
4458 out:
4459         kfree(bucket_buf);
4460         return ret;
4461 }
4462
4463 /*
4464  * prev_blkno points to the start of an existing extent.  new_blkno
4465  * points to a newly allocated extent.  Because we know each of our
4466  * clusters contains more than bucket, we can easily split one cluster
4467  * at a bucket boundary.  So we take the last cluster of the existing
4468  * extent and split it down the middle.  We move the last half of the
4469  * buckets in the last cluster of the existing extent over to the new
4470  * extent.
4471  *
4472  * first_bh is the buffer at prev_blkno so we can update the existing
4473  * extent's bucket count.  header_bh is the bucket were we were hoping
4474  * to insert our xattr.  If the bucket move places the target in the new
4475  * extent, we'll update first_bh and header_bh after modifying the old
4476  * extent.
4477  *
4478  * first_hash will be set as the 1st xe's name_hash in the new extent.
4479  */
4480 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4481                                                handle_t *handle,
4482                                                struct ocfs2_xattr_bucket *first,
4483                                                struct ocfs2_xattr_bucket *target,
4484                                                u64 new_blkno,
4485                                                u32 num_clusters,
4486                                                u32 *first_hash)
4487 {
4488         int ret;
4489         struct super_block *sb = inode->i_sb;
4490         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4491         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4492         int to_move = num_buckets / 2;
4493         u64 src_blkno;
4494         u64 last_cluster_blkno = bucket_blkno(first) +
4495                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4496
4497         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4498         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4499
4500         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4501                                 (unsigned long long)last_cluster_blkno,
4502                                 (unsigned long long)new_blkno);
4503
4504         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4505                                      last_cluster_blkno, new_blkno,
4506                                      to_move, first_hash);
4507         if (ret) {
4508                 mlog_errno(ret);
4509                 goto out;
4510         }
4511
4512         /* This is the first bucket that got moved */
4513         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4514
4515         /*
4516          * If the target bucket was part of the moved buckets, we need to
4517          * update first and target.
4518          */
4519         if (bucket_blkno(target) >= src_blkno) {
4520                 /* Find the block for the new target bucket */
4521                 src_blkno = new_blkno +
4522                         (bucket_blkno(target) - src_blkno);
4523
4524                 ocfs2_xattr_bucket_relse(first);
4525                 ocfs2_xattr_bucket_relse(target);
4526
4527                 /*
4528                  * These shouldn't fail - the buffers are in the
4529                  * journal from ocfs2_cp_xattr_bucket().
4530                  */
4531                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4532                 if (ret) {
4533                         mlog_errno(ret);
4534                         goto out;
4535                 }
4536                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4537                 if (ret)
4538                         mlog_errno(ret);
4539
4540         }
4541
4542 out:
4543         return ret;
4544 }
4545
4546 /*
4547  * Find the suitable pos when we divide a bucket into 2.
4548  * We have to make sure the xattrs with the same hash value exist
4549  * in the same bucket.
4550  *
4551  * If this ocfs2_xattr_header covers more than one hash value, find a
4552  * place where the hash value changes.  Try to find the most even split.
4553  * The most common case is that all entries have different hash values,
4554  * and the first check we make will find a place to split.
4555  */
4556 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4557 {
4558         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4559         int count = le16_to_cpu(xh->xh_count);
4560         int delta, middle = count / 2;
4561
4562         /*
4563          * We start at the middle.  Each step gets farther away in both
4564          * directions.  We therefore hit the change in hash value
4565          * nearest to the middle.  Note that this loop does not execute for
4566          * count < 2.
4567          */
4568         for (delta = 0; delta < middle; delta++) {
4569                 /* Let's check delta earlier than middle */
4570                 if (cmp_xe(&entries[middle - delta - 1],
4571                            &entries[middle - delta]))
4572                         return middle - delta;
4573
4574                 /* For even counts, don't walk off the end */
4575                 if ((middle + delta + 1) == count)
4576                         continue;
4577
4578                 /* Now try delta past middle */
4579                 if (cmp_xe(&entries[middle + delta],
4580                            &entries[middle + delta + 1]))
4581                         return middle + delta + 1;
4582         }
4583
4584         /* Every entry had the same hash */
4585         return count;
4586 }
4587
4588 /*
4589  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4590  * first_hash will record the 1st hash of the new bucket.
4591  *
4592  * Normally half of the xattrs will be moved.  But we have to make
4593  * sure that the xattrs with the same hash value are stored in the
4594  * same bucket. If all the xattrs in this bucket have the same hash
4595  * value, the new bucket will be initialized as an empty one and the
4596  * first_hash will be initialized as (hash_value+1).
4597  */
4598 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4599                                     handle_t *handle,
4600                                     u64 blk,
4601                                     u64 new_blk,
4602                                     u32 *first_hash,
4603                                     int new_bucket_head)
4604 {
4605         int ret, i;
4606         int count, start, len, name_value_len = 0, name_offset = 0;
4607         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4608         struct ocfs2_xattr_header *xh;
4609         struct ocfs2_xattr_entry *xe;
4610         int blocksize = inode->i_sb->s_blocksize;
4611
4612         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4613                                               (unsigned long long)new_blk);
4614
4615         s_bucket = ocfs2_xattr_bucket_new(inode);
4616         t_bucket = ocfs2_xattr_bucket_new(inode);
4617         if (!s_bucket || !t_bucket) {
4618                 ret = -ENOMEM;
4619                 mlog_errno(ret);
4620                 goto out;
4621         }
4622
4623         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4624         if (ret) {
4625                 mlog_errno(ret);
4626                 goto out;
4627         }
4628
4629         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4630                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4631         if (ret) {
4632                 mlog_errno(ret);
4633                 goto out;
4634         }
4635
4636         /*
4637          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4638          * there's no need to read it.
4639          */
4640         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4641         if (ret) {
4642                 mlog_errno(ret);
4643                 goto out;
4644         }
4645
4646         /*
4647          * Hey, if we're overwriting t_bucket, what difference does
4648          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4649          * same part of ocfs2_cp_xattr_bucket().
4650          */
4651         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4652                                                 new_bucket_head ?
4653                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4654                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4655         if (ret) {
4656                 mlog_errno(ret);
4657                 goto out;
4658         }
4659
4660         xh = bucket_xh(s_bucket);
4661         count = le16_to_cpu(xh->xh_count);
4662         start = ocfs2_xattr_find_divide_pos(xh);
4663
4664         if (start == count) {
4665                 xe = &xh->xh_entries[start-1];
4666
4667                 /*
4668                  * initialized a new empty bucket here.
4669                  * The hash value is set as one larger than
4670                  * that of the last entry in the previous bucket.
4671                  */
4672                 for (i = 0; i < t_bucket->bu_blocks; i++)
4673                         memset(bucket_block(t_bucket, i), 0, blocksize);
4674
4675                 xh = bucket_xh(t_bucket);
4676                 xh->xh_free_start = cpu_to_le16(blocksize);
4677                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4678                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4679
4680                 goto set_num_buckets;
4681         }
4682
4683         /* copy the whole bucket to the new first. */
4684         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4685
4686         /* update the new bucket. */
4687         xh = bucket_xh(t_bucket);
4688
4689         /*
4690          * Calculate the total name/value len and xh_free_start for
4691          * the old bucket first.
4692          */
4693         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4694         name_value_len = 0;
4695         for (i = 0; i < start; i++) {
4696                 xe = &xh->xh_entries[i];
4697                 name_value_len += namevalue_size_xe(xe);
4698                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4699                         name_offset = le16_to_cpu(xe->xe_name_offset);
4700         }
4701
4702         /*
4703          * Now begin the modification to the new bucket.
4704          *
4705          * In the new bucket, We just move the xattr entry to the beginning
4706          * and don't touch the name/value. So there will be some holes in the
4707          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4708          * called.
4709          */
4710         xe = &xh->xh_entries[start];
4711         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4712         trace_ocfs2_divide_xattr_bucket_move(len,
4713                         (int)((char *)xe - (char *)xh),
4714                         (int)((char *)xh->xh_entries - (char *)xh));
4715         memmove((char *)xh->xh_entries, (char *)xe, len);
4716         xe = &xh->xh_entries[count - start];
4717         len = sizeof(struct ocfs2_xattr_entry) * start;
4718         memset((char *)xe, 0, len);
4719
4720         le16_add_cpu(&xh->xh_count, -start);
4721         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4722
4723         /* Calculate xh_free_start for the new bucket. */
4724         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4725         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4726                 xe = &xh->xh_entries[i];
4727                 if (le16_to_cpu(xe->xe_name_offset) <
4728                     le16_to_cpu(xh->xh_free_start))
4729                         xh->xh_free_start = xe->xe_name_offset;
4730         }
4731
4732 set_num_buckets:
4733         /* set xh->xh_num_buckets for the new xh. */
4734         if (new_bucket_head)
4735                 xh->xh_num_buckets = cpu_to_le16(1);
4736         else
4737                 xh->xh_num_buckets = 0;
4738
4739         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4740
4741         /* store the first_hash of the new bucket. */
4742         if (first_hash)
4743                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4744
4745         /*
4746          * Now only update the 1st block of the old bucket.  If we
4747          * just added a new empty bucket, there is no need to modify
4748          * it.
4749          */
4750         if (start == count)
4751                 goto out;
4752
4753         xh = bucket_xh(s_bucket);
4754         memset(&xh->xh_entries[start], 0,
4755                sizeof(struct ocfs2_xattr_entry) * (count - start));
4756         xh->xh_count = cpu_to_le16(start);
4757         xh->xh_free_start = cpu_to_le16(name_offset);
4758         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4759
4760         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4761
4762 out:
4763         ocfs2_xattr_bucket_free(s_bucket);
4764         ocfs2_xattr_bucket_free(t_bucket);
4765
4766         return ret;
4767 }
4768
4769 /*
4770  * Copy xattr from one bucket to another bucket.
4771  *
4772  * The caller must make sure that the journal transaction
4773  * has enough space for journaling.
4774  */
4775 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4776                                  handle_t *handle,
4777                                  u64 s_blkno,
4778                                  u64 t_blkno,
4779                                  int t_is_new)
4780 {
4781         int ret;
4782         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4783
4784         BUG_ON(s_blkno == t_blkno);
4785
4786         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4787                                     (unsigned long long)t_blkno,
4788                                     t_is_new);
4789
4790         s_bucket = ocfs2_xattr_bucket_new(inode);
4791         t_bucket = ocfs2_xattr_bucket_new(inode);
4792         if (!s_bucket || !t_bucket) {
4793                 ret = -ENOMEM;
4794                 mlog_errno(ret);
4795                 goto out;
4796         }
4797
4798         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4799         if (ret)
4800                 goto out;
4801
4802         /*
4803          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4804          * there's no need to read it.
4805          */
4806         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4807         if (ret)
4808                 goto out;
4809
4810         /*
4811          * Hey, if we're overwriting t_bucket, what difference does
4812          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4813          * cluster to fill, we came here from
4814          * ocfs2_mv_xattr_buckets(), and it is really new -
4815          * ACCESS_CREATE is required.  But we also might have moved data
4816          * out of t_bucket before extending back into it.
4817          * ocfs2_add_new_xattr_bucket() can do this - its call to
4818          * ocfs2_add_new_xattr_cluster() may have created a new extent
4819          * and copied out the end of the old extent.  Then it re-extends
4820          * the old extent back to create space for new xattrs.  That's
4821          * how we get here, and the bucket isn't really new.
4822          */
4823         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4824                                                 t_is_new ?
4825                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4826                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4827         if (ret)
4828                 goto out;
4829
4830         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4831         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4832
4833 out:
4834         ocfs2_xattr_bucket_free(t_bucket);
4835         ocfs2_xattr_bucket_free(s_bucket);
4836
4837         return ret;
4838 }
4839
4840 /*
4841  * src_blk points to the start of an existing extent.  last_blk points to
4842  * last cluster in that extent.  to_blk points to a newly allocated
4843  * extent.  We copy the buckets from the cluster at last_blk to the new
4844  * extent.  If start_bucket is non-zero, we skip that many buckets before
4845  * we start copying.  The new extent's xh_num_buckets gets set to the
4846  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4847  * by the same amount.
4848  */
4849 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4850                                   u64 src_blk, u64 last_blk, u64 to_blk,
4851                                   unsigned int start_bucket,
4852                                   u32 *first_hash)
4853 {
4854         int i, ret, credits;
4855         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4856         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4857         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4858         struct ocfs2_xattr_bucket *old_first, *new_first;
4859
4860         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4861                                      (unsigned long long)to_blk);
4862
4863         BUG_ON(start_bucket >= num_buckets);
4864         if (start_bucket) {
4865                 num_buckets -= start_bucket;
4866                 last_blk += (start_bucket * blks_per_bucket);
4867         }
4868
4869         /* The first bucket of the original extent */
4870         old_first = ocfs2_xattr_bucket_new(inode);
4871         /* The first bucket of the new extent */
4872         new_first = ocfs2_xattr_bucket_new(inode);
4873         if (!old_first || !new_first) {
4874                 ret = -ENOMEM;
4875                 mlog_errno(ret);
4876                 goto out;
4877         }
4878
4879         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4880         if (ret) {
4881                 mlog_errno(ret);
4882                 goto out;
4883         }
4884
4885         /*
4886          * We need to update the first bucket of the old extent and all
4887          * the buckets going to the new extent.
4888          */
4889         credits = ((num_buckets + 1) * blks_per_bucket);
4890         ret = ocfs2_extend_trans(handle, credits);
4891         if (ret) {
4892                 mlog_errno(ret);
4893                 goto out;
4894         }
4895
4896         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4897                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4898         if (ret) {
4899                 mlog_errno(ret);
4900                 goto out;
4901         }
4902
4903         for (i = 0; i < num_buckets; i++) {
4904                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4905                                             last_blk + (i * blks_per_bucket),
4906                                             to_blk + (i * blks_per_bucket),
4907                                             1);
4908                 if (ret) {
4909                         mlog_errno(ret);
4910                         goto out;
4911                 }
4912         }
4913
4914         /*
4915          * Get the new bucket ready before we dirty anything
4916          * (This actually shouldn't fail, because we already dirtied
4917          * it once in ocfs2_cp_xattr_bucket()).
4918          */
4919         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4920         if (ret) {
4921                 mlog_errno(ret);
4922                 goto out;
4923         }
4924         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4925                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4926         if (ret) {
4927                 mlog_errno(ret);
4928                 goto out;
4929         }
4930
4931         /* Now update the headers */
4932         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4933         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4934
4935         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4936         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4937
4938         if (first_hash)
4939                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4940
4941 out:
4942         ocfs2_xattr_bucket_free(new_first);
4943         ocfs2_xattr_bucket_free(old_first);
4944         return ret;
4945 }
4946
4947 /*
4948  * Move some xattrs in this cluster to the new cluster.
4949  * This function should only be called when bucket size == cluster size.
4950  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4951  */
4952 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4953                                       handle_t *handle,
4954                                       u64 prev_blk,
4955                                       u64 new_blk,
4956                                       u32 *first_hash)
4957 {
4958         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4959         int ret, credits = 2 * blk_per_bucket;
4960
4961         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4962
4963         ret = ocfs2_extend_trans(handle, credits);
4964         if (ret) {
4965                 mlog_errno(ret);
4966                 return ret;
4967         }
4968
4969         /* Move half of the xattr in start_blk to the next bucket. */
4970         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4971                                           new_blk, first_hash, 1);
4972 }
4973
4974 /*
4975  * Move some xattrs from the old cluster to the new one since they are not
4976  * contiguous in ocfs2 xattr tree.
4977  *
4978  * new_blk starts a new separate cluster, and we will move some xattrs from
4979  * prev_blk to it. v_start will be set as the first name hash value in this
4980  * new cluster so that it can be used as e_cpos during tree insertion and
4981  * don't collide with our original b-tree operations. first_bh and header_bh
4982  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4983  * to extend the insert bucket.
4984  *
4985  * The problem is how much xattr should we move to the new one and when should
4986  * we update first_bh and header_bh?
4987  * 1. If cluster size > bucket size, that means the previous cluster has more
4988  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4989  *    update the first_bh and header_bh if the insert bucket has been moved
4990  *    to the new cluster.
4991  * 2. If cluster_size == bucket_size:
4992  *    a) If the previous extent rec has more than one cluster and the insert
4993  *       place isn't in the last cluster, copy the entire last cluster to the
4994  *       new one. This time, we don't need to upate the first_bh and header_bh
4995  *       since they will not be moved into the new cluster.
4996  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4997  *       the new one. And we set the extend flag to zero if the insert place is
4998  *       moved into the new allocated cluster since no extend is needed.
4999  */
5000 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5001                                             handle_t *handle,
5002                                             struct ocfs2_xattr_bucket *first,
5003                                             struct ocfs2_xattr_bucket *target,
5004                                             u64 new_blk,
5005                                             u32 prev_clusters,
5006                                             u32 *v_start,
5007                                             int *extend)
5008 {
5009         int ret;
5010
5011         trace_ocfs2_adjust_xattr_cross_cluster(
5012                         (unsigned long long)bucket_blkno(first),
5013                         (unsigned long long)new_blk, prev_clusters);
5014
5015         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5016                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5017                                                           handle,
5018                                                           first, target,
5019                                                           new_blk,
5020                                                           prev_clusters,
5021                                                           v_start);
5022                 if (ret)
5023                         mlog_errno(ret);
5024         } else {
5025                 /* The start of the last cluster in the first extent */
5026                 u64 last_blk = bucket_blkno(first) +
5027                         ((prev_clusters - 1) *
5028                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5029
5030                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5031                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5032                                                      bucket_blkno(first),
5033                                                      last_blk, new_blk, 0,
5034                                                      v_start);
5035                         if (ret)
5036                                 mlog_errno(ret);
5037                 } else {
5038                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5039                                                          last_blk, new_blk,
5040                                                          v_start);
5041                         if (ret)
5042                                 mlog_errno(ret);
5043
5044                         if ((bucket_blkno(target) == last_blk) && extend)
5045                                 *extend = 0;
5046                 }
5047         }
5048
5049         return ret;
5050 }
5051
5052 /*
5053  * Add a new cluster for xattr storage.
5054  *
5055  * If the new cluster is contiguous with the previous one, it will be
5056  * appended to the same extent record, and num_clusters will be updated.
5057  * If not, we will insert a new extent for it and move some xattrs in
5058  * the last cluster into the new allocated one.
5059  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5060  * lose the benefits of hashing because we'll have to search large leaves.
5061  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5062  * if it's bigger).
5063  *
5064  * first_bh is the first block of the previous extent rec and header_bh
5065  * indicates the bucket we will insert the new xattrs. They will be updated
5066  * when the header_bh is moved into the new cluster.
5067  */
5068 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5069                                        struct buffer_head *root_bh,
5070                                        struct ocfs2_xattr_bucket *first,
5071                                        struct ocfs2_xattr_bucket *target,
5072                                        u32 *num_clusters,
5073                                        u32 prev_cpos,
5074                                        int *extend,
5075                                        struct ocfs2_xattr_set_ctxt *ctxt)
5076 {
5077         int ret;
5078         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5079         u32 prev_clusters = *num_clusters;
5080         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5081         u64 block;
5082         handle_t *handle = ctxt->handle;
5083         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5084         struct ocfs2_extent_tree et;
5085
5086         trace_ocfs2_add_new_xattr_cluster_begin(
5087                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5088                 (unsigned long long)bucket_blkno(first),
5089                 prev_cpos, prev_clusters);
5090
5091         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5092
5093         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5094                                       OCFS2_JOURNAL_ACCESS_WRITE);
5095         if (ret < 0) {
5096                 mlog_errno(ret);
5097                 goto leave;
5098         }
5099
5100         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5101                                      clusters_to_add, &bit_off, &num_bits);
5102         if (ret < 0) {
5103                 if (ret != -ENOSPC)
5104                         mlog_errno(ret);
5105                 goto leave;
5106         }
5107
5108         BUG_ON(num_bits > clusters_to_add);
5109
5110         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5111         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5112
5113         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5114             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5115              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5116                 /*
5117                  * If this cluster is contiguous with the old one and
5118                  * adding this new cluster, we don't surpass the limit of
5119                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5120                  * initialized and used like other buckets in the previous
5121                  * cluster.
5122                  * So add it as a contiguous one. The caller will handle
5123                  * its init process.
5124                  */
5125                 v_start = prev_cpos + prev_clusters;
5126                 *num_clusters = prev_clusters + num_bits;
5127         } else {
5128                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5129                                                        handle,
5130                                                        first,
5131                                                        target,
5132                                                        block,
5133                                                        prev_clusters,
5134                                                        &v_start,
5135                                                        extend);
5136                 if (ret) {
5137                         mlog_errno(ret);
5138                         goto leave;
5139                 }
5140         }
5141
5142         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5143                                                  v_start, num_bits);
5144         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5145                                   num_bits, 0, ctxt->meta_ac);
5146         if (ret < 0) {
5147                 mlog_errno(ret);
5148                 goto leave;
5149         }
5150
5151         ocfs2_journal_dirty(handle, root_bh);
5152
5153 leave:
5154         return ret;
5155 }
5156
5157 /*
5158  * We are given an extent.  'first' is the bucket at the very front of
5159  * the extent.  The extent has space for an additional bucket past
5160  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5161  * of the target bucket.  We wish to shift every bucket past the target
5162  * down one, filling in that additional space.  When we get back to the
5163  * target, we split the target between itself and the now-empty bucket
5164  * at target+1 (aka, target_blkno + blks_per_bucket).
5165  */
5166 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5167                                      handle_t *handle,
5168                                      struct ocfs2_xattr_bucket *first,
5169                                      u64 target_blk,
5170                                      u32 num_clusters)
5171 {
5172         int ret, credits;
5173         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5174         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5175         u64 end_blk;
5176         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5177
5178         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5179                                         (unsigned long long)bucket_blkno(first),
5180                                         num_clusters, new_bucket);
5181
5182         /* The extent must have room for an additional bucket */
5183         BUG_ON(new_bucket >=
5184                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5185
5186         /* end_blk points to the last existing bucket */
5187         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5188
5189         /*
5190          * end_blk is the start of the last existing bucket.
5191          * Thus, (end_blk - target_blk) covers the target bucket and
5192          * every bucket after it up to, but not including, the last
5193          * existing bucket.  Then we add the last existing bucket, the
5194          * new bucket, and the first bucket (3 * blk_per_bucket).
5195          */
5196         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5197         ret = ocfs2_extend_trans(handle, credits);
5198         if (ret) {
5199                 mlog_errno(ret);
5200                 goto out;
5201         }
5202
5203         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5204                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5205         if (ret) {
5206                 mlog_errno(ret);
5207                 goto out;
5208         }
5209
5210         while (end_blk != target_blk) {
5211                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5212                                             end_blk + blk_per_bucket, 0);
5213                 if (ret)
5214                         goto out;
5215                 end_blk -= blk_per_bucket;
5216         }
5217
5218         /* Move half of the xattr in target_blkno to the next bucket. */
5219         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5220                                         target_blk + blk_per_bucket, NULL, 0);
5221
5222         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5223         ocfs2_xattr_bucket_journal_dirty(handle, first);
5224
5225 out:
5226         return ret;
5227 }
5228
5229 /*
5230  * Add new xattr bucket in an extent record and adjust the buckets
5231  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5232  * bucket we want to insert into.
5233  *
5234  * In the easy case, we will move all the buckets after target down by
5235  * one. Half of target's xattrs will be moved to the next bucket.
5236  *
5237  * If current cluster is full, we'll allocate a new one.  This may not
5238  * be contiguous.  The underlying calls will make sure that there is
5239  * space for the insert, shifting buckets around if necessary.
5240  * 'target' may be moved by those calls.
5241  */
5242 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5243                                       struct buffer_head *xb_bh,
5244                                       struct ocfs2_xattr_bucket *target,
5245                                       struct ocfs2_xattr_set_ctxt *ctxt)
5246 {
5247         struct ocfs2_xattr_block *xb =
5248                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5249         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5250         struct ocfs2_extent_list *el = &xb_root->xt_list;
5251         u32 name_hash =
5252                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5253         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5254         int ret, num_buckets, extend = 1;
5255         u64 p_blkno;
5256         u32 e_cpos, num_clusters;
5257         /* The bucket at the front of the extent */
5258         struct ocfs2_xattr_bucket *first;
5259
5260         trace_ocfs2_add_new_xattr_bucket(
5261                                 (unsigned long long)bucket_blkno(target));
5262
5263         /* The first bucket of the original extent */
5264         first = ocfs2_xattr_bucket_new(inode);
5265         if (!first) {
5266                 ret = -ENOMEM;
5267                 mlog_errno(ret);
5268                 goto out;
5269         }
5270
5271         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5272                                   &num_clusters, el);
5273         if (ret) {
5274                 mlog_errno(ret);
5275                 goto out;
5276         }
5277
5278         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5279         if (ret) {
5280                 mlog_errno(ret);
5281                 goto out;
5282         }
5283
5284         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5285         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5286                 /*
5287                  * This can move first+target if the target bucket moves
5288                  * to the new extent.
5289                  */
5290                 ret = ocfs2_add_new_xattr_cluster(inode,
5291                                                   xb_bh,
5292                                                   first,
5293                                                   target,
5294                                                   &num_clusters,
5295                                                   e_cpos,
5296                                                   &extend,
5297                                                   ctxt);
5298                 if (ret) {
5299                         mlog_errno(ret);
5300                         goto out;
5301                 }
5302         }
5303
5304         if (extend) {
5305                 ret = ocfs2_extend_xattr_bucket(inode,
5306                                                 ctxt->handle,
5307                                                 first,
5308                                                 bucket_blkno(target),
5309                                                 num_clusters);
5310                 if (ret)
5311                         mlog_errno(ret);
5312         }
5313
5314 out:
5315         ocfs2_xattr_bucket_free(first);
5316
5317         return ret;
5318 }
5319
5320 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5321                                         struct ocfs2_xattr_bucket *bucket,
5322                                         int offs)
5323 {
5324         int block_off = offs >> inode->i_sb->s_blocksize_bits;
5325
5326         offs = offs % inode->i_sb->s_blocksize;
5327         return bucket_block(bucket, block_off) + offs;
5328 }
5329
5330 /*
5331  * Truncate the specified xe_off entry in xattr bucket.
5332  * bucket is indicated by header_bh and len is the new length.
5333  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5334  *
5335  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5336  */
5337 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5338                                              struct ocfs2_xattr_bucket *bucket,
5339                                              int xe_off,
5340                                              int len,
5341                                              struct ocfs2_xattr_set_ctxt *ctxt)
5342 {
5343         int ret, offset;
5344         u64 value_blk;
5345         struct ocfs2_xattr_entry *xe;
5346         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5347         size_t blocksize = inode->i_sb->s_blocksize;
5348         struct ocfs2_xattr_value_buf vb = {
5349                 .vb_access = ocfs2_journal_access,
5350         };
5351
5352         xe = &xh->xh_entries[xe_off];
5353
5354         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5355
5356         offset = le16_to_cpu(xe->xe_name_offset) +
5357                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5358
5359         value_blk = offset / blocksize;
5360
5361         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5362         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5363
5364         vb.vb_bh = bucket->bu_bhs[value_blk];
5365         BUG_ON(!vb.vb_bh);
5366
5367         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5368                 (vb.vb_bh->b_data + offset % blocksize);
5369
5370         /*
5371          * From here on out we have to dirty the bucket.  The generic
5372          * value calls only modify one of the bucket's bhs, but we need
5373          * to send the bucket at once.  So if they error, they *could* have
5374          * modified something.  We have to assume they did, and dirty
5375          * the whole bucket.  This leaves us in a consistent state.
5376          */
5377         trace_ocfs2_xattr_bucket_value_truncate(
5378                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5379         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5380         if (ret) {
5381                 mlog_errno(ret);
5382                 goto out;
5383         }
5384
5385         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5386                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5387         if (ret) {
5388                 mlog_errno(ret);
5389                 goto out;
5390         }
5391
5392         xe->xe_value_size = cpu_to_le64(len);
5393
5394         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5395
5396 out:
5397         return ret;
5398 }
5399
5400 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5401                                   struct buffer_head *root_bh,
5402                                   u64 blkno,
5403                                   u32 cpos,
5404                                   u32 len,
5405                                   void *para)
5406 {
5407         int ret;
5408         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5409         struct inode *tl_inode = osb->osb_tl_inode;
5410         handle_t *handle;
5411         struct ocfs2_xattr_block *xb =
5412                         (struct ocfs2_xattr_block *)root_bh->b_data;
5413         struct ocfs2_alloc_context *meta_ac = NULL;
5414         struct ocfs2_cached_dealloc_ctxt dealloc;
5415         struct ocfs2_extent_tree et;
5416
5417         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5418                                           ocfs2_delete_xattr_in_bucket, para);
5419         if (ret) {
5420                 mlog_errno(ret);
5421                 return ret;
5422         }
5423
5424         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5425
5426         ocfs2_init_dealloc_ctxt(&dealloc);
5427
5428         trace_ocfs2_rm_xattr_cluster(
5429                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5430                         (unsigned long long)blkno, cpos, len);
5431
5432         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5433                                                len);
5434
5435         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5436         if (ret) {
5437                 mlog_errno(ret);
5438                 return ret;
5439         }
5440
5441         mutex_lock(&tl_inode->i_mutex);
5442
5443         if (ocfs2_truncate_log_needs_flush(osb)) {
5444                 ret = __ocfs2_flush_truncate_log(osb);
5445                 if (ret < 0) {
5446                         mlog_errno(ret);
5447                         goto out;
5448                 }
5449         }
5450
5451         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5452         if (IS_ERR(handle)) {
5453                 ret = -ENOMEM;
5454                 mlog_errno(ret);
5455                 goto out;
5456         }
5457
5458         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5459                                       OCFS2_JOURNAL_ACCESS_WRITE);
5460         if (ret) {
5461                 mlog_errno(ret);
5462                 goto out_commit;
5463         }
5464
5465         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5466                                   &dealloc);
5467         if (ret) {
5468                 mlog_errno(ret);
5469                 goto out_commit;
5470         }
5471
5472         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5473         ocfs2_journal_dirty(handle, root_bh);
5474
5475         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5476         if (ret)
5477                 mlog_errno(ret);
5478
5479 out_commit:
5480         ocfs2_commit_trans(osb, handle);
5481 out:
5482         ocfs2_schedule_truncate_log_flush(osb, 1);
5483
5484         mutex_unlock(&tl_inode->i_mutex);
5485
5486         if (meta_ac)
5487                 ocfs2_free_alloc_context(meta_ac);
5488
5489         ocfs2_run_deallocs(osb, &dealloc);
5490
5491         return ret;
5492 }
5493
5494 /*
5495  * check whether the xattr bucket is filled up with the same hash value.
5496  * If we want to insert the xattr with the same hash, return -ENOSPC.
5497  * If we want to insert a xattr with different hash value, go ahead
5498  * and ocfs2_divide_xattr_bucket will handle this.
5499  */
5500 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5501                                               struct ocfs2_xattr_bucket *bucket,
5502                                               const char *name)
5503 {
5504         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5505         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5506
5507         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5508                 return 0;
5509
5510         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5511             xh->xh_entries[0].xe_name_hash) {
5512                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5513                      "hash = %u\n",
5514                      (unsigned long long)bucket_blkno(bucket),
5515                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5516                 return -ENOSPC;
5517         }
5518
5519         return 0;
5520 }
5521
5522 /*
5523  * Try to set the entry in the current bucket.  If we fail, the caller
5524  * will handle getting us another bucket.
5525  */
5526 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5527                                         struct ocfs2_xattr_info *xi,
5528                                         struct ocfs2_xattr_search *xs,
5529                                         struct ocfs2_xattr_set_ctxt *ctxt)
5530 {
5531         int ret;
5532         struct ocfs2_xa_loc loc;
5533
5534         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5535
5536         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5537                                        xs->not_found ? NULL : xs->here);
5538         ret = ocfs2_xa_set(&loc, xi, ctxt);
5539         if (!ret) {
5540                 xs->here = loc.xl_entry;
5541                 goto out;
5542         }
5543         if (ret != -ENOSPC) {
5544                 mlog_errno(ret);
5545                 goto out;
5546         }
5547
5548         /* Ok, we need space.  Let's try defragmenting the bucket. */
5549         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5550                                         xs->bucket);
5551         if (ret) {
5552                 mlog_errno(ret);
5553                 goto out;
5554         }
5555
5556         ret = ocfs2_xa_set(&loc, xi, ctxt);
5557         if (!ret) {
5558                 xs->here = loc.xl_entry;
5559                 goto out;
5560         }
5561         if (ret != -ENOSPC)
5562                 mlog_errno(ret);
5563
5564
5565 out:
5566         return ret;
5567 }
5568
5569 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5570                                              struct ocfs2_xattr_info *xi,
5571                                              struct ocfs2_xattr_search *xs,
5572                                              struct ocfs2_xattr_set_ctxt *ctxt)
5573 {
5574         int ret;
5575
5576         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5577
5578         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5579         if (!ret)
5580                 goto out;
5581         if (ret != -ENOSPC) {
5582                 mlog_errno(ret);
5583                 goto out;
5584         }
5585
5586         /* Ack, need more space.  Let's try to get another bucket! */
5587
5588         /*
5589          * We do not allow for overlapping ranges between buckets. And
5590          * the maximum number of collisions we will allow for then is
5591          * one bucket's worth, so check it here whether we need to
5592          * add a new bucket for the insert.
5593          */
5594         ret = ocfs2_check_xattr_bucket_collision(inode,
5595                                                  xs->bucket,
5596                                                  xi->xi_name);
5597         if (ret) {
5598                 mlog_errno(ret);
5599                 goto out;
5600         }
5601
5602         ret = ocfs2_add_new_xattr_bucket(inode,
5603                                          xs->xattr_bh,
5604                                          xs->bucket,
5605                                          ctxt);
5606         if (ret) {
5607                 mlog_errno(ret);
5608                 goto out;
5609         }
5610
5611         /*
5612          * ocfs2_add_new_xattr_bucket() will have updated
5613          * xs->bucket if it moved, but it will not have updated
5614          * any of the other search fields.  Thus, we drop it and
5615          * re-search.  Everything should be cached, so it'll be
5616          * quick.
5617          */
5618         ocfs2_xattr_bucket_relse(xs->bucket);
5619         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5620                                            xi->xi_name_index,
5621                                            xi->xi_name, xs);
5622         if (ret && ret != -ENODATA)
5623                 goto out;
5624         xs->not_found = ret;
5625
5626         /* Ok, we have a new bucket, let's try again */
5627         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5628         if (ret && (ret != -ENOSPC))
5629                 mlog_errno(ret);
5630
5631 out:
5632         return ret;
5633 }
5634
5635 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5636                                         struct ocfs2_xattr_bucket *bucket,
5637                                         void *para)
5638 {
5639         int ret = 0, ref_credits;
5640         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5641         u16 i;
5642         struct ocfs2_xattr_entry *xe;
5643         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5644         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5645         int credits = ocfs2_remove_extent_credits(osb->sb) +
5646                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5647         struct ocfs2_xattr_value_root *xv;
5648         struct ocfs2_rm_xattr_bucket_para *args =
5649                         (struct ocfs2_rm_xattr_bucket_para *)para;
5650
5651         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5652
5653         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5654                 xe = &xh->xh_entries[i];
5655                 if (ocfs2_xattr_is_local(xe))
5656                         continue;
5657
5658                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5659                                                       i, &xv, NULL);
5660
5661                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5662                                                          args->ref_ci,
5663                                                          args->ref_root_bh,
5664                                                          &ctxt.meta_ac,
5665                                                          &ref_credits);
5666
5667                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5668                 if (IS_ERR(ctxt.handle)) {
5669                         ret = PTR_ERR(ctxt.handle);
5670                         mlog_errno(ret);
5671                         break;
5672                 }
5673
5674                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5675                                                         i, 0, &ctxt);
5676
5677                 ocfs2_commit_trans(osb, ctxt.handle);
5678                 if (ctxt.meta_ac) {
5679                         ocfs2_free_alloc_context(ctxt.meta_ac);
5680                         ctxt.meta_ac = NULL;
5681                 }
5682                 if (ret) {
5683                         mlog_errno(ret);
5684                         break;
5685                 }
5686         }
5687
5688         if (ctxt.meta_ac)
5689                 ocfs2_free_alloc_context(ctxt.meta_ac);
5690         ocfs2_schedule_truncate_log_flush(osb, 1);
5691         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5692         return ret;
5693 }
5694
5695 /*
5696  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5697  * or change the extent record flag), we need to recalculate
5698  * the metaecc for the whole bucket. So it is done here.
5699  *
5700  * Note:
5701  * We have to give the extra credits for the caller.
5702  */
5703 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5704                                             handle_t *handle,
5705                                             void *para)
5706 {
5707         int ret;
5708         struct ocfs2_xattr_bucket *bucket =
5709                         (struct ocfs2_xattr_bucket *)para;
5710
5711         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5712                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5713         if (ret) {
5714                 mlog_errno(ret);
5715                 return ret;
5716         }
5717
5718         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5719
5720         return 0;
5721 }
5722
5723 /*
5724  * Special action we need if the xattr value is refcounted.
5725  *
5726  * 1. If the xattr is refcounted, lock the tree.
5727  * 2. CoW the xattr if we are setting the new value and the value
5728  *    will be stored outside.
5729  * 3. In other case, decrease_refcount will work for us, so just
5730  *    lock the refcount tree, calculate the meta and credits is OK.
5731  *
5732  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5733  * currently CoW is a completed transaction, while this function
5734  * will also lock the allocators and let us deadlock. So we will
5735  * CoW the whole xattr value.
5736  */
5737 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5738                                         struct ocfs2_dinode *di,
5739                                         struct ocfs2_xattr_info *xi,
5740                                         struct ocfs2_xattr_search *xis,
5741                                         struct ocfs2_xattr_search *xbs,
5742                                         struct ocfs2_refcount_tree **ref_tree,
5743                                         int *meta_add,
5744                                         int *credits)
5745 {
5746         int ret = 0;
5747         struct ocfs2_xattr_block *xb;
5748         struct ocfs2_xattr_entry *xe;
5749         char *base;
5750         u32 p_cluster, num_clusters;
5751         unsigned int ext_flags;
5752         int name_offset, name_len;
5753         struct ocfs2_xattr_value_buf vb;
5754         struct ocfs2_xattr_bucket *bucket = NULL;
5755         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5756         struct ocfs2_post_refcount refcount;
5757         struct ocfs2_post_refcount *p = NULL;
5758         struct buffer_head *ref_root_bh = NULL;
5759
5760         if (!xis->not_found) {
5761                 xe = xis->here;
5762                 name_offset = le16_to_cpu(xe->xe_name_offset);
5763                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5764                 base = xis->base;
5765                 vb.vb_bh = xis->inode_bh;
5766                 vb.vb_access = ocfs2_journal_access_di;
5767         } else {
5768                 int i, block_off = 0;
5769                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5770                 xe = xbs->here;
5771                 name_offset = le16_to_cpu(xe->xe_name_offset);
5772                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5773                 i = xbs->here - xbs->header->xh_entries;
5774
5775                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5776                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5777                                                         bucket_xh(xbs->bucket),
5778                                                         i, &block_off,
5779                                                         &name_offset);
5780                         if (ret) {
5781                                 mlog_errno(ret);
5782                                 goto out;
5783                         }
5784                         base = bucket_block(xbs->bucket, block_off);
5785                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5786                         vb.vb_access = ocfs2_journal_access;
5787
5788                         if (ocfs2_meta_ecc(osb)) {
5789                                 /*create parameters for ocfs2_post_refcount. */
5790                                 bucket = xbs->bucket;
5791                                 refcount.credits = bucket->bu_blocks;
5792                                 refcount.para = bucket;
5793                                 refcount.func =
5794                                         ocfs2_xattr_bucket_post_refcount;
5795                                 p = &refcount;
5796                         }
5797                 } else {
5798                         base = xbs->base;
5799                         vb.vb_bh = xbs->xattr_bh;
5800                         vb.vb_access = ocfs2_journal_access_xb;
5801                 }
5802         }
5803
5804         if (ocfs2_xattr_is_local(xe))
5805                 goto out;
5806
5807         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5808                                 (base + name_offset + name_len);
5809
5810         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5811                                        &num_clusters, &vb.vb_xv->xr_list,
5812                                        &ext_flags);
5813         if (ret) {
5814                 mlog_errno(ret);
5815                 goto out;
5816         }
5817
5818         /*
5819          * We just need to check the 1st extent record, since we always
5820          * CoW the whole xattr. So there shouldn't be a xattr with
5821          * some REFCOUNT extent recs after the 1st one.
5822          */
5823         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5824                 goto out;
5825
5826         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5827                                        1, ref_tree, &ref_root_bh);
5828         if (ret) {
5829                 mlog_errno(ret);
5830                 goto out;
5831         }
5832
5833         /*
5834          * If we are deleting the xattr or the new size will be stored inside,
5835          * cool, leave it there, the xattr truncate process will remove them
5836          * for us(it still needs the refcount tree lock and the meta, credits).
5837          * And the worse case is that every cluster truncate will split the
5838          * refcount tree, and make the original extent become 3. So we will need
5839          * 2 * cluster more extent recs at most.
5840          */
5841         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5842
5843                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5844                                                          &(*ref_tree)->rf_ci,
5845                                                          ref_root_bh, vb.vb_xv,
5846                                                          meta_add, credits);
5847                 if (ret)
5848                         mlog_errno(ret);
5849                 goto out;
5850         }
5851
5852         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5853                                        *ref_tree, ref_root_bh, 0,
5854                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5855         if (ret)
5856                 mlog_errno(ret);
5857
5858 out:
5859         brelse(ref_root_bh);
5860         return ret;
5861 }
5862
5863 /*
5864  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5865  * The physical clusters will be added to refcount tree.
5866  */
5867 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5868                                 struct ocfs2_xattr_value_root *xv,
5869                                 struct ocfs2_extent_tree *value_et,
5870                                 struct ocfs2_caching_info *ref_ci,
5871                                 struct buffer_head *ref_root_bh,
5872                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5873                                 struct ocfs2_post_refcount *refcount)
5874 {
5875         int ret = 0;
5876         u32 clusters = le32_to_cpu(xv->xr_clusters);
5877         u32 cpos, p_cluster, num_clusters;
5878         struct ocfs2_extent_list *el = &xv->xr_list;
5879         unsigned int ext_flags;
5880
5881         cpos = 0;
5882         while (cpos < clusters) {
5883                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5884                                                &num_clusters, el, &ext_flags);
5885                 if (ret) {
5886                         mlog_errno(ret);
5887                         break;
5888                 }
5889
5890                 cpos += num_clusters;
5891                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5892                         continue;
5893
5894                 BUG_ON(!p_cluster);
5895
5896                 ret = ocfs2_add_refcount_flag(inode, value_et,
5897                                               ref_ci, ref_root_bh,
5898                                               cpos - num_clusters,
5899                                               p_cluster, num_clusters,
5900                                               dealloc, refcount);
5901                 if (ret) {
5902                         mlog_errno(ret);
5903                         break;
5904                 }
5905         }
5906
5907         return ret;
5908 }
5909
5910 /*
5911  * Given a normal ocfs2_xattr_header, refcount all the entries which
5912  * have value stored outside.
5913  * Used for xattrs stored in inode and ocfs2_xattr_block.
5914  */
5915 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5916                                 struct ocfs2_xattr_value_buf *vb,
5917                                 struct ocfs2_xattr_header *header,
5918                                 struct ocfs2_caching_info *ref_ci,
5919                                 struct buffer_head *ref_root_bh,
5920                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5921 {
5922
5923         struct ocfs2_xattr_entry *xe;
5924         struct ocfs2_xattr_value_root *xv;
5925         struct ocfs2_extent_tree et;
5926         int i, ret = 0;
5927
5928         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5929                 xe = &header->xh_entries[i];
5930
5931                 if (ocfs2_xattr_is_local(xe))
5932                         continue;
5933
5934                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5935                         le16_to_cpu(xe->xe_name_offset) +
5936                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5937
5938                 vb->vb_xv = xv;
5939                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5940
5941                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5942                                                         ref_ci, ref_root_bh,
5943                                                         dealloc, NULL);
5944                 if (ret) {
5945                         mlog_errno(ret);
5946                         break;
5947                 }
5948         }
5949
5950         return ret;
5951 }
5952
5953 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5954                                 struct buffer_head *fe_bh,
5955                                 struct ocfs2_caching_info *ref_ci,
5956                                 struct buffer_head *ref_root_bh,
5957                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5958 {
5959         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5960         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5961                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5962                                 le16_to_cpu(di->i_xattr_inline_size));
5963         struct ocfs2_xattr_value_buf vb = {
5964                 .vb_bh = fe_bh,
5965                 .vb_access = ocfs2_journal_access_di,
5966         };
5967
5968         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5969                                                   ref_ci, ref_root_bh, dealloc);
5970 }
5971
5972 struct ocfs2_xattr_tree_value_refcount_para {
5973         struct ocfs2_caching_info *ref_ci;
5974         struct buffer_head *ref_root_bh;
5975         struct ocfs2_cached_dealloc_ctxt *dealloc;
5976 };
5977
5978 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5979                                            struct ocfs2_xattr_bucket *bucket,
5980                                            int offset,
5981                                            struct ocfs2_xattr_value_root **xv,
5982                                            struct buffer_head **bh)
5983 {
5984         int ret, block_off, name_offset;
5985         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5986         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5987         void *base;
5988
5989         ret = ocfs2_xattr_bucket_get_name_value(sb,
5990                                                 bucket_xh(bucket),
5991                                                 offset,
5992                                                 &block_off,
5993                                                 &name_offset);
5994         if (ret) {
5995                 mlog_errno(ret);
5996                 goto out;
5997         }
5998
5999         base = bucket_block(bucket, block_off);
6000
6001         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6002                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6003
6004         if (bh)
6005                 *bh = bucket->bu_bhs[block_off];
6006 out:
6007         return ret;
6008 }
6009
6010 /*
6011  * For a given xattr bucket, refcount all the entries which
6012  * have value stored outside.
6013  */
6014 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6015                                              struct ocfs2_xattr_bucket *bucket,
6016                                              void *para)
6017 {
6018         int i, ret = 0;
6019         struct ocfs2_extent_tree et;
6020         struct ocfs2_xattr_tree_value_refcount_para *ref =
6021                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6022         struct ocfs2_xattr_header *xh =
6023                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6024         struct ocfs2_xattr_entry *xe;
6025         struct ocfs2_xattr_value_buf vb = {
6026                 .vb_access = ocfs2_journal_access,
6027         };
6028         struct ocfs2_post_refcount refcount = {
6029                 .credits = bucket->bu_blocks,
6030                 .para = bucket,
6031                 .func = ocfs2_xattr_bucket_post_refcount,
6032         };
6033         struct ocfs2_post_refcount *p = NULL;
6034
6035         /* We only need post_refcount if we support metaecc. */
6036         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6037                 p = &refcount;
6038
6039         trace_ocfs2_xattr_bucket_value_refcount(
6040                                 (unsigned long long)bucket_blkno(bucket),
6041                                 le16_to_cpu(xh->xh_count));
6042         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6043                 xe = &xh->xh_entries[i];
6044
6045                 if (ocfs2_xattr_is_local(xe))
6046                         continue;
6047
6048                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6049                                                       &vb.vb_xv, &vb.vb_bh);
6050                 if (ret) {
6051                         mlog_errno(ret);
6052                         break;
6053                 }
6054
6055                 ocfs2_init_xattr_value_extent_tree(&et,
6056                                                    INODE_CACHE(inode), &vb);
6057
6058                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6059                                                         &et, ref->ref_ci,
6060                                                         ref->ref_root_bh,
6061                                                         ref->dealloc, p);
6062                 if (ret) {
6063                         mlog_errno(ret);
6064                         break;
6065                 }
6066         }
6067
6068         return ret;
6069
6070 }
6071
6072 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6073                                      struct buffer_head *root_bh,
6074                                      u64 blkno, u32 cpos, u32 len, void *para)
6075 {
6076         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6077                                            ocfs2_xattr_bucket_value_refcount,
6078                                            para);
6079 }
6080
6081 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6082                                 struct buffer_head *blk_bh,
6083                                 struct ocfs2_caching_info *ref_ci,
6084                                 struct buffer_head *ref_root_bh,
6085                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6086 {
6087         int ret = 0;
6088         struct ocfs2_xattr_block *xb =
6089                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6090
6091         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6092                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6093                 struct ocfs2_xattr_value_buf vb = {
6094                         .vb_bh = blk_bh,
6095                         .vb_access = ocfs2_journal_access_xb,
6096                 };
6097
6098                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6099                                                          ref_ci, ref_root_bh,
6100                                                          dealloc);
6101         } else {
6102                 struct ocfs2_xattr_tree_value_refcount_para para = {
6103                         .ref_ci = ref_ci,
6104                         .ref_root_bh = ref_root_bh,
6105                         .dealloc = dealloc,
6106                 };
6107
6108                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6109                                                 ocfs2_refcount_xattr_tree_rec,
6110                                                 &para);
6111         }
6112
6113         return ret;
6114 }
6115
6116 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6117                                      struct buffer_head *fe_bh,
6118                                      struct ocfs2_caching_info *ref_ci,
6119                                      struct buffer_head *ref_root_bh,
6120                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6121 {
6122         int ret = 0;
6123         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6124         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6125         struct buffer_head *blk_bh = NULL;
6126
6127         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6128                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6129                                                          ref_ci, ref_root_bh,
6130                                                          dealloc);
6131                 if (ret) {
6132                         mlog_errno(ret);
6133                         goto out;
6134                 }
6135         }
6136
6137         if (!di->i_xattr_loc)
6138                 goto out;
6139
6140         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6141                                      &blk_bh);
6142         if (ret < 0) {
6143                 mlog_errno(ret);
6144                 goto out;
6145         }
6146
6147         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6148                                                 ref_root_bh, dealloc);
6149         if (ret)
6150                 mlog_errno(ret);
6151
6152         brelse(blk_bh);
6153 out:
6154
6155         return ret;
6156 }
6157
6158 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6159 /*
6160  * Store the information we need in xattr reflink.
6161  * old_bh and new_bh are inode bh for the old and new inode.
6162  */
6163 struct ocfs2_xattr_reflink {
6164         struct inode *old_inode;
6165         struct inode *new_inode;
6166         struct buffer_head *old_bh;
6167         struct buffer_head *new_bh;
6168         struct ocfs2_caching_info *ref_ci;
6169         struct buffer_head *ref_root_bh;
6170         struct ocfs2_cached_dealloc_ctxt *dealloc;
6171         should_xattr_reflinked *xattr_reflinked;
6172 };
6173
6174 /*
6175  * Given a xattr header and xe offset,
6176  * return the proper xv and the corresponding bh.
6177  * xattr in inode, block and xattr tree have different implementaions.
6178  */
6179 typedef int (get_xattr_value_root)(struct super_block *sb,
6180                                    struct buffer_head *bh,
6181                                    struct ocfs2_xattr_header *xh,
6182                                    int offset,
6183                                    struct ocfs2_xattr_value_root **xv,
6184                                    struct buffer_head **ret_bh,
6185                                    void *para);
6186
6187 /*
6188  * Calculate all the xattr value root metadata stored in this xattr header and
6189  * credits we need if we create them from the scratch.
6190  * We use get_xattr_value_root so that all types of xattr container can use it.
6191  */
6192 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6193                                              struct buffer_head *bh,
6194                                              struct ocfs2_xattr_header *xh,
6195                                              int *metas, int *credits,
6196                                              int *num_recs,
6197                                              get_xattr_value_root *func,
6198                                              void *para)
6199 {
6200         int i, ret = 0;
6201         struct ocfs2_xattr_value_root *xv;
6202         struct ocfs2_xattr_entry *xe;
6203
6204         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6205                 xe = &xh->xh_entries[i];
6206                 if (ocfs2_xattr_is_local(xe))
6207                         continue;
6208
6209                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6210                 if (ret) {
6211                         mlog_errno(ret);
6212                         break;
6213                 }
6214
6215                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6216                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6217
6218                 *credits += ocfs2_calc_extend_credits(sb,
6219                                                 &def_xv.xv.xr_list,
6220                                                 le32_to_cpu(xv->xr_clusters));
6221
6222                 /*
6223                  * If the value is a tree with depth > 1, We don't go deep
6224                  * to the extent block, so just calculate a maximum record num.
6225                  */
6226                 if (!xv->xr_list.l_tree_depth)
6227                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6228                 else
6229                         *num_recs += ocfs2_clusters_for_bytes(sb,
6230                                                               XATTR_SIZE_MAX);
6231         }
6232
6233         return ret;
6234 }
6235
6236 /* Used by xattr inode and block to return the right xv and buffer_head. */
6237 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6238                                       struct buffer_head *bh,
6239                                       struct ocfs2_xattr_header *xh,
6240                                       int offset,
6241                                       struct ocfs2_xattr_value_root **xv,
6242                                       struct buffer_head **ret_bh,
6243                                       void *para)
6244 {
6245         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6246
6247         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6248                 le16_to_cpu(xe->xe_name_offset) +
6249                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6250
6251         if (ret_bh)
6252                 *ret_bh = bh;
6253
6254         return 0;
6255 }
6256
6257 /*
6258  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6259  * It is only used for inline xattr and xattr block.
6260  */
6261 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6262                                         struct ocfs2_xattr_header *xh,
6263                                         struct buffer_head *ref_root_bh,
6264                                         int *credits,
6265                                         struct ocfs2_alloc_context **meta_ac)
6266 {
6267         int ret, meta_add = 0, num_recs = 0;
6268         struct ocfs2_refcount_block *rb =
6269                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6270
6271         *credits = 0;
6272
6273         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6274                                                 &meta_add, credits, &num_recs,
6275                                                 ocfs2_get_xattr_value_root,
6276                                                 NULL);
6277         if (ret) {
6278                 mlog_errno(ret);
6279                 goto out;
6280         }
6281
6282         /*
6283          * We need to add/modify num_recs in refcount tree, so just calculate
6284          * an approximate number we need for refcount tree change.
6285          * Sometimes we need to split the tree, and after split,  half recs
6286          * will be moved to the new block, and a new block can only provide
6287          * half number of recs. So we multiple new blocks by 2.
6288          */
6289         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6290         meta_add += num_recs;
6291         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6292         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6293                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6294                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6295         else
6296                 *credits += 1;
6297
6298         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6299         if (ret)
6300                 mlog_errno(ret);
6301
6302 out:
6303         return ret;
6304 }
6305
6306 /*
6307  * Given a xattr header, reflink all the xattrs in this container.
6308  * It can be used for inode, block and bucket.
6309  *
6310  * NOTE:
6311  * Before we call this function, the caller has memcpy the xattr in
6312  * old_xh to the new_xh.
6313  *
6314  * If args.xattr_reflinked is set, call it to decide whether the xe should
6315  * be reflinked or not. If not, remove it from the new xattr header.
6316  */
6317 static int ocfs2_reflink_xattr_header(handle_t *handle,
6318                                       struct ocfs2_xattr_reflink *args,
6319                                       struct buffer_head *old_bh,
6320                                       struct ocfs2_xattr_header *xh,
6321                                       struct buffer_head *new_bh,
6322                                       struct ocfs2_xattr_header *new_xh,
6323                                       struct ocfs2_xattr_value_buf *vb,
6324                                       struct ocfs2_alloc_context *meta_ac,
6325                                       get_xattr_value_root *func,
6326                                       void *para)
6327 {
6328         int ret = 0, i, j;
6329         struct super_block *sb = args->old_inode->i_sb;
6330         struct buffer_head *value_bh;
6331         struct ocfs2_xattr_entry *xe, *last;
6332         struct ocfs2_xattr_value_root *xv, *new_xv;
6333         struct ocfs2_extent_tree data_et;
6334         u32 clusters, cpos, p_cluster, num_clusters;
6335         unsigned int ext_flags = 0;
6336
6337         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6338                                          le16_to_cpu(xh->xh_count));
6339
6340         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6341         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6342                 xe = &xh->xh_entries[i];
6343
6344                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6345                         xe = &new_xh->xh_entries[j];
6346
6347                         le16_add_cpu(&new_xh->xh_count, -1);
6348                         if (new_xh->xh_count) {
6349                                 memmove(xe, xe + 1,
6350                                         (void *)last - (void *)xe);
6351                                 memset(last, 0,
6352                                        sizeof(struct ocfs2_xattr_entry));
6353                         }
6354
6355                         /*
6356                          * We don't want j to increase in the next round since
6357                          * it is already moved ahead.
6358                          */
6359                         j--;
6360                         continue;
6361                 }
6362
6363                 if (ocfs2_xattr_is_local(xe))
6364                         continue;
6365
6366                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6367                 if (ret) {
6368                         mlog_errno(ret);
6369                         break;
6370                 }
6371
6372                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6373                 if (ret) {
6374                         mlog_errno(ret);
6375                         break;
6376                 }
6377
6378                 /*
6379                  * For the xattr which has l_tree_depth = 0, all the extent
6380                  * recs have already be copied to the new xh with the
6381                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6382                  * increase the refount count int the refcount tree.
6383                  *
6384                  * For the xattr which has l_tree_depth > 0, we need
6385                  * to initialize it to the empty default value root,
6386                  * and then insert the extents one by one.
6387                  */
6388                 if (xv->xr_list.l_tree_depth) {
6389                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6390                         vb->vb_xv = new_xv;
6391                         vb->vb_bh = value_bh;
6392                         ocfs2_init_xattr_value_extent_tree(&data_et,
6393                                         INODE_CACHE(args->new_inode), vb);
6394                 }
6395
6396                 clusters = le32_to_cpu(xv->xr_clusters);
6397                 cpos = 0;
6398                 while (cpos < clusters) {
6399                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6400                                                        cpos,
6401                                                        &p_cluster,
6402                                                        &num_clusters,
6403                                                        &xv->xr_list,
6404                                                        &ext_flags);
6405                         if (ret) {
6406                                 mlog_errno(ret);
6407                                 goto out;
6408                         }
6409
6410                         BUG_ON(!p_cluster);
6411
6412                         if (xv->xr_list.l_tree_depth) {
6413                                 ret = ocfs2_insert_extent(handle,
6414                                                 &data_et, cpos,
6415                                                 ocfs2_clusters_to_blocks(
6416                                                         args->old_inode->i_sb,
6417                                                         p_cluster),
6418                                                 num_clusters, ext_flags,
6419                                                 meta_ac);
6420                                 if (ret) {
6421                                         mlog_errno(ret);
6422                                         goto out;
6423                                 }
6424                         }
6425
6426                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6427                                                       args->ref_root_bh,
6428                                                       p_cluster, num_clusters,
6429                                                       meta_ac, args->dealloc);
6430                         if (ret) {
6431                                 mlog_errno(ret);
6432                                 goto out;
6433                         }
6434
6435                         cpos += num_clusters;
6436                 }
6437         }
6438
6439 out:
6440         return ret;
6441 }
6442
6443 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6444 {
6445         int ret = 0, credits = 0;
6446         handle_t *handle;
6447         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6448         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6449         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6450         int header_off = osb->sb->s_blocksize - inline_size;
6451         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6452                                         (args->old_bh->b_data + header_off);
6453         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6454                                         (args->new_bh->b_data + header_off);
6455         struct ocfs2_alloc_context *meta_ac = NULL;
6456         struct ocfs2_inode_info *new_oi;
6457         struct ocfs2_dinode *new_di;
6458         struct ocfs2_xattr_value_buf vb = {
6459                 .vb_bh = args->new_bh,
6460                 .vb_access = ocfs2_journal_access_di,
6461         };
6462
6463         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6464                                                   &credits, &meta_ac);
6465         if (ret) {
6466                 mlog_errno(ret);
6467                 goto out;
6468         }
6469
6470         handle = ocfs2_start_trans(osb, credits);
6471         if (IS_ERR(handle)) {
6472                 ret = PTR_ERR(handle);
6473                 mlog_errno(ret);
6474                 goto out;
6475         }
6476
6477         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6478                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6479         if (ret) {
6480                 mlog_errno(ret);
6481                 goto out_commit;
6482         }
6483
6484         memcpy(args->new_bh->b_data + header_off,
6485                args->old_bh->b_data + header_off, inline_size);
6486
6487         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6488         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6489
6490         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6491                                          args->new_bh, new_xh, &vb, meta_ac,
6492                                          ocfs2_get_xattr_value_root, NULL);
6493         if (ret) {
6494                 mlog_errno(ret);
6495                 goto out_commit;
6496         }
6497
6498         new_oi = OCFS2_I(args->new_inode);
6499         /*
6500          * Adjust extent record count to reserve space for extended attribute.
6501          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6502          */
6503         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6504             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6505                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6506                 le16_add_cpu(&el->l_count, -(inline_size /
6507                                         sizeof(struct ocfs2_extent_rec)));
6508         }
6509         spin_lock(&new_oi->ip_lock);
6510         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6511         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6512         spin_unlock(&new_oi->ip_lock);
6513
6514         ocfs2_journal_dirty(handle, args->new_bh);
6515
6516 out_commit:
6517         ocfs2_commit_trans(osb, handle);
6518
6519 out:
6520         if (meta_ac)
6521                 ocfs2_free_alloc_context(meta_ac);
6522         return ret;
6523 }
6524
6525 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6526                                           struct buffer_head *fe_bh,
6527                                           struct buffer_head **ret_bh,
6528                                           int indexed)
6529 {
6530         int ret;
6531         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6532         struct ocfs2_xattr_set_ctxt ctxt;
6533
6534         memset(&ctxt, 0, sizeof(ctxt));
6535         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6536         if (ret < 0) {
6537                 mlog_errno(ret);
6538                 return ret;
6539         }
6540
6541         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6542         if (IS_ERR(ctxt.handle)) {
6543                 ret = PTR_ERR(ctxt.handle);
6544                 mlog_errno(ret);
6545                 goto out;
6546         }
6547
6548         trace_ocfs2_create_empty_xattr_block(
6549                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6550         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6551                                        ret_bh);
6552         if (ret)
6553                 mlog_errno(ret);
6554
6555         ocfs2_commit_trans(osb, ctxt.handle);
6556 out:
6557         ocfs2_free_alloc_context(ctxt.meta_ac);
6558         return ret;
6559 }
6560
6561 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6562                                      struct buffer_head *blk_bh,
6563                                      struct buffer_head *new_blk_bh)
6564 {
6565         int ret = 0, credits = 0;
6566         handle_t *handle;
6567         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6568         struct ocfs2_dinode *new_di;
6569         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6570         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6571         struct ocfs2_xattr_block *xb =
6572                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6573         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6574         struct ocfs2_xattr_block *new_xb =
6575                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6576         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6577         struct ocfs2_alloc_context *meta_ac;
6578         struct ocfs2_xattr_value_buf vb = {
6579                 .vb_bh = new_blk_bh,
6580                 .vb_access = ocfs2_journal_access_xb,
6581         };
6582
6583         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6584                                                   &credits, &meta_ac);
6585         if (ret) {
6586                 mlog_errno(ret);
6587                 return ret;
6588         }
6589
6590         /* One more credits in case we need to add xattr flags in new inode. */
6591         handle = ocfs2_start_trans(osb, credits + 1);
6592         if (IS_ERR(handle)) {
6593                 ret = PTR_ERR(handle);
6594                 mlog_errno(ret);
6595                 goto out;
6596         }
6597
6598         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6599                 ret = ocfs2_journal_access_di(handle,
6600                                               INODE_CACHE(args->new_inode),
6601                                               args->new_bh,
6602                                               OCFS2_JOURNAL_ACCESS_WRITE);
6603                 if (ret) {
6604                         mlog_errno(ret);
6605                         goto out_commit;
6606                 }
6607         }
6608
6609         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6610                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6611         if (ret) {
6612                 mlog_errno(ret);
6613                 goto out_commit;
6614         }
6615
6616         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6617                osb->sb->s_blocksize - header_off);
6618
6619         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6620                                          new_blk_bh, new_xh, &vb, meta_ac,
6621                                          ocfs2_get_xattr_value_root, NULL);
6622         if (ret) {
6623                 mlog_errno(ret);
6624                 goto out_commit;
6625         }
6626
6627         ocfs2_journal_dirty(handle, new_blk_bh);
6628
6629         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6630                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6631                 spin_lock(&new_oi->ip_lock);
6632                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6633                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6634                 spin_unlock(&new_oi->ip_lock);
6635
6636                 ocfs2_journal_dirty(handle, args->new_bh);
6637         }
6638
6639 out_commit:
6640         ocfs2_commit_trans(osb, handle);
6641
6642 out:
6643         ocfs2_free_alloc_context(meta_ac);
6644         return ret;
6645 }
6646
6647 struct ocfs2_reflink_xattr_tree_args {
6648         struct ocfs2_xattr_reflink *reflink;
6649         struct buffer_head *old_blk_bh;
6650         struct buffer_head *new_blk_bh;
6651         struct ocfs2_xattr_bucket *old_bucket;
6652         struct ocfs2_xattr_bucket *new_bucket;
6653 };
6654
6655 /*
6656  * NOTE:
6657  * We have to handle the case that both old bucket and new bucket
6658  * will call this function to get the right ret_bh.
6659  * So The caller must give us the right bh.
6660  */
6661 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6662                                         struct buffer_head *bh,
6663                                         struct ocfs2_xattr_header *xh,
6664                                         int offset,
6665                                         struct ocfs2_xattr_value_root **xv,
6666                                         struct buffer_head **ret_bh,
6667                                         void *para)
6668 {
6669         struct ocfs2_reflink_xattr_tree_args *args =
6670                         (struct ocfs2_reflink_xattr_tree_args *)para;
6671         struct ocfs2_xattr_bucket *bucket;
6672
6673         if (bh == args->old_bucket->bu_bhs[0])
6674                 bucket = args->old_bucket;
6675         else
6676                 bucket = args->new_bucket;
6677
6678         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6679                                                xv, ret_bh);
6680 }
6681
6682 struct ocfs2_value_tree_metas {
6683         int num_metas;
6684         int credits;
6685         int num_recs;
6686 };
6687
6688 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6689                                         struct buffer_head *bh,
6690                                         struct ocfs2_xattr_header *xh,
6691                                         int offset,
6692                                         struct ocfs2_xattr_value_root **xv,
6693                                         struct buffer_head **ret_bh,
6694                                         void *para)
6695 {
6696         struct ocfs2_xattr_bucket *bucket =
6697                                 (struct ocfs2_xattr_bucket *)para;
6698
6699         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6700                                                xv, ret_bh);
6701 }
6702
6703 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6704                                       struct ocfs2_xattr_bucket *bucket,
6705                                       void *para)
6706 {
6707         struct ocfs2_value_tree_metas *metas =
6708                         (struct ocfs2_value_tree_metas *)para;
6709         struct ocfs2_xattr_header *xh =
6710                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6711
6712         /* Add the credits for this bucket first. */
6713         metas->credits += bucket->bu_blocks;
6714         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6715                                         xh, &metas->num_metas,
6716                                         &metas->credits, &metas->num_recs,
6717                                         ocfs2_value_tree_metas_in_bucket,
6718                                         bucket);
6719 }
6720
6721 /*
6722  * Given a xattr extent rec starting from blkno and having len clusters,
6723  * iterate all the buckets calculate how much metadata we need for reflinking
6724  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6725  */
6726 static int ocfs2_lock_reflink_xattr_rec_allocators(
6727                                 struct ocfs2_reflink_xattr_tree_args *args,
6728                                 struct ocfs2_extent_tree *xt_et,
6729                                 u64 blkno, u32 len, int *credits,
6730                                 struct ocfs2_alloc_context **meta_ac,
6731                                 struct ocfs2_alloc_context **data_ac)
6732 {
6733         int ret, num_free_extents;
6734         struct ocfs2_value_tree_metas metas;
6735         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6736         struct ocfs2_refcount_block *rb;
6737
6738         memset(&metas, 0, sizeof(metas));
6739
6740         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6741                                           ocfs2_calc_value_tree_metas, &metas);
6742         if (ret) {
6743                 mlog_errno(ret);
6744                 goto out;
6745         }
6746
6747         *credits = metas.credits;
6748
6749         /*
6750          * Calculate we need for refcount tree change.
6751          *
6752          * We need to add/modify num_recs in refcount tree, so just calculate
6753          * an approximate number we need for refcount tree change.
6754          * Sometimes we need to split the tree, and after split,  half recs
6755          * will be moved to the new block, and a new block can only provide
6756          * half number of recs. So we multiple new blocks by 2.
6757          * In the end, we have to add credits for modifying the already
6758          * existed refcount block.
6759          */
6760         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6761         metas.num_recs =
6762                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6763                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6764         metas.num_metas += metas.num_recs;
6765         *credits += metas.num_recs +
6766                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6767         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6768                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6769                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6770         else
6771                 *credits += 1;
6772
6773         /* count in the xattr tree change. */
6774         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6775         if (num_free_extents < 0) {
6776                 ret = num_free_extents;
6777                 mlog_errno(ret);
6778                 goto out;
6779         }
6780
6781         if (num_free_extents < len)
6782                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6783
6784         *credits += ocfs2_calc_extend_credits(osb->sb,
6785                                               xt_et->et_root_el, len);
6786
6787         if (metas.num_metas) {
6788                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6789                                                         meta_ac);
6790                 if (ret) {
6791                         mlog_errno(ret);
6792                         goto out;
6793                 }
6794         }
6795
6796         if (len) {
6797                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6798                 if (ret)
6799                         mlog_errno(ret);
6800         }
6801 out:
6802         if (ret) {
6803                 if (*meta_ac) {
6804                         ocfs2_free_alloc_context(*meta_ac);
6805                         *meta_ac = NULL;
6806                 }
6807         }
6808
6809         return ret;
6810 }
6811
6812 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6813                                 u64 blkno, u64 new_blkno, u32 clusters,
6814                                 u32 *cpos, int num_buckets,
6815                                 struct ocfs2_alloc_context *meta_ac,
6816                                 struct ocfs2_alloc_context *data_ac,
6817                                 struct ocfs2_reflink_xattr_tree_args *args)
6818 {
6819         int i, j, ret = 0;
6820         struct super_block *sb = args->reflink->old_inode->i_sb;
6821         int bpb = args->old_bucket->bu_blocks;
6822         struct ocfs2_xattr_value_buf vb = {
6823                 .vb_access = ocfs2_journal_access,
6824         };
6825
6826         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6827                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6828                 if (ret) {
6829                         mlog_errno(ret);
6830                         break;
6831                 }
6832
6833                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6834                 if (ret) {
6835                         mlog_errno(ret);
6836                         break;
6837                 }
6838
6839                 ret = ocfs2_xattr_bucket_journal_access(handle,
6840                                                 args->new_bucket,
6841                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6842                 if (ret) {
6843                         mlog_errno(ret);
6844                         break;
6845                 }
6846
6847                 for (j = 0; j < bpb; j++)
6848                         memcpy(bucket_block(args->new_bucket, j),
6849                                bucket_block(args->old_bucket, j),
6850                                sb->s_blocksize);
6851
6852                 /*
6853                  * Record the start cpos so that we can use it to initialize
6854                  * our xattr tree we also set the xh_num_bucket for the new
6855                  * bucket.
6856                  */
6857                 if (i == 0) {
6858                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6859                                             xh_entries[0].xe_name_hash);
6860                         bucket_xh(args->new_bucket)->xh_num_buckets =
6861                                 cpu_to_le16(num_buckets);
6862                 }
6863
6864                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6865
6866                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6867                                         args->old_bucket->bu_bhs[0],
6868                                         bucket_xh(args->old_bucket),
6869                                         args->new_bucket->bu_bhs[0],
6870                                         bucket_xh(args->new_bucket),
6871                                         &vb, meta_ac,
6872                                         ocfs2_get_reflink_xattr_value_root,
6873                                         args);
6874                 if (ret) {
6875                         mlog_errno(ret);
6876                         break;
6877                 }
6878
6879                 /*
6880                  * Re-access and dirty the bucket to calculate metaecc.
6881                  * Because we may extend the transaction in reflink_xattr_header
6882                  * which will let the already accessed block gone.
6883                  */
6884                 ret = ocfs2_xattr_bucket_journal_access(handle,
6885                                                 args->new_bucket,
6886                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6887                 if (ret) {
6888                         mlog_errno(ret);
6889                         break;
6890                 }
6891
6892                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6893
6894                 ocfs2_xattr_bucket_relse(args->old_bucket);
6895                 ocfs2_xattr_bucket_relse(args->new_bucket);
6896         }
6897
6898         ocfs2_xattr_bucket_relse(args->old_bucket);
6899         ocfs2_xattr_bucket_relse(args->new_bucket);
6900         return ret;
6901 }
6902
6903 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6904                                 struct inode *inode,
6905                                 struct ocfs2_reflink_xattr_tree_args *args,
6906                                 struct ocfs2_extent_tree *et,
6907                                 struct ocfs2_alloc_context *meta_ac,
6908                                 struct ocfs2_alloc_context *data_ac,
6909                                 u64 blkno, u32 cpos, u32 len)
6910 {
6911         int ret, first_inserted = 0;
6912         u32 p_cluster, num_clusters, reflink_cpos = 0;
6913         u64 new_blkno;
6914         unsigned int num_buckets, reflink_buckets;
6915         unsigned int bpc =
6916                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6917
6918         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6919         if (ret) {
6920                 mlog_errno(ret);
6921                 goto out;
6922         }
6923         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6924         ocfs2_xattr_bucket_relse(args->old_bucket);
6925
6926         while (len && num_buckets) {
6927                 ret = ocfs2_claim_clusters(handle, data_ac,
6928                                            1, &p_cluster, &num_clusters);
6929                 if (ret) {
6930                         mlog_errno(ret);
6931                         goto out;
6932                 }
6933
6934                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6935                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6936
6937                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6938                                                  new_blkno, num_clusters,
6939                                                  &reflink_cpos, reflink_buckets,
6940                                                  meta_ac, data_ac, args);
6941                 if (ret) {
6942                         mlog_errno(ret);
6943                         goto out;
6944                 }
6945
6946                 /*
6947                  * For the 1st allocated cluster, we make it use the same cpos
6948                  * so that the xattr tree looks the same as the original one
6949                  * in the most case.
6950                  */
6951                 if (!first_inserted) {
6952                         reflink_cpos = cpos;
6953                         first_inserted = 1;
6954                 }
6955                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6956                                           num_clusters, 0, meta_ac);
6957                 if (ret)
6958                         mlog_errno(ret);
6959
6960                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6961                                                   num_clusters, reflink_cpos);
6962
6963                 len -= num_clusters;
6964                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6965                 num_buckets -= reflink_buckets;
6966         }
6967 out:
6968         return ret;
6969 }
6970
6971 /*
6972  * Create the same xattr extent record in the new inode's xattr tree.
6973  */
6974 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6975                                    struct buffer_head *root_bh,
6976                                    u64 blkno,
6977                                    u32 cpos,
6978                                    u32 len,
6979                                    void *para)
6980 {
6981         int ret, credits = 0;
6982         handle_t *handle;
6983         struct ocfs2_reflink_xattr_tree_args *args =
6984                         (struct ocfs2_reflink_xattr_tree_args *)para;
6985         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6986         struct ocfs2_alloc_context *meta_ac = NULL;
6987         struct ocfs2_alloc_context *data_ac = NULL;
6988         struct ocfs2_extent_tree et;
6989
6990         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6991
6992         ocfs2_init_xattr_tree_extent_tree(&et,
6993                                           INODE_CACHE(args->reflink->new_inode),
6994                                           args->new_blk_bh);
6995
6996         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6997                                                       len, &credits,
6998                                                       &meta_ac, &data_ac);
6999         if (ret) {
7000                 mlog_errno(ret);
7001                 goto out;
7002         }
7003
7004         handle = ocfs2_start_trans(osb, credits);
7005         if (IS_ERR(handle)) {
7006                 ret = PTR_ERR(handle);
7007                 mlog_errno(ret);
7008                 goto out;
7009         }
7010
7011         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7012                                           meta_ac, data_ac,
7013                                           blkno, cpos, len);
7014         if (ret)
7015                 mlog_errno(ret);
7016
7017         ocfs2_commit_trans(osb, handle);
7018
7019 out:
7020         if (meta_ac)
7021                 ocfs2_free_alloc_context(meta_ac);
7022         if (data_ac)
7023                 ocfs2_free_alloc_context(data_ac);
7024         return ret;
7025 }
7026
7027 /*
7028  * Create reflinked xattr buckets.
7029  * We will add bucket one by one, and refcount all the xattrs in the bucket
7030  * if they are stored outside.
7031  */
7032 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7033                                     struct buffer_head *blk_bh,
7034                                     struct buffer_head *new_blk_bh)
7035 {
7036         int ret;
7037         struct ocfs2_reflink_xattr_tree_args para;
7038
7039         memset(&para, 0, sizeof(para));
7040         para.reflink = args;
7041         para.old_blk_bh = blk_bh;
7042         para.new_blk_bh = new_blk_bh;
7043
7044         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7045         if (!para.old_bucket) {
7046                 mlog_errno(-ENOMEM);
7047                 return -ENOMEM;
7048         }
7049
7050         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7051         if (!para.new_bucket) {
7052                 ret = -ENOMEM;
7053                 mlog_errno(ret);
7054                 goto out;
7055         }
7056
7057         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7058                                               ocfs2_reflink_xattr_rec,
7059                                               &para);
7060         if (ret)
7061                 mlog_errno(ret);
7062
7063 out:
7064         ocfs2_xattr_bucket_free(para.old_bucket);
7065         ocfs2_xattr_bucket_free(para.new_bucket);
7066         return ret;
7067 }
7068
7069 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7070                                         struct buffer_head *blk_bh)
7071 {
7072         int ret, indexed = 0;
7073         struct buffer_head *new_blk_bh = NULL;
7074         struct ocfs2_xattr_block *xb =
7075                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7076
7077
7078         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7079                 indexed = 1;
7080
7081         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7082                                              &new_blk_bh, indexed);
7083         if (ret) {
7084                 mlog_errno(ret);
7085                 goto out;
7086         }
7087
7088         if (!indexed)
7089                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7090         else
7091                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7092         if (ret)
7093                 mlog_errno(ret);
7094
7095 out:
7096         brelse(new_blk_bh);
7097         return ret;
7098 }
7099
7100 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7101 {
7102         int type = ocfs2_xattr_get_type(xe);
7103
7104         return type != OCFS2_XATTR_INDEX_SECURITY &&
7105                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7106                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7107 }
7108
7109 int ocfs2_reflink_xattrs(struct inode *old_inode,
7110                          struct buffer_head *old_bh,
7111                          struct inode *new_inode,
7112                          struct buffer_head *new_bh,
7113                          bool preserve_security)
7114 {
7115         int ret;
7116         struct ocfs2_xattr_reflink args;
7117         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7118         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7119         struct buffer_head *blk_bh = NULL;
7120         struct ocfs2_cached_dealloc_ctxt dealloc;
7121         struct ocfs2_refcount_tree *ref_tree;
7122         struct buffer_head *ref_root_bh = NULL;
7123
7124         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7125                                        le64_to_cpu(di->i_refcount_loc),
7126                                        1, &ref_tree, &ref_root_bh);
7127         if (ret) {
7128                 mlog_errno(ret);
7129                 goto out;
7130         }
7131
7132         ocfs2_init_dealloc_ctxt(&dealloc);
7133
7134         args.old_inode = old_inode;
7135         args.new_inode = new_inode;
7136         args.old_bh = old_bh;
7137         args.new_bh = new_bh;
7138         args.ref_ci = &ref_tree->rf_ci;
7139         args.ref_root_bh = ref_root_bh;
7140         args.dealloc = &dealloc;
7141         if (preserve_security)
7142                 args.xattr_reflinked = NULL;
7143         else
7144                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7145
7146         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7147                 ret = ocfs2_reflink_xattr_inline(&args);
7148                 if (ret) {
7149                         mlog_errno(ret);
7150                         goto out_unlock;
7151                 }
7152         }
7153
7154         if (!di->i_xattr_loc)
7155                 goto out_unlock;
7156
7157         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7158                                      &blk_bh);
7159         if (ret < 0) {
7160                 mlog_errno(ret);
7161                 goto out_unlock;
7162         }
7163
7164         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7165         if (ret)
7166                 mlog_errno(ret);
7167
7168         brelse(blk_bh);
7169
7170 out_unlock:
7171         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7172                                    ref_tree, 1);
7173         brelse(ref_root_bh);
7174
7175         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7176                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7177                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7178         }
7179
7180 out:
7181         return ret;
7182 }
7183
7184 /*
7185  * Initialize security and acl for a already created inode.
7186  * Used for reflink a non-preserve-security file.
7187  *
7188  * It uses common api like ocfs2_xattr_set, so the caller
7189  * must not hold any lock expect i_mutex.
7190  */
7191 int ocfs2_init_security_and_acl(struct inode *dir,
7192                                 struct inode *inode,
7193                                 const struct qstr *qstr)
7194 {
7195         int ret = 0;
7196         struct buffer_head *dir_bh = NULL;
7197
7198         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7199         if (ret) {
7200                 mlog_errno(ret);
7201                 goto leave;
7202         }
7203
7204         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7205         if (ret) {
7206                 mlog_errno(ret);
7207                 goto leave;
7208         }
7209
7210         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7211         if (ret)
7212                 mlog_errno(ret);
7213
7214         ocfs2_inode_unlock(dir, 0);
7215         brelse(dir_bh);
7216 leave:
7217         return ret;
7218 }
7219 /*
7220  * 'security' attributes support
7221  */
7222 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7223                                         size_t list_size, const char *name,
7224                                         size_t name_len, int type)
7225 {
7226         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7227         const size_t total_len = prefix_len + name_len + 1;
7228
7229         if (list && total_len <= list_size) {
7230                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7231                 memcpy(list + prefix_len, name, name_len);
7232                 list[prefix_len + name_len] = '\0';
7233         }
7234         return total_len;
7235 }
7236
7237 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7238                                     void *buffer, size_t size, int type)
7239 {
7240         if (strcmp(name, "") == 0)
7241                 return -EINVAL;
7242         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7243                                name, buffer, size);
7244 }
7245
7246 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7247                 const void *value, size_t size, int flags, int type)
7248 {
7249         if (strcmp(name, "") == 0)
7250                 return -EINVAL;
7251
7252         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7253                                name, value, size, flags);
7254 }
7255
7256 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7257                      void *fs_info)
7258 {
7259         const struct xattr *xattr;
7260         int err = 0;
7261
7262         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7263                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7264                                       xattr->name, xattr->value,
7265                                       xattr->value_len, XATTR_CREATE);
7266                 if (err)
7267                         break;
7268         }
7269         return err;
7270 }
7271
7272 int ocfs2_init_security_get(struct inode *inode,
7273                             struct inode *dir,
7274                             const struct qstr *qstr,
7275                             struct ocfs2_security_xattr_info *si)
7276 {
7277         /* check whether ocfs2 support feature xattr */
7278         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7279                 return -EOPNOTSUPP;
7280         if (si)
7281                 return security_old_inode_init_security(inode, dir, qstr,
7282                                                         &si->name, &si->value,
7283                                                         &si->value_len);
7284
7285         return security_inode_init_security(inode, dir, qstr,
7286                                             &ocfs2_initxattrs, NULL);
7287 }
7288
7289 int ocfs2_init_security_set(handle_t *handle,
7290                             struct inode *inode,
7291                             struct buffer_head *di_bh,
7292                             struct ocfs2_security_xattr_info *si,
7293                             struct ocfs2_alloc_context *xattr_ac,
7294                             struct ocfs2_alloc_context *data_ac)
7295 {
7296         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7297                                      OCFS2_XATTR_INDEX_SECURITY,
7298                                      si->name, si->value, si->value_len, 0,
7299                                      xattr_ac, data_ac);
7300 }
7301
7302 const struct xattr_handler ocfs2_xattr_security_handler = {
7303         .prefix = XATTR_SECURITY_PREFIX,
7304         .list   = ocfs2_xattr_security_list,
7305         .get    = ocfs2_xattr_security_get,
7306         .set    = ocfs2_xattr_security_set,
7307 };
7308
7309 /*
7310  * 'trusted' attributes support
7311  */
7312 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7313                                        size_t list_size, const char *name,
7314                                        size_t name_len, int type)
7315 {
7316         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7317         const size_t total_len = prefix_len + name_len + 1;
7318
7319         if (list && total_len <= list_size) {
7320                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7321                 memcpy(list + prefix_len, name, name_len);
7322                 list[prefix_len + name_len] = '\0';
7323         }
7324         return total_len;
7325 }
7326
7327 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7328                 void *buffer, size_t size, int type)
7329 {
7330         if (strcmp(name, "") == 0)
7331                 return -EINVAL;
7332         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7333                                name, buffer, size);
7334 }
7335
7336 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7337                 const void *value, size_t size, int flags, int type)
7338 {
7339         if (strcmp(name, "") == 0)
7340                 return -EINVAL;
7341
7342         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7343                                name, value, size, flags);
7344 }
7345
7346 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7347         .prefix = XATTR_TRUSTED_PREFIX,
7348         .list   = ocfs2_xattr_trusted_list,
7349         .get    = ocfs2_xattr_trusted_get,
7350         .set    = ocfs2_xattr_trusted_set,
7351 };
7352
7353 /*
7354  * 'user' attributes support
7355  */
7356 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7357                                     size_t list_size, const char *name,
7358                                     size_t name_len, int type)
7359 {
7360         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7361         const size_t total_len = prefix_len + name_len + 1;
7362         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7363
7364         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7365                 return 0;
7366
7367         if (list && total_len <= list_size) {
7368                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7369                 memcpy(list + prefix_len, name, name_len);
7370                 list[prefix_len + name_len] = '\0';
7371         }
7372         return total_len;
7373 }
7374
7375 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7376                 void *buffer, size_t size, int type)
7377 {
7378         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7379
7380         if (strcmp(name, "") == 0)
7381                 return -EINVAL;
7382         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7383                 return -EOPNOTSUPP;
7384         return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7385                                buffer, size);
7386 }
7387
7388 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7389                 const void *value, size_t size, int flags, int type)
7390 {
7391         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7392
7393         if (strcmp(name, "") == 0)
7394                 return -EINVAL;
7395         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7396                 return -EOPNOTSUPP;
7397
7398         return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7399                                name, value, size, flags);
7400 }
7401
7402 const struct xattr_handler ocfs2_xattr_user_handler = {
7403         .prefix = XATTR_USER_PREFIX,
7404         .list   = ocfs2_xattr_user_list,
7405         .get    = ocfs2_xattr_user_get,
7406         .set    = ocfs2_xattr_user_set,
7407 };