fs/ocfs2/xattr.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * xattr.c
   6  *
   7  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   8  *
   9  * CREDITS:
  10  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  11  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  12  */
  13
  14 #include <linux/capability.h>
  15 #include <linux/fs.h>
  16 #include <linux/types.h>
  17 #include <linux/slab.h>
  18 #include <linux/highmem.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/uio.h>
  21 #include <linux/sched.h>
  22 #include <linux/splice.h>
  23 #include <linux/mount.h>
  24 #include <linux/writeback.h>
  25 #include <linux/falloc.h>
  26 #include <linux/sort.h>
  27 #include <linux/init.h>
  28 #include <linux/module.h>
  29 #include <linux/string.h>
  30 #include <linux/security.h>
  31
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35 #include "alloc.h"
  36 #include "blockcheck.h"
  37 #include "dlmglue.h"
  38 #include "file.h"
  39 #include "symlink.h"
  40 #include "sysfile.h"
  41 #include "inode.h"
  42 #include "journal.h"
  43 #include "ocfs2_fs.h"
  44 #include "suballoc.h"
  45 #include "uptodate.h"
  46 #include "buffer_head_io.h"
  47 #include "super.h"
  48 #include "xattr.h"
  49 #include "refcounttree.h"
  50 #include "acl.h"
  51 #include "ocfs2_trace.h"
  52
  53 struct ocfs2_xattr_def_value_root {
  54         struct ocfs2_xattr_value_root   xv;
  55         struct ocfs2_extent_rec         er;
  56 };
  57
  58 struct ocfs2_xattr_bucket {
  59         /* The inode these xattrs are associated with */
  60         struct inode *bu_inode;
  61
  62         /* The actual buffers that make up the bucket */
  63         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  64
  65         /* How many blocks make up one bucket for this filesystem */
  66         int bu_blocks;
  67 };
  68
  69 struct ocfs2_xattr_set_ctxt {
  70         handle_t *handle;
  71         struct ocfs2_alloc_context *meta_ac;
  72         struct ocfs2_alloc_context *data_ac;
  73         struct ocfs2_cached_dealloc_ctxt dealloc;
  74         int set_abort;
  75 };
  76
  77 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  78 #define OCFS2_XATTR_INLINE_SIZE 80
  79 #define OCFS2_XATTR_HEADER_GAP  4
  80 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  81                                          - sizeof(struct ocfs2_xattr_header) \
  82                                          - OCFS2_XATTR_HEADER_GAP)
  83 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  84                                          - sizeof(struct ocfs2_xattr_block) \
  85                                          - sizeof(struct ocfs2_xattr_header) \
  86                                          - OCFS2_XATTR_HEADER_GAP)
  87
  88 static struct ocfs2_xattr_def_value_root def_xv = {
  89         .xv.xr_list.l_count = cpu_to_le16(1),
  90 };
  91
  92 const struct xattr_handler *ocfs2_xattr_handlers[] = {
  93         &ocfs2_xattr_user_handler,
  94         &posix_acl_access_xattr_handler,
  95         &posix_acl_default_xattr_handler,
  96         &ocfs2_xattr_trusted_handler,
  97         &ocfs2_xattr_security_handler,
  98         NULL
  99 };
 100
 101 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 102         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 103         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 104                                         = &posix_acl_access_xattr_handler,
 105         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 106                                         = &posix_acl_default_xattr_handler,
 107         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 108         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 109 };
 110
 111 struct ocfs2_xattr_info {
 112         int             xi_name_index;
 113         const char      *xi_name;
 114         int             xi_name_len;
 115         const void      *xi_value;
 116         size_t          xi_value_len;
 117 };
 118
 119 struct ocfs2_xattr_search {
 120         struct buffer_head *inode_bh;
 121         /*
 122          * xattr_bh point to the block buffer head which has extended attribute
 123          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 124          */
 125         struct buffer_head *xattr_bh;
 126         struct ocfs2_xattr_header *header;
 127         struct ocfs2_xattr_bucket *bucket;
 128         void *base;
 129         void *end;
 130         struct ocfs2_xattr_entry *here;
 131         int not_found;
 132 };
 133
 134 /* Operations on struct ocfs2_xa_entry */
 135 struct ocfs2_xa_loc;
 136 struct ocfs2_xa_loc_operations {
 137         /*
 138          * Journal functions
 139          */
 140         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 141                                   int type);
 142         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 143
 144         /*
 145          * Return a pointer to the appropriate buffer in loc->xl_storage
 146          * at the given offset from loc->xl_header.
 147          */
 148         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 149
 150         /* Can we reuse the existing entry for the new value? */
 151         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 152                              struct ocfs2_xattr_info *xi);
 153
 154         /* How much space is needed for the new value? */
 155         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 156                                struct ocfs2_xattr_info *xi);
 157
 158         /*
 159          * Return the offset of the first name+value pair.  This is
 160          * the start of our downward-filling free space.
 161          */
 162         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 163
 164         /*
 165          * Remove the name+value at this location.  Do whatever is
 166          * appropriate with the remaining name+value pairs.
 167          */
 168         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 169
 170         /* Fill xl_entry with a new entry */
 171         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 172
 173         /* Add name+value storage to an entry */
 174         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 175
 176         /*
 177          * Initialize the value buf's access and bh fields for this entry.
 178          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 179          */
 180         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 181                                    struct ocfs2_xattr_value_buf *vb);
 182 };
 183
 184 /*
 185  * Describes an xattr entry location.  This is a memory structure
 186  * tracking the on-disk structure.
 187  */
 188 struct ocfs2_xa_loc {
 189         /* This xattr belongs to this inode */
 190         struct inode *xl_inode;
 191
 192         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 193         struct ocfs2_xattr_header *xl_header;
 194
 195         /* Bytes from xl_header to the end of the storage */
 196         int xl_size;
 197
 198         /*
 199          * The ocfs2_xattr_entry this location describes.  If this is
 200          * NULL, this location describes the on-disk structure where it
 201          * would have been.
 202          */
 203         struct ocfs2_xattr_entry *xl_entry;
 204
 205         /*
 206          * Internal housekeeping
 207          */
 208
 209         /* Buffer(s) containing this entry */
 210         void *xl_storage;
 211
 212         /* Operations on the storage backing this location */
 213         const struct ocfs2_xa_loc_operations *xl_ops;
 214 };
 215
 216 /*
 217  * Convenience functions to calculate how much space is needed for a
 218  * given name+value pair
 219  */
 220 static int namevalue_size(int name_len, uint64_t value_len)
 221 {
 222         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 223                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 224         else
 225                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 226 }
 227
 228 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 229 {
 230         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 231 }
 232
 233 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 234 {
 235         u64 value_len = le64_to_cpu(xe->xe_value_size);
 236
 237         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 238                ocfs2_xattr_is_local(xe));
 239         return namevalue_size(xe->xe_name_len, value_len);
 240 }
 241
 242
 243 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 244                                              struct ocfs2_xattr_header *xh,
 245                                              int index,
 246                                              int *block_off,
 247                                              int *new_offset);
 248
 249 static int ocfs2_xattr_block_find(struct inode *inode,
 250                                   int name_index,
 251                                   const char *name,
 252                                   struct ocfs2_xattr_search *xs);
 253 static int ocfs2_xattr_index_block_find(struct inode *inode,
 254                                         struct buffer_head *root_bh,
 255                                         int name_index,
 256                                         const char *name,
 257                                         struct ocfs2_xattr_search *xs);
 258
 259 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 260                                         struct buffer_head *blk_bh,
 261                                         char *buffer,
 262                                         size_t buffer_size);
 263
 264 static int ocfs2_xattr_create_index_block(struct inode *inode,
 265                                           struct ocfs2_xattr_search *xs,
 266                                           struct ocfs2_xattr_set_ctxt *ctxt);
 267
 268 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 269                                              struct ocfs2_xattr_info *xi,
 270                                              struct ocfs2_xattr_search *xs,
 271                                              struct ocfs2_xattr_set_ctxt *ctxt);
 272
 273 typedef int (xattr_tree_rec_func)(struct inode *inode,
 274                                   struct buffer_head *root_bh,
 275                                   u64 blkno, u32 cpos, u32 len, void *para);
 276 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 277                                            struct buffer_head *root_bh,
 278                                            xattr_tree_rec_func *rec_func,
 279                                            void *para);
 280 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 281                                         struct ocfs2_xattr_bucket *bucket,
 282                                         void *para);
 283 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 284                                   struct buffer_head *root_bh,
 285                                   u64 blkno,
 286                                   u32 cpos,
 287                                   u32 len,
 288                                   void *para);
 289
 290 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 291                                   u64 src_blk, u64 last_blk, u64 to_blk,
 292                                   unsigned int start_bucket,
 293                                   u32 *first_hash);
 294 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 295                                         struct ocfs2_dinode *di,
 296                                         struct ocfs2_xattr_info *xi,
 297                                         struct ocfs2_xattr_search *xis,
 298                                         struct ocfs2_xattr_search *xbs,
 299                                         struct ocfs2_refcount_tree **ref_tree,
 300                                         int *meta_need,
 301                                         int *credits);
 302 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 303                                            struct ocfs2_xattr_bucket *bucket,
 304                                            int offset,
 305                                            struct ocfs2_xattr_value_root **xv,
 306                                            struct buffer_head **bh);
 307
 308 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 309 {
 310         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 311 }
 312
 313 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 314 {
 315         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 316 }
 317
 318 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 319 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 320 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 321
 322 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 323 {
 324         struct ocfs2_xattr_bucket *bucket;
 325         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 326
 327         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 328
 329         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 330         if (bucket) {
 331                 bucket->bu_inode = inode;
 332                 bucket->bu_blocks = blks;
 333         }
 334
 335         return bucket;
 336 }
 337
 338 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 339 {
 340         int i;
 341
 342         for (i = 0; i < bucket->bu_blocks; i++) {
 343                 brelse(bucket->bu_bhs[i]);
 344                 bucket->bu_bhs[i] = NULL;
 345         }
 346 }
 347
 348 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 349 {
 350         if (bucket) {
 351                 ocfs2_xattr_bucket_relse(bucket);
 352                 bucket->bu_inode = NULL;
 353                 kfree(bucket);
 354         }
 355 }
 356
 357 /*
 358  * A bucket that has never been written to disk doesn't need to be
 359  * read.  We just need the buffer_heads.  Don't call this for
 360  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 361  * them fully.
 362  */
 363 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 364                                    u64 xb_blkno, int new)
 365 {
 366         int i, rc = 0;
 367
 368         for (i = 0; i < bucket->bu_blocks; i++) {
 369                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 370                                               xb_blkno + i);
 371                 if (!bucket->bu_bhs[i]) {
 372                         rc = -ENOMEM;
 373                         mlog_errno(rc);
 374                         break;
 375                 }
 376
 377                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 378                                            bucket->bu_bhs[i])) {
 379                         if (new)
 380                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 381                                                               bucket->bu_bhs[i]);
 382                         else {
 383                                 set_buffer_uptodate(bucket->bu_bhs[i]);
 384                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 385                                                           bucket->bu_bhs[i]);
 386                         }
 387                 }
 388         }
 389
 390         if (rc)
 391                 ocfs2_xattr_bucket_relse(bucket);
 392         return rc;
 393 }
 394
 395 /* Read the xattr bucket at xb_blkno */
 396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 397                                    u64 xb_blkno)
 398 {
 399         int rc;
 400
 401         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 402                                bucket->bu_blocks, bucket->bu_bhs, 0,
 403                                NULL);
 404         if (!rc) {
 405                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 406                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 407                                                  bucket->bu_bhs,
 408                                                  bucket->bu_blocks,
 409                                                  &bucket_xh(bucket)->xh_check);
 410                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 411                 if (rc)
 412                         mlog_errno(rc);
 413         }
 414
 415         if (rc)
 416                 ocfs2_xattr_bucket_relse(bucket);
 417         return rc;
 418 }
 419
 420 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 421                                              struct ocfs2_xattr_bucket *bucket,
 422                                              int type)
 423 {
 424         int i, rc = 0;
 425
 426         for (i = 0; i < bucket->bu_blocks; i++) {
 427                 rc = ocfs2_journal_access(handle,
 428                                           INODE_CACHE(bucket->bu_inode),
 429                                           bucket->bu_bhs[i], type);
 430                 if (rc) {
 431                         mlog_errno(rc);
 432                         break;
 433                 }
 434         }
 435
 436         return rc;
 437 }
 438
 439 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 440                                              struct ocfs2_xattr_bucket *bucket)
 441 {
 442         int i;
 443
 444         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 445         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 446                                    bucket->bu_bhs, bucket->bu_blocks,
 447                                    &bucket_xh(bucket)->xh_check);
 448         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 449
 450         for (i = 0; i < bucket->bu_blocks; i++)
 451                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 452 }
 453
 454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 455                                          struct ocfs2_xattr_bucket *src)
 456 {
 457         int i;
 458         int blocksize = src->bu_inode->i_sb->s_blocksize;
 459
 460         BUG_ON(dest->bu_blocks != src->bu_blocks);
 461         BUG_ON(dest->bu_inode != src->bu_inode);
 462
 463         for (i = 0; i < src->bu_blocks; i++) {
 464                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 465                        blocksize);
 466         }
 467 }
 468
 469 static int ocfs2_validate_xattr_block(struct super_block *sb,
 470                                       struct buffer_head *bh)
 471 {
 472         int rc;
 473         struct ocfs2_xattr_block *xb =
 474                 (struct ocfs2_xattr_block *)bh->b_data;
 475
 476         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 477
 478         BUG_ON(!buffer_uptodate(bh));
 479
 480         /*
 481          * If the ecc fails, we return the error but otherwise
 482          * leave the filesystem running.  We know any error is
 483          * local to this block.
 484          */
 485         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 486         if (rc)
 487                 return rc;
 488
 489         /*
 490          * Errors after here are fatal
 491          */
 492
 493         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 494                 return ocfs2_error(sb,
 495                                    "Extended attribute block #%llu has bad signature %.*s\n",
 496                                    (unsigned long long)bh->b_blocknr, 7,
 497                                    xb->xb_signature);
 498         }
 499
 500         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 501                 return ocfs2_error(sb,
 502                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 503                                    (unsigned long long)bh->b_blocknr,
 504                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
 505         }
 506
 507         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 508                 return ocfs2_error(sb,
 509                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 510                                    (unsigned long long)bh->b_blocknr,
 511                                    le32_to_cpu(xb->xb_fs_generation));
 512         }
 513
 514         return 0;
 515 }
 516
 517 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 518                                   struct buffer_head **bh)
 519 {
 520         int rc;
 521         struct buffer_head *tmp = *bh;
 522
 523         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 524                               ocfs2_validate_xattr_block);
 525
 526         /* If ocfs2_read_block() got us a new bh, pass it up. */
 527         if (!rc && !*bh)
 528                 *bh = tmp;
 529
 530         return rc;
 531 }
 532
 533 static inline const char *ocfs2_xattr_prefix(int name_index)
 534 {
 535         const struct xattr_handler *handler = NULL;
 536
 537         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 538                 handler = ocfs2_xattr_handler_map[name_index];
 539         return handler ? xattr_prefix(handler) : NULL;
 540 }
 541
 542 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 543                                  const char *name,
 544                                  int name_len)
 545 {
 546         /* Get hash value of uuid from super block */
 547         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 548         int i;
 549
 550         /* hash extended attribute name */
 551         for (i = 0; i < name_len; i++) {
 552                 hash = (hash << OCFS2_HASH_SHIFT) ^
 553                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 554                        *name++;
 555         }
 556
 557         return hash;
 558 }
 559
 560 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 561 {
 562         return namevalue_size(name_len, value_len) +
 563                 sizeof(struct ocfs2_xattr_entry);
 564 }
 565
 566 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 567 {
 568         return namevalue_size_xi(xi) +
 569                 sizeof(struct ocfs2_xattr_entry);
 570 }
 571
 572 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 573 {
 574         return namevalue_size_xe(xe) +
 575                 sizeof(struct ocfs2_xattr_entry);
 576 }
 577
 578 int ocfs2_calc_security_init(struct inode *dir,
 579                              struct ocfs2_security_xattr_info *si,
 580                              int *want_clusters,
 581                              int *xattr_credits,
 582                              struct ocfs2_alloc_context **xattr_ac)
 583 {
 584         int ret = 0;
 585         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 586         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 587                                                  si->value_len);
 588
 589         /*
 590          * The max space of security xattr taken inline is
 591          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 592          * So reserve one metadata block for it is ok.
 593          */
 594         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 595             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 596                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 597                 if (ret) {
 598                         mlog_errno(ret);
 599                         return ret;
 600                 }
 601                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 602         }
 603
 604         /* reserve clusters for xattr value which will be set in B tree*/
 605         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 606                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 607                                                             si->value_len);
 608
 609                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 610                                                            new_clusters);
 611                 *want_clusters += new_clusters;
 612         }
 613         return ret;
 614 }
 615
 616 int ocfs2_calc_xattr_init(struct inode *dir,
 617                           struct buffer_head *dir_bh,
 618                           umode_t mode,
 619                           struct ocfs2_security_xattr_info *si,
 620                           int *want_clusters,
 621                           int *xattr_credits,
 622                           int *want_meta)
 623 {
 624         int ret = 0;
 625         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 626         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 627
 628         if (si->enable)
 629                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 630                                                      si->value_len);
 631
 632         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 633                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
 634                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 635                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 636                                         "", NULL, 0);
 637                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
 638                 if (acl_len > 0) {
 639                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 640                         if (S_ISDIR(mode))
 641                                 a_size <<= 1;
 642                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 643                         ret = acl_len;
 644                         mlog_errno(ret);
 645                         return ret;
 646                 }
 647         }
 648
 649         if (!(s_size + a_size))
 650                 return ret;
 651
 652         /*
 653          * The max space of security xattr taken inline is
 654          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 655          * The max space of acl xattr taken inline is
 656          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 657          * when blocksize = 512, may reserve one more cluser for
 658          * xattr bucket, otherwise reserve one metadata block
 659          * for them is ok.
 660          * If this is a new directory with inline data,
 661          * we choose to reserve the entire inline area for
 662          * directory contents and force an external xattr block.
 663          */
 664         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 665             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 666             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 667                 *want_meta = *want_meta + 1;
 668                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 669         }
 670
 671         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 672             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 673                 *want_clusters += 1;
 674                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 675         }
 676
 677         /*
 678          * reserve credits and clusters for xattrs which has large value
 679          * and have to be set outside
 680          */
 681         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 682                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 683                                                         si->value_len);
 684                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 685                                                            new_clusters);
 686                 *want_clusters += new_clusters;
 687         }
 688         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 689             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 690                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 691                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 692                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 693                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 694                                                            new_clusters);
 695                 *want_clusters += new_clusters;
 696         }
 697
 698         return ret;
 699 }
 700
 701 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 702                                          u32 clusters_to_add,
 703                                          struct ocfs2_xattr_value_buf *vb,
 704                                          struct ocfs2_xattr_set_ctxt *ctxt)
 705 {
 706         int status = 0, credits;
 707         handle_t *handle = ctxt->handle;
 708         enum ocfs2_alloc_restarted why;
 709         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 710         struct ocfs2_extent_tree et;
 711
 712         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 713
 714         while (clusters_to_add) {
 715                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 716
 717                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 718                                        OCFS2_JOURNAL_ACCESS_WRITE);
 719                 if (status < 0) {
 720                         mlog_errno(status);
 721                         break;
 722                 }
 723
 724                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 725                 status = ocfs2_add_clusters_in_btree(handle,
 726                                                      &et,
 727                                                      &logical_start,
 728                                                      clusters_to_add,
 729                                                      0,
 730                                                      ctxt->data_ac,
 731                                                      ctxt->meta_ac,
 732                                                      &why);
 733                 if ((status < 0) && (status != -EAGAIN)) {
 734                         if (status != -ENOSPC)
 735                                 mlog_errno(status);
 736                         break;
 737                 }
 738
 739                 ocfs2_journal_dirty(handle, vb->vb_bh);
 740
 741                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 742                                          prev_clusters;
 743
 744                 if (why != RESTART_NONE && clusters_to_add) {
 745                         /*
 746                          * We can only fail in case the alloc file doesn't give
 747                          * up enough clusters.
 748                          */
 749                         BUG_ON(why == RESTART_META);
 750
 751                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 752                                                             &vb->vb_xv->xr_list);
 753                         status = ocfs2_extend_trans(handle, credits);
 754                         if (status < 0) {
 755                                 status = -ENOMEM;
 756                                 mlog_errno(status);
 757                                 break;
 758                         }
 759                 }
 760         }
 761
 762         return status;
 763 }
 764
 765 static int __ocfs2_remove_xattr_range(struct inode *inode,
 766                                       struct ocfs2_xattr_value_buf *vb,
 767                                       u32 cpos, u32 phys_cpos, u32 len,
 768                                       unsigned int ext_flags,
 769                                       struct ocfs2_xattr_set_ctxt *ctxt)
 770 {
 771         int ret;
 772         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 773         handle_t *handle = ctxt->handle;
 774         struct ocfs2_extent_tree et;
 775
 776         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 777
 778         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 779                             OCFS2_JOURNAL_ACCESS_WRITE);
 780         if (ret) {
 781                 mlog_errno(ret);
 782                 goto out;
 783         }
 784
 785         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 786                                   &ctxt->dealloc);
 787         if (ret) {
 788                 mlog_errno(ret);
 789                 goto out;
 790         }
 791
 792         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 793         ocfs2_journal_dirty(handle, vb->vb_bh);
 794
 795         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 796                 ret = ocfs2_decrease_refcount(inode, handle,
 797                                         ocfs2_blocks_to_clusters(inode->i_sb,
 798                                                                  phys_blkno),
 799                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 800         else
 801                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 802                                                   phys_blkno, len);
 803         if (ret)
 804                 mlog_errno(ret);
 805
 806 out:
 807         return ret;
 808 }
 809
 810 static int ocfs2_xattr_shrink_size(struct inode *inode,
 811                                    u32 old_clusters,
 812                                    u32 new_clusters,
 813                                    struct ocfs2_xattr_value_buf *vb,
 814                                    struct ocfs2_xattr_set_ctxt *ctxt)
 815 {
 816         int ret = 0;
 817         unsigned int ext_flags;
 818         u32 trunc_len, cpos, phys_cpos, alloc_size;
 819         u64 block;
 820
 821         if (old_clusters <= new_clusters)
 822                 return 0;
 823
 824         cpos = new_clusters;
 825         trunc_len = old_clusters - new_clusters;
 826         while (trunc_len) {
 827                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 828                                                &alloc_size,
 829                                                &vb->vb_xv->xr_list, &ext_flags);
 830                 if (ret) {
 831                         mlog_errno(ret);
 832                         goto out;
 833                 }
 834
 835                 if (alloc_size > trunc_len)
 836                         alloc_size = trunc_len;
 837
 838                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 839                                                  phys_cpos, alloc_size,
 840                                                  ext_flags, ctxt);
 841                 if (ret) {
 842                         mlog_errno(ret);
 843                         goto out;
 844                 }
 845
 846                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 847                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 848                                                        block, alloc_size);
 849                 cpos += alloc_size;
 850                 trunc_len -= alloc_size;
 851         }
 852
 853 out:
 854         return ret;
 855 }
 856
 857 static int ocfs2_xattr_value_truncate(struct inode *inode,
 858                                       struct ocfs2_xattr_value_buf *vb,
 859                                       int len,
 860                                       struct ocfs2_xattr_set_ctxt *ctxt)
 861 {
 862         int ret;
 863         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 864         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 865
 866         if (new_clusters == old_clusters)
 867                 return 0;
 868
 869         if (new_clusters > old_clusters)
 870                 ret = ocfs2_xattr_extend_allocation(inode,
 871                                                     new_clusters - old_clusters,
 872                                                     vb, ctxt);
 873         else
 874                 ret = ocfs2_xattr_shrink_size(inode,
 875                                               old_clusters, new_clusters,
 876                                               vb, ctxt);
 877
 878         return ret;
 879 }
 880
 881 static int ocfs2_xattr_list_entry(struct super_block *sb,
 882                                   char *buffer, size_t size,
 883                                   size_t *result, int type,
 884                                   const char *name, int name_len)
 885 {
 886         char *p = buffer + *result;
 887         const char *prefix;
 888         int prefix_len;
 889         int total_len;
 890
 891         switch(type) {
 892         case OCFS2_XATTR_INDEX_USER:
 893                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
 894                         return 0;
 895                 break;
 896
 897         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
 898         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
 899                 if (!(sb->s_flags & SB_POSIXACL))
 900                         return 0;
 901                 break;
 902
 903         case OCFS2_XATTR_INDEX_TRUSTED:
 904                 if (!capable(CAP_SYS_ADMIN))
 905                         return 0;
 906                 break;
 907         }
 908
 909         prefix = ocfs2_xattr_prefix(type);
 910         if (!prefix)
 911                 return 0;
 912         prefix_len = strlen(prefix);
 913         total_len = prefix_len + name_len + 1;
 914         *result += total_len;
 915
 916         /* we are just looking for how big our buffer needs to be */
 917         if (!size)
 918                 return 0;
 919
 920         if (*result > size)
 921                 return -ERANGE;
 922
 923         memcpy(p, prefix, prefix_len);
 924         memcpy(p + prefix_len, name, name_len);
 925         p[prefix_len + name_len] = '\0';
 926
 927         return 0;
 928 }
 929
 930 static int ocfs2_xattr_list_entries(struct inode *inode,
 931                                     struct ocfs2_xattr_header *header,
 932                                     char *buffer, size_t buffer_size)
 933 {
 934         size_t result = 0;
 935         int i, type, ret;
 936         const char *name;
 937
 938         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 939                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 940                 type = ocfs2_xattr_get_type(entry);
 941                 name = (const char *)header +
 942                         le16_to_cpu(entry->xe_name_offset);
 943
 944                 ret = ocfs2_xattr_list_entry(inode->i_sb,
 945                                              buffer, buffer_size,
 946                                              &result, type, name,
 947                                              entry->xe_name_len);
 948                 if (ret)
 949                         return ret;
 950         }
 951
 952         return result;
 953 }
 954
 955 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 956                                          struct ocfs2_dinode *di)
 957 {
 958         struct ocfs2_xattr_header *xh;
 959         int i;
 960
 961         xh = (struct ocfs2_xattr_header *)
 962                  ((void *)di + inode->i_sb->s_blocksize -
 963                  le16_to_cpu(di->i_xattr_inline_size));
 964
 965         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 966                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 967                         return 1;
 968
 969         return 0;
 970 }
 971
 972 static int ocfs2_xattr_ibody_list(struct inode *inode,
 973                                   struct ocfs2_dinode *di,
 974                                   char *buffer,
 975                                   size_t buffer_size)
 976 {
 977         struct ocfs2_xattr_header *header = NULL;
 978         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 979         int ret = 0;
 980
 981         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 982                 return ret;
 983
 984         header = (struct ocfs2_xattr_header *)
 985                  ((void *)di + inode->i_sb->s_blocksize -
 986                  le16_to_cpu(di->i_xattr_inline_size));
 987
 988         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 989
 990         return ret;
 991 }
 992
 993 static int ocfs2_xattr_block_list(struct inode *inode,
 994                                   struct ocfs2_dinode *di,
 995                                   char *buffer,
 996                                   size_t buffer_size)
 997 {
 998         struct buffer_head *blk_bh = NULL;
 999         struct ocfs2_xattr_block *xb;
1000         int ret = 0;
1001
1002         if (!di->i_xattr_loc)
1003                 return ret;
1004
1005         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1006                                      &blk_bh);
1007         if (ret < 0) {
1008                 mlog_errno(ret);
1009                 return ret;
1010         }
1011
1012         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1013         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1014                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1015                 ret = ocfs2_xattr_list_entries(inode, header,
1016                                                buffer, buffer_size);
1017         } else
1018                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1019                                                    buffer, buffer_size);
1020
1021         brelse(blk_bh);
1022
1023         return ret;
1024 }
1025
1026 ssize_t ocfs2_listxattr(struct dentry *dentry,
1027                         char *buffer,
1028                         size_t size)
1029 {
1030         int ret = 0, i_ret = 0, b_ret = 0;
1031         struct buffer_head *di_bh = NULL;
1032         struct ocfs2_dinode *di = NULL;
1033         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1034
1035         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1036                 return -EOPNOTSUPP;
1037
1038         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1039                 return ret;
1040
1041         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1042         if (ret < 0) {
1043                 mlog_errno(ret);
1044                 return ret;
1045         }
1046
1047         di = (struct ocfs2_dinode *)di_bh->b_data;
1048
1049         down_read(&oi->ip_xattr_sem);
1050         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1051         if (i_ret < 0)
1052                 b_ret = 0;
1053         else {
1054                 if (buffer) {
1055                         buffer += i_ret;
1056                         size -= i_ret;
1057                 }
1058                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1059                                                buffer, size);
1060                 if (b_ret < 0)
1061                         i_ret = 0;
1062         }
1063         up_read(&oi->ip_xattr_sem);
1064         ocfs2_inode_unlock(d_inode(dentry), 0);
1065
1066         brelse(di_bh);
1067
1068         return i_ret + b_ret;
1069 }
1070
1071 static int ocfs2_xattr_find_entry(int name_index,
1072                                   const char *name,
1073                                   struct ocfs2_xattr_search *xs)
1074 {
1075         struct ocfs2_xattr_entry *entry;
1076         size_t name_len;
1077         int i, cmp = 1;
1078
1079         if (name == NULL)
1080                 return -EINVAL;
1081
1082         name_len = strlen(name);
1083         entry = xs->here;
1084         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1085                 cmp = name_index - ocfs2_xattr_get_type(entry);
1086                 if (!cmp)
1087                         cmp = name_len - entry->xe_name_len;
1088                 if (!cmp)
1089                         cmp = memcmp(name, (xs->base +
1090                                      le16_to_cpu(entry->xe_name_offset)),
1091                                      name_len);
1092                 if (cmp == 0)
1093                         break;
1094                 entry += 1;
1095         }
1096         xs->here = entry;
1097
1098         return cmp ? -ENODATA : 0;
1099 }
1100
1101 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1102                                          struct ocfs2_xattr_value_root *xv,
1103                                          void *buffer,
1104                                          size_t len)
1105 {
1106         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1107         u64 blkno;
1108         int i, ret = 0;
1109         size_t cplen, blocksize;
1110         struct buffer_head *bh = NULL;
1111         struct ocfs2_extent_list *el;
1112
1113         el = &xv->xr_list;
1114         clusters = le32_to_cpu(xv->xr_clusters);
1115         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1116         blocksize = inode->i_sb->s_blocksize;
1117
1118         cpos = 0;
1119         while (cpos < clusters) {
1120                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1121                                                &num_clusters, el, NULL);
1122                 if (ret) {
1123                         mlog_errno(ret);
1124                         goto out;
1125                 }
1126
1127                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1128                 /* Copy ocfs2_xattr_value */
1129                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1130                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1131                                                &bh, NULL);
1132                         if (ret) {
1133                                 mlog_errno(ret);
1134                                 goto out;
1135                         }
1136
1137                         cplen = len >= blocksize ? blocksize : len;
1138                         memcpy(buffer, bh->b_data, cplen);
1139                         len -= cplen;
1140                         buffer += cplen;
1141
1142                         brelse(bh);
1143                         bh = NULL;
1144                         if (len == 0)
1145                                 break;
1146                 }
1147                 cpos += num_clusters;
1148         }
1149 out:
1150         return ret;
1151 }
1152
1153 static int ocfs2_xattr_ibody_get(struct inode *inode,
1154                                  int name_index,
1155                                  const char *name,
1156                                  void *buffer,
1157                                  size_t buffer_size,
1158                                  struct ocfs2_xattr_search *xs)
1159 {
1160         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1161         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1162         struct ocfs2_xattr_value_root *xv;
1163         size_t size;
1164         int ret = 0;
1165
1166         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1167                 return -ENODATA;
1168
1169         xs->end = (void *)di + inode->i_sb->s_blocksize;
1170         xs->header = (struct ocfs2_xattr_header *)
1171                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1172         xs->base = (void *)xs->header;
1173         xs->here = xs->header->xh_entries;
1174
1175         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1176         if (ret)
1177                 return ret;
1178         size = le64_to_cpu(xs->here->xe_value_size);
1179         if (buffer) {
1180                 if (size > buffer_size)
1181                         return -ERANGE;
1182                 if (ocfs2_xattr_is_local(xs->here)) {
1183                         memcpy(buffer, (void *)xs->base +
1184                                le16_to_cpu(xs->here->xe_name_offset) +
1185                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1186                 } else {
1187                         xv = (struct ocfs2_xattr_value_root *)
1188                                 (xs->base + le16_to_cpu(
1189                                  xs->here->xe_name_offset) +
1190                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1191                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1192                                                             buffer, size);
1193                         if (ret < 0) {
1194                                 mlog_errno(ret);
1195                                 return ret;
1196                         }
1197                 }
1198         }
1199
1200         return size;
1201 }
1202
1203 static int ocfs2_xattr_block_get(struct inode *inode,
1204                                  int name_index,
1205                                  const char *name,
1206                                  void *buffer,
1207                                  size_t buffer_size,
1208                                  struct ocfs2_xattr_search *xs)
1209 {
1210         struct ocfs2_xattr_block *xb;
1211         struct ocfs2_xattr_value_root *xv;
1212         size_t size;
1213         int ret = -ENODATA, name_offset, name_len, i;
1214         int block_off;
1215
1216         xs->bucket = ocfs2_xattr_bucket_new(inode);
1217         if (!xs->bucket) {
1218                 ret = -ENOMEM;
1219                 mlog_errno(ret);
1220                 goto cleanup;
1221         }
1222
1223         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1224         if (ret) {
1225                 mlog_errno(ret);
1226                 goto cleanup;
1227         }
1228
1229         if (xs->not_found) {
1230                 ret = -ENODATA;
1231                 goto cleanup;
1232         }
1233
1234         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1235         size = le64_to_cpu(xs->here->xe_value_size);
1236         if (buffer) {
1237                 ret = -ERANGE;
1238                 if (size > buffer_size)
1239                         goto cleanup;
1240
1241                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1242                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1243                 i = xs->here - xs->header->xh_entries;
1244
1245                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1246                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1247                                                                 bucket_xh(xs->bucket),
1248                                                                 i,
1249                                                                 &block_off,
1250                                                                 &name_offset);
1251                         if (ret) {
1252                                 mlog_errno(ret);
1253                                 goto cleanup;
1254                         }
1255                         xs->base = bucket_block(xs->bucket, block_off);
1256                 }
1257                 if (ocfs2_xattr_is_local(xs->here)) {
1258                         memcpy(buffer, (void *)xs->base +
1259                                name_offset + name_len, size);
1260                 } else {
1261                         xv = (struct ocfs2_xattr_value_root *)
1262                                 (xs->base + name_offset + name_len);
1263                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1264                                                             buffer, size);
1265                         if (ret < 0) {
1266                                 mlog_errno(ret);
1267                                 goto cleanup;
1268                         }
1269                 }
1270         }
1271         ret = size;
1272 cleanup:
1273         ocfs2_xattr_bucket_free(xs->bucket);
1274
1275         brelse(xs->xattr_bh);
1276         xs->xattr_bh = NULL;
1277         return ret;
1278 }
1279
1280 int ocfs2_xattr_get_nolock(struct inode *inode,
1281                            struct buffer_head *di_bh,
1282                            int name_index,
1283                            const char *name,
1284                            void *buffer,
1285                            size_t buffer_size)
1286 {
1287         int ret;
1288         struct ocfs2_dinode *di = NULL;
1289         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1290         struct ocfs2_xattr_search xis = {
1291                 .not_found = -ENODATA,
1292         };
1293         struct ocfs2_xattr_search xbs = {
1294                 .not_found = -ENODATA,
1295         };
1296
1297         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1298                 return -EOPNOTSUPP;
1299
1300         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1301                 return -ENODATA;
1302
1303         xis.inode_bh = xbs.inode_bh = di_bh;
1304         di = (struct ocfs2_dinode *)di_bh->b_data;
1305
1306         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1307                                     buffer_size, &xis);
1308         if (ret == -ENODATA && di->i_xattr_loc)
1309                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1310                                             buffer_size, &xbs);
1311
1312         return ret;
1313 }
1314
1315 /* ocfs2_xattr_get()
1316  *
1317  * Copy an extended attribute into the buffer provided.
1318  * Buffer is NULL to compute the size of buffer required.
1319  */
1320 static int ocfs2_xattr_get(struct inode *inode,
1321                            int name_index,
1322                            const char *name,
1323                            void *buffer,
1324                            size_t buffer_size)
1325 {
1326         int ret, had_lock;
1327         struct buffer_head *di_bh = NULL;
1328         struct ocfs2_lock_holder oh;
1329
1330         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1331         if (had_lock < 0) {
1332                 mlog_errno(had_lock);
1333                 return had_lock;
1334         }
1335         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1336         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1337                                      name, buffer, buffer_size);
1338         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1339
1340         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1341
1342         brelse(di_bh);
1343
1344         return ret;
1345 }
1346
1347 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1348                                            handle_t *handle,
1349                                            struct ocfs2_xattr_value_buf *vb,
1350                                            const void *value,
1351                                            int value_len)
1352 {
1353         int ret = 0, i, cp_len;
1354         u16 blocksize = inode->i_sb->s_blocksize;
1355         u32 p_cluster, num_clusters;
1356         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1357         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1358         u64 blkno;
1359         struct buffer_head *bh = NULL;
1360         unsigned int ext_flags;
1361         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1362
1363         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1364
1365         while (cpos < clusters) {
1366                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1367                                                &num_clusters, &xv->xr_list,
1368                                                &ext_flags);
1369                 if (ret) {
1370                         mlog_errno(ret);
1371                         goto out;
1372                 }
1373
1374                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1375
1376                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1377
1378                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1379                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1380                                                &bh, NULL);
1381                         if (ret) {
1382                                 mlog_errno(ret);
1383                                 goto out;
1384                         }
1385
1386                         ret = ocfs2_journal_access(handle,
1387                                                    INODE_CACHE(inode),
1388                                                    bh,
1389                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1390                         if (ret < 0) {
1391                                 mlog_errno(ret);
1392                                 goto out;
1393                         }
1394
1395                         cp_len = value_len > blocksize ? blocksize : value_len;
1396                         memcpy(bh->b_data, value, cp_len);
1397                         value_len -= cp_len;
1398                         value += cp_len;
1399                         if (cp_len < blocksize)
1400                                 memset(bh->b_data + cp_len, 0,
1401                                        blocksize - cp_len);
1402
1403                         ocfs2_journal_dirty(handle, bh);
1404                         brelse(bh);
1405                         bh = NULL;
1406
1407                         /*
1408                          * XXX: do we need to empty all the following
1409                          * blocks in this cluster?
1410                          */
1411                         if (!value_len)
1412                                 break;
1413                 }
1414                 cpos += num_clusters;
1415         }
1416 out:
1417         brelse(bh);
1418
1419         return ret;
1420 }
1421
1422 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1423                                        int num_entries)
1424 {
1425         int free_space;
1426
1427         if (!needed_space)
1428                 return 0;
1429
1430         free_space = free_start -
1431                 sizeof(struct ocfs2_xattr_header) -
1432                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1433                 OCFS2_XATTR_HEADER_GAP;
1434         if (free_space < 0)
1435                 return -EIO;
1436         if (free_space < needed_space)
1437                 return -ENOSPC;
1438
1439         return 0;
1440 }
1441
1442 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1443                                    int type)
1444 {
1445         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1446 }
1447
1448 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1449 {
1450         loc->xl_ops->xlo_journal_dirty(handle, loc);
1451 }
1452
1453 /* Give a pointer into the storage for the given offset */
1454 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1455 {
1456         BUG_ON(offset >= loc->xl_size);
1457         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1458 }
1459
1460 /*
1461  * Wipe the name+value pair and allow the storage to reclaim it.  This
1462  * must be followed by either removal of the entry or a call to
1463  * ocfs2_xa_add_namevalue().
1464  */
1465 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1466 {
1467         loc->xl_ops->xlo_wipe_namevalue(loc);
1468 }
1469
1470 /*
1471  * Find lowest offset to a name+value pair.  This is the start of our
1472  * downward-growing free space.
1473  */
1474 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1475 {
1476         return loc->xl_ops->xlo_get_free_start(loc);
1477 }
1478
1479 /* Can we reuse loc->xl_entry for xi? */
1480 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1481                                     struct ocfs2_xattr_info *xi)
1482 {
1483         return loc->xl_ops->xlo_can_reuse(loc, xi);
1484 }
1485
1486 /* How much free space is needed to set the new value */
1487 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1488                                 struct ocfs2_xattr_info *xi)
1489 {
1490         return loc->xl_ops->xlo_check_space(loc, xi);
1491 }
1492
1493 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1494 {
1495         loc->xl_ops->xlo_add_entry(loc, name_hash);
1496         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1497         /*
1498          * We can't leave the new entry's xe_name_offset at zero or
1499          * add_namevalue() will go nuts.  We set it to the size of our
1500          * storage so that it can never be less than any other entry.
1501          */
1502         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1503 }
1504
1505 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1506                                    struct ocfs2_xattr_info *xi)
1507 {
1508         int size = namevalue_size_xi(xi);
1509         int nameval_offset;
1510         char *nameval_buf;
1511
1512         loc->xl_ops->xlo_add_namevalue(loc, size);
1513         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1514         loc->xl_entry->xe_name_len = xi->xi_name_len;
1515         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1516         ocfs2_xattr_set_local(loc->xl_entry,
1517                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1518
1519         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1520         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1521         memset(nameval_buf, 0, size);
1522         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1523 }
1524
1525 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1526                                     struct ocfs2_xattr_value_buf *vb)
1527 {
1528         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1529         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1530
1531         /* Value bufs are for value trees */
1532         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1533         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1534                (name_size + OCFS2_XATTR_ROOT_SIZE));
1535
1536         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1537         vb->vb_xv =
1538                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1539                                                         nameval_offset +
1540                                                         name_size);
1541 }
1542
1543 static int ocfs2_xa_block_journal_access(handle_t *handle,
1544                                          struct ocfs2_xa_loc *loc, int type)
1545 {
1546         struct buffer_head *bh = loc->xl_storage;
1547         ocfs2_journal_access_func access;
1548
1549         if (loc->xl_size == (bh->b_size -
1550                              offsetof(struct ocfs2_xattr_block,
1551                                       xb_attrs.xb_header)))
1552                 access = ocfs2_journal_access_xb;
1553         else
1554                 access = ocfs2_journal_access_di;
1555         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1556 }
1557
1558 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1559                                          struct ocfs2_xa_loc *loc)
1560 {
1561         struct buffer_head *bh = loc->xl_storage;
1562
1563         ocfs2_journal_dirty(handle, bh);
1564 }
1565
1566 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1567                                            int offset)
1568 {
1569         return (char *)loc->xl_header + offset;
1570 }
1571
1572 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1573                                     struct ocfs2_xattr_info *xi)
1574 {
1575         /*
1576          * Block storage is strict.  If the sizes aren't exact, we will
1577          * remove the old one and reinsert the new.
1578          */
1579         return namevalue_size_xe(loc->xl_entry) ==
1580                 namevalue_size_xi(xi);
1581 }
1582
1583 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1584 {
1585         struct ocfs2_xattr_header *xh = loc->xl_header;
1586         int i, count = le16_to_cpu(xh->xh_count);
1587         int offset, free_start = loc->xl_size;
1588
1589         for (i = 0; i < count; i++) {
1590                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1591                 if (offset < free_start)
1592                         free_start = offset;
1593         }
1594
1595         return free_start;
1596 }
1597
1598 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1599                                       struct ocfs2_xattr_info *xi)
1600 {
1601         int count = le16_to_cpu(loc->xl_header->xh_count);
1602         int free_start = ocfs2_xa_get_free_start(loc);
1603         int needed_space = ocfs2_xi_entry_usage(xi);
1604
1605         /*
1606          * Block storage will reclaim the original entry before inserting
1607          * the new value, so we only need the difference.  If the new
1608          * entry is smaller than the old one, we don't need anything.
1609          */
1610         if (loc->xl_entry) {
1611                 /* Don't need space if we're reusing! */
1612                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1613                         needed_space = 0;
1614                 else
1615                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1616         }
1617         if (needed_space < 0)
1618                 needed_space = 0;
1619         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1620 }
1621
1622 /*
1623  * Block storage for xattrs keeps the name+value pairs compacted.  When
1624  * we remove one, we have to shift any that preceded it towards the end.
1625  */
1626 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1627 {
1628         int i, offset;
1629         int namevalue_offset, first_namevalue_offset, namevalue_size;
1630         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1631         struct ocfs2_xattr_header *xh = loc->xl_header;
1632         int count = le16_to_cpu(xh->xh_count);
1633
1634         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1635         namevalue_size = namevalue_size_xe(entry);
1636         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1637
1638         /* Shift the name+value pairs */
1639         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1640                 (char *)xh + first_namevalue_offset,
1641                 namevalue_offset - first_namevalue_offset);
1642         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1643
1644         /* Now tell xh->xh_entries about it */
1645         for (i = 0; i < count; i++) {
1646                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1647                 if (offset <= namevalue_offset)
1648                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1649                                      namevalue_size);
1650         }
1651
1652         /*
1653          * Note that we don't update xh_free_start or xh_name_value_len
1654          * because they're not used in block-stored xattrs.
1655          */
1656 }
1657
1658 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1659 {
1660         int count = le16_to_cpu(loc->xl_header->xh_count);
1661         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1662         le16_add_cpu(&loc->xl_header->xh_count, 1);
1663         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1664 }
1665
1666 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1667 {
1668         int free_start = ocfs2_xa_get_free_start(loc);
1669
1670         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1671 }
1672
1673 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1674                                           struct ocfs2_xattr_value_buf *vb)
1675 {
1676         struct buffer_head *bh = loc->xl_storage;
1677
1678         if (loc->xl_size == (bh->b_size -
1679                              offsetof(struct ocfs2_xattr_block,
1680                                       xb_attrs.xb_header)))
1681                 vb->vb_access = ocfs2_journal_access_xb;
1682         else
1683                 vb->vb_access = ocfs2_journal_access_di;
1684         vb->vb_bh = bh;
1685 }
1686
1687 /*
1688  * Operations for xattrs stored in blocks.  This includes inline inode
1689  * storage and unindexed ocfs2_xattr_blocks.
1690  */
1691 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1692         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1693         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1694         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1695         .xlo_check_space        = ocfs2_xa_block_check_space,
1696         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1697         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1698         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1699         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1700         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1701         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1702 };
1703
1704 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1705                                           struct ocfs2_xa_loc *loc, int type)
1706 {
1707         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1708
1709         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1710 }
1711
1712 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1713                                           struct ocfs2_xa_loc *loc)
1714 {
1715         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1716
1717         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1718 }
1719
1720 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1721                                             int offset)
1722 {
1723         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1724         int block, block_offset;
1725
1726         /* The header is at the front of the bucket */
1727         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1728         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1729
1730         return bucket_block(bucket, block) + block_offset;
1731 }
1732
1733 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1734                                      struct ocfs2_xattr_info *xi)
1735 {
1736         return namevalue_size_xe(loc->xl_entry) >=
1737                 namevalue_size_xi(xi);
1738 }
1739
1740 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1741 {
1742         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1743         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1744 }
1745
1746 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1747                                          int free_start, int size)
1748 {
1749         /*
1750          * We need to make sure that the name+value pair fits within
1751          * one block.
1752          */
1753         if (((free_start - size) >> sb->s_blocksize_bits) !=
1754             ((free_start - 1) >> sb->s_blocksize_bits))
1755                 free_start -= free_start % sb->s_blocksize;
1756
1757         return free_start;
1758 }
1759
1760 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1761                                        struct ocfs2_xattr_info *xi)
1762 {
1763         int rc;
1764         int count = le16_to_cpu(loc->xl_header->xh_count);
1765         int free_start = ocfs2_xa_get_free_start(loc);
1766         int needed_space = ocfs2_xi_entry_usage(xi);
1767         int size = namevalue_size_xi(xi);
1768         struct super_block *sb = loc->xl_inode->i_sb;
1769
1770         /*
1771          * Bucket storage does not reclaim name+value pairs it cannot
1772          * reuse.  They live as holes until the bucket fills, and then
1773          * the bucket is defragmented.  However, the bucket can reclaim
1774          * the ocfs2_xattr_entry.
1775          */
1776         if (loc->xl_entry) {
1777                 /* Don't need space if we're reusing! */
1778                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1779                         needed_space = 0;
1780                 else
1781                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1782         }
1783         BUG_ON(needed_space < 0);
1784
1785         if (free_start < size) {
1786                 if (needed_space)
1787                         return -ENOSPC;
1788         } else {
1789                 /*
1790                  * First we check if it would fit in the first place.
1791                  * Below, we align the free start to a block.  This may
1792                  * slide us below the minimum gap.  By checking unaligned
1793                  * first, we avoid that error.
1794                  */
1795                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1796                                                  count);
1797                 if (rc)
1798                         return rc;
1799                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1800                                                            size);
1801         }
1802         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1803 }
1804
1805 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1806 {
1807         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1808                      -namevalue_size_xe(loc->xl_entry));
1809 }
1810
1811 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1812 {
1813         struct ocfs2_xattr_header *xh = loc->xl_header;
1814         int count = le16_to_cpu(xh->xh_count);
1815         int low = 0, high = count - 1, tmp;
1816         struct ocfs2_xattr_entry *tmp_xe;
1817
1818         /*
1819          * We keep buckets sorted by name_hash, so we need to find
1820          * our insert place.
1821          */
1822         while (low <= high && count) {
1823                 tmp = (low + high) / 2;
1824                 tmp_xe = &xh->xh_entries[tmp];
1825
1826                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1827                         low = tmp + 1;
1828                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1829                         high = tmp - 1;
1830                 else {
1831                         low = tmp;
1832                         break;
1833                 }
1834         }
1835
1836         if (low != count)
1837                 memmove(&xh->xh_entries[low + 1],
1838                         &xh->xh_entries[low],
1839                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1840
1841         le16_add_cpu(&xh->xh_count, 1);
1842         loc->xl_entry = &xh->xh_entries[low];
1843         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1844 }
1845
1846 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1847 {
1848         int free_start = ocfs2_xa_get_free_start(loc);
1849         struct ocfs2_xattr_header *xh = loc->xl_header;
1850         struct super_block *sb = loc->xl_inode->i_sb;
1851         int nameval_offset;
1852
1853         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1854         nameval_offset = free_start - size;
1855         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1856         xh->xh_free_start = cpu_to_le16(nameval_offset);
1857         le16_add_cpu(&xh->xh_name_value_len, size);
1858
1859 }
1860
1861 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1862                                            struct ocfs2_xattr_value_buf *vb)
1863 {
1864         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1865         struct super_block *sb = loc->xl_inode->i_sb;
1866         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1867         int size = namevalue_size_xe(loc->xl_entry);
1868         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1869
1870         /* Values are not allowed to straddle block boundaries */
1871         BUG_ON(block_offset !=
1872                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1873         /* We expect the bucket to be filled in */
1874         BUG_ON(!bucket->bu_bhs[block_offset]);
1875
1876         vb->vb_access = ocfs2_journal_access;
1877         vb->vb_bh = bucket->bu_bhs[block_offset];
1878 }
1879
1880 /* Operations for xattrs stored in buckets. */
1881 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1882         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1883         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1884         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1885         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1886         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1887         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1888         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1889         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1890         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1891         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1892 };
1893
1894 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1895 {
1896         struct ocfs2_xattr_value_buf vb;
1897
1898         if (ocfs2_xattr_is_local(loc->xl_entry))
1899                 return 0;
1900
1901         ocfs2_xa_fill_value_buf(loc, &vb);
1902         return le32_to_cpu(vb.vb_xv->xr_clusters);
1903 }
1904
1905 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1906                                    struct ocfs2_xattr_set_ctxt *ctxt)
1907 {
1908         int trunc_rc, access_rc;
1909         struct ocfs2_xattr_value_buf vb;
1910
1911         ocfs2_xa_fill_value_buf(loc, &vb);
1912         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1913                                               ctxt);
1914
1915         /*
1916          * The caller of ocfs2_xa_value_truncate() has already called
1917          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1918          * calls ocfs2_extend_trans().  This may commit the previous
1919          * transaction and open a new one.  If this is a bucket, truncate
1920          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1921          * the caller is expecting to dirty the entire bucket.  So we must
1922          * reset the journal work.  We do this even if truncate has failed,
1923          * as it could have failed after committing the extend.
1924          */
1925         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1926                                             OCFS2_JOURNAL_ACCESS_WRITE);
1927
1928         /* Errors in truncate take precedence */
1929         return trunc_rc ? trunc_rc : access_rc;
1930 }
1931
1932 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1933 {
1934         int index, count;
1935         struct ocfs2_xattr_header *xh = loc->xl_header;
1936         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1937
1938         ocfs2_xa_wipe_namevalue(loc);
1939         loc->xl_entry = NULL;
1940
1941         le16_add_cpu(&xh->xh_count, -1);
1942         count = le16_to_cpu(xh->xh_count);
1943
1944         /*
1945          * Only zero out the entry if there are more remaining.  This is
1946          * important for an empty bucket, as it keeps track of the
1947          * bucket's hash value.  It doesn't hurt empty block storage.
1948          */
1949         if (count) {
1950                 index = ((char *)entry - (char *)&xh->xh_entries) /
1951                         sizeof(struct ocfs2_xattr_entry);
1952                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1953                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1954                 memset(&xh->xh_entries[count], 0,
1955                        sizeof(struct ocfs2_xattr_entry));
1956         }
1957 }
1958
1959 /*
1960  * If we have a problem adjusting the size of an external value during
1961  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1962  * in an intermediate state.  For example, the value may be partially
1963  * truncated.
1964  *
1965  * If the value tree hasn't changed, the extend/truncate went nowhere.
1966  * We have nothing to do.  The caller can treat it as a straight error.
1967  *
1968  * If the value tree got partially truncated, we now have a corrupted
1969  * extended attribute.  We're going to wipe its entry and leak the
1970  * clusters.  Better to leak some storage than leave a corrupt entry.
1971  *
1972  * If the value tree grew, it obviously didn't grow enough for the
1973  * new entry.  We're not going to try and reclaim those clusters either.
1974  * If there was already an external value there (orig_clusters != 0),
1975  * the new clusters are attached safely and we can just leave the old
1976  * value in place.  If there was no external value there, we remove
1977  * the entry.
1978  *
1979  * This way, the xattr block we store in the journal will be consistent.
1980  * If the size change broke because of the journal, no changes will hit
1981  * disk anyway.
1982  */
1983 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1984                                             const char *what,
1985                                             unsigned int orig_clusters)
1986 {
1987         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1988         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1989                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1990
1991         if (new_clusters < orig_clusters) {
1992                 mlog(ML_ERROR,
1993                      "Partial truncate while %s xattr %.*s.  Leaking "
1994                      "%u clusters and removing the entry\n",
1995                      what, loc->xl_entry->xe_name_len, nameval_buf,
1996                      orig_clusters - new_clusters);
1997                 ocfs2_xa_remove_entry(loc);
1998         } else if (!orig_clusters) {
1999                 mlog(ML_ERROR,
2000                      "Unable to allocate an external value for xattr "
2001                      "%.*s safely.  Leaking %u clusters and removing the "
2002                      "entry\n",
2003                      loc->xl_entry->xe_name_len, nameval_buf,
2004                      new_clusters - orig_clusters);
2005                 ocfs2_xa_remove_entry(loc);
2006         } else if (new_clusters > orig_clusters)
2007                 mlog(ML_ERROR,
2008                      "Unable to grow xattr %.*s safely.  %u new clusters "
2009                      "have been added, but the value will not be "
2010                      "modified\n",
2011                      loc->xl_entry->xe_name_len, nameval_buf,
2012                      new_clusters - orig_clusters);
2013 }
2014
2015 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2016                            struct ocfs2_xattr_set_ctxt *ctxt)
2017 {
2018         int rc = 0;
2019         unsigned int orig_clusters;
2020
2021         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2022                 orig_clusters = ocfs2_xa_value_clusters(loc);
2023                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2024                 if (rc) {
2025                         mlog_errno(rc);
2026                         /*
2027                          * Since this is remove, we can return 0 if
2028                          * ocfs2_xa_cleanup_value_truncate() is going to
2029                          * wipe the entry anyway.  So we check the
2030                          * cluster count as well.
2031                          */
2032                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2033                                 rc = 0;
2034                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2035                                                         orig_clusters);
2036                         if (rc)
2037                                 goto out;
2038                 }
2039         }
2040
2041         ocfs2_xa_remove_entry(loc);
2042
2043 out:
2044         return rc;
2045 }
2046
2047 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2048 {
2049         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2050         char *nameval_buf;
2051
2052         nameval_buf = ocfs2_xa_offset_pointer(loc,
2053                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2054         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2055 }
2056
2057 /*
2058  * Take an existing entry and make it ready for the new value.  This
2059  * won't allocate space, but it may free space.  It should be ready for
2060  * ocfs2_xa_prepare_entry() to finish the work.
2061  */
2062 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2063                                 struct ocfs2_xattr_info *xi,
2064                                 struct ocfs2_xattr_set_ctxt *ctxt)
2065 {
2066         int rc = 0;
2067         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2068         unsigned int orig_clusters;
2069         char *nameval_buf;
2070         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2071         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2072
2073         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2074                name_size);
2075
2076         nameval_buf = ocfs2_xa_offset_pointer(loc,
2077                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2078         if (xe_local) {
2079                 memset(nameval_buf + name_size, 0,
2080                        namevalue_size_xe(loc->xl_entry) - name_size);
2081                 if (!xi_local)
2082                         ocfs2_xa_install_value_root(loc);
2083         } else {
2084                 orig_clusters = ocfs2_xa_value_clusters(loc);
2085                 if (xi_local) {
2086                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2087                         if (rc < 0)
2088                                 mlog_errno(rc);
2089                         else
2090                                 memset(nameval_buf + name_size, 0,
2091                                        namevalue_size_xe(loc->xl_entry) -
2092                                        name_size);
2093                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2094                            xi->xi_value_len) {
2095                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2096                                                      ctxt);
2097                         if (rc < 0)
2098                                 mlog_errno(rc);
2099                 }
2100
2101                 if (rc) {
2102                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2103                                                         orig_clusters);
2104                         goto out;
2105                 }
2106         }
2107
2108         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2109         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2110
2111 out:
2112         return rc;
2113 }
2114
2115 /*
2116  * Prepares loc->xl_entry to receive the new xattr.  This includes
2117  * properly setting up the name+value pair region.  If loc->xl_entry
2118  * already exists, it will take care of modifying it appropriately.
2119  *
2120  * Note that this modifies the data.  You did journal_access already,
2121  * right?
2122  */
2123 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2124                                   struct ocfs2_xattr_info *xi,
2125                                   u32 name_hash,
2126                                   struct ocfs2_xattr_set_ctxt *ctxt)
2127 {
2128         int rc = 0;
2129         unsigned int orig_clusters;
2130         __le64 orig_value_size = 0;
2131
2132         rc = ocfs2_xa_check_space(loc, xi);
2133         if (rc)
2134                 goto out;
2135
2136         if (loc->xl_entry) {
2137                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2138                         orig_value_size = loc->xl_entry->xe_value_size;
2139                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2140                         if (rc)
2141                                 goto out;
2142                         goto alloc_value;
2143                 }
2144
2145                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2146                         orig_clusters = ocfs2_xa_value_clusters(loc);
2147                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2148                         if (rc) {
2149                                 mlog_errno(rc);
2150                                 ocfs2_xa_cleanup_value_truncate(loc,
2151                                                                 "overwriting",
2152                                                                 orig_clusters);
2153                                 goto out;
2154                         }
2155                 }
2156                 ocfs2_xa_wipe_namevalue(loc);
2157         } else
2158                 ocfs2_xa_add_entry(loc, name_hash);
2159
2160         /*
2161          * If we get here, we have a blank entry.  Fill it.  We grow our
2162          * name+value pair back from the end.
2163          */
2164         ocfs2_xa_add_namevalue(loc, xi);
2165         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2166                 ocfs2_xa_install_value_root(loc);
2167
2168 alloc_value:
2169         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2170                 orig_clusters = ocfs2_xa_value_clusters(loc);
2171                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2172                 if (rc < 0) {
2173                         ctxt->set_abort = 1;
2174                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2175                                                         orig_clusters);
2176                         /*
2177                          * If we were growing an existing value,
2178                          * ocfs2_xa_cleanup_value_truncate() won't remove
2179                          * the entry. We need to restore the original value
2180                          * size.
2181                          */
2182                         if (loc->xl_entry) {
2183                                 BUG_ON(!orig_value_size);
2184                                 loc->xl_entry->xe_value_size = orig_value_size;
2185                         }
2186                         mlog_errno(rc);
2187                 }
2188         }
2189
2190 out:
2191         return rc;
2192 }
2193
2194 /*
2195  * Store the value portion of the name+value pair.  This will skip
2196  * values that are stored externally.  Their tree roots were set up
2197  * by ocfs2_xa_prepare_entry().
2198  */
2199 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2200                                 struct ocfs2_xattr_info *xi,
2201                                 struct ocfs2_xattr_set_ctxt *ctxt)
2202 {
2203         int rc = 0;
2204         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2205         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2206         char *nameval_buf;
2207         struct ocfs2_xattr_value_buf vb;
2208
2209         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2210         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2211                 ocfs2_xa_fill_value_buf(loc, &vb);
2212                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2213                                                      ctxt->handle, &vb,
2214                                                      xi->xi_value,
2215                                                      xi->xi_value_len);
2216         } else
2217                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2218
2219         return rc;
2220 }
2221
2222 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2223                         struct ocfs2_xattr_info *xi,
2224                         struct ocfs2_xattr_set_ctxt *ctxt)
2225 {
2226         int ret;
2227         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2228                                               xi->xi_name_len);
2229
2230         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2231                                       OCFS2_JOURNAL_ACCESS_WRITE);
2232         if (ret) {
2233                 mlog_errno(ret);
2234                 goto out;
2235         }
2236
2237         /*
2238          * From here on out, everything is going to modify the buffer a
2239          * little.  Errors are going to leave the xattr header in a
2240          * sane state.  Thus, even with errors we dirty the sucker.
2241          */
2242
2243         /* Don't worry, we are never called with !xi_value and !xl_entry */
2244         if (!xi->xi_value) {
2245                 ret = ocfs2_xa_remove(loc, ctxt);
2246                 goto out_dirty;
2247         }
2248
2249         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2250         if (ret) {
2251                 if (ret != -ENOSPC)
2252                         mlog_errno(ret);
2253                 goto out_dirty;
2254         }
2255
2256         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2257         if (ret)
2258                 mlog_errno(ret);
2259
2260 out_dirty:
2261         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2262
2263 out:
2264         return ret;
2265 }
2266
2267 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2268                                      struct inode *inode,
2269                                      struct buffer_head *bh,
2270                                      struct ocfs2_xattr_entry *entry)
2271 {
2272         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2273
2274         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2275
2276         loc->xl_inode = inode;
2277         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2278         loc->xl_storage = bh;
2279         loc->xl_entry = entry;
2280         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2281         loc->xl_header =
2282                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2283                                               loc->xl_size);
2284 }
2285
2286 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2287                                           struct inode *inode,
2288                                           struct buffer_head *bh,
2289                                           struct ocfs2_xattr_entry *entry)
2290 {
2291         struct ocfs2_xattr_block *xb =
2292                 (struct ocfs2_xattr_block *)bh->b_data;
2293
2294         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2295
2296         loc->xl_inode = inode;
2297         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2298         loc->xl_storage = bh;
2299         loc->xl_header = &(xb->xb_attrs.xb_header);
2300         loc->xl_entry = entry;
2301         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2302                                              xb_attrs.xb_header);
2303 }
2304
2305 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2306                                            struct ocfs2_xattr_bucket *bucket,
2307                                            struct ocfs2_xattr_entry *entry)
2308 {
2309         loc->xl_inode = bucket->bu_inode;
2310         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2311         loc->xl_storage = bucket;
2312         loc->xl_header = bucket_xh(bucket);
2313         loc->xl_entry = entry;
2314         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2315 }
2316
2317 /*
2318  * In xattr remove, if it is stored outside and refcounted, we may have
2319  * the chance to split the refcount tree. So need the allocators.
2320  */
2321 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2322                                         struct ocfs2_xattr_value_root *xv,
2323                                         struct ocfs2_caching_info *ref_ci,
2324                                         struct buffer_head *ref_root_bh,
2325                                         struct ocfs2_alloc_context **meta_ac,
2326                                         int *ref_credits)
2327 {
2328         int ret, meta_add = 0;
2329         u32 p_cluster, num_clusters;
2330         unsigned int ext_flags;
2331
2332         *ref_credits = 0;
2333         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2334                                        &num_clusters,
2335                                        &xv->xr_list,
2336                                        &ext_flags);
2337         if (ret) {
2338                 mlog_errno(ret);
2339                 goto out;
2340         }
2341
2342         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2343                 goto out;
2344
2345         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2346                                                  ref_root_bh, xv,
2347                                                  &meta_add, ref_credits);
2348         if (ret) {
2349                 mlog_errno(ret);
2350                 goto out;
2351         }
2352
2353         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2354                                                 meta_add, meta_ac);
2355         if (ret)
2356                 mlog_errno(ret);
2357
2358 out:
2359         return ret;
2360 }
2361
2362 static int ocfs2_remove_value_outside(struct inode*inode,
2363                                       struct ocfs2_xattr_value_buf *vb,
2364                                       struct ocfs2_xattr_header *header,
2365                                       struct ocfs2_caching_info *ref_ci,
2366                                       struct buffer_head *ref_root_bh)
2367 {
2368         int ret = 0, i, ref_credits;
2369         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2370         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2371         void *val;
2372
2373         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2374
2375         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2376                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2377
2378                 if (ocfs2_xattr_is_local(entry))
2379                         continue;
2380
2381                 val = (void *)header +
2382                         le16_to_cpu(entry->xe_name_offset);
2383                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2384                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2385
2386                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2387                                                          ref_ci, ref_root_bh,
2388                                                          &ctxt.meta_ac,
2389                                                          &ref_credits);
2390
2391                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2392                                         ocfs2_remove_extent_credits(osb->sb));
2393                 if (IS_ERR(ctxt.handle)) {
2394                         ret = PTR_ERR(ctxt.handle);
2395                         mlog_errno(ret);
2396                         break;
2397                 }
2398
2399                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2400
2401                 ocfs2_commit_trans(osb, ctxt.handle);
2402                 if (ctxt.meta_ac) {
2403                         ocfs2_free_alloc_context(ctxt.meta_ac);
2404                         ctxt.meta_ac = NULL;
2405                 }
2406
2407                 if (ret < 0) {
2408                         mlog_errno(ret);
2409                         break;
2410                 }
2411
2412         }
2413
2414         if (ctxt.meta_ac)
2415                 ocfs2_free_alloc_context(ctxt.meta_ac);
2416         ocfs2_schedule_truncate_log_flush(osb, 1);
2417         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2418         return ret;
2419 }
2420
2421 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2422                                     struct buffer_head *di_bh,
2423                                     struct ocfs2_caching_info *ref_ci,
2424                                     struct buffer_head *ref_root_bh)
2425 {
2426
2427         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2428         struct ocfs2_xattr_header *header;
2429         int ret;
2430         struct ocfs2_xattr_value_buf vb = {
2431                 .vb_bh = di_bh,
2432                 .vb_access = ocfs2_journal_access_di,
2433         };
2434
2435         header = (struct ocfs2_xattr_header *)
2436                  ((void *)di + inode->i_sb->s_blocksize -
2437                  le16_to_cpu(di->i_xattr_inline_size));
2438
2439         ret = ocfs2_remove_value_outside(inode, &vb, header,
2440                                          ref_ci, ref_root_bh);
2441
2442         return ret;
2443 }
2444
2445 struct ocfs2_rm_xattr_bucket_para {
2446         struct ocfs2_caching_info *ref_ci;
2447         struct buffer_head *ref_root_bh;
2448 };
2449
2450 static int ocfs2_xattr_block_remove(struct inode *inode,
2451                                     struct buffer_head *blk_bh,
2452                                     struct ocfs2_caching_info *ref_ci,
2453                                     struct buffer_head *ref_root_bh)
2454 {
2455         struct ocfs2_xattr_block *xb;
2456         int ret = 0;
2457         struct ocfs2_xattr_value_buf vb = {
2458                 .vb_bh = blk_bh,
2459                 .vb_access = ocfs2_journal_access_xb,
2460         };
2461         struct ocfs2_rm_xattr_bucket_para args = {
2462                 .ref_ci = ref_ci,
2463                 .ref_root_bh = ref_root_bh,
2464         };
2465
2466         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2467         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2468                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2469                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2470                                                  ref_ci, ref_root_bh);
2471         } else
2472                 ret = ocfs2_iterate_xattr_index_block(inode,
2473                                                 blk_bh,
2474                                                 ocfs2_rm_xattr_cluster,
2475                                                 &args);
2476
2477         return ret;
2478 }
2479
2480 static int ocfs2_xattr_free_block(struct inode *inode,
2481                                   u64 block,
2482                                   struct ocfs2_caching_info *ref_ci,
2483                                   struct buffer_head *ref_root_bh)
2484 {
2485         struct inode *xb_alloc_inode;
2486         struct buffer_head *xb_alloc_bh = NULL;
2487         struct buffer_head *blk_bh = NULL;
2488         struct ocfs2_xattr_block *xb;
2489         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2490         handle_t *handle;
2491         int ret = 0;
2492         u64 blk, bg_blkno;
2493         u16 bit;
2494
2495         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2496         if (ret < 0) {
2497                 mlog_errno(ret);
2498                 goto out;
2499         }
2500
2501         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2502         if (ret < 0) {
2503                 mlog_errno(ret);
2504                 goto out;
2505         }
2506
2507         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2508         blk = le64_to_cpu(xb->xb_blkno);
2509         bit = le16_to_cpu(xb->xb_suballoc_bit);
2510         if (xb->xb_suballoc_loc)
2511                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2512         else
2513                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2514
2515         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2516                                 EXTENT_ALLOC_SYSTEM_INODE,
2517                                 le16_to_cpu(xb->xb_suballoc_slot));
2518         if (!xb_alloc_inode) {
2519                 ret = -ENOMEM;
2520                 mlog_errno(ret);
2521                 goto out;
2522         }
2523         inode_lock(xb_alloc_inode);
2524
2525         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2526         if (ret < 0) {
2527                 mlog_errno(ret);
2528                 goto out_mutex;
2529         }
2530
2531         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2532         if (IS_ERR(handle)) {
2533                 ret = PTR_ERR(handle);
2534                 mlog_errno(ret);
2535                 goto out_unlock;
2536         }
2537
2538         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2539                                        bit, bg_blkno, 1);
2540         if (ret < 0)
2541                 mlog_errno(ret);
2542
2543         ocfs2_commit_trans(osb, handle);
2544 out_unlock:
2545         ocfs2_inode_unlock(xb_alloc_inode, 1);
2546         brelse(xb_alloc_bh);
2547 out_mutex:
2548         inode_unlock(xb_alloc_inode);
2549         iput(xb_alloc_inode);
2550 out:
2551         brelse(blk_bh);
2552         return ret;
2553 }
2554
2555 /*
2556  * ocfs2_xattr_remove()
2557  *
2558  * Free extended attribute resources associated with this inode.
2559  */
2560 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2561 {
2562         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2563         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2564         struct ocfs2_refcount_tree *ref_tree = NULL;
2565         struct buffer_head *ref_root_bh = NULL;
2566         struct ocfs2_caching_info *ref_ci = NULL;
2567         handle_t *handle;
2568         int ret;
2569
2570         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2571                 return 0;
2572
2573         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2574                 return 0;
2575
2576         if (ocfs2_is_refcount_inode(inode)) {
2577                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2578                                                le64_to_cpu(di->i_refcount_loc),
2579                                                1, &ref_tree, &ref_root_bh);
2580                 if (ret) {
2581                         mlog_errno(ret);
2582                         goto out;
2583                 }
2584                 ref_ci = &ref_tree->rf_ci;
2585
2586         }
2587
2588         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2589                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2590                                                ref_ci, ref_root_bh);
2591                 if (ret < 0) {
2592                         mlog_errno(ret);
2593                         goto out;
2594                 }
2595         }
2596
2597         if (di->i_xattr_loc) {
2598                 ret = ocfs2_xattr_free_block(inode,
2599                                              le64_to_cpu(di->i_xattr_loc),
2600                                              ref_ci, ref_root_bh);
2601                 if (ret < 0) {
2602                         mlog_errno(ret);
2603                         goto out;
2604                 }
2605         }
2606
2607         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2608                                    OCFS2_INODE_UPDATE_CREDITS);
2609         if (IS_ERR(handle)) {
2610                 ret = PTR_ERR(handle);
2611                 mlog_errno(ret);
2612                 goto out;
2613         }
2614         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2615                                       OCFS2_JOURNAL_ACCESS_WRITE);
2616         if (ret) {
2617                 mlog_errno(ret);
2618                 goto out_commit;
2619         }
2620
2621         di->i_xattr_loc = 0;
2622
2623         spin_lock(&oi->ip_lock);
2624         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2625         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2626         spin_unlock(&oi->ip_lock);
2627         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2628
2629         ocfs2_journal_dirty(handle, di_bh);
2630 out_commit:
2631         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2632 out:
2633         if (ref_tree)
2634                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2635         brelse(ref_root_bh);
2636         return ret;
2637 }
2638
2639 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2640                                         struct ocfs2_dinode *di)
2641 {
2642         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2643         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2644         int free;
2645
2646         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2647                 return 0;
2648
2649         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2650                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2651                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2652         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2653                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2654                         le64_to_cpu(di->i_size);
2655         } else {
2656                 struct ocfs2_extent_list *el = &di->id2.i_list;
2657                 free = (le16_to_cpu(el->l_count) -
2658                         le16_to_cpu(el->l_next_free_rec)) *
2659                         sizeof(struct ocfs2_extent_rec);
2660         }
2661         if (free >= xattrsize)
2662                 return 1;
2663
2664         return 0;
2665 }
2666
2667 /*
2668  * ocfs2_xattr_ibody_find()
2669  *
2670  * Find extended attribute in inode block and
2671  * fill search info into struct ocfs2_xattr_search.
2672  */
2673 static int ocfs2_xattr_ibody_find(struct inode *inode,
2674                                   int name_index,
2675                                   const char *name,
2676                                   struct ocfs2_xattr_search *xs)
2677 {
2678         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2679         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2680         int ret;
2681         int has_space = 0;
2682
2683         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2684                 return 0;
2685
2686         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2687                 down_read(&oi->ip_alloc_sem);
2688                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2689                 up_read(&oi->ip_alloc_sem);
2690                 if (!has_space)
2691                         return 0;
2692         }
2693
2694         xs->xattr_bh = xs->inode_bh;
2695         xs->end = (void *)di + inode->i_sb->s_blocksize;
2696         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2697                 xs->header = (struct ocfs2_xattr_header *)
2698                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2699         else
2700                 xs->header = (struct ocfs2_xattr_header *)
2701                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2702         xs->base = (void *)xs->header;
2703         xs->here = xs->header->xh_entries;
2704
2705         /* Find the named attribute. */
2706         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2707                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2708                 if (ret && ret != -ENODATA)
2709                         return ret;
2710                 xs->not_found = ret;
2711         }
2712
2713         return 0;
2714 }
2715
2716 static int ocfs2_xattr_ibody_init(struct inode *inode,
2717                                   struct buffer_head *di_bh,
2718                                   struct ocfs2_xattr_set_ctxt *ctxt)
2719 {
2720         int ret;
2721         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2722         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2723         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2724         unsigned int xattrsize = osb->s_xattr_inline_size;
2725
2726         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2727                 ret = -ENOSPC;
2728                 goto out;
2729         }
2730
2731         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2732                                       OCFS2_JOURNAL_ACCESS_WRITE);
2733         if (ret) {
2734                 mlog_errno(ret);
2735                 goto out;
2736         }
2737
2738         /*
2739          * Adjust extent record count or inline data size
2740          * to reserve space for extended attribute.
2741          */
2742         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2743                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2744                 le16_add_cpu(&idata->id_count, -xattrsize);
2745         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2746                 struct ocfs2_extent_list *el = &di->id2.i_list;
2747                 le16_add_cpu(&el->l_count, -(xattrsize /
2748                                              sizeof(struct ocfs2_extent_rec)));
2749         }
2750         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2751
2752         spin_lock(&oi->ip_lock);
2753         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2754         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2755         spin_unlock(&oi->ip_lock);
2756
2757         ocfs2_journal_dirty(ctxt->handle, di_bh);
2758
2759 out:
2760         return ret;
2761 }
2762
2763 /*
2764  * ocfs2_xattr_ibody_set()
2765  *
2766  * Set, replace or remove an extended attribute into inode block.
2767  *
2768  */
2769 static int ocfs2_xattr_ibody_set(struct inode *inode,
2770                                  struct ocfs2_xattr_info *xi,
2771                                  struct ocfs2_xattr_search *xs,
2772                                  struct ocfs2_xattr_set_ctxt *ctxt)
2773 {
2774         int ret;
2775         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2776         struct ocfs2_xa_loc loc;
2777
2778         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2779                 return -ENOSPC;
2780
2781         down_write(&oi->ip_alloc_sem);
2782         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2783                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2784                 if (ret) {
2785                         if (ret != -ENOSPC)
2786                                 mlog_errno(ret);
2787                         goto out;
2788                 }
2789         }
2790
2791         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2792                                  xs->not_found ? NULL : xs->here);
2793         ret = ocfs2_xa_set(&loc, xi, ctxt);
2794         if (ret) {
2795                 if (ret != -ENOSPC)
2796                         mlog_errno(ret);
2797                 goto out;
2798         }
2799         xs->here = loc.xl_entry;
2800
2801 out:
2802         up_write(&oi->ip_alloc_sem);
2803
2804         return ret;
2805 }
2806
2807 /*
2808  * ocfs2_xattr_block_find()
2809  *
2810  * Find extended attribute in external block and
2811  * fill search info into struct ocfs2_xattr_search.
2812  */
2813 static int ocfs2_xattr_block_find(struct inode *inode,
2814                                   int name_index,
2815                                   const char *name,
2816                                   struct ocfs2_xattr_search *xs)
2817 {
2818         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2819         struct buffer_head *blk_bh = NULL;
2820         struct ocfs2_xattr_block *xb;
2821         int ret = 0;
2822
2823         if (!di->i_xattr_loc)
2824                 return ret;
2825
2826         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2827                                      &blk_bh);
2828         if (ret < 0) {
2829                 mlog_errno(ret);
2830                 return ret;
2831         }
2832
2833         xs->xattr_bh = blk_bh;
2834         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2835
2836         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2837                 xs->header = &xb->xb_attrs.xb_header;
2838                 xs->base = (void *)xs->header;
2839                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2840                 xs->here = xs->header->xh_entries;
2841
2842                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2843         } else
2844                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2845                                                    name_index,
2846                                                    name, xs);
2847
2848         if (ret && ret != -ENODATA) {
2849                 xs->xattr_bh = NULL;
2850                 goto cleanup;
2851         }
2852         xs->not_found = ret;
2853         return 0;
2854 cleanup:
2855         brelse(blk_bh);
2856
2857         return ret;
2858 }
2859
2860 static int ocfs2_create_xattr_block(struct inode *inode,
2861                                     struct buffer_head *inode_bh,
2862                                     struct ocfs2_xattr_set_ctxt *ctxt,
2863                                     int indexed,
2864                                     struct buffer_head **ret_bh)
2865 {
2866         int ret;
2867         u16 suballoc_bit_start;
2868         u32 num_got;
2869         u64 suballoc_loc, first_blkno;
2870         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2871         struct buffer_head *new_bh = NULL;
2872         struct ocfs2_xattr_block *xblk;
2873
2874         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2875                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2876         if (ret < 0) {
2877                 mlog_errno(ret);
2878                 goto end;
2879         }
2880
2881         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2882                                    &suballoc_loc, &suballoc_bit_start,
2883                                    &num_got, &first_blkno);
2884         if (ret < 0) {
2885                 mlog_errno(ret);
2886                 goto end;
2887         }
2888
2889         new_bh = sb_getblk(inode->i_sb, first_blkno);
2890         if (!new_bh) {
2891                 ret = -ENOMEM;
2892                 mlog_errno(ret);
2893                 goto end;
2894         }
2895
2896         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2897
2898         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2899                                       new_bh,
2900                                       OCFS2_JOURNAL_ACCESS_CREATE);
2901         if (ret < 0) {
2902                 mlog_errno(ret);
2903                 goto end;
2904         }
2905
2906         /* Initialize ocfs2_xattr_block */
2907         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2908         memset(xblk, 0, inode->i_sb->s_blocksize);
2909         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2910         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2911         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2912         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2913         xblk->xb_fs_generation =
2914                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2915         xblk->xb_blkno = cpu_to_le64(first_blkno);
2916         if (indexed) {
2917                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2918                 xr->xt_clusters = cpu_to_le32(1);
2919                 xr->xt_last_eb_blk = 0;
2920                 xr->xt_list.l_tree_depth = 0;
2921                 xr->xt_list.l_count = cpu_to_le16(
2922                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2923                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2924                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2925         }
2926         ocfs2_journal_dirty(ctxt->handle, new_bh);
2927
2928         /* Add it to the inode */
2929         di->i_xattr_loc = cpu_to_le64(first_blkno);
2930
2931         spin_lock(&OCFS2_I(inode)->ip_lock);
2932         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2933         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2934         spin_unlock(&OCFS2_I(inode)->ip_lock);
2935
2936         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2937
2938         *ret_bh = new_bh;
2939         new_bh = NULL;
2940
2941 end:
2942         brelse(new_bh);
2943         return ret;
2944 }
2945
2946 /*
2947  * ocfs2_xattr_block_set()
2948  *
2949  * Set, replace or remove an extended attribute into external block.
2950  *
2951  */
2952 static int ocfs2_xattr_block_set(struct inode *inode,
2953                                  struct ocfs2_xattr_info *xi,
2954                                  struct ocfs2_xattr_search *xs,
2955                                  struct ocfs2_xattr_set_ctxt *ctxt)
2956 {
2957         struct buffer_head *new_bh = NULL;
2958         struct ocfs2_xattr_block *xblk = NULL;
2959         int ret;
2960         struct ocfs2_xa_loc loc;
2961
2962         if (!xs->xattr_bh) {
2963                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2964                                                0, &new_bh);
2965                 if (ret) {
2966                         mlog_errno(ret);
2967                         goto end;
2968                 }
2969
2970                 xs->xattr_bh = new_bh;
2971                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2972                 xs->header = &xblk->xb_attrs.xb_header;
2973                 xs->base = (void *)xs->header;
2974                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2975                 xs->here = xs->header->xh_entries;
2976         } else
2977                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2978
2979         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2980                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2981                                               xs->not_found ? NULL : xs->here);
2982
2983                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2984                 if (!ret)
2985                         xs->here = loc.xl_entry;
2986                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2987                         goto end;
2988                 else {
2989                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2990                         if (ret)
2991                                 goto end;
2992                 }
2993         }
2994
2995         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2996                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2997
2998 end:
2999         return ret;
3000 }
3001
3002 /* Check whether the new xattr can be inserted into the inode. */
3003 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3004                                        struct ocfs2_xattr_info *xi,
3005                                        struct ocfs2_xattr_search *xs)
3006 {
3007         struct ocfs2_xattr_entry *last;
3008         int free, i;
3009         size_t min_offs = xs->end - xs->base;
3010
3011         if (!xs->header)
3012                 return 0;
3013
3014         last = xs->header->xh_entries;
3015
3016         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3017                 size_t offs = le16_to_cpu(last->xe_name_offset);
3018                 if (offs < min_offs)
3019                         min_offs = offs;
3020                 last += 1;
3021         }
3022
3023         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3024         if (free < 0)
3025                 return 0;
3026
3027         BUG_ON(!xs->not_found);
3028
3029         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3030                 return 1;
3031
3032         return 0;
3033 }
3034
3035 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3036                                      struct ocfs2_dinode *di,
3037                                      struct ocfs2_xattr_info *xi,
3038                                      struct ocfs2_xattr_search *xis,
3039                                      struct ocfs2_xattr_search *xbs,
3040                                      int *clusters_need,
3041                                      int *meta_need,
3042                                      int *credits_need)
3043 {
3044         int ret = 0, old_in_xb = 0;
3045         int clusters_add = 0, meta_add = 0, credits = 0;
3046         struct buffer_head *bh = NULL;
3047         struct ocfs2_xattr_block *xb = NULL;
3048         struct ocfs2_xattr_entry *xe = NULL;
3049         struct ocfs2_xattr_value_root *xv = NULL;
3050         char *base = NULL;
3051         int name_offset, name_len = 0;
3052         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3053                                                     xi->xi_value_len);
3054         u64 value_size;
3055
3056         /*
3057          * Calculate the clusters we need to write.
3058          * No matter whether we replace an old one or add a new one,
3059          * we need this for writing.
3060          */
3061         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3062                 credits += new_clusters *
3063                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3064
3065         if (xis->not_found && xbs->not_found) {
3066                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3067
3068                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3069                         clusters_add += new_clusters;
3070                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3071                                                         &def_xv.xv.xr_list);
3072                 }
3073
3074                 goto meta_guess;
3075         }
3076
3077         if (!xis->not_found) {
3078                 xe = xis->here;
3079                 name_offset = le16_to_cpu(xe->xe_name_offset);
3080                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3081                 base = xis->base;
3082                 credits += OCFS2_INODE_UPDATE_CREDITS;
3083         } else {
3084                 int i, block_off = 0;
3085                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3086                 xe = xbs->here;
3087                 name_offset = le16_to_cpu(xe->xe_name_offset);
3088                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3089                 i = xbs->here - xbs->header->xh_entries;
3090                 old_in_xb = 1;
3091
3092                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3093                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3094                                                         bucket_xh(xbs->bucket),
3095                                                         i, &block_off,
3096                                                         &name_offset);
3097                         base = bucket_block(xbs->bucket, block_off);
3098                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3099                 } else {
3100                         base = xbs->base;
3101                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3102                 }
3103         }
3104
3105         /*
3106          * delete a xattr doesn't need metadata and cluster allocation.
3107          * so just calculate the credits and return.
3108          *
3109          * The credits for removing the value tree will be extended
3110          * by ocfs2_remove_extent itself.
3111          */
3112         if (!xi->xi_value) {
3113                 if (!ocfs2_xattr_is_local(xe))
3114                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3115
3116                 goto out;
3117         }
3118
3119         /* do cluster allocation guess first. */
3120         value_size = le64_to_cpu(xe->xe_value_size);
3121
3122         if (old_in_xb) {
3123                 /*
3124                  * In xattr set, we always try to set the xe in inode first,
3125                  * so if it can be inserted into inode successfully, the old
3126                  * one will be removed from the xattr block, and this xattr
3127                  * will be inserted into inode as a new xattr in inode.
3128                  */
3129                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3130                         clusters_add += new_clusters;
3131                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3132                                     OCFS2_INODE_UPDATE_CREDITS;
3133                         if (!ocfs2_xattr_is_local(xe))
3134                                 credits += ocfs2_calc_extend_credits(
3135                                                         inode->i_sb,
3136                                                         &def_xv.xv.xr_list);
3137                         goto out;
3138                 }
3139         }
3140
3141         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3142                 /* the new values will be stored outside. */
3143                 u32 old_clusters = 0;
3144
3145                 if (!ocfs2_xattr_is_local(xe)) {
3146                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3147                                                                  value_size);
3148                         xv = (struct ocfs2_xattr_value_root *)
3149                              (base + name_offset + name_len);
3150                         value_size = OCFS2_XATTR_ROOT_SIZE;
3151                 } else
3152                         xv = &def_xv.xv;
3153
3154                 if (old_clusters >= new_clusters) {
3155                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3156                         goto out;
3157                 } else {
3158                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3159                         clusters_add += new_clusters - old_clusters;
3160                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3161                                                              &xv->xr_list);
3162                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3163                                 goto out;
3164                 }
3165         } else {
3166                 /*
3167                  * Now the new value will be stored inside. So if the new
3168                  * value is smaller than the size of value root or the old
3169                  * value, we don't need any allocation, otherwise we have
3170                  * to guess metadata allocation.
3171                  */
3172                 if ((ocfs2_xattr_is_local(xe) &&
3173                      (value_size >= xi->xi_value_len)) ||
3174                     (!ocfs2_xattr_is_local(xe) &&
3175                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3176                         goto out;
3177         }
3178
3179 meta_guess:
3180         /* calculate metadata allocation. */
3181         if (di->i_xattr_loc) {
3182                 if (!xbs->xattr_bh) {
3183                         ret = ocfs2_read_xattr_block(inode,
3184                                                      le64_to_cpu(di->i_xattr_loc),
3185                                                      &bh);
3186                         if (ret) {
3187                                 mlog_errno(ret);
3188                                 goto out;
3189                         }
3190
3191                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3192                 } else
3193                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3194
3195                 /*
3196                  * If there is already an xattr tree, good, we can calculate
3197                  * like other b-trees. Otherwise we may have the chance of
3198                  * create a tree, the credit calculation is borrowed from
3199                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3200                  * new tree will be cluster based, so no meta is needed.
3201                  */
3202                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3203                         struct ocfs2_extent_list *el =
3204                                  &xb->xb_attrs.xb_root.xt_list;
3205                         meta_add += ocfs2_extend_meta_needed(el);
3206                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3207                                                              el);
3208                 } else
3209                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3210
3211                 /*
3212                  * This cluster will be used either for new bucket or for
3213                  * new xattr block.
3214                  * If the cluster size is the same as the bucket size, one
3215                  * more is needed since we may need to extend the bucket
3216                  * also.
3217                  */
3218                 clusters_add += 1;
3219                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3220                 if (OCFS2_XATTR_BUCKET_SIZE ==
3221                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3222                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3223                         clusters_add += 1;
3224                 }
3225         } else {
3226                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3227                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3228                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3229                         meta_add += ocfs2_extend_meta_needed(el);
3230                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3231                                                              el);
3232                 } else {
3233                         meta_add += 1;
3234                 }
3235         }
3236 out:
3237         if (clusters_need)
3238                 *clusters_need = clusters_add;
3239         if (meta_need)
3240                 *meta_need = meta_add;
3241         if (credits_need)
3242                 *credits_need = credits;
3243         brelse(bh);
3244         return ret;
3245 }
3246
3247 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3248                                      struct ocfs2_dinode *di,
3249                                      struct ocfs2_xattr_info *xi,
3250                                      struct ocfs2_xattr_search *xis,
3251                                      struct ocfs2_xattr_search *xbs,
3252                                      struct ocfs2_xattr_set_ctxt *ctxt,
3253                                      int extra_meta,
3254                                      int *credits)
3255 {
3256         int clusters_add, meta_add, ret;
3257         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3258
3259         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3260
3261         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3262
3263         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3264                                         &clusters_add, &meta_add, credits);
3265         if (ret) {
3266                 mlog_errno(ret);
3267                 return ret;
3268         }
3269
3270         meta_add += extra_meta;
3271         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3272                                         clusters_add, *credits);
3273
3274         if (meta_add) {
3275                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3276                                                         &ctxt->meta_ac);
3277                 if (ret) {
3278                         mlog_errno(ret);
3279                         goto out;
3280                 }
3281         }
3282
3283         if (clusters_add) {
3284                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3285                 if (ret)
3286                         mlog_errno(ret);
3287         }
3288 out:
3289         if (ret) {
3290                 if (ctxt->meta_ac) {
3291                         ocfs2_free_alloc_context(ctxt->meta_ac);
3292                         ctxt->meta_ac = NULL;
3293                 }
3294
3295                 /*
3296                  * We cannot have an error and a non null ctxt->data_ac.
3297                  */
3298         }
3299
3300         return ret;
3301 }
3302
3303 static int __ocfs2_xattr_set_handle(struct inode *inode,
3304                                     struct ocfs2_dinode *di,
3305                                     struct ocfs2_xattr_info *xi,
3306                                     struct ocfs2_xattr_search *xis,
3307                                     struct ocfs2_xattr_search *xbs,
3308                                     struct ocfs2_xattr_set_ctxt *ctxt)
3309 {
3310         int ret = 0, credits, old_found;
3311
3312         if (!xi->xi_value) {
3313                 /* Remove existing extended attribute */
3314                 if (!xis->not_found)
3315                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3316                 else if (!xbs->not_found)
3317                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3318         } else {
3319                 /* We always try to set extended attribute into inode first*/
3320                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3321                 if (!ret && !xbs->not_found) {
3322                         /*
3323                          * If succeed and that extended attribute existing in
3324                          * external block, then we will remove it.
3325                          */
3326                         xi->xi_value = NULL;
3327                         xi->xi_value_len = 0;
3328
3329                         old_found = xis->not_found;
3330                         xis->not_found = -ENODATA;
3331                         ret = ocfs2_calc_xattr_set_need(inode,
3332                                                         di,
3333                                                         xi,
3334                                                         xis,
3335                                                         xbs,
3336                                                         NULL,
3337                                                         NULL,
3338                                                         &credits);
3339                         xis->not_found = old_found;
3340                         if (ret) {
3341                                 mlog_errno(ret);
3342                                 goto out;
3343                         }
3344
3345                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3346                         if (ret) {
3347                                 mlog_errno(ret);
3348                                 goto out;
3349                         }
3350                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3351                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3352                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3353                                 ret = ocfs2_xattr_block_find(inode,
3354                                                              xi->xi_name_index,
3355                                                              xi->xi_name, xbs);
3356                                 if (ret)
3357                                         goto out;
3358
3359                                 old_found = xis->not_found;
3360                                 xis->not_found = -ENODATA;
3361                                 ret = ocfs2_calc_xattr_set_need(inode,
3362                                                                 di,
3363                                                                 xi,
3364                                                                 xis,
3365                                                                 xbs,
3366                                                                 NULL,
3367                                                                 NULL,
3368                                                                 &credits);
3369                                 xis->not_found = old_found;
3370                                 if (ret) {
3371                                         mlog_errno(ret);
3372                                         goto out;
3373                                 }
3374
3375                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3376                                 if (ret) {
3377                                         mlog_errno(ret);
3378                                         goto out;
3379                                 }
3380                         }
3381                         /*
3382                          * If no space in inode, we will set extended attribute
3383                          * into external block.
3384                          */
3385                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3386                         if (ret)
3387                                 goto out;
3388                         if (!xis->not_found) {
3389                                 /*
3390                                  * If succeed and that extended attribute
3391                                  * existing in inode, we will remove it.
3392                                  */
3393                                 xi->xi_value = NULL;
3394                                 xi->xi_value_len = 0;
3395                                 xbs->not_found = -ENODATA;
3396                                 ret = ocfs2_calc_xattr_set_need(inode,
3397                                                                 di,
3398                                                                 xi,
3399                                                                 xis,
3400                                                                 xbs,
3401                                                                 NULL,
3402                                                                 NULL,
3403                                                                 &credits);
3404                                 if (ret) {
3405                                         mlog_errno(ret);
3406                                         goto out;
3407                                 }
3408
3409                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3410                                 if (ret) {
3411                                         mlog_errno(ret);
3412                                         goto out;
3413                                 }
3414                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3415                                                             xis, ctxt);
3416                         }
3417                 }
3418         }
3419
3420         if (!ret) {
3421                 /* Update inode ctime. */
3422                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3423                                               xis->inode_bh,
3424                                               OCFS2_JOURNAL_ACCESS_WRITE);
3425                 if (ret) {
3426                         mlog_errno(ret);
3427                         goto out;
3428                 }
3429
3430                 inode->i_ctime = current_time(inode);
3431                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3432                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3433                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3434         }
3435 out:
3436         return ret;
3437 }
3438
3439 /*
3440  * This function only called duing creating inode
3441  * for init security/acl xattrs of the new inode.
3442  * All transanction credits have been reserved in mknod.
3443  */
3444 int ocfs2_xattr_set_handle(handle_t *handle,
3445                            struct inode *inode,
3446                            struct buffer_head *di_bh,
3447                            int name_index,
3448                            const char *name,
3449                            const void *value,
3450                            size_t value_len,
3451                            int flags,
3452                            struct ocfs2_alloc_context *meta_ac,
3453                            struct ocfs2_alloc_context *data_ac)
3454 {
3455         struct ocfs2_dinode *di;
3456         int ret;
3457
3458         struct ocfs2_xattr_info xi = {
3459                 .xi_name_index = name_index,
3460                 .xi_name = name,
3461                 .xi_name_len = strlen(name),
3462                 .xi_value = value,
3463                 .xi_value_len = value_len,
3464         };
3465
3466         struct ocfs2_xattr_search xis = {
3467                 .not_found = -ENODATA,
3468         };
3469
3470         struct ocfs2_xattr_search xbs = {
3471                 .not_found = -ENODATA,
3472         };
3473
3474         struct ocfs2_xattr_set_ctxt ctxt = {
3475                 .handle = handle,
3476                 .meta_ac = meta_ac,
3477                 .data_ac = data_ac,
3478         };
3479
3480         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3481                 return -EOPNOTSUPP;
3482
3483         /*
3484          * In extreme situation, may need xattr bucket when
3485          * block size is too small. And we have already reserved
3486          * the credits for bucket in mknod.
3487          */
3488         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3489                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3490                 if (!xbs.bucket) {
3491                         mlog_errno(-ENOMEM);
3492                         return -ENOMEM;
3493                 }
3494         }
3495
3496         xis.inode_bh = xbs.inode_bh = di_bh;
3497         di = (struct ocfs2_dinode *)di_bh->b_data;
3498
3499         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3500
3501         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3502         if (ret)
3503                 goto cleanup;
3504         if (xis.not_found) {
3505                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3506                 if (ret)
3507                         goto cleanup;
3508         }
3509
3510         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3511
3512 cleanup:
3513         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3514         brelse(xbs.xattr_bh);
3515         ocfs2_xattr_bucket_free(xbs.bucket);
3516
3517         return ret;
3518 }
3519
3520 /*
3521  * ocfs2_xattr_set()
3522  *
3523  * Set, replace or remove an extended attribute for this inode.
3524  * value is NULL to remove an existing extended attribute, else either
3525  * create or replace an extended attribute.
3526  */
3527 int ocfs2_xattr_set(struct inode *inode,
3528                     int name_index,
3529                     const char *name,
3530                     const void *value,
3531                     size_t value_len,
3532                     int flags)
3533 {
3534         struct buffer_head *di_bh = NULL;
3535         struct ocfs2_dinode *di;
3536         int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3537         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3538         struct inode *tl_inode = osb->osb_tl_inode;
3539         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3540         struct ocfs2_refcount_tree *ref_tree = NULL;
3541         struct ocfs2_lock_holder oh;
3542
3543         struct ocfs2_xattr_info xi = {
3544                 .xi_name_index = name_index,
3545                 .xi_name = name,
3546                 .xi_name_len = strlen(name),
3547                 .xi_value = value,
3548                 .xi_value_len = value_len,
3549         };
3550
3551         struct ocfs2_xattr_search xis = {
3552                 .not_found = -ENODATA,
3553         };
3554
3555         struct ocfs2_xattr_search xbs = {
3556                 .not_found = -ENODATA,
3557         };
3558
3559         if (!ocfs2_supports_xattr(osb))
3560                 return -EOPNOTSUPP;
3561
3562         /*
3563          * Only xbs will be used on indexed trees.  xis doesn't need a
3564          * bucket.
3565          */
3566         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3567         if (!xbs.bucket) {
3568                 mlog_errno(-ENOMEM);
3569                 return -ENOMEM;
3570         }
3571
3572         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3573         if (had_lock < 0) {
3574                 ret = had_lock;
3575                 mlog_errno(ret);
3576                 goto cleanup_nolock;
3577         }
3578         xis.inode_bh = xbs.inode_bh = di_bh;
3579         di = (struct ocfs2_dinode *)di_bh->b_data;
3580
3581         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3582         /*
3583          * Scan inode and external block to find the same name
3584          * extended attribute and collect search information.
3585          */
3586         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3587         if (ret)
3588                 goto cleanup;
3589         if (xis.not_found) {
3590                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3591                 if (ret)
3592                         goto cleanup;
3593         }
3594
3595         if (xis.not_found && xbs.not_found) {
3596                 ret = -ENODATA;
3597                 if (flags & XATTR_REPLACE)
3598                         goto cleanup;
3599                 ret = 0;
3600                 if (!value)
3601                         goto cleanup;
3602         } else {
3603                 ret = -EEXIST;
3604                 if (flags & XATTR_CREATE)
3605                         goto cleanup;
3606         }
3607
3608         /* Check whether the value is refcounted and do some preparation. */
3609         if (ocfs2_is_refcount_inode(inode) &&
3610             (!xis.not_found || !xbs.not_found)) {
3611                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3612                                                    &xis, &xbs, &ref_tree,
3613                                                    &ref_meta, &ref_credits);
3614                 if (ret) {
3615                         mlog_errno(ret);
3616                         goto cleanup;
3617                 }
3618         }
3619
3620         inode_lock(tl_inode);
3621
3622         if (ocfs2_truncate_log_needs_flush(osb)) {
3623                 ret = __ocfs2_flush_truncate_log(osb);
3624                 if (ret < 0) {
3625                         inode_unlock(tl_inode);
3626                         mlog_errno(ret);
3627                         goto cleanup;
3628                 }
3629         }
3630         inode_unlock(tl_inode);
3631
3632         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3633                                         &xbs, &ctxt, ref_meta, &credits);
3634         if (ret) {
3635                 mlog_errno(ret);
3636                 goto cleanup;
3637         }
3638
3639         /* we need to update inode's ctime field, so add credit for it. */
3640         credits += OCFS2_INODE_UPDATE_CREDITS;
3641         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3642         if (IS_ERR(ctxt.handle)) {
3643                 ret = PTR_ERR(ctxt.handle);
3644                 mlog_errno(ret);
3645                 goto out_free_ac;
3646         }
3647
3648         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3649         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3650
3651         ocfs2_commit_trans(osb, ctxt.handle);
3652
3653 out_free_ac:
3654         if (ctxt.data_ac)
3655                 ocfs2_free_alloc_context(ctxt.data_ac);
3656         if (ctxt.meta_ac)
3657                 ocfs2_free_alloc_context(ctxt.meta_ac);
3658         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3659                 ocfs2_schedule_truncate_log_flush(osb, 1);
3660         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3661
3662 cleanup:
3663         if (ref_tree)
3664                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3665         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3666         if (!value && !ret) {
3667                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3668                 if (ret)
3669                         mlog_errno(ret);
3670         }
3671         ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3672 cleanup_nolock:
3673         brelse(di_bh);
3674         brelse(xbs.xattr_bh);
3675         ocfs2_xattr_bucket_free(xbs.bucket);
3676
3677         return ret;
3678 }
3679
3680 /*
3681  * Find the xattr extent rec which may contains name_hash.
3682  * e_cpos will be the first name hash of the xattr rec.
3683  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3684  */
3685 static int ocfs2_xattr_get_rec(struct inode *inode,
3686                                u32 name_hash,
3687                                u64 *p_blkno,
3688                                u32 *e_cpos,
3689                                u32 *num_clusters,
3690                                struct ocfs2_extent_list *el)
3691 {
3692         int ret = 0, i;
3693         struct buffer_head *eb_bh = NULL;
3694         struct ocfs2_extent_block *eb;
3695         struct ocfs2_extent_rec *rec = NULL;
3696         u64 e_blkno = 0;
3697
3698         if (el->l_tree_depth) {
3699                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3700                                       &eb_bh);
3701                 if (ret) {
3702                         mlog_errno(ret);
3703                         goto out;
3704                 }
3705
3706                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3707                 el = &eb->h_list;
3708
3709                 if (el->l_tree_depth) {
3710                         ret = ocfs2_error(inode->i_sb,
3711                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3712                                           inode->i_ino,
3713                                           (unsigned long long)eb_bh->b_blocknr);
3714                         goto out;
3715                 }
3716         }
3717
3718         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3719                 rec = &el->l_recs[i];
3720
3721                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3722                         e_blkno = le64_to_cpu(rec->e_blkno);
3723                         break;
3724                 }
3725         }
3726
3727         if (!e_blkno) {
3728                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3729                                   inode->i_ino,
3730                                   le32_to_cpu(rec->e_cpos),
3731                                   ocfs2_rec_clusters(el, rec));
3732                 goto out;
3733         }
3734
3735         *p_blkno = le64_to_cpu(rec->e_blkno);
3736         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3737         if (e_cpos)
3738                 *e_cpos = le32_to_cpu(rec->e_cpos);
3739 out:
3740         brelse(eb_bh);
3741         return ret;
3742 }
3743
3744 typedef int (xattr_bucket_func)(struct inode *inode,
3745                                 struct ocfs2_xattr_bucket *bucket,
3746                                 void *para);
3747
3748 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3749                                    struct ocfs2_xattr_bucket *bucket,
3750                                    int name_index,
3751                                    const char *name,
3752                                    u32 name_hash,
3753                                    u16 *xe_index,
3754                                    int *found)
3755 {
3756         int i, ret = 0, cmp = 1, block_off, new_offset;
3757         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3758         size_t name_len = strlen(name);
3759         struct ocfs2_xattr_entry *xe = NULL;
3760         char *xe_name;
3761
3762         /*
3763          * We don't use binary search in the bucket because there
3764          * may be multiple entries with the same name hash.
3765          */
3766         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3767                 xe = &xh->xh_entries[i];
3768
3769                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3770                         continue;
3771                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3772                         break;
3773
3774                 cmp = name_index - ocfs2_xattr_get_type(xe);
3775                 if (!cmp)
3776                         cmp = name_len - xe->xe_name_len;
3777                 if (cmp)
3778                         continue;
3779
3780                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3781                                                         xh,
3782                                                         i,
3783                                                         &block_off,
3784                                                         &new_offset);
3785                 if (ret) {
3786                         mlog_errno(ret);
3787                         break;
3788                 }
3789
3790
3791                 xe_name = bucket_block(bucket, block_off) + new_offset;
3792                 if (!memcmp(name, xe_name, name_len)) {
3793                         *xe_index = i;
3794                         *found = 1;
3795                         ret = 0;
3796                         break;
3797                 }
3798         }
3799
3800         return ret;
3801 }
3802
3803 /*
3804  * Find the specified xattr entry in a series of buckets.
3805  * This series start from p_blkno and last for num_clusters.
3806  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3807  * the num of the valid buckets.
3808  *
3809  * Return the buffer_head this xattr should reside in. And if the xattr's
3810  * hash is in the gap of 2 buckets, return the lower bucket.
3811  */
3812 static int ocfs2_xattr_bucket_find(struct inode *inode,
3813                                    int name_index,
3814                                    const char *name,
3815                                    u32 name_hash,
3816                                    u64 p_blkno,
3817                                    u32 first_hash,
3818                                    u32 num_clusters,
3819                                    struct ocfs2_xattr_search *xs)
3820 {
3821         int ret, found = 0;
3822         struct ocfs2_xattr_header *xh = NULL;
3823         struct ocfs2_xattr_entry *xe = NULL;
3824         u16 index = 0;
3825         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3826         int low_bucket = 0, bucket, high_bucket;
3827         struct ocfs2_xattr_bucket *search;
3828         u64 blkno, lower_blkno = 0;
3829
3830         search = ocfs2_xattr_bucket_new(inode);
3831         if (!search) {
3832                 ret = -ENOMEM;
3833                 mlog_errno(ret);
3834                 goto out;
3835         }
3836
3837         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3838         if (ret) {
3839                 mlog_errno(ret);
3840                 goto out;
3841         }
3842
3843         xh = bucket_xh(search);
3844         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3845         while (low_bucket <= high_bucket) {
3846                 ocfs2_xattr_bucket_relse(search);
3847
3848                 bucket = (low_bucket + high_bucket) / 2;
3849                 blkno = p_blkno + bucket * blk_per_bucket;
3850                 ret = ocfs2_read_xattr_bucket(search, blkno);
3851                 if (ret) {
3852                         mlog_errno(ret);
3853                         goto out;
3854                 }
3855
3856                 xh = bucket_xh(search);
3857                 xe = &xh->xh_entries[0];
3858                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3859                         high_bucket = bucket - 1;
3860                         continue;
3861                 }
3862
3863                 /*
3864                  * Check whether the hash of the last entry in our
3865                  * bucket is larger than the search one. for an empty
3866                  * bucket, the last one is also the first one.
3867                  */
3868                 if (xh->xh_count)
3869                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3870
3871                 /* record lower_blkno which may be the insert place. */
3872                 lower_blkno = blkno;
3873
3874                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3875                         low_bucket = bucket + 1;
3876                         continue;
3877                 }
3878
3879                 /* the searched xattr should reside in this bucket if exists. */
3880                 ret = ocfs2_find_xe_in_bucket(inode, search,
3881                                               name_index, name, name_hash,
3882                                               &index, &found);
3883                 if (ret) {
3884                         mlog_errno(ret);
3885                         goto out;
3886                 }
3887                 break;
3888         }
3889
3890         /*
3891          * Record the bucket we have found.
3892          * When the xattr's hash value is in the gap of 2 buckets, we will
3893          * always set it to the previous bucket.
3894          */
3895         if (!lower_blkno)
3896                 lower_blkno = p_blkno;
3897
3898         /* This should be in cache - we just read it during the search */
3899         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3900         if (ret) {
3901                 mlog_errno(ret);
3902                 goto out;
3903         }
3904
3905         xs->header = bucket_xh(xs->bucket);
3906         xs->base = bucket_block(xs->bucket, 0);
3907         xs->end = xs->base + inode->i_sb->s_blocksize;
3908
3909         if (found) {
3910                 xs->here = &xs->header->xh_entries[index];
3911                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3912                         name, name_index, name_hash,
3913                         (unsigned long long)bucket_blkno(xs->bucket),
3914                         index);
3915         } else
3916                 ret = -ENODATA;
3917
3918 out:
3919         ocfs2_xattr_bucket_free(search);
3920         return ret;
3921 }
3922
3923 static int ocfs2_xattr_index_block_find(struct inode *inode,
3924                                         struct buffer_head *root_bh,
3925                                         int name_index,
3926                                         const char *name,
3927                                         struct ocfs2_xattr_search *xs)
3928 {
3929         int ret;
3930         struct ocfs2_xattr_block *xb =
3931                         (struct ocfs2_xattr_block *)root_bh->b_data;
3932         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3933         struct ocfs2_extent_list *el = &xb_root->xt_list;
3934         u64 p_blkno = 0;
3935         u32 first_hash, num_clusters = 0;
3936         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3937
3938         if (le16_to_cpu(el->l_next_free_rec) == 0)
3939                 return -ENODATA;
3940
3941         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3942                                         name, name_index, name_hash,
3943                                         (unsigned long long)root_bh->b_blocknr,
3944                                         -1);
3945
3946         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3947                                   &num_clusters, el);
3948         if (ret) {
3949                 mlog_errno(ret);
3950                 goto out;
3951         }
3952
3953         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3954
3955         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3956                                         name, name_index, first_hash,
3957                                         (unsigned long long)p_blkno,
3958                                         num_clusters);
3959
3960         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3961                                       p_blkno, first_hash, num_clusters, xs);
3962
3963 out:
3964         return ret;
3965 }
3966
3967 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3968                                        u64 blkno,
3969                                        u32 clusters,
3970                                        xattr_bucket_func *func,
3971                                        void *para)
3972 {
3973         int i, ret = 0;
3974         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3975         u32 num_buckets = clusters * bpc;
3976         struct ocfs2_xattr_bucket *bucket;
3977
3978         bucket = ocfs2_xattr_bucket_new(inode);
3979         if (!bucket) {
3980                 mlog_errno(-ENOMEM);
3981                 return -ENOMEM;
3982         }
3983
3984         trace_ocfs2_iterate_xattr_buckets(
3985                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3986                 (unsigned long long)blkno, clusters);
3987
3988         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3989                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3990                 if (ret) {
3991                         mlog_errno(ret);
3992                         break;
3993                 }
3994
3995                 /*
3996                  * The real bucket num in this series of blocks is stored
3997                  * in the 1st bucket.
3998                  */
3999                 if (i == 0)
4000                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4001
4002                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4003                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4004                 if (func) {
4005                         ret = func(inode, bucket, para);
4006                         if (ret && ret != -ERANGE)
4007                                 mlog_errno(ret);
4008                         /* Fall through to bucket_relse() */
4009                 }
4010
4011                 ocfs2_xattr_bucket_relse(bucket);
4012                 if (ret)
4013                         break;
4014         }
4015
4016         ocfs2_xattr_bucket_free(bucket);
4017         return ret;
4018 }
4019
4020 struct ocfs2_xattr_tree_list {
4021         char *buffer;
4022         size_t buffer_size;
4023         size_t result;
4024 };
4025
4026 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4027                                              struct ocfs2_xattr_header *xh,
4028                                              int index,
4029                                              int *block_off,
4030                                              int *new_offset)
4031 {
4032         u16 name_offset;
4033
4034         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4035                 return -EINVAL;
4036
4037         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4038
4039         *block_off = name_offset >> sb->s_blocksize_bits;
4040         *new_offset = name_offset % sb->s_blocksize;
4041
4042         return 0;
4043 }
4044
4045 static int ocfs2_list_xattr_bucket(struct inode *inode,
4046                                    struct ocfs2_xattr_bucket *bucket,
4047                                    void *para)
4048 {
4049         int ret = 0, type;
4050         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4051         int i, block_off, new_offset;
4052         const char *name;
4053
4054         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4055                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4056                 type = ocfs2_xattr_get_type(entry);
4057
4058                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4059                                                         bucket_xh(bucket),
4060                                                         i,
4061                                                         &block_off,
4062                                                         &new_offset);
4063                 if (ret)
4064                         break;
4065
4066                 name = (const char *)bucket_block(bucket, block_off) +
4067                         new_offset;
4068                 ret = ocfs2_xattr_list_entry(inode->i_sb,
4069                                              xl->buffer,
4070                                              xl->buffer_size,
4071                                              &xl->result,
4072                                              type, name,
4073                                              entry->xe_name_len);
4074                 if (ret)
4075                         break;
4076         }
4077
4078         return ret;
4079 }
4080
4081 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4082                                            struct buffer_head *blk_bh,
4083                                            xattr_tree_rec_func *rec_func,
4084                                            void *para)
4085 {
4086         struct ocfs2_xattr_block *xb =
4087                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4088         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4089         int ret = 0;
4090         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4091         u64 p_blkno = 0;
4092
4093         if (!el->l_next_free_rec || !rec_func)
4094                 return 0;
4095
4096         while (name_hash > 0) {
4097                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4098                                           &e_cpos, &num_clusters, el);
4099                 if (ret) {
4100                         mlog_errno(ret);
4101                         break;
4102                 }
4103
4104                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4105                                num_clusters, para);
4106                 if (ret) {
4107                         if (ret != -ERANGE)
4108                                 mlog_errno(ret);
4109                         break;
4110                 }
4111
4112                 if (e_cpos == 0)
4113                         break;
4114
4115                 name_hash = e_cpos - 1;
4116         }
4117
4118         return ret;
4119
4120 }
4121
4122 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4123                                      struct buffer_head *root_bh,
4124                                      u64 blkno, u32 cpos, u32 len, void *para)
4125 {
4126         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4127                                            ocfs2_list_xattr_bucket, para);
4128 }
4129
4130 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4131                                              struct buffer_head *blk_bh,
4132                                              char *buffer,
4133                                              size_t buffer_size)
4134 {
4135         int ret;
4136         struct ocfs2_xattr_tree_list xl = {
4137                 .buffer = buffer,
4138                 .buffer_size = buffer_size,
4139                 .result = 0,
4140         };
4141
4142         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4143                                               ocfs2_list_xattr_tree_rec, &xl);
4144         if (ret) {
4145                 mlog_errno(ret);
4146                 goto out;
4147         }
4148
4149         ret = xl.result;
4150 out:
4151         return ret;
4152 }
4153
4154 static int cmp_xe(const void *a, const void *b)
4155 {
4156         const struct ocfs2_xattr_entry *l = a, *r = b;
4157         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4158         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4159
4160         if (l_hash > r_hash)
4161                 return 1;
4162         if (l_hash < r_hash)
4163                 return -1;
4164         return 0;
4165 }
4166
4167 static void swap_xe(void *a, void *b, int size)
4168 {
4169         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4170
4171         tmp = *l;
4172         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4173         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4174 }
4175
4176 /*
4177  * When the ocfs2_xattr_block is filled up, new bucket will be created
4178  * and all the xattr entries will be moved to the new bucket.
4179  * The header goes at the start of the bucket, and the names+values are
4180  * filled from the end.  This is why *target starts as the last buffer.
4181  * Note: we need to sort the entries since they are not saved in order
4182  * in the ocfs2_xattr_block.
4183  */
4184 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4185                                            struct buffer_head *xb_bh,
4186                                            struct ocfs2_xattr_bucket *bucket)
4187 {
4188         int i, blocksize = inode->i_sb->s_blocksize;
4189         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4190         u16 offset, size, off_change;
4191         struct ocfs2_xattr_entry *xe;
4192         struct ocfs2_xattr_block *xb =
4193                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4194         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4195         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4196         u16 count = le16_to_cpu(xb_xh->xh_count);
4197         char *src = xb_bh->b_data;
4198         char *target = bucket_block(bucket, blks - 1);
4199
4200         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4201                                 (unsigned long long)xb_bh->b_blocknr,
4202                                 (unsigned long long)bucket_blkno(bucket));
4203
4204         for (i = 0; i < blks; i++)
4205                 memset(bucket_block(bucket, i), 0, blocksize);
4206
4207         /*
4208          * Since the xe_name_offset is based on ocfs2_xattr_header,
4209          * there is a offset change corresponding to the change of
4210          * ocfs2_xattr_header's position.
4211          */
4212         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4213         xe = &xb_xh->xh_entries[count - 1];
4214         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4215         size = blocksize - offset;
4216
4217         /* copy all the names and values. */
4218         memcpy(target + offset, src + offset, size);
4219
4220         /* Init new header now. */
4221         xh->xh_count = xb_xh->xh_count;
4222         xh->xh_num_buckets = cpu_to_le16(1);
4223         xh->xh_name_value_len = cpu_to_le16(size);
4224         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4225
4226         /* copy all the entries. */
4227         target = bucket_block(bucket, 0);
4228         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4229         size = count * sizeof(struct ocfs2_xattr_entry);
4230         memcpy(target + offset, (char *)xb_xh + offset, size);
4231
4232         /* Change the xe offset for all the xe because of the move. */
4233         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4234                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4235         for (i = 0; i < count; i++)
4236                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4237
4238         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4239
4240         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4241              cmp_xe, swap_xe);
4242 }
4243
4244 /*
4245  * After we move xattr from block to index btree, we have to
4246  * update ocfs2_xattr_search to the new xe and base.
4247  *
4248  * When the entry is in xattr block, xattr_bh indicates the storage place.
4249  * While if the entry is in index b-tree, "bucket" indicates the
4250  * real place of the xattr.
4251  */
4252 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4253                                             struct ocfs2_xattr_search *xs,
4254                                             struct buffer_head *old_bh)
4255 {
4256         char *buf = old_bh->b_data;
4257         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4258         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4259         int i;
4260
4261         xs->header = bucket_xh(xs->bucket);
4262         xs->base = bucket_block(xs->bucket, 0);
4263         xs->end = xs->base + inode->i_sb->s_blocksize;
4264
4265         if (xs->not_found)
4266                 return;
4267
4268         i = xs->here - old_xh->xh_entries;
4269         xs->here = &xs->header->xh_entries[i];
4270 }
4271
4272 static int ocfs2_xattr_create_index_block(struct inode *inode,
4273                                           struct ocfs2_xattr_search *xs,
4274                                           struct ocfs2_xattr_set_ctxt *ctxt)
4275 {
4276         int ret;
4277         u32 bit_off, len;
4278         u64 blkno;
4279         handle_t *handle = ctxt->handle;
4280         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4281         struct buffer_head *xb_bh = xs->xattr_bh;
4282         struct ocfs2_xattr_block *xb =
4283                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4284         struct ocfs2_xattr_tree_root *xr;
4285         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4286
4287         trace_ocfs2_xattr_create_index_block_begin(
4288                                 (unsigned long long)xb_bh->b_blocknr);
4289
4290         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4291         BUG_ON(!xs->bucket);
4292
4293         /*
4294          * XXX:
4295          * We can use this lock for now, and maybe move to a dedicated mutex
4296          * if performance becomes a problem later.
4297          */
4298         down_write(&oi->ip_alloc_sem);
4299
4300         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4301                                       OCFS2_JOURNAL_ACCESS_WRITE);
4302         if (ret) {
4303                 mlog_errno(ret);
4304                 goto out;
4305         }
4306
4307         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4308                                      1, 1, &bit_off, &len);
4309         if (ret) {
4310                 mlog_errno(ret);
4311                 goto out;
4312         }
4313
4314         /*
4315          * The bucket may spread in many blocks, and
4316          * we will only touch the 1st block and the last block
4317          * in the whole bucket(one for entry and one for data).
4318          */
4319         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4320
4321         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4322
4323         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4324         if (ret) {
4325                 mlog_errno(ret);
4326                 goto out;
4327         }
4328
4329         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4330                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4331         if (ret) {
4332                 mlog_errno(ret);
4333                 goto out;
4334         }
4335
4336         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4337         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4338
4339         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4340
4341         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4342         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4343                offsetof(struct ocfs2_xattr_block, xb_attrs));
4344
4345         xr = &xb->xb_attrs.xb_root;
4346         xr->xt_clusters = cpu_to_le32(1);
4347         xr->xt_last_eb_blk = 0;
4348         xr->xt_list.l_tree_depth = 0;
4349         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4350         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4351
4352         xr->xt_list.l_recs[0].e_cpos = 0;
4353         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4354         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4355
4356         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4357
4358         ocfs2_journal_dirty(handle, xb_bh);
4359
4360 out:
4361         up_write(&oi->ip_alloc_sem);
4362
4363         return ret;
4364 }
4365
4366 static int cmp_xe_offset(const void *a, const void *b)
4367 {
4368         const struct ocfs2_xattr_entry *l = a, *r = b;
4369         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4370         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4371
4372         if (l_name_offset < r_name_offset)
4373                 return 1;
4374         if (l_name_offset > r_name_offset)
4375                 return -1;
4376         return 0;
4377 }
4378
4379 /*
4380  * defrag a xattr bucket if we find that the bucket has some
4381  * holes beteen name/value pairs.
4382  * We will move all the name/value pairs to the end of the bucket
4383  * so that we can spare some space for insertion.
4384  */
4385 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4386                                      handle_t *handle,
4387                                      struct ocfs2_xattr_bucket *bucket)
4388 {
4389         int ret, i;
4390         size_t end, offset, len;
4391         struct ocfs2_xattr_header *xh;
4392         char *entries, *buf, *bucket_buf = NULL;
4393         u64 blkno = bucket_blkno(bucket);
4394         u16 xh_free_start;
4395         size_t blocksize = inode->i_sb->s_blocksize;
4396         struct ocfs2_xattr_entry *xe;
4397
4398         /*
4399          * In order to make the operation more efficient and generic,
4400          * we copy all the blocks into a contiguous memory and do the
4401          * defragment there, so if anything is error, we will not touch
4402          * the real block.
4403          */
4404         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4405         if (!bucket_buf) {
4406                 ret = -EIO;
4407                 goto out;
4408         }
4409
4410         buf = bucket_buf;
4411         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4412                 memcpy(buf, bucket_block(bucket, i), blocksize);
4413
4414         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4415                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4416         if (ret < 0) {
4417                 mlog_errno(ret);
4418                 goto out;
4419         }
4420
4421         xh = (struct ocfs2_xattr_header *)bucket_buf;
4422         entries = (char *)xh->xh_entries;
4423         xh_free_start = le16_to_cpu(xh->xh_free_start);
4424
4425         trace_ocfs2_defrag_xattr_bucket(
4426              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4427              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4428
4429         /*
4430          * sort all the entries by their offset.
4431          * the largest will be the first, so that we can
4432          * move them to the end one by one.
4433          */
4434         sort(entries, le16_to_cpu(xh->xh_count),
4435              sizeof(struct ocfs2_xattr_entry),
4436              cmp_xe_offset, swap_xe);
4437
4438         /* Move all name/values to the end of the bucket. */
4439         xe = xh->xh_entries;
4440         end = OCFS2_XATTR_BUCKET_SIZE;
4441         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4442                 offset = le16_to_cpu(xe->xe_name_offset);
4443                 len = namevalue_size_xe(xe);
4444
4445                 /*
4446                  * We must make sure that the name/value pair
4447                  * exist in the same block. So adjust end to
4448                  * the previous block end if needed.
4449                  */
4450                 if (((end - len) / blocksize !=
4451                         (end - 1) / blocksize))
4452                         end = end - end % blocksize;
4453
4454                 if (end > offset + len) {
4455                         memmove(bucket_buf + end - len,
4456                                 bucket_buf + offset, len);
4457                         xe->xe_name_offset = cpu_to_le16(end - len);
4458                 }
4459
4460                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4461                                 "bucket %llu\n", (unsigned long long)blkno);
4462
4463                 end -= len;
4464         }
4465
4466         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4467                         "bucket %llu\n", (unsigned long long)blkno);
4468
4469         if (xh_free_start == end)
4470                 goto out;
4471
4472         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4473         xh->xh_free_start = cpu_to_le16(end);
4474
4475         /* sort the entries by their name_hash. */
4476         sort(entries, le16_to_cpu(xh->xh_count),
4477              sizeof(struct ocfs2_xattr_entry),
4478              cmp_xe, swap_xe);
4479
4480         buf = bucket_buf;
4481         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4482                 memcpy(bucket_block(bucket, i), buf, blocksize);
4483         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4484
4485 out:
4486         kfree(bucket_buf);
4487         return ret;
4488 }
4489
4490 /*
4491  * prev_blkno points to the start of an existing extent.  new_blkno
4492  * points to a newly allocated extent.  Because we know each of our
4493  * clusters contains more than bucket, we can easily split one cluster
4494  * at a bucket boundary.  So we take the last cluster of the existing
4495  * extent and split it down the middle.  We move the last half of the
4496  * buckets in the last cluster of the existing extent over to the new
4497  * extent.
4498  *
4499  * first_bh is the buffer at prev_blkno so we can update the existing
4500  * extent's bucket count.  header_bh is the bucket were we were hoping
4501  * to insert our xattr.  If the bucket move places the target in the new
4502  * extent, we'll update first_bh and header_bh after modifying the old
4503  * extent.
4504  *
4505  * first_hash will be set as the 1st xe's name_hash in the new extent.
4506  */
4507 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4508                                                handle_t *handle,
4509                                                struct ocfs2_xattr_bucket *first,
4510                                                struct ocfs2_xattr_bucket *target,
4511                                                u64 new_blkno,
4512                                                u32 num_clusters,
4513                                                u32 *first_hash)
4514 {
4515         int ret;
4516         struct super_block *sb = inode->i_sb;
4517         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4518         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4519         int to_move = num_buckets / 2;
4520         u64 src_blkno;
4521         u64 last_cluster_blkno = bucket_blkno(first) +
4522                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4523
4524         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4525         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4526
4527         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4528                                 (unsigned long long)last_cluster_blkno,
4529                                 (unsigned long long)new_blkno);
4530
4531         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4532                                      last_cluster_blkno, new_blkno,
4533                                      to_move, first_hash);
4534         if (ret) {
4535                 mlog_errno(ret);
4536                 goto out;
4537         }
4538
4539         /* This is the first bucket that got moved */
4540         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4541
4542         /*
4543          * If the target bucket was part of the moved buckets, we need to
4544          * update first and target.
4545          */
4546         if (bucket_blkno(target) >= src_blkno) {
4547                 /* Find the block for the new target bucket */
4548                 src_blkno = new_blkno +
4549                         (bucket_blkno(target) - src_blkno);
4550
4551                 ocfs2_xattr_bucket_relse(first);
4552                 ocfs2_xattr_bucket_relse(target);
4553
4554                 /*
4555                  * These shouldn't fail - the buffers are in the
4556                  * journal from ocfs2_cp_xattr_bucket().
4557                  */
4558                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4559                 if (ret) {
4560                         mlog_errno(ret);
4561                         goto out;
4562                 }
4563                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4564                 if (ret)
4565                         mlog_errno(ret);
4566
4567         }
4568
4569 out:
4570         return ret;
4571 }
4572
4573 /*
4574  * Find the suitable pos when we divide a bucket into 2.
4575  * We have to make sure the xattrs with the same hash value exist
4576  * in the same bucket.
4577  *
4578  * If this ocfs2_xattr_header covers more than one hash value, find a
4579  * place where the hash value changes.  Try to find the most even split.
4580  * The most common case is that all entries have different hash values,
4581  * and the first check we make will find a place to split.
4582  */
4583 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4584 {
4585         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4586         int count = le16_to_cpu(xh->xh_count);
4587         int delta, middle = count / 2;
4588
4589         /*
4590          * We start at the middle.  Each step gets farther away in both
4591          * directions.  We therefore hit the change in hash value
4592          * nearest to the middle.  Note that this loop does not execute for
4593          * count < 2.
4594          */
4595         for (delta = 0; delta < middle; delta++) {
4596                 /* Let's check delta earlier than middle */
4597                 if (cmp_xe(&entries[middle - delta - 1],
4598                            &entries[middle - delta]))
4599                         return middle - delta;
4600
4601                 /* For even counts, don't walk off the end */
4602                 if ((middle + delta + 1) == count)
4603                         continue;
4604
4605                 /* Now try delta past middle */
4606                 if (cmp_xe(&entries[middle + delta],
4607                            &entries[middle + delta + 1]))
4608                         return middle + delta + 1;
4609         }
4610
4611         /* Every entry had the same hash */
4612         return count;
4613 }
4614
4615 /*
4616  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4617  * first_hash will record the 1st hash of the new bucket.
4618  *
4619  * Normally half of the xattrs will be moved.  But we have to make
4620  * sure that the xattrs with the same hash value are stored in the
4621  * same bucket. If all the xattrs in this bucket have the same hash
4622  * value, the new bucket will be initialized as an empty one and the
4623  * first_hash will be initialized as (hash_value+1).
4624  */
4625 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4626                                     handle_t *handle,
4627                                     u64 blk,
4628                                     u64 new_blk,
4629                                     u32 *first_hash,
4630                                     int new_bucket_head)
4631 {
4632         int ret, i;
4633         int count, start, len, name_value_len = 0, name_offset = 0;
4634         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4635         struct ocfs2_xattr_header *xh;
4636         struct ocfs2_xattr_entry *xe;
4637         int blocksize = inode->i_sb->s_blocksize;
4638
4639         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4640                                               (unsigned long long)new_blk);
4641
4642         s_bucket = ocfs2_xattr_bucket_new(inode);
4643         t_bucket = ocfs2_xattr_bucket_new(inode);
4644         if (!s_bucket || !t_bucket) {
4645                 ret = -ENOMEM;
4646                 mlog_errno(ret);
4647                 goto out;
4648         }
4649
4650         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4651         if (ret) {
4652                 mlog_errno(ret);
4653                 goto out;
4654         }
4655
4656         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4657                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4658         if (ret) {
4659                 mlog_errno(ret);
4660                 goto out;
4661         }
4662
4663         /*
4664          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4665          * there's no need to read it.
4666          */
4667         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4668         if (ret) {
4669                 mlog_errno(ret);
4670                 goto out;
4671         }
4672
4673         /*
4674          * Hey, if we're overwriting t_bucket, what difference does
4675          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4676          * same part of ocfs2_cp_xattr_bucket().
4677          */
4678         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4679                                                 new_bucket_head ?
4680                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4681                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4682         if (ret) {
4683                 mlog_errno(ret);
4684                 goto out;
4685         }
4686
4687         xh = bucket_xh(s_bucket);
4688         count = le16_to_cpu(xh->xh_count);
4689         start = ocfs2_xattr_find_divide_pos(xh);
4690
4691         if (start == count) {
4692                 xe = &xh->xh_entries[start-1];
4693
4694                 /*
4695                  * initialized a new empty bucket here.
4696                  * The hash value is set as one larger than
4697                  * that of the last entry in the previous bucket.
4698                  */
4699                 for (i = 0; i < t_bucket->bu_blocks; i++)
4700                         memset(bucket_block(t_bucket, i), 0, blocksize);
4701
4702                 xh = bucket_xh(t_bucket);
4703                 xh->xh_free_start = cpu_to_le16(blocksize);
4704                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4705                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4706
4707                 goto set_num_buckets;
4708         }
4709
4710         /* copy the whole bucket to the new first. */
4711         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4712
4713         /* update the new bucket. */
4714         xh = bucket_xh(t_bucket);
4715
4716         /*
4717          * Calculate the total name/value len and xh_free_start for
4718          * the old bucket first.
4719          */
4720         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4721         name_value_len = 0;
4722         for (i = 0; i < start; i++) {
4723                 xe = &xh->xh_entries[i];
4724                 name_value_len += namevalue_size_xe(xe);
4725                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4726                         name_offset = le16_to_cpu(xe->xe_name_offset);
4727         }
4728
4729         /*
4730          * Now begin the modification to the new bucket.
4731          *
4732          * In the new bucket, We just move the xattr entry to the beginning
4733          * and don't touch the name/value. So there will be some holes in the
4734          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4735          * called.
4736          */
4737         xe = &xh->xh_entries[start];
4738         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4739         trace_ocfs2_divide_xattr_bucket_move(len,
4740                         (int)((char *)xe - (char *)xh),
4741                         (int)((char *)xh->xh_entries - (char *)xh));
4742         memmove((char *)xh->xh_entries, (char *)xe, len);
4743         xe = &xh->xh_entries[count - start];
4744         len = sizeof(struct ocfs2_xattr_entry) * start;
4745         memset((char *)xe, 0, len);
4746
4747         le16_add_cpu(&xh->xh_count, -start);
4748         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4749
4750         /* Calculate xh_free_start for the new bucket. */
4751         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4752         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4753                 xe = &xh->xh_entries[i];
4754                 if (le16_to_cpu(xe->xe_name_offset) <
4755                     le16_to_cpu(xh->xh_free_start))
4756                         xh->xh_free_start = xe->xe_name_offset;
4757         }
4758
4759 set_num_buckets:
4760         /* set xh->xh_num_buckets for the new xh. */
4761         if (new_bucket_head)
4762                 xh->xh_num_buckets = cpu_to_le16(1);
4763         else
4764                 xh->xh_num_buckets = 0;
4765
4766         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4767
4768         /* store the first_hash of the new bucket. */
4769         if (first_hash)
4770                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4771
4772         /*
4773          * Now only update the 1st block of the old bucket.  If we
4774          * just added a new empty bucket, there is no need to modify
4775          * it.
4776          */
4777         if (start == count)
4778                 goto out;
4779
4780         xh = bucket_xh(s_bucket);
4781         memset(&xh->xh_entries[start], 0,
4782                sizeof(struct ocfs2_xattr_entry) * (count - start));
4783         xh->xh_count = cpu_to_le16(start);
4784         xh->xh_free_start = cpu_to_le16(name_offset);
4785         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4786
4787         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4788
4789 out:
4790         ocfs2_xattr_bucket_free(s_bucket);
4791         ocfs2_xattr_bucket_free(t_bucket);
4792
4793         return ret;
4794 }
4795
4796 /*
4797  * Copy xattr from one bucket to another bucket.
4798  *
4799  * The caller must make sure that the journal transaction
4800  * has enough space for journaling.
4801  */
4802 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4803                                  handle_t *handle,
4804                                  u64 s_blkno,
4805                                  u64 t_blkno,
4806                                  int t_is_new)
4807 {
4808         int ret;
4809         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4810
4811         BUG_ON(s_blkno == t_blkno);
4812
4813         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4814                                     (unsigned long long)t_blkno,
4815                                     t_is_new);
4816
4817         s_bucket = ocfs2_xattr_bucket_new(inode);
4818         t_bucket = ocfs2_xattr_bucket_new(inode);
4819         if (!s_bucket || !t_bucket) {
4820                 ret = -ENOMEM;
4821                 mlog_errno(ret);
4822                 goto out;
4823         }
4824
4825         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4826         if (ret)
4827                 goto out;
4828
4829         /*
4830          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4831          * there's no need to read it.
4832          */
4833         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4834         if (ret)
4835                 goto out;
4836
4837         /*
4838          * Hey, if we're overwriting t_bucket, what difference does
4839          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4840          * cluster to fill, we came here from
4841          * ocfs2_mv_xattr_buckets(), and it is really new -
4842          * ACCESS_CREATE is required.  But we also might have moved data
4843          * out of t_bucket before extending back into it.
4844          * ocfs2_add_new_xattr_bucket() can do this - its call to
4845          * ocfs2_add_new_xattr_cluster() may have created a new extent
4846          * and copied out the end of the old extent.  Then it re-extends
4847          * the old extent back to create space for new xattrs.  That's
4848          * how we get here, and the bucket isn't really new.
4849          */
4850         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4851                                                 t_is_new ?
4852                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4853                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4854         if (ret)
4855                 goto out;
4856
4857         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4858         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4859
4860 out:
4861         ocfs2_xattr_bucket_free(t_bucket);
4862         ocfs2_xattr_bucket_free(s_bucket);
4863
4864         return ret;
4865 }
4866
4867 /*
4868  * src_blk points to the start of an existing extent.  last_blk points to
4869  * last cluster in that extent.  to_blk points to a newly allocated
4870  * extent.  We copy the buckets from the cluster at last_blk to the new
4871  * extent.  If start_bucket is non-zero, we skip that many buckets before
4872  * we start copying.  The new extent's xh_num_buckets gets set to the
4873  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4874  * by the same amount.
4875  */
4876 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4877                                   u64 src_blk, u64 last_blk, u64 to_blk,
4878                                   unsigned int start_bucket,
4879                                   u32 *first_hash)
4880 {
4881         int i, ret, credits;
4882         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4883         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4884         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4885         struct ocfs2_xattr_bucket *old_first, *new_first;
4886
4887         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4888                                      (unsigned long long)to_blk);
4889
4890         BUG_ON(start_bucket >= num_buckets);
4891         if (start_bucket) {
4892                 num_buckets -= start_bucket;
4893                 last_blk += (start_bucket * blks_per_bucket);
4894         }
4895
4896         /* The first bucket of the original extent */
4897         old_first = ocfs2_xattr_bucket_new(inode);
4898         /* The first bucket of the new extent */
4899         new_first = ocfs2_xattr_bucket_new(inode);
4900         if (!old_first || !new_first) {
4901                 ret = -ENOMEM;
4902                 mlog_errno(ret);
4903                 goto out;
4904         }
4905
4906         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4907         if (ret) {
4908                 mlog_errno(ret);
4909                 goto out;
4910         }
4911
4912         /*
4913          * We need to update the first bucket of the old extent and all
4914          * the buckets going to the new extent.
4915          */
4916         credits = ((num_buckets + 1) * blks_per_bucket);
4917         ret = ocfs2_extend_trans(handle, credits);
4918         if (ret) {
4919                 mlog_errno(ret);
4920                 goto out;
4921         }
4922
4923         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4924                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4925         if (ret) {
4926                 mlog_errno(ret);
4927                 goto out;
4928         }
4929
4930         for (i = 0; i < num_buckets; i++) {
4931                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4932                                             last_blk + (i * blks_per_bucket),
4933                                             to_blk + (i * blks_per_bucket),
4934                                             1);
4935                 if (ret) {
4936                         mlog_errno(ret);
4937                         goto out;
4938                 }
4939         }
4940
4941         /*
4942          * Get the new bucket ready before we dirty anything
4943          * (This actually shouldn't fail, because we already dirtied
4944          * it once in ocfs2_cp_xattr_bucket()).
4945          */
4946         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4947         if (ret) {
4948                 mlog_errno(ret);
4949                 goto out;
4950         }
4951         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4952                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4953         if (ret) {
4954                 mlog_errno(ret);
4955                 goto out;
4956         }
4957
4958         /* Now update the headers */
4959         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4960         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4961
4962         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4963         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4964
4965         if (first_hash)
4966                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4967
4968 out:
4969         ocfs2_xattr_bucket_free(new_first);
4970         ocfs2_xattr_bucket_free(old_first);
4971         return ret;
4972 }
4973
4974 /*
4975  * Move some xattrs in this cluster to the new cluster.
4976  * This function should only be called when bucket size == cluster size.
4977  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4978  */
4979 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4980                                       handle_t *handle,
4981                                       u64 prev_blk,
4982                                       u64 new_blk,
4983                                       u32 *first_hash)
4984 {
4985         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4986         int ret, credits = 2 * blk_per_bucket;
4987
4988         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4989
4990         ret = ocfs2_extend_trans(handle, credits);
4991         if (ret) {
4992                 mlog_errno(ret);
4993                 return ret;
4994         }
4995
4996         /* Move half of the xattr in start_blk to the next bucket. */
4997         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4998                                           new_blk, first_hash, 1);
4999 }
5000
5001 /*
5002  * Move some xattrs from the old cluster to the new one since they are not
5003  * contiguous in ocfs2 xattr tree.
5004  *
5005  * new_blk starts a new separate cluster, and we will move some xattrs from
5006  * prev_blk to it. v_start will be set as the first name hash value in this
5007  * new cluster so that it can be used as e_cpos during tree insertion and
5008  * don't collide with our original b-tree operations. first_bh and header_bh
5009  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5010  * to extend the insert bucket.
5011  *
5012  * The problem is how much xattr should we move to the new one and when should
5013  * we update first_bh and header_bh?
5014  * 1. If cluster size > bucket size, that means the previous cluster has more
5015  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5016  *    update the first_bh and header_bh if the insert bucket has been moved
5017  *    to the new cluster.
5018  * 2. If cluster_size == bucket_size:
5019  *    a) If the previous extent rec has more than one cluster and the insert
5020  *       place isn't in the last cluster, copy the entire last cluster to the
5021  *       new one. This time, we don't need to upate the first_bh and header_bh
5022  *       since they will not be moved into the new cluster.
5023  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5024  *       the new one. And we set the extend flag to zero if the insert place is
5025  *       moved into the new allocated cluster since no extend is needed.
5026  */
5027 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5028                                             handle_t *handle,
5029                                             struct ocfs2_xattr_bucket *first,
5030                                             struct ocfs2_xattr_bucket *target,
5031                                             u64 new_blk,
5032                                             u32 prev_clusters,
5033                                             u32 *v_start,
5034                                             int *extend)
5035 {
5036         int ret;
5037
5038         trace_ocfs2_adjust_xattr_cross_cluster(
5039                         (unsigned long long)bucket_blkno(first),
5040                         (unsigned long long)new_blk, prev_clusters);
5041
5042         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5043                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5044                                                           handle,
5045                                                           first, target,
5046                                                           new_blk,
5047                                                           prev_clusters,
5048                                                           v_start);
5049                 if (ret)
5050                         mlog_errno(ret);
5051         } else {
5052                 /* The start of the last cluster in the first extent */
5053                 u64 last_blk = bucket_blkno(first) +
5054                         ((prev_clusters - 1) *
5055                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5056
5057                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5058                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5059                                                      bucket_blkno(first),
5060                                                      last_blk, new_blk, 0,
5061                                                      v_start);
5062                         if (ret)
5063                                 mlog_errno(ret);
5064                 } else {
5065                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5066                                                          last_blk, new_blk,
5067                                                          v_start);
5068                         if (ret)
5069                                 mlog_errno(ret);
5070
5071                         if ((bucket_blkno(target) == last_blk) && extend)
5072                                 *extend = 0;
5073                 }
5074         }
5075
5076         return ret;
5077 }
5078
5079 /*
5080  * Add a new cluster for xattr storage.
5081  *
5082  * If the new cluster is contiguous with the previous one, it will be
5083  * appended to the same extent record, and num_clusters will be updated.
5084  * If not, we will insert a new extent for it and move some xattrs in
5085  * the last cluster into the new allocated one.
5086  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5087  * lose the benefits of hashing because we'll have to search large leaves.
5088  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5089  * if it's bigger).
5090  *
5091  * first_bh is the first block of the previous extent rec and header_bh
5092  * indicates the bucket we will insert the new xattrs. They will be updated
5093  * when the header_bh is moved into the new cluster.
5094  */
5095 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5096                                        struct buffer_head *root_bh,
5097                                        struct ocfs2_xattr_bucket *first,
5098                                        struct ocfs2_xattr_bucket *target,
5099                                        u32 *num_clusters,
5100                                        u32 prev_cpos,
5101                                        int *extend,
5102                                        struct ocfs2_xattr_set_ctxt *ctxt)
5103 {
5104         int ret;
5105         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5106         u32 prev_clusters = *num_clusters;
5107         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5108         u64 block;
5109         handle_t *handle = ctxt->handle;
5110         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5111         struct ocfs2_extent_tree et;
5112
5113         trace_ocfs2_add_new_xattr_cluster_begin(
5114                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5115                 (unsigned long long)bucket_blkno(first),
5116                 prev_cpos, prev_clusters);
5117
5118         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5119
5120         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5121                                       OCFS2_JOURNAL_ACCESS_WRITE);
5122         if (ret < 0) {
5123                 mlog_errno(ret);
5124                 goto leave;
5125         }
5126
5127         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5128                                      clusters_to_add, &bit_off, &num_bits);
5129         if (ret < 0) {
5130                 if (ret != -ENOSPC)
5131                         mlog_errno(ret);
5132                 goto leave;
5133         }
5134
5135         BUG_ON(num_bits > clusters_to_add);
5136
5137         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5138         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5139
5140         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5141             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5142              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5143                 /*
5144                  * If this cluster is contiguous with the old one and
5145                  * adding this new cluster, we don't surpass the limit of
5146                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5147                  * initialized and used like other buckets in the previous
5148                  * cluster.
5149                  * So add it as a contiguous one. The caller will handle
5150                  * its init process.
5151                  */
5152                 v_start = prev_cpos + prev_clusters;
5153                 *num_clusters = prev_clusters + num_bits;
5154         } else {
5155                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5156                                                        handle,
5157                                                        first,
5158                                                        target,
5159                                                        block,
5160                                                        prev_clusters,
5161                                                        &v_start,
5162                                                        extend);
5163                 if (ret) {
5164                         mlog_errno(ret);
5165                         goto leave;
5166                 }
5167         }
5168
5169         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5170                                                  v_start, num_bits);
5171         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5172                                   num_bits, 0, ctxt->meta_ac);
5173         if (ret < 0) {
5174                 mlog_errno(ret);
5175                 goto leave;
5176         }
5177
5178         ocfs2_journal_dirty(handle, root_bh);
5179
5180 leave:
5181         return ret;
5182 }
5183
5184 /*
5185  * We are given an extent.  'first' is the bucket at the very front of
5186  * the extent.  The extent has space for an additional bucket past
5187  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5188  * of the target bucket.  We wish to shift every bucket past the target
5189  * down one, filling in that additional space.  When we get back to the
5190  * target, we split the target between itself and the now-empty bucket
5191  * at target+1 (aka, target_blkno + blks_per_bucket).
5192  */
5193 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5194                                      handle_t *handle,
5195                                      struct ocfs2_xattr_bucket *first,
5196                                      u64 target_blk,
5197                                      u32 num_clusters)
5198 {
5199         int ret, credits;
5200         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5201         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5202         u64 end_blk;
5203         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5204
5205         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5206                                         (unsigned long long)bucket_blkno(first),
5207                                         num_clusters, new_bucket);
5208
5209         /* The extent must have room for an additional bucket */
5210         BUG_ON(new_bucket >=
5211                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5212
5213         /* end_blk points to the last existing bucket */
5214         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5215
5216         /*
5217          * end_blk is the start of the last existing bucket.
5218          * Thus, (end_blk - target_blk) covers the target bucket and
5219          * every bucket after it up to, but not including, the last
5220          * existing bucket.  Then we add the last existing bucket, the
5221          * new bucket, and the first bucket (3 * blk_per_bucket).
5222          */
5223         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5224         ret = ocfs2_extend_trans(handle, credits);
5225         if (ret) {
5226                 mlog_errno(ret);
5227                 goto out;
5228         }
5229
5230         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5231                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5232         if (ret) {
5233                 mlog_errno(ret);
5234                 goto out;
5235         }
5236
5237         while (end_blk != target_blk) {
5238                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5239                                             end_blk + blk_per_bucket, 0);
5240                 if (ret)
5241                         goto out;
5242                 end_blk -= blk_per_bucket;
5243         }
5244
5245         /* Move half of the xattr in target_blkno to the next bucket. */
5246         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5247                                         target_blk + blk_per_bucket, NULL, 0);
5248
5249         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5250         ocfs2_xattr_bucket_journal_dirty(handle, first);
5251
5252 out:
5253         return ret;
5254 }
5255
5256 /*
5257  * Add new xattr bucket in an extent record and adjust the buckets
5258  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5259  * bucket we want to insert into.
5260  *
5261  * In the easy case, we will move all the buckets after target down by
5262  * one. Half of target's xattrs will be moved to the next bucket.
5263  *
5264  * If current cluster is full, we'll allocate a new one.  This may not
5265  * be contiguous.  The underlying calls will make sure that there is
5266  * space for the insert, shifting buckets around if necessary.
5267  * 'target' may be moved by those calls.
5268  */
5269 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5270                                       struct buffer_head *xb_bh,
5271                                       struct ocfs2_xattr_bucket *target,
5272                                       struct ocfs2_xattr_set_ctxt *ctxt)
5273 {
5274         struct ocfs2_xattr_block *xb =
5275                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5276         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5277         struct ocfs2_extent_list *el = &xb_root->xt_list;
5278         u32 name_hash =
5279                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5280         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5281         int ret, num_buckets, extend = 1;
5282         u64 p_blkno;
5283         u32 e_cpos, num_clusters;
5284         /* The bucket at the front of the extent */
5285         struct ocfs2_xattr_bucket *first;
5286
5287         trace_ocfs2_add_new_xattr_bucket(
5288                                 (unsigned long long)bucket_blkno(target));
5289
5290         /* The first bucket of the original extent */
5291         first = ocfs2_xattr_bucket_new(inode);
5292         if (!first) {
5293                 ret = -ENOMEM;
5294                 mlog_errno(ret);
5295                 goto out;
5296         }
5297
5298         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5299                                   &num_clusters, el);
5300         if (ret) {
5301                 mlog_errno(ret);
5302                 goto out;
5303         }
5304
5305         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5306         if (ret) {
5307                 mlog_errno(ret);
5308                 goto out;
5309         }
5310
5311         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5312         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5313                 /*
5314                  * This can move first+target if the target bucket moves
5315                  * to the new extent.
5316                  */
5317                 ret = ocfs2_add_new_xattr_cluster(inode,
5318                                                   xb_bh,
5319                                                   first,
5320                                                   target,
5321                                                   &num_clusters,
5322                                                   e_cpos,
5323                                                   &extend,
5324                                                   ctxt);
5325                 if (ret) {
5326                         mlog_errno(ret);
5327                         goto out;
5328                 }
5329         }
5330
5331         if (extend) {
5332                 ret = ocfs2_extend_xattr_bucket(inode,
5333                                                 ctxt->handle,
5334                                                 first,
5335                                                 bucket_blkno(target),
5336                                                 num_clusters);
5337                 if (ret)
5338                         mlog_errno(ret);
5339         }
5340
5341 out:
5342         ocfs2_xattr_bucket_free(first);
5343
5344         return ret;
5345 }
5346
5347 /*
5348  * Truncate the specified xe_off entry in xattr bucket.
5349  * bucket is indicated by header_bh and len is the new length.
5350  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5351  *
5352  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5353  */
5354 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5355                                              struct ocfs2_xattr_bucket *bucket,
5356                                              int xe_off,
5357                                              int len,
5358                                              struct ocfs2_xattr_set_ctxt *ctxt)
5359 {
5360         int ret, offset;
5361         u64 value_blk;
5362         struct ocfs2_xattr_entry *xe;
5363         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5364         size_t blocksize = inode->i_sb->s_blocksize;
5365         struct ocfs2_xattr_value_buf vb = {
5366                 .vb_access = ocfs2_journal_access,
5367         };
5368
5369         xe = &xh->xh_entries[xe_off];
5370
5371         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5372
5373         offset = le16_to_cpu(xe->xe_name_offset) +
5374                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5375
5376         value_blk = offset / blocksize;
5377
5378         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5379         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5380
5381         vb.vb_bh = bucket->bu_bhs[value_blk];
5382         BUG_ON(!vb.vb_bh);
5383
5384         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5385                 (vb.vb_bh->b_data + offset % blocksize);
5386
5387         /*
5388          * From here on out we have to dirty the bucket.  The generic
5389          * value calls only modify one of the bucket's bhs, but we need
5390          * to send the bucket at once.  So if they error, they *could* have
5391          * modified something.  We have to assume they did, and dirty
5392          * the whole bucket.  This leaves us in a consistent state.
5393          */
5394         trace_ocfs2_xattr_bucket_value_truncate(
5395                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5396         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5397         if (ret) {
5398                 mlog_errno(ret);
5399                 goto out;
5400         }
5401
5402         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5403                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5404         if (ret) {
5405                 mlog_errno(ret);
5406                 goto out;
5407         }
5408
5409         xe->xe_value_size = cpu_to_le64(len);
5410
5411         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5412
5413 out:
5414         return ret;
5415 }
5416
5417 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5418                                   struct buffer_head *root_bh,
5419                                   u64 blkno,
5420                                   u32 cpos,
5421                                   u32 len,
5422                                   void *para)
5423 {
5424         int ret;
5425         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5426         struct inode *tl_inode = osb->osb_tl_inode;
5427         handle_t *handle;
5428         struct ocfs2_xattr_block *xb =
5429                         (struct ocfs2_xattr_block *)root_bh->b_data;
5430         struct ocfs2_alloc_context *meta_ac = NULL;
5431         struct ocfs2_cached_dealloc_ctxt dealloc;
5432         struct ocfs2_extent_tree et;
5433
5434         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5435                                           ocfs2_delete_xattr_in_bucket, para);
5436         if (ret) {
5437                 mlog_errno(ret);
5438                 return ret;
5439         }
5440
5441         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5442
5443         ocfs2_init_dealloc_ctxt(&dealloc);
5444
5445         trace_ocfs2_rm_xattr_cluster(
5446                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5447                         (unsigned long long)blkno, cpos, len);
5448
5449         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5450                                                len);
5451
5452         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5453         if (ret) {
5454                 mlog_errno(ret);
5455                 return ret;
5456         }
5457
5458         inode_lock(tl_inode);
5459
5460         if (ocfs2_truncate_log_needs_flush(osb)) {
5461                 ret = __ocfs2_flush_truncate_log(osb);
5462                 if (ret < 0) {
5463                         mlog_errno(ret);
5464                         goto out;
5465                 }
5466         }
5467
5468         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5469         if (IS_ERR(handle)) {
5470                 ret = -ENOMEM;
5471                 mlog_errno(ret);
5472                 goto out;
5473         }
5474
5475         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5476                                       OCFS2_JOURNAL_ACCESS_WRITE);
5477         if (ret) {
5478                 mlog_errno(ret);
5479                 goto out_commit;
5480         }
5481
5482         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5483                                   &dealloc);
5484         if (ret) {
5485                 mlog_errno(ret);
5486                 goto out_commit;
5487         }
5488
5489         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5490         ocfs2_journal_dirty(handle, root_bh);
5491
5492         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5493         if (ret)
5494                 mlog_errno(ret);
5495         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5496
5497 out_commit:
5498         ocfs2_commit_trans(osb, handle);
5499 out:
5500         ocfs2_schedule_truncate_log_flush(osb, 1);
5501
5502         inode_unlock(tl_inode);
5503
5504         if (meta_ac)
5505                 ocfs2_free_alloc_context(meta_ac);
5506
5507         ocfs2_run_deallocs(osb, &dealloc);
5508
5509         return ret;
5510 }
5511
5512 /*
5513  * check whether the xattr bucket is filled up with the same hash value.
5514  * If we want to insert the xattr with the same hash, return -ENOSPC.
5515  * If we want to insert a xattr with different hash value, go ahead
5516  * and ocfs2_divide_xattr_bucket will handle this.
5517  */
5518 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5519                                               struct ocfs2_xattr_bucket *bucket,
5520                                               const char *name)
5521 {
5522         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5523         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5524
5525         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5526                 return 0;
5527
5528         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5529             xh->xh_entries[0].xe_name_hash) {
5530                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5531                      "hash = %u\n",
5532                      (unsigned long long)bucket_blkno(bucket),
5533                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5534                 return -ENOSPC;
5535         }
5536
5537         return 0;
5538 }
5539
5540 /*
5541  * Try to set the entry in the current bucket.  If we fail, the caller
5542  * will handle getting us another bucket.
5543  */
5544 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5545                                         struct ocfs2_xattr_info *xi,
5546                                         struct ocfs2_xattr_search *xs,
5547                                         struct ocfs2_xattr_set_ctxt *ctxt)
5548 {
5549         int ret;
5550         struct ocfs2_xa_loc loc;
5551
5552         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5553
5554         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5555                                        xs->not_found ? NULL : xs->here);
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                 goto out;
5564         }
5565
5566         /* Ok, we need space.  Let's try defragmenting the bucket. */
5567         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5568                                         xs->bucket);
5569         if (ret) {
5570                 mlog_errno(ret);
5571                 goto out;
5572         }
5573
5574         ret = ocfs2_xa_set(&loc, xi, ctxt);
5575         if (!ret) {
5576                 xs->here = loc.xl_entry;
5577                 goto out;
5578         }
5579         if (ret != -ENOSPC)
5580                 mlog_errno(ret);
5581
5582
5583 out:
5584         return ret;
5585 }
5586
5587 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5588                                              struct ocfs2_xattr_info *xi,
5589                                              struct ocfs2_xattr_search *xs,
5590                                              struct ocfs2_xattr_set_ctxt *ctxt)
5591 {
5592         int ret;
5593
5594         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5595
5596         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5597         if (!ret)
5598                 goto out;
5599         if (ret != -ENOSPC) {
5600                 mlog_errno(ret);
5601                 goto out;
5602         }
5603
5604         /* Ack, need more space.  Let's try to get another bucket! */
5605
5606         /*
5607          * We do not allow for overlapping ranges between buckets. And
5608          * the maximum number of collisions we will allow for then is
5609          * one bucket's worth, so check it here whether we need to
5610          * add a new bucket for the insert.
5611          */
5612         ret = ocfs2_check_xattr_bucket_collision(inode,
5613                                                  xs->bucket,
5614                                                  xi->xi_name);
5615         if (ret) {
5616                 mlog_errno(ret);
5617                 goto out;
5618         }
5619
5620         ret = ocfs2_add_new_xattr_bucket(inode,
5621                                          xs->xattr_bh,
5622                                          xs->bucket,
5623                                          ctxt);
5624         if (ret) {
5625                 mlog_errno(ret);
5626                 goto out;
5627         }
5628
5629         /*
5630          * ocfs2_add_new_xattr_bucket() will have updated
5631          * xs->bucket if it moved, but it will not have updated
5632          * any of the other search fields.  Thus, we drop it and
5633          * re-search.  Everything should be cached, so it'll be
5634          * quick.
5635          */
5636         ocfs2_xattr_bucket_relse(xs->bucket);
5637         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5638                                            xi->xi_name_index,
5639                                            xi->xi_name, xs);
5640         if (ret && ret != -ENODATA)
5641                 goto out;
5642         xs->not_found = ret;
5643
5644         /* Ok, we have a new bucket, let's try again */
5645         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5646         if (ret && (ret != -ENOSPC))
5647                 mlog_errno(ret);
5648
5649 out:
5650         return ret;
5651 }
5652
5653 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5654                                         struct ocfs2_xattr_bucket *bucket,
5655                                         void *para)
5656 {
5657         int ret = 0, ref_credits;
5658         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5659         u16 i;
5660         struct ocfs2_xattr_entry *xe;
5661         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5662         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5663         int credits = ocfs2_remove_extent_credits(osb->sb) +
5664                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5665         struct ocfs2_xattr_value_root *xv;
5666         struct ocfs2_rm_xattr_bucket_para *args =
5667                         (struct ocfs2_rm_xattr_bucket_para *)para;
5668
5669         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5670
5671         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5672                 xe = &xh->xh_entries[i];
5673                 if (ocfs2_xattr_is_local(xe))
5674                         continue;
5675
5676                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5677                                                       i, &xv, NULL);
5678                 if (ret) {
5679                         mlog_errno(ret);
5680                         break;
5681                 }
5682
5683                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5684                                                          args->ref_ci,
5685                                                          args->ref_root_bh,
5686                                                          &ctxt.meta_ac,
5687                                                          &ref_credits);
5688
5689                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5690                 if (IS_ERR(ctxt.handle)) {
5691                         ret = PTR_ERR(ctxt.handle);
5692                         mlog_errno(ret);
5693                         break;
5694                 }
5695
5696                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5697                                                         i, 0, &ctxt);
5698
5699                 ocfs2_commit_trans(osb, ctxt.handle);
5700                 if (ctxt.meta_ac) {
5701                         ocfs2_free_alloc_context(ctxt.meta_ac);
5702                         ctxt.meta_ac = NULL;
5703                 }
5704                 if (ret) {
5705                         mlog_errno(ret);
5706                         break;
5707                 }
5708         }
5709
5710         if (ctxt.meta_ac)
5711                 ocfs2_free_alloc_context(ctxt.meta_ac);
5712         ocfs2_schedule_truncate_log_flush(osb, 1);
5713         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5714         return ret;
5715 }
5716
5717 /*
5718  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5719  * or change the extent record flag), we need to recalculate
5720  * the metaecc for the whole bucket. So it is done here.
5721  *
5722  * Note:
5723  * We have to give the extra credits for the caller.
5724  */
5725 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5726                                             handle_t *handle,
5727                                             void *para)
5728 {
5729         int ret;
5730         struct ocfs2_xattr_bucket *bucket =
5731                         (struct ocfs2_xattr_bucket *)para;
5732
5733         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5734                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5735         if (ret) {
5736                 mlog_errno(ret);
5737                 return ret;
5738         }
5739
5740         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5741
5742         return 0;
5743 }
5744
5745 /*
5746  * Special action we need if the xattr value is refcounted.
5747  *
5748  * 1. If the xattr is refcounted, lock the tree.
5749  * 2. CoW the xattr if we are setting the new value and the value
5750  *    will be stored outside.
5751  * 3. In other case, decrease_refcount will work for us, so just
5752  *    lock the refcount tree, calculate the meta and credits is OK.
5753  *
5754  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5755  * currently CoW is a completed transaction, while this function
5756  * will also lock the allocators and let us deadlock. So we will
5757  * CoW the whole xattr value.
5758  */
5759 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5760                                         struct ocfs2_dinode *di,
5761                                         struct ocfs2_xattr_info *xi,
5762                                         struct ocfs2_xattr_search *xis,
5763                                         struct ocfs2_xattr_search *xbs,
5764                                         struct ocfs2_refcount_tree **ref_tree,
5765                                         int *meta_add,
5766                                         int *credits)
5767 {
5768         int ret = 0;
5769         struct ocfs2_xattr_block *xb;
5770         struct ocfs2_xattr_entry *xe;
5771         char *base;
5772         u32 p_cluster, num_clusters;
5773         unsigned int ext_flags;
5774         int name_offset, name_len;
5775         struct ocfs2_xattr_value_buf vb;
5776         struct ocfs2_xattr_bucket *bucket = NULL;
5777         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5778         struct ocfs2_post_refcount refcount;
5779         struct ocfs2_post_refcount *p = NULL;
5780         struct buffer_head *ref_root_bh = NULL;
5781
5782         if (!xis->not_found) {
5783                 xe = xis->here;
5784                 name_offset = le16_to_cpu(xe->xe_name_offset);
5785                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5786                 base = xis->base;
5787                 vb.vb_bh = xis->inode_bh;
5788                 vb.vb_access = ocfs2_journal_access_di;
5789         } else {
5790                 int i, block_off = 0;
5791                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5792                 xe = xbs->here;
5793                 name_offset = le16_to_cpu(xe->xe_name_offset);
5794                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5795                 i = xbs->here - xbs->header->xh_entries;
5796
5797                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5798                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5799                                                         bucket_xh(xbs->bucket),
5800                                                         i, &block_off,
5801                                                         &name_offset);
5802                         if (ret) {
5803                                 mlog_errno(ret);
5804                                 goto out;
5805                         }
5806                         base = bucket_block(xbs->bucket, block_off);
5807                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5808                         vb.vb_access = ocfs2_journal_access;
5809
5810                         if (ocfs2_meta_ecc(osb)) {
5811                                 /*create parameters for ocfs2_post_refcount. */
5812                                 bucket = xbs->bucket;
5813                                 refcount.credits = bucket->bu_blocks;
5814                                 refcount.para = bucket;
5815                                 refcount.func =
5816                                         ocfs2_xattr_bucket_post_refcount;
5817                                 p = &refcount;
5818                         }
5819                 } else {
5820                         base = xbs->base;
5821                         vb.vb_bh = xbs->xattr_bh;
5822                         vb.vb_access = ocfs2_journal_access_xb;
5823                 }
5824         }
5825
5826         if (ocfs2_xattr_is_local(xe))
5827                 goto out;
5828
5829         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5830                                 (base + name_offset + name_len);
5831
5832         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5833                                        &num_clusters, &vb.vb_xv->xr_list,
5834                                        &ext_flags);
5835         if (ret) {
5836                 mlog_errno(ret);
5837                 goto out;
5838         }
5839
5840         /*
5841          * We just need to check the 1st extent record, since we always
5842          * CoW the whole xattr. So there shouldn't be a xattr with
5843          * some REFCOUNT extent recs after the 1st one.
5844          */
5845         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5846                 goto out;
5847
5848         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5849                                        1, ref_tree, &ref_root_bh);
5850         if (ret) {
5851                 mlog_errno(ret);
5852                 goto out;
5853         }
5854
5855         /*
5856          * If we are deleting the xattr or the new size will be stored inside,
5857          * cool, leave it there, the xattr truncate process will remove them
5858          * for us(it still needs the refcount tree lock and the meta, credits).
5859          * And the worse case is that every cluster truncate will split the
5860          * refcount tree, and make the original extent become 3. So we will need
5861          * 2 * cluster more extent recs at most.
5862          */
5863         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5864
5865                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5866                                                          &(*ref_tree)->rf_ci,
5867                                                          ref_root_bh, vb.vb_xv,
5868                                                          meta_add, credits);
5869                 if (ret)
5870                         mlog_errno(ret);
5871                 goto out;
5872         }
5873
5874         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5875                                        *ref_tree, ref_root_bh, 0,
5876                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5877         if (ret)
5878                 mlog_errno(ret);
5879
5880 out:
5881         brelse(ref_root_bh);
5882         return ret;
5883 }
5884
5885 /*
5886  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5887  * The physical clusters will be added to refcount tree.
5888  */
5889 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5890                                 struct ocfs2_xattr_value_root *xv,
5891                                 struct ocfs2_extent_tree *value_et,
5892                                 struct ocfs2_caching_info *ref_ci,
5893                                 struct buffer_head *ref_root_bh,
5894                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5895                                 struct ocfs2_post_refcount *refcount)
5896 {
5897         int ret = 0;
5898         u32 clusters = le32_to_cpu(xv->xr_clusters);
5899         u32 cpos, p_cluster, num_clusters;
5900         struct ocfs2_extent_list *el = &xv->xr_list;
5901         unsigned int ext_flags;
5902
5903         cpos = 0;
5904         while (cpos < clusters) {
5905                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5906                                                &num_clusters, el, &ext_flags);
5907                 if (ret) {
5908                         mlog_errno(ret);
5909                         break;
5910                 }
5911
5912                 cpos += num_clusters;
5913                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5914                         continue;
5915
5916                 BUG_ON(!p_cluster);
5917
5918                 ret = ocfs2_add_refcount_flag(inode, value_et,
5919                                               ref_ci, ref_root_bh,
5920                                               cpos - num_clusters,
5921                                               p_cluster, num_clusters,
5922                                               dealloc, refcount);
5923                 if (ret) {
5924                         mlog_errno(ret);
5925                         break;
5926                 }
5927         }
5928
5929         return ret;
5930 }
5931
5932 /*
5933  * Given a normal ocfs2_xattr_header, refcount all the entries which
5934  * have value stored outside.
5935  * Used for xattrs stored in inode and ocfs2_xattr_block.
5936  */
5937 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5938                                 struct ocfs2_xattr_value_buf *vb,
5939                                 struct ocfs2_xattr_header *header,
5940                                 struct ocfs2_caching_info *ref_ci,
5941                                 struct buffer_head *ref_root_bh,
5942                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5943 {
5944
5945         struct ocfs2_xattr_entry *xe;
5946         struct ocfs2_xattr_value_root *xv;
5947         struct ocfs2_extent_tree et;
5948         int i, ret = 0;
5949
5950         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5951                 xe = &header->xh_entries[i];
5952
5953                 if (ocfs2_xattr_is_local(xe))
5954                         continue;
5955
5956                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5957                         le16_to_cpu(xe->xe_name_offset) +
5958                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5959
5960                 vb->vb_xv = xv;
5961                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5962
5963                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5964                                                         ref_ci, ref_root_bh,
5965                                                         dealloc, NULL);
5966                 if (ret) {
5967                         mlog_errno(ret);
5968                         break;
5969                 }
5970         }
5971
5972         return ret;
5973 }
5974
5975 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5976                                 struct buffer_head *fe_bh,
5977                                 struct ocfs2_caching_info *ref_ci,
5978                                 struct buffer_head *ref_root_bh,
5979                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5980 {
5981         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5982         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5983                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5984                                 le16_to_cpu(di->i_xattr_inline_size));
5985         struct ocfs2_xattr_value_buf vb = {
5986                 .vb_bh = fe_bh,
5987                 .vb_access = ocfs2_journal_access_di,
5988         };
5989
5990         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5991                                                   ref_ci, ref_root_bh, dealloc);
5992 }
5993
5994 struct ocfs2_xattr_tree_value_refcount_para {
5995         struct ocfs2_caching_info *ref_ci;
5996         struct buffer_head *ref_root_bh;
5997         struct ocfs2_cached_dealloc_ctxt *dealloc;
5998 };
5999
6000 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6001                                            struct ocfs2_xattr_bucket *bucket,
6002                                            int offset,
6003                                            struct ocfs2_xattr_value_root **xv,
6004                                            struct buffer_head **bh)
6005 {
6006         int ret, block_off, name_offset;
6007         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6008         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6009         void *base;
6010
6011         ret = ocfs2_xattr_bucket_get_name_value(sb,
6012                                                 bucket_xh(bucket),
6013                                                 offset,
6014                                                 &block_off,
6015                                                 &name_offset);
6016         if (ret) {
6017                 mlog_errno(ret);
6018                 goto out;
6019         }
6020
6021         base = bucket_block(bucket, block_off);
6022
6023         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6024                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6025
6026         if (bh)
6027                 *bh = bucket->bu_bhs[block_off];
6028 out:
6029         return ret;
6030 }
6031
6032 /*
6033  * For a given xattr bucket, refcount all the entries which
6034  * have value stored outside.
6035  */
6036 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6037                                              struct ocfs2_xattr_bucket *bucket,
6038                                              void *para)
6039 {
6040         int i, ret = 0;
6041         struct ocfs2_extent_tree et;
6042         struct ocfs2_xattr_tree_value_refcount_para *ref =
6043                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6044         struct ocfs2_xattr_header *xh =
6045                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6046         struct ocfs2_xattr_entry *xe;
6047         struct ocfs2_xattr_value_buf vb = {
6048                 .vb_access = ocfs2_journal_access,
6049         };
6050         struct ocfs2_post_refcount refcount = {
6051                 .credits = bucket->bu_blocks,
6052                 .para = bucket,
6053                 .func = ocfs2_xattr_bucket_post_refcount,
6054         };
6055         struct ocfs2_post_refcount *p = NULL;
6056
6057         /* We only need post_refcount if we support metaecc. */
6058         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6059                 p = &refcount;
6060
6061         trace_ocfs2_xattr_bucket_value_refcount(
6062                                 (unsigned long long)bucket_blkno(bucket),
6063                                 le16_to_cpu(xh->xh_count));
6064         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6065                 xe = &xh->xh_entries[i];
6066
6067                 if (ocfs2_xattr_is_local(xe))
6068                         continue;
6069
6070                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6071                                                       &vb.vb_xv, &vb.vb_bh);
6072                 if (ret) {
6073                         mlog_errno(ret);
6074                         break;
6075                 }
6076
6077                 ocfs2_init_xattr_value_extent_tree(&et,
6078                                                    INODE_CACHE(inode), &vb);
6079
6080                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6081                                                         &et, ref->ref_ci,
6082                                                         ref->ref_root_bh,
6083                                                         ref->dealloc, p);
6084                 if (ret) {
6085                         mlog_errno(ret);
6086                         break;
6087                 }
6088         }
6089
6090         return ret;
6091
6092 }
6093
6094 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6095                                      struct buffer_head *root_bh,
6096                                      u64 blkno, u32 cpos, u32 len, void *para)
6097 {
6098         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6099                                            ocfs2_xattr_bucket_value_refcount,
6100                                            para);
6101 }
6102
6103 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6104                                 struct buffer_head *blk_bh,
6105                                 struct ocfs2_caching_info *ref_ci,
6106                                 struct buffer_head *ref_root_bh,
6107                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6108 {
6109         int ret = 0;
6110         struct ocfs2_xattr_block *xb =
6111                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6112
6113         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6114                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6115                 struct ocfs2_xattr_value_buf vb = {
6116                         .vb_bh = blk_bh,
6117                         .vb_access = ocfs2_journal_access_xb,
6118                 };
6119
6120                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6121                                                          ref_ci, ref_root_bh,
6122                                                          dealloc);
6123         } else {
6124                 struct ocfs2_xattr_tree_value_refcount_para para = {
6125                         .ref_ci = ref_ci,
6126                         .ref_root_bh = ref_root_bh,
6127                         .dealloc = dealloc,
6128                 };
6129
6130                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6131                                                 ocfs2_refcount_xattr_tree_rec,
6132                                                 &para);
6133         }
6134
6135         return ret;
6136 }
6137
6138 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6139                                      struct buffer_head *fe_bh,
6140                                      struct ocfs2_caching_info *ref_ci,
6141                                      struct buffer_head *ref_root_bh,
6142                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6143 {
6144         int ret = 0;
6145         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6146         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6147         struct buffer_head *blk_bh = NULL;
6148
6149         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6150                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6151                                                          ref_ci, ref_root_bh,
6152                                                          dealloc);
6153                 if (ret) {
6154                         mlog_errno(ret);
6155                         goto out;
6156                 }
6157         }
6158
6159         if (!di->i_xattr_loc)
6160                 goto out;
6161
6162         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6163                                      &blk_bh);
6164         if (ret < 0) {
6165                 mlog_errno(ret);
6166                 goto out;
6167         }
6168
6169         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6170                                                 ref_root_bh, dealloc);
6171         if (ret)
6172                 mlog_errno(ret);
6173
6174         brelse(blk_bh);
6175 out:
6176
6177         return ret;
6178 }
6179
6180 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6181 /*
6182  * Store the information we need in xattr reflink.
6183  * old_bh and new_bh are inode bh for the old and new inode.
6184  */
6185 struct ocfs2_xattr_reflink {
6186         struct inode *old_inode;
6187         struct inode *new_inode;
6188         struct buffer_head *old_bh;
6189         struct buffer_head *new_bh;
6190         struct ocfs2_caching_info *ref_ci;
6191         struct buffer_head *ref_root_bh;
6192         struct ocfs2_cached_dealloc_ctxt *dealloc;
6193         should_xattr_reflinked *xattr_reflinked;
6194 };
6195
6196 /*
6197  * Given a xattr header and xe offset,
6198  * return the proper xv and the corresponding bh.
6199  * xattr in inode, block and xattr tree have different implementaions.
6200  */
6201 typedef int (get_xattr_value_root)(struct super_block *sb,
6202                                    struct buffer_head *bh,
6203                                    struct ocfs2_xattr_header *xh,
6204                                    int offset,
6205                                    struct ocfs2_xattr_value_root **xv,
6206                                    struct buffer_head **ret_bh,
6207                                    void *para);
6208
6209 /*
6210  * Calculate all the xattr value root metadata stored in this xattr header and
6211  * credits we need if we create them from the scratch.
6212  * We use get_xattr_value_root so that all types of xattr container can use it.
6213  */
6214 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6215                                              struct buffer_head *bh,
6216                                              struct ocfs2_xattr_header *xh,
6217                                              int *metas, int *credits,
6218                                              int *num_recs,
6219                                              get_xattr_value_root *func,
6220                                              void *para)
6221 {
6222         int i, ret = 0;
6223         struct ocfs2_xattr_value_root *xv;
6224         struct ocfs2_xattr_entry *xe;
6225
6226         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6227                 xe = &xh->xh_entries[i];
6228                 if (ocfs2_xattr_is_local(xe))
6229                         continue;
6230
6231                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6232                 if (ret) {
6233                         mlog_errno(ret);
6234                         break;
6235                 }
6236
6237                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6238                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6239
6240                 *credits += ocfs2_calc_extend_credits(sb,
6241                                                 &def_xv.xv.xr_list);
6242
6243                 /*
6244                  * If the value is a tree with depth > 1, We don't go deep
6245                  * to the extent block, so just calculate a maximum record num.
6246                  */
6247                 if (!xv->xr_list.l_tree_depth)
6248                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6249                 else
6250                         *num_recs += ocfs2_clusters_for_bytes(sb,
6251                                                               XATTR_SIZE_MAX);
6252         }
6253
6254         return ret;
6255 }
6256
6257 /* Used by xattr inode and block to return the right xv and buffer_head. */
6258 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6259                                       struct buffer_head *bh,
6260                                       struct ocfs2_xattr_header *xh,
6261                                       int offset,
6262                                       struct ocfs2_xattr_value_root **xv,
6263                                       struct buffer_head **ret_bh,
6264                                       void *para)
6265 {
6266         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6267
6268         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6269                 le16_to_cpu(xe->xe_name_offset) +
6270                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6271
6272         if (ret_bh)
6273                 *ret_bh = bh;
6274
6275         return 0;
6276 }
6277
6278 /*
6279  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6280  * It is only used for inline xattr and xattr block.
6281  */
6282 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6283                                         struct ocfs2_xattr_header *xh,
6284                                         struct buffer_head *ref_root_bh,
6285                                         int *credits,
6286                                         struct ocfs2_alloc_context **meta_ac)
6287 {
6288         int ret, meta_add = 0, num_recs = 0;
6289         struct ocfs2_refcount_block *rb =
6290                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6291
6292         *credits = 0;
6293
6294         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6295                                                 &meta_add, credits, &num_recs,
6296                                                 ocfs2_get_xattr_value_root,
6297                                                 NULL);
6298         if (ret) {
6299                 mlog_errno(ret);
6300                 goto out;
6301         }
6302
6303         /*
6304          * We need to add/modify num_recs in refcount tree, so just calculate
6305          * an approximate number we need for refcount tree change.
6306          * Sometimes we need to split the tree, and after split,  half recs
6307          * will be moved to the new block, and a new block can only provide
6308          * half number of recs. So we multiple new blocks by 2.
6309          */
6310         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6311         meta_add += num_recs;
6312         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6313         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6314                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6315                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6316         else
6317                 *credits += 1;
6318
6319         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6320         if (ret)
6321                 mlog_errno(ret);
6322
6323 out:
6324         return ret;
6325 }
6326
6327 /*
6328  * Given a xattr header, reflink all the xattrs in this container.
6329  * It can be used for inode, block and bucket.
6330  *
6331  * NOTE:
6332  * Before we call this function, the caller has memcpy the xattr in
6333  * old_xh to the new_xh.
6334  *
6335  * If args.xattr_reflinked is set, call it to decide whether the xe should
6336  * be reflinked or not. If not, remove it from the new xattr header.
6337  */
6338 static int ocfs2_reflink_xattr_header(handle_t *handle,
6339                                       struct ocfs2_xattr_reflink *args,
6340                                       struct buffer_head *old_bh,
6341                                       struct ocfs2_xattr_header *xh,
6342                                       struct buffer_head *new_bh,
6343                                       struct ocfs2_xattr_header *new_xh,
6344                                       struct ocfs2_xattr_value_buf *vb,
6345                                       struct ocfs2_alloc_context *meta_ac,
6346                                       get_xattr_value_root *func,
6347                                       void *para)
6348 {
6349         int ret = 0, i, j;
6350         struct super_block *sb = args->old_inode->i_sb;
6351         struct buffer_head *value_bh;
6352         struct ocfs2_xattr_entry *xe, *last;
6353         struct ocfs2_xattr_value_root *xv, *new_xv;
6354         struct ocfs2_extent_tree data_et;
6355         u32 clusters, cpos, p_cluster, num_clusters;
6356         unsigned int ext_flags = 0;
6357
6358         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6359                                          le16_to_cpu(xh->xh_count));
6360
6361         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6362         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6363                 xe = &xh->xh_entries[i];
6364
6365                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6366                         xe = &new_xh->xh_entries[j];
6367
6368                         le16_add_cpu(&new_xh->xh_count, -1);
6369                         if (new_xh->xh_count) {
6370                                 memmove(xe, xe + 1,
6371                                         (void *)last - (void *)xe);
6372                                 memset(last, 0,
6373                                        sizeof(struct ocfs2_xattr_entry));
6374                         }
6375
6376                         /*
6377                          * We don't want j to increase in the next round since
6378                          * it is already moved ahead.
6379                          */
6380                         j--;
6381                         continue;
6382                 }
6383
6384                 if (ocfs2_xattr_is_local(xe))
6385                         continue;
6386
6387                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6388                 if (ret) {
6389                         mlog_errno(ret);
6390                         break;
6391                 }
6392
6393                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6394                 if (ret) {
6395                         mlog_errno(ret);
6396                         break;
6397                 }
6398
6399                 /*
6400                  * For the xattr which has l_tree_depth = 0, all the extent
6401                  * recs have already be copied to the new xh with the
6402                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6403                  * increase the refount count int the refcount tree.
6404                  *
6405                  * For the xattr which has l_tree_depth > 0, we need
6406                  * to initialize it to the empty default value root,
6407                  * and then insert the extents one by one.
6408                  */
6409                 if (xv->xr_list.l_tree_depth) {
6410                         memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6411                         vb->vb_xv = new_xv;
6412                         vb->vb_bh = value_bh;
6413                         ocfs2_init_xattr_value_extent_tree(&data_et,
6414                                         INODE_CACHE(args->new_inode), vb);
6415                 }
6416
6417                 clusters = le32_to_cpu(xv->xr_clusters);
6418                 cpos = 0;
6419                 while (cpos < clusters) {
6420                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6421                                                        cpos,
6422                                                        &p_cluster,
6423                                                        &num_clusters,
6424                                                        &xv->xr_list,
6425                                                        &ext_flags);
6426                         if (ret) {
6427                                 mlog_errno(ret);
6428                                 goto out;
6429                         }
6430
6431                         BUG_ON(!p_cluster);
6432
6433                         if (xv->xr_list.l_tree_depth) {
6434                                 ret = ocfs2_insert_extent(handle,
6435                                                 &data_et, cpos,
6436                                                 ocfs2_clusters_to_blocks(
6437                                                         args->old_inode->i_sb,
6438                                                         p_cluster),
6439                                                 num_clusters, ext_flags,
6440                                                 meta_ac);
6441                                 if (ret) {
6442                                         mlog_errno(ret);
6443                                         goto out;
6444                                 }
6445                         }
6446
6447                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6448                                                       args->ref_root_bh,
6449                                                       p_cluster, num_clusters,
6450                                                       meta_ac, args->dealloc);
6451                         if (ret) {
6452                                 mlog_errno(ret);
6453                                 goto out;
6454                         }
6455
6456                         cpos += num_clusters;
6457                 }
6458         }
6459
6460 out:
6461         return ret;
6462 }
6463
6464 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6465 {
6466         int ret = 0, credits = 0;
6467         handle_t *handle;
6468         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6469         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6470         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6471         int header_off = osb->sb->s_blocksize - inline_size;
6472         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6473                                         (args->old_bh->b_data + header_off);
6474         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6475                                         (args->new_bh->b_data + header_off);
6476         struct ocfs2_alloc_context *meta_ac = NULL;
6477         struct ocfs2_inode_info *new_oi;
6478         struct ocfs2_dinode *new_di;
6479         struct ocfs2_xattr_value_buf vb = {
6480                 .vb_bh = args->new_bh,
6481                 .vb_access = ocfs2_journal_access_di,
6482         };
6483
6484         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6485                                                   &credits, &meta_ac);
6486         if (ret) {
6487                 mlog_errno(ret);
6488                 goto out;
6489         }
6490
6491         handle = ocfs2_start_trans(osb, credits);
6492         if (IS_ERR(handle)) {
6493                 ret = PTR_ERR(handle);
6494                 mlog_errno(ret);
6495                 goto out;
6496         }
6497
6498         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6499                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6500         if (ret) {
6501                 mlog_errno(ret);
6502                 goto out_commit;
6503         }
6504
6505         memcpy(args->new_bh->b_data + header_off,
6506                args->old_bh->b_data + header_off, inline_size);
6507
6508         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6509         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6510
6511         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6512                                          args->new_bh, new_xh, &vb, meta_ac,
6513                                          ocfs2_get_xattr_value_root, NULL);
6514         if (ret) {
6515                 mlog_errno(ret);
6516                 goto out_commit;
6517         }
6518
6519         new_oi = OCFS2_I(args->new_inode);
6520         /*
6521          * Adjust extent record count to reserve space for extended attribute.
6522          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6523          */
6524         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6525             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6526                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6527                 le16_add_cpu(&el->l_count, -(inline_size /
6528                                         sizeof(struct ocfs2_extent_rec)));
6529         }
6530         spin_lock(&new_oi->ip_lock);
6531         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6532         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6533         spin_unlock(&new_oi->ip_lock);
6534
6535         ocfs2_journal_dirty(handle, args->new_bh);
6536
6537 out_commit:
6538         ocfs2_commit_trans(osb, handle);
6539
6540 out:
6541         if (meta_ac)
6542                 ocfs2_free_alloc_context(meta_ac);
6543         return ret;
6544 }
6545
6546 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6547                                           struct buffer_head *fe_bh,
6548                                           struct buffer_head **ret_bh,
6549                                           int indexed)
6550 {
6551         int ret;
6552         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6553         struct ocfs2_xattr_set_ctxt ctxt;
6554
6555         memset(&ctxt, 0, sizeof(ctxt));
6556         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6557         if (ret < 0) {
6558                 mlog_errno(ret);
6559                 return ret;
6560         }
6561
6562         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6563         if (IS_ERR(ctxt.handle)) {
6564                 ret = PTR_ERR(ctxt.handle);
6565                 mlog_errno(ret);
6566                 goto out;
6567         }
6568
6569         trace_ocfs2_create_empty_xattr_block(
6570                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6571         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6572                                        ret_bh);
6573         if (ret)
6574                 mlog_errno(ret);
6575
6576         ocfs2_commit_trans(osb, ctxt.handle);
6577 out:
6578         ocfs2_free_alloc_context(ctxt.meta_ac);
6579         return ret;
6580 }
6581
6582 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6583                                      struct buffer_head *blk_bh,
6584                                      struct buffer_head *new_blk_bh)
6585 {
6586         int ret = 0, credits = 0;
6587         handle_t *handle;
6588         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6589         struct ocfs2_dinode *new_di;
6590         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6591         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6592         struct ocfs2_xattr_block *xb =
6593                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6594         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6595         struct ocfs2_xattr_block *new_xb =
6596                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6597         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6598         struct ocfs2_alloc_context *meta_ac;
6599         struct ocfs2_xattr_value_buf vb = {
6600                 .vb_bh = new_blk_bh,
6601                 .vb_access = ocfs2_journal_access_xb,
6602         };
6603
6604         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6605                                                   &credits, &meta_ac);
6606         if (ret) {
6607                 mlog_errno(ret);
6608                 return ret;
6609         }
6610
6611         /* One more credits in case we need to add xattr flags in new inode. */
6612         handle = ocfs2_start_trans(osb, credits + 1);
6613         if (IS_ERR(handle)) {
6614                 ret = PTR_ERR(handle);
6615                 mlog_errno(ret);
6616                 goto out;
6617         }
6618
6619         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6620                 ret = ocfs2_journal_access_di(handle,
6621                                               INODE_CACHE(args->new_inode),
6622                                               args->new_bh,
6623                                               OCFS2_JOURNAL_ACCESS_WRITE);
6624                 if (ret) {
6625                         mlog_errno(ret);
6626                         goto out_commit;
6627                 }
6628         }
6629
6630         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6631                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6632         if (ret) {
6633                 mlog_errno(ret);
6634                 goto out_commit;
6635         }
6636
6637         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6638                osb->sb->s_blocksize - header_off);
6639
6640         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6641                                          new_blk_bh, new_xh, &vb, meta_ac,
6642                                          ocfs2_get_xattr_value_root, NULL);
6643         if (ret) {
6644                 mlog_errno(ret);
6645                 goto out_commit;
6646         }
6647
6648         ocfs2_journal_dirty(handle, new_blk_bh);
6649
6650         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6651                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6652                 spin_lock(&new_oi->ip_lock);
6653                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6654                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6655                 spin_unlock(&new_oi->ip_lock);
6656
6657                 ocfs2_journal_dirty(handle, args->new_bh);
6658         }
6659
6660 out_commit:
6661         ocfs2_commit_trans(osb, handle);
6662
6663 out:
6664         ocfs2_free_alloc_context(meta_ac);
6665         return ret;
6666 }
6667
6668 struct ocfs2_reflink_xattr_tree_args {
6669         struct ocfs2_xattr_reflink *reflink;
6670         struct buffer_head *old_blk_bh;
6671         struct buffer_head *new_blk_bh;
6672         struct ocfs2_xattr_bucket *old_bucket;
6673         struct ocfs2_xattr_bucket *new_bucket;
6674 };
6675
6676 /*
6677  * NOTE:
6678  * We have to handle the case that both old bucket and new bucket
6679  * will call this function to get the right ret_bh.
6680  * So The caller must give us the right bh.
6681  */
6682 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6683                                         struct buffer_head *bh,
6684                                         struct ocfs2_xattr_header *xh,
6685                                         int offset,
6686                                         struct ocfs2_xattr_value_root **xv,
6687                                         struct buffer_head **ret_bh,
6688                                         void *para)
6689 {
6690         struct ocfs2_reflink_xattr_tree_args *args =
6691                         (struct ocfs2_reflink_xattr_tree_args *)para;
6692         struct ocfs2_xattr_bucket *bucket;
6693
6694         if (bh == args->old_bucket->bu_bhs[0])
6695                 bucket = args->old_bucket;
6696         else
6697                 bucket = args->new_bucket;
6698
6699         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6700                                                xv, ret_bh);
6701 }
6702
6703 struct ocfs2_value_tree_metas {
6704         int num_metas;
6705         int credits;
6706         int num_recs;
6707 };
6708
6709 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6710                                         struct buffer_head *bh,
6711                                         struct ocfs2_xattr_header *xh,
6712                                         int offset,
6713                                         struct ocfs2_xattr_value_root **xv,
6714                                         struct buffer_head **ret_bh,
6715                                         void *para)
6716 {
6717         struct ocfs2_xattr_bucket *bucket =
6718                                 (struct ocfs2_xattr_bucket *)para;
6719
6720         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6721                                                xv, ret_bh);
6722 }
6723
6724 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6725                                       struct ocfs2_xattr_bucket *bucket,
6726                                       void *para)
6727 {
6728         struct ocfs2_value_tree_metas *metas =
6729                         (struct ocfs2_value_tree_metas *)para;
6730         struct ocfs2_xattr_header *xh =
6731                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6732
6733         /* Add the credits for this bucket first. */
6734         metas->credits += bucket->bu_blocks;
6735         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6736                                         xh, &metas->num_metas,
6737                                         &metas->credits, &metas->num_recs,
6738                                         ocfs2_value_tree_metas_in_bucket,
6739                                         bucket);
6740 }
6741
6742 /*
6743  * Given a xattr extent rec starting from blkno and having len clusters,
6744  * iterate all the buckets calculate how much metadata we need for reflinking
6745  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6746  */
6747 static int ocfs2_lock_reflink_xattr_rec_allocators(
6748                                 struct ocfs2_reflink_xattr_tree_args *args,
6749                                 struct ocfs2_extent_tree *xt_et,
6750                                 u64 blkno, u32 len, int *credits,
6751                                 struct ocfs2_alloc_context **meta_ac,
6752                                 struct ocfs2_alloc_context **data_ac)
6753 {
6754         int ret, num_free_extents;
6755         struct ocfs2_value_tree_metas metas;
6756         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6757         struct ocfs2_refcount_block *rb;
6758
6759         memset(&metas, 0, sizeof(metas));
6760
6761         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6762                                           ocfs2_calc_value_tree_metas, &metas);
6763         if (ret) {
6764                 mlog_errno(ret);
6765                 goto out;
6766         }
6767
6768         *credits = metas.credits;
6769
6770         /*
6771          * Calculate we need for refcount tree change.
6772          *
6773          * We need to add/modify num_recs in refcount tree, so just calculate
6774          * an approximate number we need for refcount tree change.
6775          * Sometimes we need to split the tree, and after split,  half recs
6776          * will be moved to the new block, and a new block can only provide
6777          * half number of recs. So we multiple new blocks by 2.
6778          * In the end, we have to add credits for modifying the already
6779          * existed refcount block.
6780          */
6781         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6782         metas.num_recs =
6783                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6784                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6785         metas.num_metas += metas.num_recs;
6786         *credits += metas.num_recs +
6787                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6788         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6789                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6790                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6791         else
6792                 *credits += 1;
6793
6794         /* count in the xattr tree change. */
6795         num_free_extents = ocfs2_num_free_extents(xt_et);
6796         if (num_free_extents < 0) {
6797                 ret = num_free_extents;
6798                 mlog_errno(ret);
6799                 goto out;
6800         }
6801
6802         if (num_free_extents < len)
6803                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6804
6805         *credits += ocfs2_calc_extend_credits(osb->sb,
6806                                               xt_et->et_root_el);
6807
6808         if (metas.num_metas) {
6809                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6810                                                         meta_ac);
6811                 if (ret) {
6812                         mlog_errno(ret);
6813                         goto out;
6814                 }
6815         }
6816
6817         if (len) {
6818                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6819                 if (ret)
6820                         mlog_errno(ret);
6821         }
6822 out:
6823         if (ret) {
6824                 if (*meta_ac) {
6825                         ocfs2_free_alloc_context(*meta_ac);
6826                         *meta_ac = NULL;
6827                 }
6828         }
6829
6830         return ret;
6831 }
6832
6833 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6834                                 u64 blkno, u64 new_blkno, u32 clusters,
6835                                 u32 *cpos, int num_buckets,
6836                                 struct ocfs2_alloc_context *meta_ac,
6837                                 struct ocfs2_alloc_context *data_ac,
6838                                 struct ocfs2_reflink_xattr_tree_args *args)
6839 {
6840         int i, j, ret = 0;
6841         struct super_block *sb = args->reflink->old_inode->i_sb;
6842         int bpb = args->old_bucket->bu_blocks;
6843         struct ocfs2_xattr_value_buf vb = {
6844                 .vb_access = ocfs2_journal_access,
6845         };
6846
6847         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6848                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6849                 if (ret) {
6850                         mlog_errno(ret);
6851                         break;
6852                 }
6853
6854                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6855                 if (ret) {
6856                         mlog_errno(ret);
6857                         break;
6858                 }
6859
6860                 ret = ocfs2_xattr_bucket_journal_access(handle,
6861                                                 args->new_bucket,
6862                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6863                 if (ret) {
6864                         mlog_errno(ret);
6865                         break;
6866                 }
6867
6868                 for (j = 0; j < bpb; j++)
6869                         memcpy(bucket_block(args->new_bucket, j),
6870                                bucket_block(args->old_bucket, j),
6871                                sb->s_blocksize);
6872
6873                 /*
6874                  * Record the start cpos so that we can use it to initialize
6875                  * our xattr tree we also set the xh_num_bucket for the new
6876                  * bucket.
6877                  */
6878                 if (i == 0) {
6879                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6880                                             xh_entries[0].xe_name_hash);
6881                         bucket_xh(args->new_bucket)->xh_num_buckets =
6882                                 cpu_to_le16(num_buckets);
6883                 }
6884
6885                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6886
6887                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6888                                         args->old_bucket->bu_bhs[0],
6889                                         bucket_xh(args->old_bucket),
6890                                         args->new_bucket->bu_bhs[0],
6891                                         bucket_xh(args->new_bucket),
6892                                         &vb, meta_ac,
6893                                         ocfs2_get_reflink_xattr_value_root,
6894                                         args);
6895                 if (ret) {
6896                         mlog_errno(ret);
6897                         break;
6898                 }
6899
6900                 /*
6901                  * Re-access and dirty the bucket to calculate metaecc.
6902                  * Because we may extend the transaction in reflink_xattr_header
6903                  * which will let the already accessed block gone.
6904                  */
6905                 ret = ocfs2_xattr_bucket_journal_access(handle,
6906                                                 args->new_bucket,
6907                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6908                 if (ret) {
6909                         mlog_errno(ret);
6910                         break;
6911                 }
6912
6913                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6914
6915                 ocfs2_xattr_bucket_relse(args->old_bucket);
6916                 ocfs2_xattr_bucket_relse(args->new_bucket);
6917         }
6918
6919         ocfs2_xattr_bucket_relse(args->old_bucket);
6920         ocfs2_xattr_bucket_relse(args->new_bucket);
6921         return ret;
6922 }
6923
6924 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6925                                 struct inode *inode,
6926                                 struct ocfs2_reflink_xattr_tree_args *args,
6927                                 struct ocfs2_extent_tree *et,
6928                                 struct ocfs2_alloc_context *meta_ac,
6929                                 struct ocfs2_alloc_context *data_ac,
6930                                 u64 blkno, u32 cpos, u32 len)
6931 {
6932         int ret, first_inserted = 0;
6933         u32 p_cluster, num_clusters, reflink_cpos = 0;
6934         u64 new_blkno;
6935         unsigned int num_buckets, reflink_buckets;
6936         unsigned int bpc =
6937                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6938
6939         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6940         if (ret) {
6941                 mlog_errno(ret);
6942                 goto out;
6943         }
6944         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6945         ocfs2_xattr_bucket_relse(args->old_bucket);
6946
6947         while (len && num_buckets) {
6948                 ret = ocfs2_claim_clusters(handle, data_ac,
6949                                            1, &p_cluster, &num_clusters);
6950                 if (ret) {
6951                         mlog_errno(ret);
6952                         goto out;
6953                 }
6954
6955                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6956                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6957
6958                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6959                                                  new_blkno, num_clusters,
6960                                                  &reflink_cpos, reflink_buckets,
6961                                                  meta_ac, data_ac, args);
6962                 if (ret) {
6963                         mlog_errno(ret);
6964                         goto out;
6965                 }
6966
6967                 /*
6968                  * For the 1st allocated cluster, we make it use the same cpos
6969                  * so that the xattr tree looks the same as the original one
6970                  * in the most case.
6971                  */
6972                 if (!first_inserted) {
6973                         reflink_cpos = cpos;
6974                         first_inserted = 1;
6975                 }
6976                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6977                                           num_clusters, 0, meta_ac);
6978                 if (ret)
6979                         mlog_errno(ret);
6980
6981                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6982                                                   num_clusters, reflink_cpos);
6983
6984                 len -= num_clusters;
6985                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6986                 num_buckets -= reflink_buckets;
6987         }
6988 out:
6989         return ret;
6990 }
6991
6992 /*
6993  * Create the same xattr extent record in the new inode's xattr tree.
6994  */
6995 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6996                                    struct buffer_head *root_bh,
6997                                    u64 blkno,
6998                                    u32 cpos,
6999                                    u32 len,
7000                                    void *para)
7001 {
7002         int ret, credits = 0;
7003         handle_t *handle;
7004         struct ocfs2_reflink_xattr_tree_args *args =
7005                         (struct ocfs2_reflink_xattr_tree_args *)para;
7006         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7007         struct ocfs2_alloc_context *meta_ac = NULL;
7008         struct ocfs2_alloc_context *data_ac = NULL;
7009         struct ocfs2_extent_tree et;
7010
7011         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7012
7013         ocfs2_init_xattr_tree_extent_tree(&et,
7014                                           INODE_CACHE(args->reflink->new_inode),
7015                                           args->new_blk_bh);
7016
7017         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7018                                                       len, &credits,
7019                                                       &meta_ac, &data_ac);
7020         if (ret) {
7021                 mlog_errno(ret);
7022                 goto out;
7023         }
7024
7025         handle = ocfs2_start_trans(osb, credits);
7026         if (IS_ERR(handle)) {
7027                 ret = PTR_ERR(handle);
7028                 mlog_errno(ret);
7029                 goto out;
7030         }
7031
7032         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7033                                           meta_ac, data_ac,
7034                                           blkno, cpos, len);
7035         if (ret)
7036                 mlog_errno(ret);
7037
7038         ocfs2_commit_trans(osb, handle);
7039
7040 out:
7041         if (meta_ac)
7042                 ocfs2_free_alloc_context(meta_ac);
7043         if (data_ac)
7044                 ocfs2_free_alloc_context(data_ac);
7045         return ret;
7046 }
7047
7048 /*
7049  * Create reflinked xattr buckets.
7050  * We will add bucket one by one, and refcount all the xattrs in the bucket
7051  * if they are stored outside.
7052  */
7053 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7054                                     struct buffer_head *blk_bh,
7055                                     struct buffer_head *new_blk_bh)
7056 {
7057         int ret;
7058         struct ocfs2_reflink_xattr_tree_args para;
7059
7060         memset(&para, 0, sizeof(para));
7061         para.reflink = args;
7062         para.old_blk_bh = blk_bh;
7063         para.new_blk_bh = new_blk_bh;
7064
7065         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7066         if (!para.old_bucket) {
7067                 mlog_errno(-ENOMEM);
7068                 return -ENOMEM;
7069         }
7070
7071         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7072         if (!para.new_bucket) {
7073                 ret = -ENOMEM;
7074                 mlog_errno(ret);
7075                 goto out;
7076         }
7077
7078         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7079                                               ocfs2_reflink_xattr_rec,
7080                                               &para);
7081         if (ret)
7082                 mlog_errno(ret);
7083
7084 out:
7085         ocfs2_xattr_bucket_free(para.old_bucket);
7086         ocfs2_xattr_bucket_free(para.new_bucket);
7087         return ret;
7088 }
7089
7090 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7091                                         struct buffer_head *blk_bh)
7092 {
7093         int ret, indexed = 0;
7094         struct buffer_head *new_blk_bh = NULL;
7095         struct ocfs2_xattr_block *xb =
7096                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7097
7098
7099         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7100                 indexed = 1;
7101
7102         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7103                                              &new_blk_bh, indexed);
7104         if (ret) {
7105                 mlog_errno(ret);
7106                 goto out;
7107         }
7108
7109         if (!indexed)
7110                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7111         else
7112                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7113         if (ret)
7114                 mlog_errno(ret);
7115
7116 out:
7117         brelse(new_blk_bh);
7118         return ret;
7119 }
7120
7121 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7122 {
7123         int type = ocfs2_xattr_get_type(xe);
7124
7125         return type != OCFS2_XATTR_INDEX_SECURITY &&
7126                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7127                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7128 }
7129
7130 int ocfs2_reflink_xattrs(struct inode *old_inode,
7131                          struct buffer_head *old_bh,
7132                          struct inode *new_inode,
7133                          struct buffer_head *new_bh,
7134                          bool preserve_security)
7135 {
7136         int ret;
7137         struct ocfs2_xattr_reflink args;
7138         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7139         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7140         struct buffer_head *blk_bh = NULL;
7141         struct ocfs2_cached_dealloc_ctxt dealloc;
7142         struct ocfs2_refcount_tree *ref_tree;
7143         struct buffer_head *ref_root_bh = NULL;
7144
7145         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7146                                        le64_to_cpu(di->i_refcount_loc),
7147                                        1, &ref_tree, &ref_root_bh);
7148         if (ret) {
7149                 mlog_errno(ret);
7150                 goto out;
7151         }
7152
7153         ocfs2_init_dealloc_ctxt(&dealloc);
7154
7155         args.old_inode = old_inode;
7156         args.new_inode = new_inode;
7157         args.old_bh = old_bh;
7158         args.new_bh = new_bh;
7159         args.ref_ci = &ref_tree->rf_ci;
7160         args.ref_root_bh = ref_root_bh;
7161         args.dealloc = &dealloc;
7162         if (preserve_security)
7163                 args.xattr_reflinked = NULL;
7164         else
7165                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7166
7167         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7168                 ret = ocfs2_reflink_xattr_inline(&args);
7169                 if (ret) {
7170                         mlog_errno(ret);
7171                         goto out_unlock;
7172                 }
7173         }
7174
7175         if (!di->i_xattr_loc)
7176                 goto out_unlock;
7177
7178         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7179                                      &blk_bh);
7180         if (ret < 0) {
7181                 mlog_errno(ret);
7182                 goto out_unlock;
7183         }
7184
7185         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7186         if (ret)
7187                 mlog_errno(ret);
7188
7189         brelse(blk_bh);
7190
7191 out_unlock:
7192         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7193                                    ref_tree, 1);
7194         brelse(ref_root_bh);
7195
7196         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7197                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7198                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7199         }
7200
7201 out:
7202         return ret;
7203 }
7204
7205 /*
7206  * Initialize security and acl for a already created inode.
7207  * Used for reflink a non-preserve-security file.
7208  *
7209  * It uses common api like ocfs2_xattr_set, so the caller
7210  * must not hold any lock expect i_mutex.
7211  */
7212 int ocfs2_init_security_and_acl(struct inode *dir,
7213                                 struct inode *inode,
7214                                 const struct qstr *qstr)
7215 {
7216         int ret = 0;
7217         struct buffer_head *dir_bh = NULL;
7218
7219         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7220         if (ret) {
7221                 mlog_errno(ret);
7222                 goto leave;
7223         }
7224
7225         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7226         if (ret) {
7227                 mlog_errno(ret);
7228                 goto leave;
7229         }
7230         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7231         if (ret)
7232                 mlog_errno(ret);
7233
7234         ocfs2_inode_unlock(dir, 0);
7235         brelse(dir_bh);
7236 leave:
7237         return ret;
7238 }
7239
7240 /*
7241  * 'security' attributes support
7242  */
7243 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7244                                     struct dentry *unused, struct inode *inode,
7245                                     const char *name, void *buffer, size_t size)
7246 {
7247         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7248                                name, buffer, size);
7249 }
7250
7251 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7252                                     struct dentry *unused, struct inode *inode,
7253                                     const char *name, const void *value,
7254                                     size_t size, int flags)
7255 {
7256         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7257                                name, value, size, flags);
7258 }
7259
7260 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7261                      void *fs_info)
7262 {
7263         const struct xattr *xattr;
7264         int err = 0;
7265
7266         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7267                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7268                                       xattr->name, xattr->value,
7269                                       xattr->value_len, XATTR_CREATE);
7270                 if (err)
7271                         break;
7272         }
7273         return err;
7274 }
7275
7276 int ocfs2_init_security_get(struct inode *inode,
7277                             struct inode *dir,
7278                             const struct qstr *qstr,
7279                             struct ocfs2_security_xattr_info *si)
7280 {
7281         /* check whether ocfs2 support feature xattr */
7282         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7283                 return -EOPNOTSUPP;
7284         if (si)
7285                 return security_old_inode_init_security(inode, dir, qstr,
7286                                                         &si->name, &si->value,
7287                                                         &si->value_len);
7288
7289         return security_inode_init_security(inode, dir, qstr,
7290                                             &ocfs2_initxattrs, NULL);
7291 }
7292
7293 int ocfs2_init_security_set(handle_t *handle,
7294                             struct inode *inode,
7295                             struct buffer_head *di_bh,
7296                             struct ocfs2_security_xattr_info *si,
7297                             struct ocfs2_alloc_context *xattr_ac,
7298                             struct ocfs2_alloc_context *data_ac)
7299 {
7300         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7301                                      OCFS2_XATTR_INDEX_SECURITY,
7302                                      si->name, si->value, si->value_len, 0,
7303                                      xattr_ac, data_ac);
7304 }
7305
7306 const struct xattr_handler ocfs2_xattr_security_handler = {
7307         .prefix = XATTR_SECURITY_PREFIX,
7308         .get    = ocfs2_xattr_security_get,
7309         .set    = ocfs2_xattr_security_set,
7310 };
7311
7312 /*
7313  * 'trusted' attributes support
7314  */
7315 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7316                                    struct dentry *unused, struct inode *inode,
7317                                    const char *name, void *buffer, size_t size)
7318 {
7319         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7320                                name, buffer, size);
7321 }
7322
7323 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7324                                    struct dentry *unused, struct inode *inode,
7325                                    const char *name, const void *value,
7326                                    size_t size, int flags)
7327 {
7328         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7329                                name, value, size, flags);
7330 }
7331
7332 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7333         .prefix = XATTR_TRUSTED_PREFIX,
7334         .get    = ocfs2_xattr_trusted_get,
7335         .set    = ocfs2_xattr_trusted_set,
7336 };
7337
7338 /*
7339  * 'user' attributes support
7340  */
7341 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7342                                 struct dentry *unused, struct inode *inode,
7343                                 const char *name, void *buffer, size_t size)
7344 {
7345         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7346
7347         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7348                 return -EOPNOTSUPP;
7349         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7350                                buffer, size);
7351 }
7352
7353 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7354                                 struct dentry *unused, struct inode *inode,
7355                                 const char *name, const void *value,
7356                                 size_t size, int flags)
7357 {
7358         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7359
7360         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7361                 return -EOPNOTSUPP;
7362
7363         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7364                                name, value, size, flags);
7365 }
7366
7367 const struct xattr_handler ocfs2_xattr_user_handler = {
7368         .prefix = XATTR_USER_PREFIX,
7369         .get    = ocfs2_xattr_user_get,
7370         .set    = ocfs2_xattr_user_set,
7371 };