fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #include <cluster/masklog.h>
  41
  42 #include "ocfs2.h"
  43 #include "alloc.h"
  44 #include "blockcheck.h"
  45 #include "dlmglue.h"
  46 #include "file.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "inode.h"
  50 #include "journal.h"
  51 #include "ocfs2_fs.h"
  52 #include "suballoc.h"
  53 #include "uptodate.h"
  54 #include "buffer_head_io.h"
  55 #include "super.h"
  56 #include "xattr.h"
  57 #include "refcounttree.h"
  58 #include "acl.h"
  59 #include "ocfs2_trace.h"
  60
  61 struct ocfs2_xattr_def_value_root {
  62         struct ocfs2_xattr_value_root   xv;
  63         struct ocfs2_extent_rec         er;
  64 };
  65
  66 struct ocfs2_xattr_bucket {
  67         /* The inode these xattrs are associated with */
  68         struct inode *bu_inode;
  69
  70         /* The actual buffers that make up the bucket */
  71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  72
  73         /* How many blocks make up one bucket for this filesystem */
  74         int bu_blocks;
  75 };
  76
  77 struct ocfs2_xattr_set_ctxt {
  78         handle_t *handle;
  79         struct ocfs2_alloc_context *meta_ac;
  80         struct ocfs2_alloc_context *data_ac;
  81         struct ocfs2_cached_dealloc_ctxt dealloc;
  82         int set_abort;
  83 };
  84
  85 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  86 #define OCFS2_XATTR_INLINE_SIZE 80
  87 #define OCFS2_XATTR_HEADER_GAP  4
  88 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  89                                          - sizeof(struct ocfs2_xattr_header) \
  90                                          - OCFS2_XATTR_HEADER_GAP)
  91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  92                                          - sizeof(struct ocfs2_xattr_block) \
  93                                          - sizeof(struct ocfs2_xattr_header) \
  94                                          - OCFS2_XATTR_HEADER_GAP)
  95
  96 static struct ocfs2_xattr_def_value_root def_xv = {
  97         .xv.xr_list.l_count = cpu_to_le16(1),
  98 };
  99
 100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
 101         &ocfs2_xattr_user_handler,
 102         &posix_acl_access_xattr_handler,
 103         &posix_acl_default_xattr_handler,
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &posix_acl_access_xattr_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &posix_acl_default_xattr_handler,
 115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 117 };
 118
 119 struct ocfs2_xattr_info {
 120         int             xi_name_index;
 121         const char      *xi_name;
 122         int             xi_name_len;
 123         const void      *xi_value;
 124         size_t          xi_value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 /* Operations on struct ocfs2_xa_entry */
 143 struct ocfs2_xa_loc;
 144 struct ocfs2_xa_loc_operations {
 145         /*
 146          * Journal functions
 147          */
 148         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
 149                                   int type);
 150         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
 151
 152         /*
 153          * Return a pointer to the appropriate buffer in loc->xl_storage
 154          * at the given offset from loc->xl_header.
 155          */
 156         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
 157
 158         /* Can we reuse the existing entry for the new value? */
 159         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
 160                              struct ocfs2_xattr_info *xi);
 161
 162         /* How much space is needed for the new value? */
 163         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
 164                                struct ocfs2_xattr_info *xi);
 165
 166         /*
 167          * Return the offset of the first name+value pair.  This is
 168          * the start of our downward-filling free space.
 169          */
 170         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
 171
 172         /*
 173          * Remove the name+value at this location.  Do whatever is
 174          * appropriate with the remaining name+value pairs.
 175          */
 176         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
 177
 178         /* Fill xl_entry with a new entry */
 179         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
 180
 181         /* Add name+value storage to an entry */
 182         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
 183
 184         /*
 185          * Initialize the value buf's access and bh fields for this entry.
 186          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
 187          */
 188         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
 189                                    struct ocfs2_xattr_value_buf *vb);
 190 };
 191
 192 /*
 193  * Describes an xattr entry location.  This is a memory structure
 194  * tracking the on-disk structure.
 195  */
 196 struct ocfs2_xa_loc {
 197         /* This xattr belongs to this inode */
 198         struct inode *xl_inode;
 199
 200         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
 201         struct ocfs2_xattr_header *xl_header;
 202
 203         /* Bytes from xl_header to the end of the storage */
 204         int xl_size;
 205
 206         /*
 207          * The ocfs2_xattr_entry this location describes.  If this is
 208          * NULL, this location describes the on-disk structure where it
 209          * would have been.
 210          */
 211         struct ocfs2_xattr_entry *xl_entry;
 212
 213         /*
 214          * Internal housekeeping
 215          */
 216
 217         /* Buffer(s) containing this entry */
 218         void *xl_storage;
 219
 220         /* Operations on the storage backing this location */
 221         const struct ocfs2_xa_loc_operations *xl_ops;
 222 };
 223
 224 /*
 225  * Convenience functions to calculate how much space is needed for a
 226  * given name+value pair
 227  */
 228 static int namevalue_size(int name_len, uint64_t value_len)
 229 {
 230         if (value_len > OCFS2_XATTR_INLINE_SIZE)
 231                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 232         else
 233                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 234 }
 235
 236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
 237 {
 238         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
 239 }
 240
 241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 242 {
 243         u64 value_len = le64_to_cpu(xe->xe_value_size);
 244
 245         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
 246                ocfs2_xattr_is_local(xe));
 247         return namevalue_size(xe->xe_name_len, value_len);
 248 }
 249
 250
 251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 252                                              struct ocfs2_xattr_header *xh,
 253                                              int index,
 254                                              int *block_off,
 255                                              int *new_offset);
 256
 257 static int ocfs2_xattr_block_find(struct inode *inode,
 258                                   int name_index,
 259                                   const char *name,
 260                                   struct ocfs2_xattr_search *xs);
 261 static int ocfs2_xattr_index_block_find(struct inode *inode,
 262                                         struct buffer_head *root_bh,
 263                                         int name_index,
 264                                         const char *name,
 265                                         struct ocfs2_xattr_search *xs);
 266
 267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 268                                         struct buffer_head *blk_bh,
 269                                         char *buffer,
 270                                         size_t buffer_size);
 271
 272 static int ocfs2_xattr_create_index_block(struct inode *inode,
 273                                           struct ocfs2_xattr_search *xs,
 274                                           struct ocfs2_xattr_set_ctxt *ctxt);
 275
 276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 277                                              struct ocfs2_xattr_info *xi,
 278                                              struct ocfs2_xattr_search *xs,
 279                                              struct ocfs2_xattr_set_ctxt *ctxt);
 280
 281 typedef int (xattr_tree_rec_func)(struct inode *inode,
 282                                   struct buffer_head *root_bh,
 283                                   u64 blkno, u32 cpos, u32 len, void *para);
 284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
 285                                            struct buffer_head *root_bh,
 286                                            xattr_tree_rec_func *rec_func,
 287                                            void *para);
 288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
 289                                         struct ocfs2_xattr_bucket *bucket,
 290                                         void *para);
 291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
 292                                   struct buffer_head *root_bh,
 293                                   u64 blkno,
 294                                   u32 cpos,
 295                                   u32 len,
 296                                   void *para);
 297
 298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 299                                   u64 src_blk, u64 last_blk, u64 to_blk,
 300                                   unsigned int start_bucket,
 301                                   u32 *first_hash);
 302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 303                                         struct ocfs2_dinode *di,
 304                                         struct ocfs2_xattr_info *xi,
 305                                         struct ocfs2_xattr_search *xis,
 306                                         struct ocfs2_xattr_search *xbs,
 307                                         struct ocfs2_refcount_tree **ref_tree,
 308                                         int *meta_need,
 309                                         int *credits);
 310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
 311                                            struct ocfs2_xattr_bucket *bucket,
 312                                            int offset,
 313                                            struct ocfs2_xattr_value_root **xv,
 314                                            struct buffer_head **bh);
 315
 316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 317 {
 318         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 319 }
 320
 321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 322 {
 323         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 324 }
 325
 326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 329
 330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 331 {
 332         struct ocfs2_xattr_bucket *bucket;
 333         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 334
 335         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 336
 337         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 338         if (bucket) {
 339                 bucket->bu_inode = inode;
 340                 bucket->bu_blocks = blks;
 341         }
 342
 343         return bucket;
 344 }
 345
 346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 347 {
 348         int i;
 349
 350         for (i = 0; i < bucket->bu_blocks; i++) {
 351                 brelse(bucket->bu_bhs[i]);
 352                 bucket->bu_bhs[i] = NULL;
 353         }
 354 }
 355
 356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 357 {
 358         if (bucket) {
 359                 ocfs2_xattr_bucket_relse(bucket);
 360                 bucket->bu_inode = NULL;
 361                 kfree(bucket);
 362         }
 363 }
 364
 365 /*
 366  * A bucket that has never been written to disk doesn't need to be
 367  * read.  We just need the buffer_heads.  Don't call this for
 368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 369  * them fully.
 370  */
 371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 372                                    u64 xb_blkno, int new)
 373 {
 374         int i, rc = 0;
 375
 376         for (i = 0; i < bucket->bu_blocks; i++) {
 377                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 378                                               xb_blkno + i);
 379                 if (!bucket->bu_bhs[i]) {
 380                         rc = -ENOMEM;
 381                         mlog_errno(rc);
 382                         break;
 383                 }
 384
 385                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 386                                            bucket->bu_bhs[i])) {
 387                         if (new)
 388                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 389                                                               bucket->bu_bhs[i]);
 390                         else {
 391                                 set_buffer_uptodate(bucket->bu_bhs[i]);
 392                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 393                                                           bucket->bu_bhs[i]);
 394                         }
 395                 }
 396         }
 397
 398         if (rc)
 399                 ocfs2_xattr_bucket_relse(bucket);
 400         return rc;
 401 }
 402
 403 /* Read the xattr bucket at xb_blkno */
 404 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 405                                    u64 xb_blkno)
 406 {
 407         int rc;
 408
 409         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 410                                bucket->bu_blocks, bucket->bu_bhs, 0,
 411                                NULL);
 412         if (!rc) {
 413                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 414                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 415                                                  bucket->bu_bhs,
 416                                                  bucket->bu_blocks,
 417                                                  &bucket_xh(bucket)->xh_check);
 418                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 419                 if (rc)
 420                         mlog_errno(rc);
 421         }
 422
 423         if (rc)
 424                 ocfs2_xattr_bucket_relse(bucket);
 425         return rc;
 426 }
 427
 428 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 429                                              struct ocfs2_xattr_bucket *bucket,
 430                                              int type)
 431 {
 432         int i, rc = 0;
 433
 434         for (i = 0; i < bucket->bu_blocks; i++) {
 435                 rc = ocfs2_journal_access(handle,
 436                                           INODE_CACHE(bucket->bu_inode),
 437                                           bucket->bu_bhs[i], type);
 438                 if (rc) {
 439                         mlog_errno(rc);
 440                         break;
 441                 }
 442         }
 443
 444         return rc;
 445 }
 446
 447 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 448                                              struct ocfs2_xattr_bucket *bucket)
 449 {
 450         int i;
 451
 452         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 453         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 454                                    bucket->bu_bhs, bucket->bu_blocks,
 455                                    &bucket_xh(bucket)->xh_check);
 456         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 457
 458         for (i = 0; i < bucket->bu_blocks; i++)
 459                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 460 }
 461
 462 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 463                                          struct ocfs2_xattr_bucket *src)
 464 {
 465         int i;
 466         int blocksize = src->bu_inode->i_sb->s_blocksize;
 467
 468         BUG_ON(dest->bu_blocks != src->bu_blocks);
 469         BUG_ON(dest->bu_inode != src->bu_inode);
 470
 471         for (i = 0; i < src->bu_blocks; i++) {
 472                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 473                        blocksize);
 474         }
 475 }
 476
 477 static int ocfs2_validate_xattr_block(struct super_block *sb,
 478                                       struct buffer_head *bh)
 479 {
 480         int rc;
 481         struct ocfs2_xattr_block *xb =
 482                 (struct ocfs2_xattr_block *)bh->b_data;
 483
 484         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
 485
 486         BUG_ON(!buffer_uptodate(bh));
 487
 488         /*
 489          * If the ecc fails, we return the error but otherwise
 490          * leave the filesystem running.  We know any error is
 491          * local to this block.
 492          */
 493         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 494         if (rc)
 495                 return rc;
 496
 497         /*
 498          * Errors after here are fatal
 499          */
 500
 501         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 502                 return ocfs2_error(sb,
 503                                    "Extended attribute block #%llu has bad signature %.*s\n",
 504                                    (unsigned long long)bh->b_blocknr, 7,
 505                                    xb->xb_signature);
 506         }
 507
 508         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 509                 return ocfs2_error(sb,
 510                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
 511                                    (unsigned long long)bh->b_blocknr,
 512                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
 513         }
 514
 515         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 516                 return ocfs2_error(sb,
 517                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
 518                                    (unsigned long long)bh->b_blocknr,
 519                                    le32_to_cpu(xb->xb_fs_generation));
 520         }
 521
 522         return 0;
 523 }
 524
 525 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 526                                   struct buffer_head **bh)
 527 {
 528         int rc;
 529         struct buffer_head *tmp = *bh;
 530
 531         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 532                               ocfs2_validate_xattr_block);
 533
 534         /* If ocfs2_read_block() got us a new bh, pass it up. */
 535         if (!rc && !*bh)
 536                 *bh = tmp;
 537
 538         return rc;
 539 }
 540
 541 static inline const char *ocfs2_xattr_prefix(int name_index)
 542 {
 543         const struct xattr_handler *handler = NULL;
 544
 545         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 546                 handler = ocfs2_xattr_handler_map[name_index];
 547
 548         return handler ? handler->prefix : NULL;
 549 }
 550
 551 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 552                                  const char *name,
 553                                  int name_len)
 554 {
 555         /* Get hash value of uuid from super block */
 556         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 557         int i;
 558
 559         /* hash extended attribute name */
 560         for (i = 0; i < name_len; i++) {
 561                 hash = (hash << OCFS2_HASH_SHIFT) ^
 562                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 563                        *name++;
 564         }
 565
 566         return hash;
 567 }
 568
 569 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 570 {
 571         return namevalue_size(name_len, value_len) +
 572                 sizeof(struct ocfs2_xattr_entry);
 573 }
 574
 575 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
 576 {
 577         return namevalue_size_xi(xi) +
 578                 sizeof(struct ocfs2_xattr_entry);
 579 }
 580
 581 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
 582 {
 583         return namevalue_size_xe(xe) +
 584                 sizeof(struct ocfs2_xattr_entry);
 585 }
 586
 587 int ocfs2_calc_security_init(struct inode *dir,
 588                              struct ocfs2_security_xattr_info *si,
 589                              int *want_clusters,
 590                              int *xattr_credits,
 591                              struct ocfs2_alloc_context **xattr_ac)
 592 {
 593         int ret = 0;
 594         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 595         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 596                                                  si->value_len);
 597
 598         /*
 599          * The max space of security xattr taken inline is
 600          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 601          * So reserve one metadata block for it is ok.
 602          */
 603         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 604             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 605                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 606                 if (ret) {
 607                         mlog_errno(ret);
 608                         return ret;
 609                 }
 610                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 611         }
 612
 613         /* reserve clusters for xattr value which will be set in B tree*/
 614         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 615                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 616                                                             si->value_len);
 617
 618                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 619                                                            new_clusters);
 620                 *want_clusters += new_clusters;
 621         }
 622         return ret;
 623 }
 624
 625 int ocfs2_calc_xattr_init(struct inode *dir,
 626                           struct buffer_head *dir_bh,
 627                           umode_t mode,
 628                           struct ocfs2_security_xattr_info *si,
 629                           int *want_clusters,
 630                           int *xattr_credits,
 631                           int *want_meta)
 632 {
 633         int ret = 0;
 634         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 635         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 636
 637         if (si->enable)
 638                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 639                                                      si->value_len);
 640
 641         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 642                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
 643                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 644                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 645                                         "", NULL, 0);
 646                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
 647                 if (acl_len > 0) {
 648                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 649                         if (S_ISDIR(mode))
 650                                 a_size <<= 1;
 651                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 652                         mlog_errno(ret);
 653                         return ret;
 654                 }
 655         }
 656
 657         if (!(s_size + a_size))
 658                 return ret;
 659
 660         /*
 661          * The max space of security xattr taken inline is
 662          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 663          * The max space of acl xattr taken inline is
 664          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 665          * when blocksize = 512, may reserve one more cluser for
 666          * xattr bucket, otherwise reserve one metadata block
 667          * for them is ok.
 668          * If this is a new directory with inline data,
 669          * we choose to reserve the entire inline area for
 670          * directory contents and force an external xattr block.
 671          */
 672         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 673             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 674             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 675                 *want_meta = *want_meta + 1;
 676                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 677         }
 678
 679         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 680             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 681                 *want_clusters += 1;
 682                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 683         }
 684
 685         /*
 686          * reserve credits and clusters for xattrs which has large value
 687          * and have to be set outside
 688          */
 689         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 690                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 691                                                         si->value_len);
 692                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 693                                                            new_clusters);
 694                 *want_clusters += new_clusters;
 695         }
 696         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 697             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 698                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 699                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 700                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 701                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 702                                                            new_clusters);
 703                 *want_clusters += new_clusters;
 704         }
 705
 706         return ret;
 707 }
 708
 709 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 710                                          u32 clusters_to_add,
 711                                          struct ocfs2_xattr_value_buf *vb,
 712                                          struct ocfs2_xattr_set_ctxt *ctxt)
 713 {
 714         int status = 0, credits;
 715         handle_t *handle = ctxt->handle;
 716         enum ocfs2_alloc_restarted why;
 717         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 718         struct ocfs2_extent_tree et;
 719
 720         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 721
 722         while (clusters_to_add) {
 723                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
 724
 725                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 726                                        OCFS2_JOURNAL_ACCESS_WRITE);
 727                 if (status < 0) {
 728                         mlog_errno(status);
 729                         break;
 730                 }
 731
 732                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 733                 status = ocfs2_add_clusters_in_btree(handle,
 734                                                      &et,
 735                                                      &logical_start,
 736                                                      clusters_to_add,
 737                                                      0,
 738                                                      ctxt->data_ac,
 739                                                      ctxt->meta_ac,
 740                                                      &why);
 741                 if ((status < 0) && (status != -EAGAIN)) {
 742                         if (status != -ENOSPC)
 743                                 mlog_errno(status);
 744                         break;
 745                 }
 746
 747                 ocfs2_journal_dirty(handle, vb->vb_bh);
 748
 749                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
 750                                          prev_clusters;
 751
 752                 if (why != RESTART_NONE && clusters_to_add) {
 753                         /*
 754                          * We can only fail in case the alloc file doesn't give
 755                          * up enough clusters.
 756                          */
 757                         BUG_ON(why == RESTART_META);
 758
 759                         credits = ocfs2_calc_extend_credits(inode->i_sb,
 760                                                             &vb->vb_xv->xr_list);
 761                         status = ocfs2_extend_trans(handle, credits);
 762                         if (status < 0) {
 763                                 status = -ENOMEM;
 764                                 mlog_errno(status);
 765                                 break;
 766                         }
 767                 }
 768         }
 769
 770         return status;
 771 }
 772
 773 static int __ocfs2_remove_xattr_range(struct inode *inode,
 774                                       struct ocfs2_xattr_value_buf *vb,
 775                                       u32 cpos, u32 phys_cpos, u32 len,
 776                                       unsigned int ext_flags,
 777                                       struct ocfs2_xattr_set_ctxt *ctxt)
 778 {
 779         int ret;
 780         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 781         handle_t *handle = ctxt->handle;
 782         struct ocfs2_extent_tree et;
 783
 784         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 785
 786         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 787                             OCFS2_JOURNAL_ACCESS_WRITE);
 788         if (ret) {
 789                 mlog_errno(ret);
 790                 goto out;
 791         }
 792
 793         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 794                                   &ctxt->dealloc);
 795         if (ret) {
 796                 mlog_errno(ret);
 797                 goto out;
 798         }
 799
 800         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 801         ocfs2_journal_dirty(handle, vb->vb_bh);
 802
 803         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 804                 ret = ocfs2_decrease_refcount(inode, handle,
 805                                         ocfs2_blocks_to_clusters(inode->i_sb,
 806                                                                  phys_blkno),
 807                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 808         else
 809                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 810                                                   phys_blkno, len);
 811         if (ret)
 812                 mlog_errno(ret);
 813
 814 out:
 815         return ret;
 816 }
 817
 818 static int ocfs2_xattr_shrink_size(struct inode *inode,
 819                                    u32 old_clusters,
 820                                    u32 new_clusters,
 821                                    struct ocfs2_xattr_value_buf *vb,
 822                                    struct ocfs2_xattr_set_ctxt *ctxt)
 823 {
 824         int ret = 0;
 825         unsigned int ext_flags;
 826         u32 trunc_len, cpos, phys_cpos, alloc_size;
 827         u64 block;
 828
 829         if (old_clusters <= new_clusters)
 830                 return 0;
 831
 832         cpos = new_clusters;
 833         trunc_len = old_clusters - new_clusters;
 834         while (trunc_len) {
 835                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 836                                                &alloc_size,
 837                                                &vb->vb_xv->xr_list, &ext_flags);
 838                 if (ret) {
 839                         mlog_errno(ret);
 840                         goto out;
 841                 }
 842
 843                 if (alloc_size > trunc_len)
 844                         alloc_size = trunc_len;
 845
 846                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 847                                                  phys_cpos, alloc_size,
 848                                                  ext_flags, ctxt);
 849                 if (ret) {
 850                         mlog_errno(ret);
 851                         goto out;
 852                 }
 853
 854                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 855                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 856                                                        block, alloc_size);
 857                 cpos += alloc_size;
 858                 trunc_len -= alloc_size;
 859         }
 860
 861 out:
 862         return ret;
 863 }
 864
 865 static int ocfs2_xattr_value_truncate(struct inode *inode,
 866                                       struct ocfs2_xattr_value_buf *vb,
 867                                       int len,
 868                                       struct ocfs2_xattr_set_ctxt *ctxt)
 869 {
 870         int ret;
 871         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 872         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 873
 874         if (new_clusters == old_clusters)
 875                 return 0;
 876
 877         if (new_clusters > old_clusters)
 878                 ret = ocfs2_xattr_extend_allocation(inode,
 879                                                     new_clusters - old_clusters,
 880                                                     vb, ctxt);
 881         else
 882                 ret = ocfs2_xattr_shrink_size(inode,
 883                                               old_clusters, new_clusters,
 884                                               vb, ctxt);
 885
 886         return ret;
 887 }
 888
 889 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 890                                   size_t *result, const char *prefix,
 891                                   const char *name, int name_len)
 892 {
 893         char *p = buffer + *result;
 894         int prefix_len = strlen(prefix);
 895         int total_len = prefix_len + name_len + 1;
 896
 897         *result += total_len;
 898
 899         /* we are just looking for how big our buffer needs to be */
 900         if (!size)
 901                 return 0;
 902
 903         if (*result > size)
 904                 return -ERANGE;
 905
 906         memcpy(p, prefix, prefix_len);
 907         memcpy(p + prefix_len, name, name_len);
 908         p[prefix_len + name_len] = '\0';
 909
 910         return 0;
 911 }
 912
 913 static int ocfs2_xattr_list_entries(struct inode *inode,
 914                                     struct ocfs2_xattr_header *header,
 915                                     char *buffer, size_t buffer_size)
 916 {
 917         size_t result = 0;
 918         int i, type, ret;
 919         const char *prefix, *name;
 920
 921         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 922                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 923                 type = ocfs2_xattr_get_type(entry);
 924                 prefix = ocfs2_xattr_prefix(type);
 925
 926                 if (prefix) {
 927                         name = (const char *)header +
 928                                 le16_to_cpu(entry->xe_name_offset);
 929
 930                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 931                                                      &result, prefix, name,
 932                                                      entry->xe_name_len);
 933                         if (ret)
 934                                 return ret;
 935                 }
 936         }
 937
 938         return result;
 939 }
 940
 941 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
 942                                          struct ocfs2_dinode *di)
 943 {
 944         struct ocfs2_xattr_header *xh;
 945         int i;
 946
 947         xh = (struct ocfs2_xattr_header *)
 948                  ((void *)di + inode->i_sb->s_blocksize -
 949                  le16_to_cpu(di->i_xattr_inline_size));
 950
 951         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
 952                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
 953                         return 1;
 954
 955         return 0;
 956 }
 957
 958 static int ocfs2_xattr_ibody_list(struct inode *inode,
 959                                   struct ocfs2_dinode *di,
 960                                   char *buffer,
 961                                   size_t buffer_size)
 962 {
 963         struct ocfs2_xattr_header *header = NULL;
 964         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 965         int ret = 0;
 966
 967         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 968                 return ret;
 969
 970         header = (struct ocfs2_xattr_header *)
 971                  ((void *)di + inode->i_sb->s_blocksize -
 972                  le16_to_cpu(di->i_xattr_inline_size));
 973
 974         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 975
 976         return ret;
 977 }
 978
 979 static int ocfs2_xattr_block_list(struct inode *inode,
 980                                   struct ocfs2_dinode *di,
 981                                   char *buffer,
 982                                   size_t buffer_size)
 983 {
 984         struct buffer_head *blk_bh = NULL;
 985         struct ocfs2_xattr_block *xb;
 986         int ret = 0;
 987
 988         if (!di->i_xattr_loc)
 989                 return ret;
 990
 991         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 992                                      &blk_bh);
 993         if (ret < 0) {
 994                 mlog_errno(ret);
 995                 return ret;
 996         }
 997
 998         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 999         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1000                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1001                 ret = ocfs2_xattr_list_entries(inode, header,
1002                                                buffer, buffer_size);
1003         } else
1004                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1005                                                    buffer, buffer_size);
1006
1007         brelse(blk_bh);
1008
1009         return ret;
1010 }
1011
1012 ssize_t ocfs2_listxattr(struct dentry *dentry,
1013                         char *buffer,
1014                         size_t size)
1015 {
1016         int ret = 0, i_ret = 0, b_ret = 0;
1017         struct buffer_head *di_bh = NULL;
1018         struct ocfs2_dinode *di = NULL;
1019         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1020
1021         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1022                 return -EOPNOTSUPP;
1023
1024         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1025                 return ret;
1026
1027         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1028         if (ret < 0) {
1029                 mlog_errno(ret);
1030                 return ret;
1031         }
1032
1033         di = (struct ocfs2_dinode *)di_bh->b_data;
1034
1035         down_read(&oi->ip_xattr_sem);
1036         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1037         if (i_ret < 0)
1038                 b_ret = 0;
1039         else {
1040                 if (buffer) {
1041                         buffer += i_ret;
1042                         size -= i_ret;
1043                 }
1044                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1045                                                buffer, size);
1046                 if (b_ret < 0)
1047                         i_ret = 0;
1048         }
1049         up_read(&oi->ip_xattr_sem);
1050         ocfs2_inode_unlock(d_inode(dentry), 0);
1051
1052         brelse(di_bh);
1053
1054         return i_ret + b_ret;
1055 }
1056
1057 static int ocfs2_xattr_find_entry(int name_index,
1058                                   const char *name,
1059                                   struct ocfs2_xattr_search *xs)
1060 {
1061         struct ocfs2_xattr_entry *entry;
1062         size_t name_len;
1063         int i, cmp = 1;
1064
1065         if (name == NULL)
1066                 return -EINVAL;
1067
1068         name_len = strlen(name);
1069         entry = xs->here;
1070         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1071                 cmp = name_index - ocfs2_xattr_get_type(entry);
1072                 if (!cmp)
1073                         cmp = name_len - entry->xe_name_len;
1074                 if (!cmp)
1075                         cmp = memcmp(name, (xs->base +
1076                                      le16_to_cpu(entry->xe_name_offset)),
1077                                      name_len);
1078                 if (cmp == 0)
1079                         break;
1080                 entry += 1;
1081         }
1082         xs->here = entry;
1083
1084         return cmp ? -ENODATA : 0;
1085 }
1086
1087 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1088                                          struct ocfs2_xattr_value_root *xv,
1089                                          void *buffer,
1090                                          size_t len)
1091 {
1092         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1093         u64 blkno;
1094         int i, ret = 0;
1095         size_t cplen, blocksize;
1096         struct buffer_head *bh = NULL;
1097         struct ocfs2_extent_list *el;
1098
1099         el = &xv->xr_list;
1100         clusters = le32_to_cpu(xv->xr_clusters);
1101         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1102         blocksize = inode->i_sb->s_blocksize;
1103
1104         cpos = 0;
1105         while (cpos < clusters) {
1106                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1107                                                &num_clusters, el, NULL);
1108                 if (ret) {
1109                         mlog_errno(ret);
1110                         goto out;
1111                 }
1112
1113                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1114                 /* Copy ocfs2_xattr_value */
1115                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1116                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1117                                                &bh, NULL);
1118                         if (ret) {
1119                                 mlog_errno(ret);
1120                                 goto out;
1121                         }
1122
1123                         cplen = len >= blocksize ? blocksize : len;
1124                         memcpy(buffer, bh->b_data, cplen);
1125                         len -= cplen;
1126                         buffer += cplen;
1127
1128                         brelse(bh);
1129                         bh = NULL;
1130                         if (len == 0)
1131                                 break;
1132                 }
1133                 cpos += num_clusters;
1134         }
1135 out:
1136         return ret;
1137 }
1138
1139 static int ocfs2_xattr_ibody_get(struct inode *inode,
1140                                  int name_index,
1141                                  const char *name,
1142                                  void *buffer,
1143                                  size_t buffer_size,
1144                                  struct ocfs2_xattr_search *xs)
1145 {
1146         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1147         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1148         struct ocfs2_xattr_value_root *xv;
1149         size_t size;
1150         int ret = 0;
1151
1152         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1153                 return -ENODATA;
1154
1155         xs->end = (void *)di + inode->i_sb->s_blocksize;
1156         xs->header = (struct ocfs2_xattr_header *)
1157                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1158         xs->base = (void *)xs->header;
1159         xs->here = xs->header->xh_entries;
1160
1161         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1162         if (ret)
1163                 return ret;
1164         size = le64_to_cpu(xs->here->xe_value_size);
1165         if (buffer) {
1166                 if (size > buffer_size)
1167                         return -ERANGE;
1168                 if (ocfs2_xattr_is_local(xs->here)) {
1169                         memcpy(buffer, (void *)xs->base +
1170                                le16_to_cpu(xs->here->xe_name_offset) +
1171                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1172                 } else {
1173                         xv = (struct ocfs2_xattr_value_root *)
1174                                 (xs->base + le16_to_cpu(
1175                                  xs->here->xe_name_offset) +
1176                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1177                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1178                                                             buffer, size);
1179                         if (ret < 0) {
1180                                 mlog_errno(ret);
1181                                 return ret;
1182                         }
1183                 }
1184         }
1185
1186         return size;
1187 }
1188
1189 static int ocfs2_xattr_block_get(struct inode *inode,
1190                                  int name_index,
1191                                  const char *name,
1192                                  void *buffer,
1193                                  size_t buffer_size,
1194                                  struct ocfs2_xattr_search *xs)
1195 {
1196         struct ocfs2_xattr_block *xb;
1197         struct ocfs2_xattr_value_root *xv;
1198         size_t size;
1199         int ret = -ENODATA, name_offset, name_len, i;
1200         int uninitialized_var(block_off);
1201
1202         xs->bucket = ocfs2_xattr_bucket_new(inode);
1203         if (!xs->bucket) {
1204                 ret = -ENOMEM;
1205                 mlog_errno(ret);
1206                 goto cleanup;
1207         }
1208
1209         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1210         if (ret) {
1211                 mlog_errno(ret);
1212                 goto cleanup;
1213         }
1214
1215         if (xs->not_found) {
1216                 ret = -ENODATA;
1217                 goto cleanup;
1218         }
1219
1220         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1221         size = le64_to_cpu(xs->here->xe_value_size);
1222         if (buffer) {
1223                 ret = -ERANGE;
1224                 if (size > buffer_size)
1225                         goto cleanup;
1226
1227                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1228                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1229                 i = xs->here - xs->header->xh_entries;
1230
1231                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1232                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1233                                                                 bucket_xh(xs->bucket),
1234                                                                 i,
1235                                                                 &block_off,
1236                                                                 &name_offset);
1237                         if (ret) {
1238                                 mlog_errno(ret);
1239                                 goto cleanup;
1240                         }
1241                         xs->base = bucket_block(xs->bucket, block_off);
1242                 }
1243                 if (ocfs2_xattr_is_local(xs->here)) {
1244                         memcpy(buffer, (void *)xs->base +
1245                                name_offset + name_len, size);
1246                 } else {
1247                         xv = (struct ocfs2_xattr_value_root *)
1248                                 (xs->base + name_offset + name_len);
1249                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1250                                                             buffer, size);
1251                         if (ret < 0) {
1252                                 mlog_errno(ret);
1253                                 goto cleanup;
1254                         }
1255                 }
1256         }
1257         ret = size;
1258 cleanup:
1259         ocfs2_xattr_bucket_free(xs->bucket);
1260
1261         brelse(xs->xattr_bh);
1262         xs->xattr_bh = NULL;
1263         return ret;
1264 }
1265
1266 int ocfs2_xattr_get_nolock(struct inode *inode,
1267                            struct buffer_head *di_bh,
1268                            int name_index,
1269                            const char *name,
1270                            void *buffer,
1271                            size_t buffer_size)
1272 {
1273         int ret;
1274         struct ocfs2_dinode *di = NULL;
1275         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1276         struct ocfs2_xattr_search xis = {
1277                 .not_found = -ENODATA,
1278         };
1279         struct ocfs2_xattr_search xbs = {
1280                 .not_found = -ENODATA,
1281         };
1282
1283         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1284                 return -EOPNOTSUPP;
1285
1286         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1287                 return -ENODATA;
1288
1289         xis.inode_bh = xbs.inode_bh = di_bh;
1290         di = (struct ocfs2_dinode *)di_bh->b_data;
1291
1292         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1293                                     buffer_size, &xis);
1294         if (ret == -ENODATA && di->i_xattr_loc)
1295                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1296                                             buffer_size, &xbs);
1297
1298         return ret;
1299 }
1300
1301 /* ocfs2_xattr_get()
1302  *
1303  * Copy an extended attribute into the buffer provided.
1304  * Buffer is NULL to compute the size of buffer required.
1305  */
1306 static int ocfs2_xattr_get(struct inode *inode,
1307                            int name_index,
1308                            const char *name,
1309                            void *buffer,
1310                            size_t buffer_size)
1311 {
1312         int ret;
1313         struct buffer_head *di_bh = NULL;
1314
1315         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1316         if (ret < 0) {
1317                 mlog_errno(ret);
1318                 return ret;
1319         }
1320         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1321         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1322                                      name, buffer, buffer_size);
1323         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1324
1325         ocfs2_inode_unlock(inode, 0);
1326
1327         brelse(di_bh);
1328
1329         return ret;
1330 }
1331
1332 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1333                                            handle_t *handle,
1334                                            struct ocfs2_xattr_value_buf *vb,
1335                                            const void *value,
1336                                            int value_len)
1337 {
1338         int ret = 0, i, cp_len;
1339         u16 blocksize = inode->i_sb->s_blocksize;
1340         u32 p_cluster, num_clusters;
1341         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1342         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1343         u64 blkno;
1344         struct buffer_head *bh = NULL;
1345         unsigned int ext_flags;
1346         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1347
1348         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1349
1350         while (cpos < clusters) {
1351                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1352                                                &num_clusters, &xv->xr_list,
1353                                                &ext_flags);
1354                 if (ret) {
1355                         mlog_errno(ret);
1356                         goto out;
1357                 }
1358
1359                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1360
1361                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1362
1363                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1364                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1365                                                &bh, NULL);
1366                         if (ret) {
1367                                 mlog_errno(ret);
1368                                 goto out;
1369                         }
1370
1371                         ret = ocfs2_journal_access(handle,
1372                                                    INODE_CACHE(inode),
1373                                                    bh,
1374                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1375                         if (ret < 0) {
1376                                 mlog_errno(ret);
1377                                 goto out;
1378                         }
1379
1380                         cp_len = value_len > blocksize ? blocksize : value_len;
1381                         memcpy(bh->b_data, value, cp_len);
1382                         value_len -= cp_len;
1383                         value += cp_len;
1384                         if (cp_len < blocksize)
1385                                 memset(bh->b_data + cp_len, 0,
1386                                        blocksize - cp_len);
1387
1388                         ocfs2_journal_dirty(handle, bh);
1389                         brelse(bh);
1390                         bh = NULL;
1391
1392                         /*
1393                          * XXX: do we need to empty all the following
1394                          * blocks in this cluster?
1395                          */
1396                         if (!value_len)
1397                                 break;
1398                 }
1399                 cpos += num_clusters;
1400         }
1401 out:
1402         brelse(bh);
1403
1404         return ret;
1405 }
1406
1407 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1408                                        int num_entries)
1409 {
1410         int free_space;
1411
1412         if (!needed_space)
1413                 return 0;
1414
1415         free_space = free_start -
1416                 sizeof(struct ocfs2_xattr_header) -
1417                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1418                 OCFS2_XATTR_HEADER_GAP;
1419         if (free_space < 0)
1420                 return -EIO;
1421         if (free_space < needed_space)
1422                 return -ENOSPC;
1423
1424         return 0;
1425 }
1426
1427 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1428                                    int type)
1429 {
1430         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1431 }
1432
1433 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1434 {
1435         loc->xl_ops->xlo_journal_dirty(handle, loc);
1436 }
1437
1438 /* Give a pointer into the storage for the given offset */
1439 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1440 {
1441         BUG_ON(offset >= loc->xl_size);
1442         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1443 }
1444
1445 /*
1446  * Wipe the name+value pair and allow the storage to reclaim it.  This
1447  * must be followed by either removal of the entry or a call to
1448  * ocfs2_xa_add_namevalue().
1449  */
1450 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1451 {
1452         loc->xl_ops->xlo_wipe_namevalue(loc);
1453 }
1454
1455 /*
1456  * Find lowest offset to a name+value pair.  This is the start of our
1457  * downward-growing free space.
1458  */
1459 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1460 {
1461         return loc->xl_ops->xlo_get_free_start(loc);
1462 }
1463
1464 /* Can we reuse loc->xl_entry for xi? */
1465 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1466                                     struct ocfs2_xattr_info *xi)
1467 {
1468         return loc->xl_ops->xlo_can_reuse(loc, xi);
1469 }
1470
1471 /* How much free space is needed to set the new value */
1472 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1473                                 struct ocfs2_xattr_info *xi)
1474 {
1475         return loc->xl_ops->xlo_check_space(loc, xi);
1476 }
1477
1478 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1479 {
1480         loc->xl_ops->xlo_add_entry(loc, name_hash);
1481         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1482         /*
1483          * We can't leave the new entry's xe_name_offset at zero or
1484          * add_namevalue() will go nuts.  We set it to the size of our
1485          * storage so that it can never be less than any other entry.
1486          */
1487         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1488 }
1489
1490 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1491                                    struct ocfs2_xattr_info *xi)
1492 {
1493         int size = namevalue_size_xi(xi);
1494         int nameval_offset;
1495         char *nameval_buf;
1496
1497         loc->xl_ops->xlo_add_namevalue(loc, size);
1498         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1499         loc->xl_entry->xe_name_len = xi->xi_name_len;
1500         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1501         ocfs2_xattr_set_local(loc->xl_entry,
1502                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1503
1504         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1505         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1506         memset(nameval_buf, 0, size);
1507         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1508 }
1509
1510 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1511                                     struct ocfs2_xattr_value_buf *vb)
1512 {
1513         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1514         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1515
1516         /* Value bufs are for value trees */
1517         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1518         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1519                (name_size + OCFS2_XATTR_ROOT_SIZE));
1520
1521         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1522         vb->vb_xv =
1523                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1524                                                         nameval_offset +
1525                                                         name_size);
1526 }
1527
1528 static int ocfs2_xa_block_journal_access(handle_t *handle,
1529                                          struct ocfs2_xa_loc *loc, int type)
1530 {
1531         struct buffer_head *bh = loc->xl_storage;
1532         ocfs2_journal_access_func access;
1533
1534         if (loc->xl_size == (bh->b_size -
1535                              offsetof(struct ocfs2_xattr_block,
1536                                       xb_attrs.xb_header)))
1537                 access = ocfs2_journal_access_xb;
1538         else
1539                 access = ocfs2_journal_access_di;
1540         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1541 }
1542
1543 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1544                                          struct ocfs2_xa_loc *loc)
1545 {
1546         struct buffer_head *bh = loc->xl_storage;
1547
1548         ocfs2_journal_dirty(handle, bh);
1549 }
1550
1551 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1552                                            int offset)
1553 {
1554         return (char *)loc->xl_header + offset;
1555 }
1556
1557 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1558                                     struct ocfs2_xattr_info *xi)
1559 {
1560         /*
1561          * Block storage is strict.  If the sizes aren't exact, we will
1562          * remove the old one and reinsert the new.
1563          */
1564         return namevalue_size_xe(loc->xl_entry) ==
1565                 namevalue_size_xi(xi);
1566 }
1567
1568 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1569 {
1570         struct ocfs2_xattr_header *xh = loc->xl_header;
1571         int i, count = le16_to_cpu(xh->xh_count);
1572         int offset, free_start = loc->xl_size;
1573
1574         for (i = 0; i < count; i++) {
1575                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1576                 if (offset < free_start)
1577                         free_start = offset;
1578         }
1579
1580         return free_start;
1581 }
1582
1583 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1584                                       struct ocfs2_xattr_info *xi)
1585 {
1586         int count = le16_to_cpu(loc->xl_header->xh_count);
1587         int free_start = ocfs2_xa_get_free_start(loc);
1588         int needed_space = ocfs2_xi_entry_usage(xi);
1589
1590         /*
1591          * Block storage will reclaim the original entry before inserting
1592          * the new value, so we only need the difference.  If the new
1593          * entry is smaller than the old one, we don't need anything.
1594          */
1595         if (loc->xl_entry) {
1596                 /* Don't need space if we're reusing! */
1597                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1598                         needed_space = 0;
1599                 else
1600                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1601         }
1602         if (needed_space < 0)
1603                 needed_space = 0;
1604         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1605 }
1606
1607 /*
1608  * Block storage for xattrs keeps the name+value pairs compacted.  When
1609  * we remove one, we have to shift any that preceded it towards the end.
1610  */
1611 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1612 {
1613         int i, offset;
1614         int namevalue_offset, first_namevalue_offset, namevalue_size;
1615         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1616         struct ocfs2_xattr_header *xh = loc->xl_header;
1617         int count = le16_to_cpu(xh->xh_count);
1618
1619         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1620         namevalue_size = namevalue_size_xe(entry);
1621         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1622
1623         /* Shift the name+value pairs */
1624         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1625                 (char *)xh + first_namevalue_offset,
1626                 namevalue_offset - first_namevalue_offset);
1627         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1628
1629         /* Now tell xh->xh_entries about it */
1630         for (i = 0; i < count; i++) {
1631                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1632                 if (offset <= namevalue_offset)
1633                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1634                                      namevalue_size);
1635         }
1636
1637         /*
1638          * Note that we don't update xh_free_start or xh_name_value_len
1639          * because they're not used in block-stored xattrs.
1640          */
1641 }
1642
1643 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1644 {
1645         int count = le16_to_cpu(loc->xl_header->xh_count);
1646         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1647         le16_add_cpu(&loc->xl_header->xh_count, 1);
1648         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1649 }
1650
1651 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1652 {
1653         int free_start = ocfs2_xa_get_free_start(loc);
1654
1655         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1656 }
1657
1658 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1659                                           struct ocfs2_xattr_value_buf *vb)
1660 {
1661         struct buffer_head *bh = loc->xl_storage;
1662
1663         if (loc->xl_size == (bh->b_size -
1664                              offsetof(struct ocfs2_xattr_block,
1665                                       xb_attrs.xb_header)))
1666                 vb->vb_access = ocfs2_journal_access_xb;
1667         else
1668                 vb->vb_access = ocfs2_journal_access_di;
1669         vb->vb_bh = bh;
1670 }
1671
1672 /*
1673  * Operations for xattrs stored in blocks.  This includes inline inode
1674  * storage and unindexed ocfs2_xattr_blocks.
1675  */
1676 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1677         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1678         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1679         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1680         .xlo_check_space        = ocfs2_xa_block_check_space,
1681         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1682         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1683         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1684         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1685         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1686         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1687 };
1688
1689 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1690                                           struct ocfs2_xa_loc *loc, int type)
1691 {
1692         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1693
1694         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1695 }
1696
1697 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1698                                           struct ocfs2_xa_loc *loc)
1699 {
1700         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1701
1702         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1703 }
1704
1705 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1706                                             int offset)
1707 {
1708         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1709         int block, block_offset;
1710
1711         /* The header is at the front of the bucket */
1712         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1713         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1714
1715         return bucket_block(bucket, block) + block_offset;
1716 }
1717
1718 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1719                                      struct ocfs2_xattr_info *xi)
1720 {
1721         return namevalue_size_xe(loc->xl_entry) >=
1722                 namevalue_size_xi(xi);
1723 }
1724
1725 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1726 {
1727         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1728         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1729 }
1730
1731 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1732                                          int free_start, int size)
1733 {
1734         /*
1735          * We need to make sure that the name+value pair fits within
1736          * one block.
1737          */
1738         if (((free_start - size) >> sb->s_blocksize_bits) !=
1739             ((free_start - 1) >> sb->s_blocksize_bits))
1740                 free_start -= free_start % sb->s_blocksize;
1741
1742         return free_start;
1743 }
1744
1745 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1746                                        struct ocfs2_xattr_info *xi)
1747 {
1748         int rc;
1749         int count = le16_to_cpu(loc->xl_header->xh_count);
1750         int free_start = ocfs2_xa_get_free_start(loc);
1751         int needed_space = ocfs2_xi_entry_usage(xi);
1752         int size = namevalue_size_xi(xi);
1753         struct super_block *sb = loc->xl_inode->i_sb;
1754
1755         /*
1756          * Bucket storage does not reclaim name+value pairs it cannot
1757          * reuse.  They live as holes until the bucket fills, and then
1758          * the bucket is defragmented.  However, the bucket can reclaim
1759          * the ocfs2_xattr_entry.
1760          */
1761         if (loc->xl_entry) {
1762                 /* Don't need space if we're reusing! */
1763                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1764                         needed_space = 0;
1765                 else
1766                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1767         }
1768         BUG_ON(needed_space < 0);
1769
1770         if (free_start < size) {
1771                 if (needed_space)
1772                         return -ENOSPC;
1773         } else {
1774                 /*
1775                  * First we check if it would fit in the first place.
1776                  * Below, we align the free start to a block.  This may
1777                  * slide us below the minimum gap.  By checking unaligned
1778                  * first, we avoid that error.
1779                  */
1780                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1781                                                  count);
1782                 if (rc)
1783                         return rc;
1784                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1785                                                            size);
1786         }
1787         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1788 }
1789
1790 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1791 {
1792         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1793                      -namevalue_size_xe(loc->xl_entry));
1794 }
1795
1796 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1797 {
1798         struct ocfs2_xattr_header *xh = loc->xl_header;
1799         int count = le16_to_cpu(xh->xh_count);
1800         int low = 0, high = count - 1, tmp;
1801         struct ocfs2_xattr_entry *tmp_xe;
1802
1803         /*
1804          * We keep buckets sorted by name_hash, so we need to find
1805          * our insert place.
1806          */
1807         while (low <= high && count) {
1808                 tmp = (low + high) / 2;
1809                 tmp_xe = &xh->xh_entries[tmp];
1810
1811                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1812                         low = tmp + 1;
1813                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1814                         high = tmp - 1;
1815                 else {
1816                         low = tmp;
1817                         break;
1818                 }
1819         }
1820
1821         if (low != count)
1822                 memmove(&xh->xh_entries[low + 1],
1823                         &xh->xh_entries[low],
1824                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1825
1826         le16_add_cpu(&xh->xh_count, 1);
1827         loc->xl_entry = &xh->xh_entries[low];
1828         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1829 }
1830
1831 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1832 {
1833         int free_start = ocfs2_xa_get_free_start(loc);
1834         struct ocfs2_xattr_header *xh = loc->xl_header;
1835         struct super_block *sb = loc->xl_inode->i_sb;
1836         int nameval_offset;
1837
1838         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1839         nameval_offset = free_start - size;
1840         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1841         xh->xh_free_start = cpu_to_le16(nameval_offset);
1842         le16_add_cpu(&xh->xh_name_value_len, size);
1843
1844 }
1845
1846 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1847                                            struct ocfs2_xattr_value_buf *vb)
1848 {
1849         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1850         struct super_block *sb = loc->xl_inode->i_sb;
1851         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1852         int size = namevalue_size_xe(loc->xl_entry);
1853         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1854
1855         /* Values are not allowed to straddle block boundaries */
1856         BUG_ON(block_offset !=
1857                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1858         /* We expect the bucket to be filled in */
1859         BUG_ON(!bucket->bu_bhs[block_offset]);
1860
1861         vb->vb_access = ocfs2_journal_access;
1862         vb->vb_bh = bucket->bu_bhs[block_offset];
1863 }
1864
1865 /* Operations for xattrs stored in buckets. */
1866 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1867         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1868         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1869         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1870         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1871         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1872         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1873         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1874         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1875         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1876         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1877 };
1878
1879 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1880 {
1881         struct ocfs2_xattr_value_buf vb;
1882
1883         if (ocfs2_xattr_is_local(loc->xl_entry))
1884                 return 0;
1885
1886         ocfs2_xa_fill_value_buf(loc, &vb);
1887         return le32_to_cpu(vb.vb_xv->xr_clusters);
1888 }
1889
1890 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1891                                    struct ocfs2_xattr_set_ctxt *ctxt)
1892 {
1893         int trunc_rc, access_rc;
1894         struct ocfs2_xattr_value_buf vb;
1895
1896         ocfs2_xa_fill_value_buf(loc, &vb);
1897         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1898                                               ctxt);
1899
1900         /*
1901          * The caller of ocfs2_xa_value_truncate() has already called
1902          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1903          * calls ocfs2_extend_trans().  This may commit the previous
1904          * transaction and open a new one.  If this is a bucket, truncate
1905          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1906          * the caller is expecting to dirty the entire bucket.  So we must
1907          * reset the journal work.  We do this even if truncate has failed,
1908          * as it could have failed after committing the extend.
1909          */
1910         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1911                                             OCFS2_JOURNAL_ACCESS_WRITE);
1912
1913         /* Errors in truncate take precedence */
1914         return trunc_rc ? trunc_rc : access_rc;
1915 }
1916
1917 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1918 {
1919         int index, count;
1920         struct ocfs2_xattr_header *xh = loc->xl_header;
1921         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1922
1923         ocfs2_xa_wipe_namevalue(loc);
1924         loc->xl_entry = NULL;
1925
1926         le16_add_cpu(&xh->xh_count, -1);
1927         count = le16_to_cpu(xh->xh_count);
1928
1929         /*
1930          * Only zero out the entry if there are more remaining.  This is
1931          * important for an empty bucket, as it keeps track of the
1932          * bucket's hash value.  It doesn't hurt empty block storage.
1933          */
1934         if (count) {
1935                 index = ((char *)entry - (char *)&xh->xh_entries) /
1936                         sizeof(struct ocfs2_xattr_entry);
1937                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1938                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1939                 memset(&xh->xh_entries[count], 0,
1940                        sizeof(struct ocfs2_xattr_entry));
1941         }
1942 }
1943
1944 /*
1945  * If we have a problem adjusting the size of an external value during
1946  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1947  * in an intermediate state.  For example, the value may be partially
1948  * truncated.
1949  *
1950  * If the value tree hasn't changed, the extend/truncate went nowhere.
1951  * We have nothing to do.  The caller can treat it as a straight error.
1952  *
1953  * If the value tree got partially truncated, we now have a corrupted
1954  * extended attribute.  We're going to wipe its entry and leak the
1955  * clusters.  Better to leak some storage than leave a corrupt entry.
1956  *
1957  * If the value tree grew, it obviously didn't grow enough for the
1958  * new entry.  We're not going to try and reclaim those clusters either.
1959  * If there was already an external value there (orig_clusters != 0),
1960  * the new clusters are attached safely and we can just leave the old
1961  * value in place.  If there was no external value there, we remove
1962  * the entry.
1963  *
1964  * This way, the xattr block we store in the journal will be consistent.
1965  * If the size change broke because of the journal, no changes will hit
1966  * disk anyway.
1967  */
1968 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1969                                             const char *what,
1970                                             unsigned int orig_clusters)
1971 {
1972         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1973         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1974                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1975
1976         if (new_clusters < orig_clusters) {
1977                 mlog(ML_ERROR,
1978                      "Partial truncate while %s xattr %.*s.  Leaking "
1979                      "%u clusters and removing the entry\n",
1980                      what, loc->xl_entry->xe_name_len, nameval_buf,
1981                      orig_clusters - new_clusters);
1982                 ocfs2_xa_remove_entry(loc);
1983         } else if (!orig_clusters) {
1984                 mlog(ML_ERROR,
1985                      "Unable to allocate an external value for xattr "
1986                      "%.*s safely.  Leaking %u clusters and removing the "
1987                      "entry\n",
1988                      loc->xl_entry->xe_name_len, nameval_buf,
1989                      new_clusters - orig_clusters);
1990                 ocfs2_xa_remove_entry(loc);
1991         } else if (new_clusters > orig_clusters)
1992                 mlog(ML_ERROR,
1993                      "Unable to grow xattr %.*s safely.  %u new clusters "
1994                      "have been added, but the value will not be "
1995                      "modified\n",
1996                      loc->xl_entry->xe_name_len, nameval_buf,
1997                      new_clusters - orig_clusters);
1998 }
1999
2000 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2001                            struct ocfs2_xattr_set_ctxt *ctxt)
2002 {
2003         int rc = 0;
2004         unsigned int orig_clusters;
2005
2006         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2007                 orig_clusters = ocfs2_xa_value_clusters(loc);
2008                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2009                 if (rc) {
2010                         mlog_errno(rc);
2011                         /*
2012                          * Since this is remove, we can return 0 if
2013                          * ocfs2_xa_cleanup_value_truncate() is going to
2014                          * wipe the entry anyway.  So we check the
2015                          * cluster count as well.
2016                          */
2017                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2018                                 rc = 0;
2019                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2020                                                         orig_clusters);
2021                         if (rc)
2022                                 goto out;
2023                 }
2024         }
2025
2026         ocfs2_xa_remove_entry(loc);
2027
2028 out:
2029         return rc;
2030 }
2031
2032 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2033 {
2034         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2035         char *nameval_buf;
2036
2037         nameval_buf = ocfs2_xa_offset_pointer(loc,
2038                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2039         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2040 }
2041
2042 /*
2043  * Take an existing entry and make it ready for the new value.  This
2044  * won't allocate space, but it may free space.  It should be ready for
2045  * ocfs2_xa_prepare_entry() to finish the work.
2046  */
2047 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2048                                 struct ocfs2_xattr_info *xi,
2049                                 struct ocfs2_xattr_set_ctxt *ctxt)
2050 {
2051         int rc = 0;
2052         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2053         unsigned int orig_clusters;
2054         char *nameval_buf;
2055         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2056         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2057
2058         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2059                name_size);
2060
2061         nameval_buf = ocfs2_xa_offset_pointer(loc,
2062                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2063         if (xe_local) {
2064                 memset(nameval_buf + name_size, 0,
2065                        namevalue_size_xe(loc->xl_entry) - name_size);
2066                 if (!xi_local)
2067                         ocfs2_xa_install_value_root(loc);
2068         } else {
2069                 orig_clusters = ocfs2_xa_value_clusters(loc);
2070                 if (xi_local) {
2071                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2072                         if (rc < 0)
2073                                 mlog_errno(rc);
2074                         else
2075                                 memset(nameval_buf + name_size, 0,
2076                                        namevalue_size_xe(loc->xl_entry) -
2077                                        name_size);
2078                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2079                            xi->xi_value_len) {
2080                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2081                                                      ctxt);
2082                         if (rc < 0)
2083                                 mlog_errno(rc);
2084                 }
2085
2086                 if (rc) {
2087                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2088                                                         orig_clusters);
2089                         goto out;
2090                 }
2091         }
2092
2093         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2094         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2095
2096 out:
2097         return rc;
2098 }
2099
2100 /*
2101  * Prepares loc->xl_entry to receive the new xattr.  This includes
2102  * properly setting up the name+value pair region.  If loc->xl_entry
2103  * already exists, it will take care of modifying it appropriately.
2104  *
2105  * Note that this modifies the data.  You did journal_access already,
2106  * right?
2107  */
2108 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2109                                   struct ocfs2_xattr_info *xi,
2110                                   u32 name_hash,
2111                                   struct ocfs2_xattr_set_ctxt *ctxt)
2112 {
2113         int rc = 0;
2114         unsigned int orig_clusters;
2115         __le64 orig_value_size = 0;
2116
2117         rc = ocfs2_xa_check_space(loc, xi);
2118         if (rc)
2119                 goto out;
2120
2121         if (loc->xl_entry) {
2122                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2123                         orig_value_size = loc->xl_entry->xe_value_size;
2124                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2125                         if (rc)
2126                                 goto out;
2127                         goto alloc_value;
2128                 }
2129
2130                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2131                         orig_clusters = ocfs2_xa_value_clusters(loc);
2132                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2133                         if (rc) {
2134                                 mlog_errno(rc);
2135                                 ocfs2_xa_cleanup_value_truncate(loc,
2136                                                                 "overwriting",
2137                                                                 orig_clusters);
2138                                 goto out;
2139                         }
2140                 }
2141                 ocfs2_xa_wipe_namevalue(loc);
2142         } else
2143                 ocfs2_xa_add_entry(loc, name_hash);
2144
2145         /*
2146          * If we get here, we have a blank entry.  Fill it.  We grow our
2147          * name+value pair back from the end.
2148          */
2149         ocfs2_xa_add_namevalue(loc, xi);
2150         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2151                 ocfs2_xa_install_value_root(loc);
2152
2153 alloc_value:
2154         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2155                 orig_clusters = ocfs2_xa_value_clusters(loc);
2156                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2157                 if (rc < 0) {
2158                         ctxt->set_abort = 1;
2159                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2160                                                         orig_clusters);
2161                         /*
2162                          * If we were growing an existing value,
2163                          * ocfs2_xa_cleanup_value_truncate() won't remove
2164                          * the entry. We need to restore the original value
2165                          * size.
2166                          */
2167                         if (loc->xl_entry) {
2168                                 BUG_ON(!orig_value_size);
2169                                 loc->xl_entry->xe_value_size = orig_value_size;
2170                         }
2171                         mlog_errno(rc);
2172                 }
2173         }
2174
2175 out:
2176         return rc;
2177 }
2178
2179 /*
2180  * Store the value portion of the name+value pair.  This will skip
2181  * values that are stored externally.  Their tree roots were set up
2182  * by ocfs2_xa_prepare_entry().
2183  */
2184 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2185                                 struct ocfs2_xattr_info *xi,
2186                                 struct ocfs2_xattr_set_ctxt *ctxt)
2187 {
2188         int rc = 0;
2189         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2190         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2191         char *nameval_buf;
2192         struct ocfs2_xattr_value_buf vb;
2193
2194         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2195         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2196                 ocfs2_xa_fill_value_buf(loc, &vb);
2197                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2198                                                      ctxt->handle, &vb,
2199                                                      xi->xi_value,
2200                                                      xi->xi_value_len);
2201         } else
2202                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2203
2204         return rc;
2205 }
2206
2207 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2208                         struct ocfs2_xattr_info *xi,
2209                         struct ocfs2_xattr_set_ctxt *ctxt)
2210 {
2211         int ret;
2212         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2213                                               xi->xi_name_len);
2214
2215         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2216                                       OCFS2_JOURNAL_ACCESS_WRITE);
2217         if (ret) {
2218                 mlog_errno(ret);
2219                 goto out;
2220         }
2221
2222         /*
2223          * From here on out, everything is going to modify the buffer a
2224          * little.  Errors are going to leave the xattr header in a
2225          * sane state.  Thus, even with errors we dirty the sucker.
2226          */
2227
2228         /* Don't worry, we are never called with !xi_value and !xl_entry */
2229         if (!xi->xi_value) {
2230                 ret = ocfs2_xa_remove(loc, ctxt);
2231                 goto out_dirty;
2232         }
2233
2234         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2235         if (ret) {
2236                 if (ret != -ENOSPC)
2237                         mlog_errno(ret);
2238                 goto out_dirty;
2239         }
2240
2241         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2242         if (ret)
2243                 mlog_errno(ret);
2244
2245 out_dirty:
2246         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2247
2248 out:
2249         return ret;
2250 }
2251
2252 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2253                                      struct inode *inode,
2254                                      struct buffer_head *bh,
2255                                      struct ocfs2_xattr_entry *entry)
2256 {
2257         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2258
2259         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2260
2261         loc->xl_inode = inode;
2262         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2263         loc->xl_storage = bh;
2264         loc->xl_entry = entry;
2265         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2266         loc->xl_header =
2267                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2268                                               loc->xl_size);
2269 }
2270
2271 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2272                                           struct inode *inode,
2273                                           struct buffer_head *bh,
2274                                           struct ocfs2_xattr_entry *entry)
2275 {
2276         struct ocfs2_xattr_block *xb =
2277                 (struct ocfs2_xattr_block *)bh->b_data;
2278
2279         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2280
2281         loc->xl_inode = inode;
2282         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2283         loc->xl_storage = bh;
2284         loc->xl_header = &(xb->xb_attrs.xb_header);
2285         loc->xl_entry = entry;
2286         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2287                                              xb_attrs.xb_header);
2288 }
2289
2290 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2291                                            struct ocfs2_xattr_bucket *bucket,
2292                                            struct ocfs2_xattr_entry *entry)
2293 {
2294         loc->xl_inode = bucket->bu_inode;
2295         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2296         loc->xl_storage = bucket;
2297         loc->xl_header = bucket_xh(bucket);
2298         loc->xl_entry = entry;
2299         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2300 }
2301
2302 /*
2303  * In xattr remove, if it is stored outside and refcounted, we may have
2304  * the chance to split the refcount tree. So need the allocators.
2305  */
2306 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2307                                         struct ocfs2_xattr_value_root *xv,
2308                                         struct ocfs2_caching_info *ref_ci,
2309                                         struct buffer_head *ref_root_bh,
2310                                         struct ocfs2_alloc_context **meta_ac,
2311                                         int *ref_credits)
2312 {
2313         int ret, meta_add = 0;
2314         u32 p_cluster, num_clusters;
2315         unsigned int ext_flags;
2316
2317         *ref_credits = 0;
2318         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2319                                        &num_clusters,
2320                                        &xv->xr_list,
2321                                        &ext_flags);
2322         if (ret) {
2323                 mlog_errno(ret);
2324                 goto out;
2325         }
2326
2327         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2328                 goto out;
2329
2330         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2331                                                  ref_root_bh, xv,
2332                                                  &meta_add, ref_credits);
2333         if (ret) {
2334                 mlog_errno(ret);
2335                 goto out;
2336         }
2337
2338         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2339                                                 meta_add, meta_ac);
2340         if (ret)
2341                 mlog_errno(ret);
2342
2343 out:
2344         return ret;
2345 }
2346
2347 static int ocfs2_remove_value_outside(struct inode*inode,
2348                                       struct ocfs2_xattr_value_buf *vb,
2349                                       struct ocfs2_xattr_header *header,
2350                                       struct ocfs2_caching_info *ref_ci,
2351                                       struct buffer_head *ref_root_bh)
2352 {
2353         int ret = 0, i, ref_credits;
2354         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2355         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2356         void *val;
2357
2358         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2359
2360         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2361                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2362
2363                 if (ocfs2_xattr_is_local(entry))
2364                         continue;
2365
2366                 val = (void *)header +
2367                         le16_to_cpu(entry->xe_name_offset);
2368                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2369                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2370
2371                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2372                                                          ref_ci, ref_root_bh,
2373                                                          &ctxt.meta_ac,
2374                                                          &ref_credits);
2375
2376                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2377                                         ocfs2_remove_extent_credits(osb->sb));
2378                 if (IS_ERR(ctxt.handle)) {
2379                         ret = PTR_ERR(ctxt.handle);
2380                         mlog_errno(ret);
2381                         break;
2382                 }
2383
2384                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2385
2386                 ocfs2_commit_trans(osb, ctxt.handle);
2387                 if (ctxt.meta_ac) {
2388                         ocfs2_free_alloc_context(ctxt.meta_ac);
2389                         ctxt.meta_ac = NULL;
2390                 }
2391
2392                 if (ret < 0) {
2393                         mlog_errno(ret);
2394                         break;
2395                 }
2396
2397         }
2398
2399         if (ctxt.meta_ac)
2400                 ocfs2_free_alloc_context(ctxt.meta_ac);
2401         ocfs2_schedule_truncate_log_flush(osb, 1);
2402         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2403         return ret;
2404 }
2405
2406 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2407                                     struct buffer_head *di_bh,
2408                                     struct ocfs2_caching_info *ref_ci,
2409                                     struct buffer_head *ref_root_bh)
2410 {
2411
2412         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2413         struct ocfs2_xattr_header *header;
2414         int ret;
2415         struct ocfs2_xattr_value_buf vb = {
2416                 .vb_bh = di_bh,
2417                 .vb_access = ocfs2_journal_access_di,
2418         };
2419
2420         header = (struct ocfs2_xattr_header *)
2421                  ((void *)di + inode->i_sb->s_blocksize -
2422                  le16_to_cpu(di->i_xattr_inline_size));
2423
2424         ret = ocfs2_remove_value_outside(inode, &vb, header,
2425                                          ref_ci, ref_root_bh);
2426
2427         return ret;
2428 }
2429
2430 struct ocfs2_rm_xattr_bucket_para {
2431         struct ocfs2_caching_info *ref_ci;
2432         struct buffer_head *ref_root_bh;
2433 };
2434
2435 static int ocfs2_xattr_block_remove(struct inode *inode,
2436                                     struct buffer_head *blk_bh,
2437                                     struct ocfs2_caching_info *ref_ci,
2438                                     struct buffer_head *ref_root_bh)
2439 {
2440         struct ocfs2_xattr_block *xb;
2441         int ret = 0;
2442         struct ocfs2_xattr_value_buf vb = {
2443                 .vb_bh = blk_bh,
2444                 .vb_access = ocfs2_journal_access_xb,
2445         };
2446         struct ocfs2_rm_xattr_bucket_para args = {
2447                 .ref_ci = ref_ci,
2448                 .ref_root_bh = ref_root_bh,
2449         };
2450
2451         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2452         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2453                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2454                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2455                                                  ref_ci, ref_root_bh);
2456         } else
2457                 ret = ocfs2_iterate_xattr_index_block(inode,
2458                                                 blk_bh,
2459                                                 ocfs2_rm_xattr_cluster,
2460                                                 &args);
2461
2462         return ret;
2463 }
2464
2465 static int ocfs2_xattr_free_block(struct inode *inode,
2466                                   u64 block,
2467                                   struct ocfs2_caching_info *ref_ci,
2468                                   struct buffer_head *ref_root_bh)
2469 {
2470         struct inode *xb_alloc_inode;
2471         struct buffer_head *xb_alloc_bh = NULL;
2472         struct buffer_head *blk_bh = NULL;
2473         struct ocfs2_xattr_block *xb;
2474         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2475         handle_t *handle;
2476         int ret = 0;
2477         u64 blk, bg_blkno;
2478         u16 bit;
2479
2480         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2481         if (ret < 0) {
2482                 mlog_errno(ret);
2483                 goto out;
2484         }
2485
2486         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2487         if (ret < 0) {
2488                 mlog_errno(ret);
2489                 goto out;
2490         }
2491
2492         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2493         blk = le64_to_cpu(xb->xb_blkno);
2494         bit = le16_to_cpu(xb->xb_suballoc_bit);
2495         if (xb->xb_suballoc_loc)
2496                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2497         else
2498                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2499
2500         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2501                                 EXTENT_ALLOC_SYSTEM_INODE,
2502                                 le16_to_cpu(xb->xb_suballoc_slot));
2503         if (!xb_alloc_inode) {
2504                 ret = -ENOMEM;
2505                 mlog_errno(ret);
2506                 goto out;
2507         }
2508         mutex_lock(&xb_alloc_inode->i_mutex);
2509
2510         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2511         if (ret < 0) {
2512                 mlog_errno(ret);
2513                 goto out_mutex;
2514         }
2515
2516         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2517         if (IS_ERR(handle)) {
2518                 ret = PTR_ERR(handle);
2519                 mlog_errno(ret);
2520                 goto out_unlock;
2521         }
2522
2523         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2524                                        bit, bg_blkno, 1);
2525         if (ret < 0)
2526                 mlog_errno(ret);
2527
2528         ocfs2_commit_trans(osb, handle);
2529 out_unlock:
2530         ocfs2_inode_unlock(xb_alloc_inode, 1);
2531         brelse(xb_alloc_bh);
2532 out_mutex:
2533         mutex_unlock(&xb_alloc_inode->i_mutex);
2534         iput(xb_alloc_inode);
2535 out:
2536         brelse(blk_bh);
2537         return ret;
2538 }
2539
2540 /*
2541  * ocfs2_xattr_remove()
2542  *
2543  * Free extended attribute resources associated with this inode.
2544  */
2545 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2546 {
2547         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2548         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2549         struct ocfs2_refcount_tree *ref_tree = NULL;
2550         struct buffer_head *ref_root_bh = NULL;
2551         struct ocfs2_caching_info *ref_ci = NULL;
2552         handle_t *handle;
2553         int ret;
2554
2555         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2556                 return 0;
2557
2558         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2559                 return 0;
2560
2561         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2562                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2563                                                le64_to_cpu(di->i_refcount_loc),
2564                                                1, &ref_tree, &ref_root_bh);
2565                 if (ret) {
2566                         mlog_errno(ret);
2567                         goto out;
2568                 }
2569                 ref_ci = &ref_tree->rf_ci;
2570
2571         }
2572
2573         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2574                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2575                                                ref_ci, ref_root_bh);
2576                 if (ret < 0) {
2577                         mlog_errno(ret);
2578                         goto out;
2579                 }
2580         }
2581
2582         if (di->i_xattr_loc) {
2583                 ret = ocfs2_xattr_free_block(inode,
2584                                              le64_to_cpu(di->i_xattr_loc),
2585                                              ref_ci, ref_root_bh);
2586                 if (ret < 0) {
2587                         mlog_errno(ret);
2588                         goto out;
2589                 }
2590         }
2591
2592         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2593                                    OCFS2_INODE_UPDATE_CREDITS);
2594         if (IS_ERR(handle)) {
2595                 ret = PTR_ERR(handle);
2596                 mlog_errno(ret);
2597                 goto out;
2598         }
2599         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2600                                       OCFS2_JOURNAL_ACCESS_WRITE);
2601         if (ret) {
2602                 mlog_errno(ret);
2603                 goto out_commit;
2604         }
2605
2606         di->i_xattr_loc = 0;
2607
2608         spin_lock(&oi->ip_lock);
2609         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2610         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2611         spin_unlock(&oi->ip_lock);
2612         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2613
2614         ocfs2_journal_dirty(handle, di_bh);
2615 out_commit:
2616         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2617 out:
2618         if (ref_tree)
2619                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2620         brelse(ref_root_bh);
2621         return ret;
2622 }
2623
2624 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2625                                         struct ocfs2_dinode *di)
2626 {
2627         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2628         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2629         int free;
2630
2631         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2632                 return 0;
2633
2634         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2635                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2636                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2637         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2638                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2639                         le64_to_cpu(di->i_size);
2640         } else {
2641                 struct ocfs2_extent_list *el = &di->id2.i_list;
2642                 free = (le16_to_cpu(el->l_count) -
2643                         le16_to_cpu(el->l_next_free_rec)) *
2644                         sizeof(struct ocfs2_extent_rec);
2645         }
2646         if (free >= xattrsize)
2647                 return 1;
2648
2649         return 0;
2650 }
2651
2652 /*
2653  * ocfs2_xattr_ibody_find()
2654  *
2655  * Find extended attribute in inode block and
2656  * fill search info into struct ocfs2_xattr_search.
2657  */
2658 static int ocfs2_xattr_ibody_find(struct inode *inode,
2659                                   int name_index,
2660                                   const char *name,
2661                                   struct ocfs2_xattr_search *xs)
2662 {
2663         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2664         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2665         int ret;
2666         int has_space = 0;
2667
2668         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2669                 return 0;
2670
2671         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2672                 down_read(&oi->ip_alloc_sem);
2673                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2674                 up_read(&oi->ip_alloc_sem);
2675                 if (!has_space)
2676                         return 0;
2677         }
2678
2679         xs->xattr_bh = xs->inode_bh;
2680         xs->end = (void *)di + inode->i_sb->s_blocksize;
2681         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2682                 xs->header = (struct ocfs2_xattr_header *)
2683                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2684         else
2685                 xs->header = (struct ocfs2_xattr_header *)
2686                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2687         xs->base = (void *)xs->header;
2688         xs->here = xs->header->xh_entries;
2689
2690         /* Find the named attribute. */
2691         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2692                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2693                 if (ret && ret != -ENODATA)
2694                         return ret;
2695                 xs->not_found = ret;
2696         }
2697
2698         return 0;
2699 }
2700
2701 static int ocfs2_xattr_ibody_init(struct inode *inode,
2702                                   struct buffer_head *di_bh,
2703                                   struct ocfs2_xattr_set_ctxt *ctxt)
2704 {
2705         int ret;
2706         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2707         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2709         unsigned int xattrsize = osb->s_xattr_inline_size;
2710
2711         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2712                 ret = -ENOSPC;
2713                 goto out;
2714         }
2715
2716         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2717                                       OCFS2_JOURNAL_ACCESS_WRITE);
2718         if (ret) {
2719                 mlog_errno(ret);
2720                 goto out;
2721         }
2722
2723         /*
2724          * Adjust extent record count or inline data size
2725          * to reserve space for extended attribute.
2726          */
2727         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2728                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2729                 le16_add_cpu(&idata->id_count, -xattrsize);
2730         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2731                 struct ocfs2_extent_list *el = &di->id2.i_list;
2732                 le16_add_cpu(&el->l_count, -(xattrsize /
2733                                              sizeof(struct ocfs2_extent_rec)));
2734         }
2735         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2736
2737         spin_lock(&oi->ip_lock);
2738         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2739         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2740         spin_unlock(&oi->ip_lock);
2741
2742         ocfs2_journal_dirty(ctxt->handle, di_bh);
2743
2744 out:
2745         return ret;
2746 }
2747
2748 /*
2749  * ocfs2_xattr_ibody_set()
2750  *
2751  * Set, replace or remove an extended attribute into inode block.
2752  *
2753  */
2754 static int ocfs2_xattr_ibody_set(struct inode *inode,
2755                                  struct ocfs2_xattr_info *xi,
2756                                  struct ocfs2_xattr_search *xs,
2757                                  struct ocfs2_xattr_set_ctxt *ctxt)
2758 {
2759         int ret;
2760         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2761         struct ocfs2_xa_loc loc;
2762
2763         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2764                 return -ENOSPC;
2765
2766         down_write(&oi->ip_alloc_sem);
2767         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2768                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2769                 if (ret) {
2770                         if (ret != -ENOSPC)
2771                                 mlog_errno(ret);
2772                         goto out;
2773                 }
2774         }
2775
2776         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2777                                  xs->not_found ? NULL : xs->here);
2778         ret = ocfs2_xa_set(&loc, xi, ctxt);
2779         if (ret) {
2780                 if (ret != -ENOSPC)
2781                         mlog_errno(ret);
2782                 goto out;
2783         }
2784         xs->here = loc.xl_entry;
2785
2786 out:
2787         up_write(&oi->ip_alloc_sem);
2788
2789         return ret;
2790 }
2791
2792 /*
2793  * ocfs2_xattr_block_find()
2794  *
2795  * Find extended attribute in external block and
2796  * fill search info into struct ocfs2_xattr_search.
2797  */
2798 static int ocfs2_xattr_block_find(struct inode *inode,
2799                                   int name_index,
2800                                   const char *name,
2801                                   struct ocfs2_xattr_search *xs)
2802 {
2803         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2804         struct buffer_head *blk_bh = NULL;
2805         struct ocfs2_xattr_block *xb;
2806         int ret = 0;
2807
2808         if (!di->i_xattr_loc)
2809                 return ret;
2810
2811         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2812                                      &blk_bh);
2813         if (ret < 0) {
2814                 mlog_errno(ret);
2815                 return ret;
2816         }
2817
2818         xs->xattr_bh = blk_bh;
2819         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2820
2821         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2822                 xs->header = &xb->xb_attrs.xb_header;
2823                 xs->base = (void *)xs->header;
2824                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2825                 xs->here = xs->header->xh_entries;
2826
2827                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2828         } else
2829                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2830                                                    name_index,
2831                                                    name, xs);
2832
2833         if (ret && ret != -ENODATA) {
2834                 xs->xattr_bh = NULL;
2835                 goto cleanup;
2836         }
2837         xs->not_found = ret;
2838         return 0;
2839 cleanup:
2840         brelse(blk_bh);
2841
2842         return ret;
2843 }
2844
2845 static int ocfs2_create_xattr_block(struct inode *inode,
2846                                     struct buffer_head *inode_bh,
2847                                     struct ocfs2_xattr_set_ctxt *ctxt,
2848                                     int indexed,
2849                                     struct buffer_head **ret_bh)
2850 {
2851         int ret;
2852         u16 suballoc_bit_start;
2853         u32 num_got;
2854         u64 suballoc_loc, first_blkno;
2855         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2856         struct buffer_head *new_bh = NULL;
2857         struct ocfs2_xattr_block *xblk;
2858
2859         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2860                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2861         if (ret < 0) {
2862                 mlog_errno(ret);
2863                 goto end;
2864         }
2865
2866         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2867                                    &suballoc_loc, &suballoc_bit_start,
2868                                    &num_got, &first_blkno);
2869         if (ret < 0) {
2870                 mlog_errno(ret);
2871                 goto end;
2872         }
2873
2874         new_bh = sb_getblk(inode->i_sb, first_blkno);
2875         if (!new_bh) {
2876                 ret = -ENOMEM;
2877                 mlog_errno(ret);
2878                 goto end;
2879         }
2880
2881         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2882
2883         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2884                                       new_bh,
2885                                       OCFS2_JOURNAL_ACCESS_CREATE);
2886         if (ret < 0) {
2887                 mlog_errno(ret);
2888                 goto end;
2889         }
2890
2891         /* Initialize ocfs2_xattr_block */
2892         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2893         memset(xblk, 0, inode->i_sb->s_blocksize);
2894         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2895         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2896         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2897         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2898         xblk->xb_fs_generation =
2899                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2900         xblk->xb_blkno = cpu_to_le64(first_blkno);
2901         if (indexed) {
2902                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2903                 xr->xt_clusters = cpu_to_le32(1);
2904                 xr->xt_last_eb_blk = 0;
2905                 xr->xt_list.l_tree_depth = 0;
2906                 xr->xt_list.l_count = cpu_to_le16(
2907                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2908                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2909                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2910         }
2911         ocfs2_journal_dirty(ctxt->handle, new_bh);
2912
2913         /* Add it to the inode */
2914         di->i_xattr_loc = cpu_to_le64(first_blkno);
2915
2916         spin_lock(&OCFS2_I(inode)->ip_lock);
2917         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2918         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2919         spin_unlock(&OCFS2_I(inode)->ip_lock);
2920
2921         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2922
2923         *ret_bh = new_bh;
2924         new_bh = NULL;
2925
2926 end:
2927         brelse(new_bh);
2928         return ret;
2929 }
2930
2931 /*
2932  * ocfs2_xattr_block_set()
2933  *
2934  * Set, replace or remove an extended attribute into external block.
2935  *
2936  */
2937 static int ocfs2_xattr_block_set(struct inode *inode,
2938                                  struct ocfs2_xattr_info *xi,
2939                                  struct ocfs2_xattr_search *xs,
2940                                  struct ocfs2_xattr_set_ctxt *ctxt)
2941 {
2942         struct buffer_head *new_bh = NULL;
2943         struct ocfs2_xattr_block *xblk = NULL;
2944         int ret;
2945         struct ocfs2_xa_loc loc;
2946
2947         if (!xs->xattr_bh) {
2948                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2949                                                0, &new_bh);
2950                 if (ret) {
2951                         mlog_errno(ret);
2952                         goto end;
2953                 }
2954
2955                 xs->xattr_bh = new_bh;
2956                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2957                 xs->header = &xblk->xb_attrs.xb_header;
2958                 xs->base = (void *)xs->header;
2959                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2960                 xs->here = xs->header->xh_entries;
2961         } else
2962                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2963
2964         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2965                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2966                                               xs->not_found ? NULL : xs->here);
2967
2968                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2969                 if (!ret)
2970                         xs->here = loc.xl_entry;
2971                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2972                         goto end;
2973                 else {
2974                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2975                         if (ret)
2976                                 goto end;
2977                 }
2978         }
2979
2980         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2981                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2982
2983 end:
2984         return ret;
2985 }
2986
2987 /* Check whether the new xattr can be inserted into the inode. */
2988 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2989                                        struct ocfs2_xattr_info *xi,
2990                                        struct ocfs2_xattr_search *xs)
2991 {
2992         struct ocfs2_xattr_entry *last;
2993         int free, i;
2994         size_t min_offs = xs->end - xs->base;
2995
2996         if (!xs->header)
2997                 return 0;
2998
2999         last = xs->header->xh_entries;
3000
3001         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3002                 size_t offs = le16_to_cpu(last->xe_name_offset);
3003                 if (offs < min_offs)
3004                         min_offs = offs;
3005                 last += 1;
3006         }
3007
3008         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3009         if (free < 0)
3010                 return 0;
3011
3012         BUG_ON(!xs->not_found);
3013
3014         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3015                 return 1;
3016
3017         return 0;
3018 }
3019
3020 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3021                                      struct ocfs2_dinode *di,
3022                                      struct ocfs2_xattr_info *xi,
3023                                      struct ocfs2_xattr_search *xis,
3024                                      struct ocfs2_xattr_search *xbs,
3025                                      int *clusters_need,
3026                                      int *meta_need,
3027                                      int *credits_need)
3028 {
3029         int ret = 0, old_in_xb = 0;
3030         int clusters_add = 0, meta_add = 0, credits = 0;
3031         struct buffer_head *bh = NULL;
3032         struct ocfs2_xattr_block *xb = NULL;
3033         struct ocfs2_xattr_entry *xe = NULL;
3034         struct ocfs2_xattr_value_root *xv = NULL;
3035         char *base = NULL;
3036         int name_offset, name_len = 0;
3037         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3038                                                     xi->xi_value_len);
3039         u64 value_size;
3040
3041         /*
3042          * Calculate the clusters we need to write.
3043          * No matter whether we replace an old one or add a new one,
3044          * we need this for writing.
3045          */
3046         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3047                 credits += new_clusters *
3048                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3049
3050         if (xis->not_found && xbs->not_found) {
3051                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3052
3053                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3054                         clusters_add += new_clusters;
3055                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3056                                                         &def_xv.xv.xr_list);
3057                 }
3058
3059                 goto meta_guess;
3060         }
3061
3062         if (!xis->not_found) {
3063                 xe = xis->here;
3064                 name_offset = le16_to_cpu(xe->xe_name_offset);
3065                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3066                 base = xis->base;
3067                 credits += OCFS2_INODE_UPDATE_CREDITS;
3068         } else {
3069                 int i, block_off = 0;
3070                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3071                 xe = xbs->here;
3072                 name_offset = le16_to_cpu(xe->xe_name_offset);
3073                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3074                 i = xbs->here - xbs->header->xh_entries;
3075                 old_in_xb = 1;
3076
3077                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3078                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3079                                                         bucket_xh(xbs->bucket),
3080                                                         i, &block_off,
3081                                                         &name_offset);
3082                         base = bucket_block(xbs->bucket, block_off);
3083                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3084                 } else {
3085                         base = xbs->base;
3086                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3087                 }
3088         }
3089
3090         /*
3091          * delete a xattr doesn't need metadata and cluster allocation.
3092          * so just calculate the credits and return.
3093          *
3094          * The credits for removing the value tree will be extended
3095          * by ocfs2_remove_extent itself.
3096          */
3097         if (!xi->xi_value) {
3098                 if (!ocfs2_xattr_is_local(xe))
3099                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3100
3101                 goto out;
3102         }
3103
3104         /* do cluster allocation guess first. */
3105         value_size = le64_to_cpu(xe->xe_value_size);
3106
3107         if (old_in_xb) {
3108                 /*
3109                  * In xattr set, we always try to set the xe in inode first,
3110                  * so if it can be inserted into inode successfully, the old
3111                  * one will be removed from the xattr block, and this xattr
3112                  * will be inserted into inode as a new xattr in inode.
3113                  */
3114                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3115                         clusters_add += new_clusters;
3116                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3117                                     OCFS2_INODE_UPDATE_CREDITS;
3118                         if (!ocfs2_xattr_is_local(xe))
3119                                 credits += ocfs2_calc_extend_credits(
3120                                                         inode->i_sb,
3121                                                         &def_xv.xv.xr_list);
3122                         goto out;
3123                 }
3124         }
3125
3126         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3127                 /* the new values will be stored outside. */
3128                 u32 old_clusters = 0;
3129
3130                 if (!ocfs2_xattr_is_local(xe)) {
3131                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3132                                                                  value_size);
3133                         xv = (struct ocfs2_xattr_value_root *)
3134                              (base + name_offset + name_len);
3135                         value_size = OCFS2_XATTR_ROOT_SIZE;
3136                 } else
3137                         xv = &def_xv.xv;
3138
3139                 if (old_clusters >= new_clusters) {
3140                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3141                         goto out;
3142                 } else {
3143                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3144                         clusters_add += new_clusters - old_clusters;
3145                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3146                                                              &xv->xr_list);
3147                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3148                                 goto out;
3149                 }
3150         } else {
3151                 /*
3152                  * Now the new value will be stored inside. So if the new
3153                  * value is smaller than the size of value root or the old
3154                  * value, we don't need any allocation, otherwise we have
3155                  * to guess metadata allocation.
3156                  */
3157                 if ((ocfs2_xattr_is_local(xe) &&
3158                      (value_size >= xi->xi_value_len)) ||
3159                     (!ocfs2_xattr_is_local(xe) &&
3160                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3161                         goto out;
3162         }
3163
3164 meta_guess:
3165         /* calculate metadata allocation. */
3166         if (di->i_xattr_loc) {
3167                 if (!xbs->xattr_bh) {
3168                         ret = ocfs2_read_xattr_block(inode,
3169                                                      le64_to_cpu(di->i_xattr_loc),
3170                                                      &bh);
3171                         if (ret) {
3172                                 mlog_errno(ret);
3173                                 goto out;
3174                         }
3175
3176                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3177                 } else
3178                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3179
3180                 /*
3181                  * If there is already an xattr tree, good, we can calculate
3182                  * like other b-trees. Otherwise we may have the chance of
3183                  * create a tree, the credit calculation is borrowed from
3184                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3185                  * new tree will be cluster based, so no meta is needed.
3186                  */
3187                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3188                         struct ocfs2_extent_list *el =
3189                                  &xb->xb_attrs.xb_root.xt_list;
3190                         meta_add += ocfs2_extend_meta_needed(el);
3191                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3192                                                              el);
3193                 } else
3194                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3195
3196                 /*
3197                  * This cluster will be used either for new bucket or for
3198                  * new xattr block.
3199                  * If the cluster size is the same as the bucket size, one
3200                  * more is needed since we may need to extend the bucket
3201                  * also.
3202                  */
3203                 clusters_add += 1;
3204                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3205                 if (OCFS2_XATTR_BUCKET_SIZE ==
3206                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3207                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3208                         clusters_add += 1;
3209                 }
3210         } else {
3211                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3212                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3213                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3214                         meta_add += ocfs2_extend_meta_needed(el);
3215                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3216                                                              el);
3217                 } else {
3218                         meta_add += 1;
3219                 }
3220         }
3221 out:
3222         if (clusters_need)
3223                 *clusters_need = clusters_add;
3224         if (meta_need)
3225                 *meta_need = meta_add;
3226         if (credits_need)
3227                 *credits_need = credits;
3228         brelse(bh);
3229         return ret;
3230 }
3231
3232 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3233                                      struct ocfs2_dinode *di,
3234                                      struct ocfs2_xattr_info *xi,
3235                                      struct ocfs2_xattr_search *xis,
3236                                      struct ocfs2_xattr_search *xbs,
3237                                      struct ocfs2_xattr_set_ctxt *ctxt,
3238                                      int extra_meta,
3239                                      int *credits)
3240 {
3241         int clusters_add, meta_add, ret;
3242         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3243
3244         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3245
3246         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3247
3248         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3249                                         &clusters_add, &meta_add, credits);
3250         if (ret) {
3251                 mlog_errno(ret);
3252                 return ret;
3253         }
3254
3255         meta_add += extra_meta;
3256         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3257                                         clusters_add, *credits);
3258
3259         if (meta_add) {
3260                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3261                                                         &ctxt->meta_ac);
3262                 if (ret) {
3263                         mlog_errno(ret);
3264                         goto out;
3265                 }
3266         }
3267
3268         if (clusters_add) {
3269                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3270                 if (ret)
3271                         mlog_errno(ret);
3272         }
3273 out:
3274         if (ret) {
3275                 if (ctxt->meta_ac) {
3276                         ocfs2_free_alloc_context(ctxt->meta_ac);
3277                         ctxt->meta_ac = NULL;
3278                 }
3279
3280                 /*
3281                  * We cannot have an error and a non null ctxt->data_ac.
3282                  */
3283         }
3284
3285         return ret;
3286 }
3287
3288 static int __ocfs2_xattr_set_handle(struct inode *inode,
3289                                     struct ocfs2_dinode *di,
3290                                     struct ocfs2_xattr_info *xi,
3291                                     struct ocfs2_xattr_search *xis,
3292                                     struct ocfs2_xattr_search *xbs,
3293                                     struct ocfs2_xattr_set_ctxt *ctxt)
3294 {
3295         int ret = 0, credits, old_found;
3296
3297         if (!xi->xi_value) {
3298                 /* Remove existing extended attribute */
3299                 if (!xis->not_found)
3300                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3301                 else if (!xbs->not_found)
3302                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3303         } else {
3304                 /* We always try to set extended attribute into inode first*/
3305                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3306                 if (!ret && !xbs->not_found) {
3307                         /*
3308                          * If succeed and that extended attribute existing in
3309                          * external block, then we will remove it.
3310                          */
3311                         xi->xi_value = NULL;
3312                         xi->xi_value_len = 0;
3313
3314                         old_found = xis->not_found;
3315                         xis->not_found = -ENODATA;
3316                         ret = ocfs2_calc_xattr_set_need(inode,
3317                                                         di,
3318                                                         xi,
3319                                                         xis,
3320                                                         xbs,
3321                                                         NULL,
3322                                                         NULL,
3323                                                         &credits);
3324                         xis->not_found = old_found;
3325                         if (ret) {
3326                                 mlog_errno(ret);
3327                                 goto out;
3328                         }
3329
3330                         ret = ocfs2_extend_trans(ctxt->handle, credits);
3331                         if (ret) {
3332                                 mlog_errno(ret);
3333                                 goto out;
3334                         }
3335                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3336                 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3337                         if (di->i_xattr_loc && !xbs->xattr_bh) {
3338                                 ret = ocfs2_xattr_block_find(inode,
3339                                                              xi->xi_name_index,
3340                                                              xi->xi_name, xbs);
3341                                 if (ret)
3342                                         goto out;
3343
3344                                 old_found = xis->not_found;
3345                                 xis->not_found = -ENODATA;
3346                                 ret = ocfs2_calc_xattr_set_need(inode,
3347                                                                 di,
3348                                                                 xi,
3349                                                                 xis,
3350                                                                 xbs,
3351                                                                 NULL,
3352                                                                 NULL,
3353                                                                 &credits);
3354                                 xis->not_found = old_found;
3355                                 if (ret) {
3356                                         mlog_errno(ret);
3357                                         goto out;
3358                                 }
3359
3360                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3361                                 if (ret) {
3362                                         mlog_errno(ret);
3363                                         goto out;
3364                                 }
3365                         }
3366                         /*
3367                          * If no space in inode, we will set extended attribute
3368                          * into external block.
3369                          */
3370                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3371                         if (ret)
3372                                 goto out;
3373                         if (!xis->not_found) {
3374                                 /*
3375                                  * If succeed and that extended attribute
3376                                  * existing in inode, we will remove it.
3377                                  */
3378                                 xi->xi_value = NULL;
3379                                 xi->xi_value_len = 0;
3380                                 xbs->not_found = -ENODATA;
3381                                 ret = ocfs2_calc_xattr_set_need(inode,
3382                                                                 di,
3383                                                                 xi,
3384                                                                 xis,
3385                                                                 xbs,
3386                                                                 NULL,
3387                                                                 NULL,
3388                                                                 &credits);
3389                                 if (ret) {
3390                                         mlog_errno(ret);
3391                                         goto out;
3392                                 }
3393
3394                                 ret = ocfs2_extend_trans(ctxt->handle, credits);
3395                                 if (ret) {
3396                                         mlog_errno(ret);
3397                                         goto out;
3398                                 }
3399                                 ret = ocfs2_xattr_ibody_set(inode, xi,
3400                                                             xis, ctxt);
3401                         }
3402                 }
3403         }
3404
3405         if (!ret) {
3406                 /* Update inode ctime. */
3407                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3408                                               xis->inode_bh,
3409                                               OCFS2_JOURNAL_ACCESS_WRITE);
3410                 if (ret) {
3411                         mlog_errno(ret);
3412                         goto out;
3413                 }
3414
3415                 inode->i_ctime = CURRENT_TIME;
3416                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3417                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3418                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3419         }
3420 out:
3421         return ret;
3422 }
3423
3424 /*
3425  * This function only called duing creating inode
3426  * for init security/acl xattrs of the new inode.
3427  * All transanction credits have been reserved in mknod.
3428  */
3429 int ocfs2_xattr_set_handle(handle_t *handle,
3430                            struct inode *inode,
3431                            struct buffer_head *di_bh,
3432                            int name_index,
3433                            const char *name,
3434                            const void *value,
3435                            size_t value_len,
3436                            int flags,
3437                            struct ocfs2_alloc_context *meta_ac,
3438                            struct ocfs2_alloc_context *data_ac)
3439 {
3440         struct ocfs2_dinode *di;
3441         int ret;
3442
3443         struct ocfs2_xattr_info xi = {
3444                 .xi_name_index = name_index,
3445                 .xi_name = name,
3446                 .xi_name_len = strlen(name),
3447                 .xi_value = value,
3448                 .xi_value_len = value_len,
3449         };
3450
3451         struct ocfs2_xattr_search xis = {
3452                 .not_found = -ENODATA,
3453         };
3454
3455         struct ocfs2_xattr_search xbs = {
3456                 .not_found = -ENODATA,
3457         };
3458
3459         struct ocfs2_xattr_set_ctxt ctxt = {
3460                 .handle = handle,
3461                 .meta_ac = meta_ac,
3462                 .data_ac = data_ac,
3463         };
3464
3465         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3466                 return -EOPNOTSUPP;
3467
3468         /*
3469          * In extreme situation, may need xattr bucket when
3470          * block size is too small. And we have already reserved
3471          * the credits for bucket in mknod.
3472          */
3473         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3474                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3475                 if (!xbs.bucket) {
3476                         mlog_errno(-ENOMEM);
3477                         return -ENOMEM;
3478                 }
3479         }
3480
3481         xis.inode_bh = xbs.inode_bh = di_bh;
3482         di = (struct ocfs2_dinode *)di_bh->b_data;
3483
3484         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3485
3486         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3487         if (ret)
3488                 goto cleanup;
3489         if (xis.not_found) {
3490                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3491                 if (ret)
3492                         goto cleanup;
3493         }
3494
3495         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3496
3497 cleanup:
3498         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3499         brelse(xbs.xattr_bh);
3500         ocfs2_xattr_bucket_free(xbs.bucket);
3501
3502         return ret;
3503 }
3504
3505 /*
3506  * ocfs2_xattr_set()
3507  *
3508  * Set, replace or remove an extended attribute for this inode.
3509  * value is NULL to remove an existing extended attribute, else either
3510  * create or replace an extended attribute.
3511  */
3512 int ocfs2_xattr_set(struct inode *inode,
3513                     int name_index,
3514                     const char *name,
3515                     const void *value,
3516                     size_t value_len,
3517                     int flags)
3518 {
3519         struct buffer_head *di_bh = NULL;
3520         struct ocfs2_dinode *di;
3521         int ret, credits, ref_meta = 0, ref_credits = 0;
3522         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3523         struct inode *tl_inode = osb->osb_tl_inode;
3524         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3525         struct ocfs2_refcount_tree *ref_tree = NULL;
3526
3527         struct ocfs2_xattr_info xi = {
3528                 .xi_name_index = name_index,
3529                 .xi_name = name,
3530                 .xi_name_len = strlen(name),
3531                 .xi_value = value,
3532                 .xi_value_len = value_len,
3533         };
3534
3535         struct ocfs2_xattr_search xis = {
3536                 .not_found = -ENODATA,
3537         };
3538
3539         struct ocfs2_xattr_search xbs = {
3540                 .not_found = -ENODATA,
3541         };
3542
3543         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3544                 return -EOPNOTSUPP;
3545
3546         /*
3547          * Only xbs will be used on indexed trees.  xis doesn't need a
3548          * bucket.
3549          */
3550         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3551         if (!xbs.bucket) {
3552                 mlog_errno(-ENOMEM);
3553                 return -ENOMEM;
3554         }
3555
3556         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3557         if (ret < 0) {
3558                 mlog_errno(ret);
3559                 goto cleanup_nolock;
3560         }
3561         xis.inode_bh = xbs.inode_bh = di_bh;
3562         di = (struct ocfs2_dinode *)di_bh->b_data;
3563
3564         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3565         /*
3566          * Scan inode and external block to find the same name
3567          * extended attribute and collect search information.
3568          */
3569         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3570         if (ret)
3571                 goto cleanup;
3572         if (xis.not_found) {
3573                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3574                 if (ret)
3575                         goto cleanup;
3576         }
3577
3578         if (xis.not_found && xbs.not_found) {
3579                 ret = -ENODATA;
3580                 if (flags & XATTR_REPLACE)
3581                         goto cleanup;
3582                 ret = 0;
3583                 if (!value)
3584                         goto cleanup;
3585         } else {
3586                 ret = -EEXIST;
3587                 if (flags & XATTR_CREATE)
3588                         goto cleanup;
3589         }
3590
3591         /* Check whether the value is refcounted and do some preparation. */
3592         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3593             (!xis.not_found || !xbs.not_found)) {
3594                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3595                                                    &xis, &xbs, &ref_tree,
3596                                                    &ref_meta, &ref_credits);
3597                 if (ret) {
3598                         mlog_errno(ret);
3599                         goto cleanup;
3600                 }
3601         }
3602
3603         mutex_lock(&tl_inode->i_mutex);
3604
3605         if (ocfs2_truncate_log_needs_flush(osb)) {
3606                 ret = __ocfs2_flush_truncate_log(osb);
3607                 if (ret < 0) {
3608                         mutex_unlock(&tl_inode->i_mutex);
3609                         mlog_errno(ret);
3610                         goto cleanup;
3611                 }
3612         }
3613         mutex_unlock(&tl_inode->i_mutex);
3614
3615         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3616                                         &xbs, &ctxt, ref_meta, &credits);
3617         if (ret) {
3618                 mlog_errno(ret);
3619                 goto cleanup;
3620         }
3621
3622         /* we need to update inode's ctime field, so add credit for it. */
3623         credits += OCFS2_INODE_UPDATE_CREDITS;
3624         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3625         if (IS_ERR(ctxt.handle)) {
3626                 ret = PTR_ERR(ctxt.handle);
3627                 mlog_errno(ret);
3628                 goto out_free_ac;
3629         }
3630
3631         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3632         ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3633
3634         ocfs2_commit_trans(osb, ctxt.handle);
3635
3636 out_free_ac:
3637         if (ctxt.data_ac)
3638                 ocfs2_free_alloc_context(ctxt.data_ac);
3639         if (ctxt.meta_ac)
3640                 ocfs2_free_alloc_context(ctxt.meta_ac);
3641         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3642                 ocfs2_schedule_truncate_log_flush(osb, 1);
3643         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3644
3645 cleanup:
3646         if (ref_tree)
3647                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3648         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3649         if (!value && !ret) {
3650                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3651                 if (ret)
3652                         mlog_errno(ret);
3653         }
3654         ocfs2_inode_unlock(inode, 1);
3655 cleanup_nolock:
3656         brelse(di_bh);
3657         brelse(xbs.xattr_bh);
3658         ocfs2_xattr_bucket_free(xbs.bucket);
3659
3660         return ret;
3661 }
3662
3663 /*
3664  * Find the xattr extent rec which may contains name_hash.
3665  * e_cpos will be the first name hash of the xattr rec.
3666  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3667  */
3668 static int ocfs2_xattr_get_rec(struct inode *inode,
3669                                u32 name_hash,
3670                                u64 *p_blkno,
3671                                u32 *e_cpos,
3672                                u32 *num_clusters,
3673                                struct ocfs2_extent_list *el)
3674 {
3675         int ret = 0, i;
3676         struct buffer_head *eb_bh = NULL;
3677         struct ocfs2_extent_block *eb;
3678         struct ocfs2_extent_rec *rec = NULL;
3679         u64 e_blkno = 0;
3680
3681         if (el->l_tree_depth) {
3682                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3683                                       &eb_bh);
3684                 if (ret) {
3685                         mlog_errno(ret);
3686                         goto out;
3687                 }
3688
3689                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3690                 el = &eb->h_list;
3691
3692                 if (el->l_tree_depth) {
3693                         ret = ocfs2_error(inode->i_sb,
3694                                           "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3695                                           inode->i_ino,
3696                                           (unsigned long long)eb_bh->b_blocknr);
3697                         goto out;
3698                 }
3699         }
3700
3701         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3702                 rec = &el->l_recs[i];
3703
3704                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3705                         e_blkno = le64_to_cpu(rec->e_blkno);
3706                         break;
3707                 }
3708         }
3709
3710         if (!e_blkno) {
3711                 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3712                                   inode->i_ino,
3713                                   le32_to_cpu(rec->e_cpos),
3714                                   ocfs2_rec_clusters(el, rec));
3715                 goto out;
3716         }
3717
3718         *p_blkno = le64_to_cpu(rec->e_blkno);
3719         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3720         if (e_cpos)
3721                 *e_cpos = le32_to_cpu(rec->e_cpos);
3722 out:
3723         brelse(eb_bh);
3724         return ret;
3725 }
3726
3727 typedef int (xattr_bucket_func)(struct inode *inode,
3728                                 struct ocfs2_xattr_bucket *bucket,
3729                                 void *para);
3730
3731 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3732                                    struct ocfs2_xattr_bucket *bucket,
3733                                    int name_index,
3734                                    const char *name,
3735                                    u32 name_hash,
3736                                    u16 *xe_index,
3737                                    int *found)
3738 {
3739         int i, ret = 0, cmp = 1, block_off, new_offset;
3740         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3741         size_t name_len = strlen(name);
3742         struct ocfs2_xattr_entry *xe = NULL;
3743         char *xe_name;
3744
3745         /*
3746          * We don't use binary search in the bucket because there
3747          * may be multiple entries with the same name hash.
3748          */
3749         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3750                 xe = &xh->xh_entries[i];
3751
3752                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3753                         continue;
3754                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3755                         break;
3756
3757                 cmp = name_index - ocfs2_xattr_get_type(xe);
3758                 if (!cmp)
3759                         cmp = name_len - xe->xe_name_len;
3760                 if (cmp)
3761                         continue;
3762
3763                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3764                                                         xh,
3765                                                         i,
3766                                                         &block_off,
3767                                                         &new_offset);
3768                 if (ret) {
3769                         mlog_errno(ret);
3770                         break;
3771                 }
3772
3773
3774                 xe_name = bucket_block(bucket, block_off) + new_offset;
3775                 if (!memcmp(name, xe_name, name_len)) {
3776                         *xe_index = i;
3777                         *found = 1;
3778                         ret = 0;
3779                         break;
3780                 }
3781         }
3782
3783         return ret;
3784 }
3785
3786 /*
3787  * Find the specified xattr entry in a series of buckets.
3788  * This series start from p_blkno and last for num_clusters.
3789  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3790  * the num of the valid buckets.
3791  *
3792  * Return the buffer_head this xattr should reside in. And if the xattr's
3793  * hash is in the gap of 2 buckets, return the lower bucket.
3794  */
3795 static int ocfs2_xattr_bucket_find(struct inode *inode,
3796                                    int name_index,
3797                                    const char *name,
3798                                    u32 name_hash,
3799                                    u64 p_blkno,
3800                                    u32 first_hash,
3801                                    u32 num_clusters,
3802                                    struct ocfs2_xattr_search *xs)
3803 {
3804         int ret, found = 0;
3805         struct ocfs2_xattr_header *xh = NULL;
3806         struct ocfs2_xattr_entry *xe = NULL;
3807         u16 index = 0;
3808         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3809         int low_bucket = 0, bucket, high_bucket;
3810         struct ocfs2_xattr_bucket *search;
3811         u64 blkno, lower_blkno = 0;
3812
3813         search = ocfs2_xattr_bucket_new(inode);
3814         if (!search) {
3815                 ret = -ENOMEM;
3816                 mlog_errno(ret);
3817                 goto out;
3818         }
3819
3820         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3821         if (ret) {
3822                 mlog_errno(ret);
3823                 goto out;
3824         }
3825
3826         xh = bucket_xh(search);
3827         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3828         while (low_bucket <= high_bucket) {
3829                 ocfs2_xattr_bucket_relse(search);
3830
3831                 bucket = (low_bucket + high_bucket) / 2;
3832                 blkno = p_blkno + bucket * blk_per_bucket;
3833                 ret = ocfs2_read_xattr_bucket(search, blkno);
3834                 if (ret) {
3835                         mlog_errno(ret);
3836                         goto out;
3837                 }
3838
3839                 xh = bucket_xh(search);
3840                 xe = &xh->xh_entries[0];
3841                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3842                         high_bucket = bucket - 1;
3843                         continue;
3844                 }
3845
3846                 /*
3847                  * Check whether the hash of the last entry in our
3848                  * bucket is larger than the search one. for an empty
3849                  * bucket, the last one is also the first one.
3850                  */
3851                 if (xh->xh_count)
3852                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3853
3854                 /* record lower_blkno which may be the insert place. */
3855                 lower_blkno = blkno;
3856
3857                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3858                         low_bucket = bucket + 1;
3859                         continue;
3860                 }
3861
3862                 /* the searched xattr should reside in this bucket if exists. */
3863                 ret = ocfs2_find_xe_in_bucket(inode, search,
3864                                               name_index, name, name_hash,
3865                                               &index, &found);
3866                 if (ret) {
3867                         mlog_errno(ret);
3868                         goto out;
3869                 }
3870                 break;
3871         }
3872
3873         /*
3874          * Record the bucket we have found.
3875          * When the xattr's hash value is in the gap of 2 buckets, we will
3876          * always set it to the previous bucket.
3877          */
3878         if (!lower_blkno)
3879                 lower_blkno = p_blkno;
3880
3881         /* This should be in cache - we just read it during the search */
3882         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3883         if (ret) {
3884                 mlog_errno(ret);
3885                 goto out;
3886         }
3887
3888         xs->header = bucket_xh(xs->bucket);
3889         xs->base = bucket_block(xs->bucket, 0);
3890         xs->end = xs->base + inode->i_sb->s_blocksize;
3891
3892         if (found) {
3893                 xs->here = &xs->header->xh_entries[index];
3894                 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3895                         name, name_index, name_hash,
3896                         (unsigned long long)bucket_blkno(xs->bucket),
3897                         index);
3898         } else
3899                 ret = -ENODATA;
3900
3901 out:
3902         ocfs2_xattr_bucket_free(search);
3903         return ret;
3904 }
3905
3906 static int ocfs2_xattr_index_block_find(struct inode *inode,
3907                                         struct buffer_head *root_bh,
3908                                         int name_index,
3909                                         const char *name,
3910                                         struct ocfs2_xattr_search *xs)
3911 {
3912         int ret;
3913         struct ocfs2_xattr_block *xb =
3914                         (struct ocfs2_xattr_block *)root_bh->b_data;
3915         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3916         struct ocfs2_extent_list *el = &xb_root->xt_list;
3917         u64 p_blkno = 0;
3918         u32 first_hash, num_clusters = 0;
3919         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3920
3921         if (le16_to_cpu(el->l_next_free_rec) == 0)
3922                 return -ENODATA;
3923
3924         trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3925                                         name, name_index, name_hash,
3926                                         (unsigned long long)root_bh->b_blocknr,
3927                                         -1);
3928
3929         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3930                                   &num_clusters, el);
3931         if (ret) {
3932                 mlog_errno(ret);
3933                 goto out;
3934         }
3935
3936         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3937
3938         trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3939                                         name, name_index, first_hash,
3940                                         (unsigned long long)p_blkno,
3941                                         num_clusters);
3942
3943         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3944                                       p_blkno, first_hash, num_clusters, xs);
3945
3946 out:
3947         return ret;
3948 }
3949
3950 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3951                                        u64 blkno,
3952                                        u32 clusters,
3953                                        xattr_bucket_func *func,
3954                                        void *para)
3955 {
3956         int i, ret = 0;
3957         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3958         u32 num_buckets = clusters * bpc;
3959         struct ocfs2_xattr_bucket *bucket;
3960
3961         bucket = ocfs2_xattr_bucket_new(inode);
3962         if (!bucket) {
3963                 mlog_errno(-ENOMEM);
3964                 return -ENOMEM;
3965         }
3966
3967         trace_ocfs2_iterate_xattr_buckets(
3968                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3969                 (unsigned long long)blkno, clusters);
3970
3971         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3972                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3973                 if (ret) {
3974                         mlog_errno(ret);
3975                         break;
3976                 }
3977
3978                 /*
3979                  * The real bucket num in this series of blocks is stored
3980                  * in the 1st bucket.
3981                  */
3982                 if (i == 0)
3983                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3984
3985                 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3986                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3987                 if (func) {
3988                         ret = func(inode, bucket, para);
3989                         if (ret && ret != -ERANGE)
3990                                 mlog_errno(ret);
3991                         /* Fall through to bucket_relse() */
3992                 }
3993
3994                 ocfs2_xattr_bucket_relse(bucket);
3995                 if (ret)
3996                         break;
3997         }
3998
3999         ocfs2_xattr_bucket_free(bucket);
4000         return ret;
4001 }
4002
4003 struct ocfs2_xattr_tree_list {
4004         char *buffer;
4005         size_t buffer_size;
4006         size_t result;
4007 };
4008
4009 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4010                                              struct ocfs2_xattr_header *xh,
4011                                              int index,
4012                                              int *block_off,
4013                                              int *new_offset)
4014 {
4015         u16 name_offset;
4016
4017         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4018                 return -EINVAL;
4019
4020         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4021
4022         *block_off = name_offset >> sb->s_blocksize_bits;
4023         *new_offset = name_offset % sb->s_blocksize;
4024
4025         return 0;
4026 }
4027
4028 static int ocfs2_list_xattr_bucket(struct inode *inode,
4029                                    struct ocfs2_xattr_bucket *bucket,
4030                                    void *para)
4031 {
4032         int ret = 0, type;
4033         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4034         int i, block_off, new_offset;
4035         const char *prefix, *name;
4036
4037         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4038                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4039                 type = ocfs2_xattr_get_type(entry);
4040                 prefix = ocfs2_xattr_prefix(type);
4041
4042                 if (prefix) {
4043                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4044                                                                 bucket_xh(bucket),
4045                                                                 i,
4046                                                                 &block_off,
4047                                                                 &new_offset);
4048                         if (ret)
4049                                 break;
4050
4051                         name = (const char *)bucket_block(bucket, block_off) +
4052                                 new_offset;
4053                         ret = ocfs2_xattr_list_entry(xl->buffer,
4054                                                      xl->buffer_size,
4055                                                      &xl->result,
4056                                                      prefix, name,
4057                                                      entry->xe_name_len);
4058                         if (ret)
4059                                 break;
4060                 }
4061         }
4062
4063         return ret;
4064 }
4065
4066 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4067                                            struct buffer_head *blk_bh,
4068                                            xattr_tree_rec_func *rec_func,
4069                                            void *para)
4070 {
4071         struct ocfs2_xattr_block *xb =
4072                         (struct ocfs2_xattr_block *)blk_bh->b_data;
4073         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4074         int ret = 0;
4075         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4076         u64 p_blkno = 0;
4077
4078         if (!el->l_next_free_rec || !rec_func)
4079                 return 0;
4080
4081         while (name_hash > 0) {
4082                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4083                                           &e_cpos, &num_clusters, el);
4084                 if (ret) {
4085                         mlog_errno(ret);
4086                         break;
4087                 }
4088
4089                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4090                                num_clusters, para);
4091                 if (ret) {
4092                         if (ret != -ERANGE)
4093                                 mlog_errno(ret);
4094                         break;
4095                 }
4096
4097                 if (e_cpos == 0)
4098                         break;
4099
4100                 name_hash = e_cpos - 1;
4101         }
4102
4103         return ret;
4104
4105 }
4106
4107 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4108                                      struct buffer_head *root_bh,
4109                                      u64 blkno, u32 cpos, u32 len, void *para)
4110 {
4111         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4112                                            ocfs2_list_xattr_bucket, para);
4113 }
4114
4115 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4116                                              struct buffer_head *blk_bh,
4117                                              char *buffer,
4118                                              size_t buffer_size)
4119 {
4120         int ret;
4121         struct ocfs2_xattr_tree_list xl = {
4122                 .buffer = buffer,
4123                 .buffer_size = buffer_size,
4124                 .result = 0,
4125         };
4126
4127         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4128                                               ocfs2_list_xattr_tree_rec, &xl);
4129         if (ret) {
4130                 mlog_errno(ret);
4131                 goto out;
4132         }
4133
4134         ret = xl.result;
4135 out:
4136         return ret;
4137 }
4138
4139 static int cmp_xe(const void *a, const void *b)
4140 {
4141         const struct ocfs2_xattr_entry *l = a, *r = b;
4142         u32 l_hash = le32_to_cpu(l->xe_name_hash);
4143         u32 r_hash = le32_to_cpu(r->xe_name_hash);
4144
4145         if (l_hash > r_hash)
4146                 return 1;
4147         if (l_hash < r_hash)
4148                 return -1;
4149         return 0;
4150 }
4151
4152 static void swap_xe(void *a, void *b, int size)
4153 {
4154         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4155
4156         tmp = *l;
4157         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4158         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4159 }
4160
4161 /*
4162  * When the ocfs2_xattr_block is filled up, new bucket will be created
4163  * and all the xattr entries will be moved to the new bucket.
4164  * The header goes at the start of the bucket, and the names+values are
4165  * filled from the end.  This is why *target starts as the last buffer.
4166  * Note: we need to sort the entries since they are not saved in order
4167  * in the ocfs2_xattr_block.
4168  */
4169 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4170                                            struct buffer_head *xb_bh,
4171                                            struct ocfs2_xattr_bucket *bucket)
4172 {
4173         int i, blocksize = inode->i_sb->s_blocksize;
4174         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4175         u16 offset, size, off_change;
4176         struct ocfs2_xattr_entry *xe;
4177         struct ocfs2_xattr_block *xb =
4178                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
4179         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4180         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4181         u16 count = le16_to_cpu(xb_xh->xh_count);
4182         char *src = xb_bh->b_data;
4183         char *target = bucket_block(bucket, blks - 1);
4184
4185         trace_ocfs2_cp_xattr_block_to_bucket_begin(
4186                                 (unsigned long long)xb_bh->b_blocknr,
4187                                 (unsigned long long)bucket_blkno(bucket));
4188
4189         for (i = 0; i < blks; i++)
4190                 memset(bucket_block(bucket, i), 0, blocksize);
4191
4192         /*
4193          * Since the xe_name_offset is based on ocfs2_xattr_header,
4194          * there is a offset change corresponding to the change of
4195          * ocfs2_xattr_header's position.
4196          */
4197         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4198         xe = &xb_xh->xh_entries[count - 1];
4199         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4200         size = blocksize - offset;
4201
4202         /* copy all the names and values. */
4203         memcpy(target + offset, src + offset, size);
4204
4205         /* Init new header now. */
4206         xh->xh_count = xb_xh->xh_count;
4207         xh->xh_num_buckets = cpu_to_le16(1);
4208         xh->xh_name_value_len = cpu_to_le16(size);
4209         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4210
4211         /* copy all the entries. */
4212         target = bucket_block(bucket, 0);
4213         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4214         size = count * sizeof(struct ocfs2_xattr_entry);
4215         memcpy(target + offset, (char *)xb_xh + offset, size);
4216
4217         /* Change the xe offset for all the xe because of the move. */
4218         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4219                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4220         for (i = 0; i < count; i++)
4221                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4222
4223         trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4224
4225         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4226              cmp_xe, swap_xe);
4227 }
4228
4229 /*
4230  * After we move xattr from block to index btree, we have to
4231  * update ocfs2_xattr_search to the new xe and base.
4232  *
4233  * When the entry is in xattr block, xattr_bh indicates the storage place.
4234  * While if the entry is in index b-tree, "bucket" indicates the
4235  * real place of the xattr.
4236  */
4237 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4238                                             struct ocfs2_xattr_search *xs,
4239                                             struct buffer_head *old_bh)
4240 {
4241         char *buf = old_bh->b_data;
4242         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4243         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4244         int i;
4245
4246         xs->header = bucket_xh(xs->bucket);
4247         xs->base = bucket_block(xs->bucket, 0);
4248         xs->end = xs->base + inode->i_sb->s_blocksize;
4249
4250         if (xs->not_found)
4251                 return;
4252
4253         i = xs->here - old_xh->xh_entries;
4254         xs->here = &xs->header->xh_entries[i];
4255 }
4256
4257 static int ocfs2_xattr_create_index_block(struct inode *inode,
4258                                           struct ocfs2_xattr_search *xs,
4259                                           struct ocfs2_xattr_set_ctxt *ctxt)
4260 {
4261         int ret;
4262         u32 bit_off, len;
4263         u64 blkno;
4264         handle_t *handle = ctxt->handle;
4265         struct ocfs2_inode_info *oi = OCFS2_I(inode);
4266         struct buffer_head *xb_bh = xs->xattr_bh;
4267         struct ocfs2_xattr_block *xb =
4268                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4269         struct ocfs2_xattr_tree_root *xr;
4270         u16 xb_flags = le16_to_cpu(xb->xb_flags);
4271
4272         trace_ocfs2_xattr_create_index_block_begin(
4273                                 (unsigned long long)xb_bh->b_blocknr);
4274
4275         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4276         BUG_ON(!xs->bucket);
4277
4278         /*
4279          * XXX:
4280          * We can use this lock for now, and maybe move to a dedicated mutex
4281          * if performance becomes a problem later.
4282          */
4283         down_write(&oi->ip_alloc_sem);
4284
4285         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4286                                       OCFS2_JOURNAL_ACCESS_WRITE);
4287         if (ret) {
4288                 mlog_errno(ret);
4289                 goto out;
4290         }
4291
4292         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4293                                      1, 1, &bit_off, &len);
4294         if (ret) {
4295                 mlog_errno(ret);
4296                 goto out;
4297         }
4298
4299         /*
4300          * The bucket may spread in many blocks, and
4301          * we will only touch the 1st block and the last block
4302          * in the whole bucket(one for entry and one for data).
4303          */
4304         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4305
4306         trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4307
4308         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4309         if (ret) {
4310                 mlog_errno(ret);
4311                 goto out;
4312         }
4313
4314         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4315                                                 OCFS2_JOURNAL_ACCESS_CREATE);
4316         if (ret) {
4317                 mlog_errno(ret);
4318                 goto out;
4319         }
4320
4321         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4322         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4323
4324         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4325
4326         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4327         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4328                offsetof(struct ocfs2_xattr_block, xb_attrs));
4329
4330         xr = &xb->xb_attrs.xb_root;
4331         xr->xt_clusters = cpu_to_le32(1);
4332         xr->xt_last_eb_blk = 0;
4333         xr->xt_list.l_tree_depth = 0;
4334         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4335         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4336
4337         xr->xt_list.l_recs[0].e_cpos = 0;
4338         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4339         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4340
4341         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4342
4343         ocfs2_journal_dirty(handle, xb_bh);
4344
4345 out:
4346         up_write(&oi->ip_alloc_sem);
4347
4348         return ret;
4349 }
4350
4351 static int cmp_xe_offset(const void *a, const void *b)
4352 {
4353         const struct ocfs2_xattr_entry *l = a, *r = b;
4354         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4355         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4356
4357         if (l_name_offset < r_name_offset)
4358                 return 1;
4359         if (l_name_offset > r_name_offset)
4360                 return -1;
4361         return 0;
4362 }
4363
4364 /*
4365  * defrag a xattr bucket if we find that the bucket has some
4366  * holes beteen name/value pairs.
4367  * We will move all the name/value pairs to the end of the bucket
4368  * so that we can spare some space for insertion.
4369  */
4370 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4371                                      handle_t *handle,
4372                                      struct ocfs2_xattr_bucket *bucket)
4373 {
4374         int ret, i;
4375         size_t end, offset, len;
4376         struct ocfs2_xattr_header *xh;
4377         char *entries, *buf, *bucket_buf = NULL;
4378         u64 blkno = bucket_blkno(bucket);
4379         u16 xh_free_start;
4380         size_t blocksize = inode->i_sb->s_blocksize;
4381         struct ocfs2_xattr_entry *xe;
4382
4383         /*
4384          * In order to make the operation more efficient and generic,
4385          * we copy all the blocks into a contiguous memory and do the
4386          * defragment there, so if anything is error, we will not touch
4387          * the real block.
4388          */
4389         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4390         if (!bucket_buf) {
4391                 ret = -EIO;
4392                 goto out;
4393         }
4394
4395         buf = bucket_buf;
4396         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4397                 memcpy(buf, bucket_block(bucket, i), blocksize);
4398
4399         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4400                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4401         if (ret < 0) {
4402                 mlog_errno(ret);
4403                 goto out;
4404         }
4405
4406         xh = (struct ocfs2_xattr_header *)bucket_buf;
4407         entries = (char *)xh->xh_entries;
4408         xh_free_start = le16_to_cpu(xh->xh_free_start);
4409
4410         trace_ocfs2_defrag_xattr_bucket(
4411              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4412              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4413
4414         /*
4415          * sort all the entries by their offset.
4416          * the largest will be the first, so that we can
4417          * move them to the end one by one.
4418          */
4419         sort(entries, le16_to_cpu(xh->xh_count),
4420              sizeof(struct ocfs2_xattr_entry),
4421              cmp_xe_offset, swap_xe);
4422
4423         /* Move all name/values to the end of the bucket. */
4424         xe = xh->xh_entries;
4425         end = OCFS2_XATTR_BUCKET_SIZE;
4426         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4427                 offset = le16_to_cpu(xe->xe_name_offset);
4428                 len = namevalue_size_xe(xe);
4429
4430                 /*
4431                  * We must make sure that the name/value pair
4432                  * exist in the same block. So adjust end to
4433                  * the previous block end if needed.
4434                  */
4435                 if (((end - len) / blocksize !=
4436                         (end - 1) / blocksize))
4437                         end = end - end % blocksize;
4438
4439                 if (end > offset + len) {
4440                         memmove(bucket_buf + end - len,
4441                                 bucket_buf + offset, len);
4442                         xe->xe_name_offset = cpu_to_le16(end - len);
4443                 }
4444
4445                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4446                                 "bucket %llu\n", (unsigned long long)blkno);
4447
4448                 end -= len;
4449         }
4450
4451         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4452                         "bucket %llu\n", (unsigned long long)blkno);
4453
4454         if (xh_free_start == end)
4455                 goto out;
4456
4457         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4458         xh->xh_free_start = cpu_to_le16(end);
4459
4460         /* sort the entries by their name_hash. */
4461         sort(entries, le16_to_cpu(xh->xh_count),
4462              sizeof(struct ocfs2_xattr_entry),
4463              cmp_xe, swap_xe);
4464
4465         buf = bucket_buf;
4466         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4467                 memcpy(bucket_block(bucket, i), buf, blocksize);
4468         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4469
4470 out:
4471         kfree(bucket_buf);
4472         return ret;
4473 }
4474
4475 /*
4476  * prev_blkno points to the start of an existing extent.  new_blkno
4477  * points to a newly allocated extent.  Because we know each of our
4478  * clusters contains more than bucket, we can easily split one cluster
4479  * at a bucket boundary.  So we take the last cluster of the existing
4480  * extent and split it down the middle.  We move the last half of the
4481  * buckets in the last cluster of the existing extent over to the new
4482  * extent.
4483  *
4484  * first_bh is the buffer at prev_blkno so we can update the existing
4485  * extent's bucket count.  header_bh is the bucket were we were hoping
4486  * to insert our xattr.  If the bucket move places the target in the new
4487  * extent, we'll update first_bh and header_bh after modifying the old
4488  * extent.
4489  *
4490  * first_hash will be set as the 1st xe's name_hash in the new extent.
4491  */
4492 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4493                                                handle_t *handle,
4494                                                struct ocfs2_xattr_bucket *first,
4495                                                struct ocfs2_xattr_bucket *target,
4496                                                u64 new_blkno,
4497                                                u32 num_clusters,
4498                                                u32 *first_hash)
4499 {
4500         int ret;
4501         struct super_block *sb = inode->i_sb;
4502         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4503         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4504         int to_move = num_buckets / 2;
4505         u64 src_blkno;
4506         u64 last_cluster_blkno = bucket_blkno(first) +
4507                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4508
4509         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4510         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4511
4512         trace_ocfs2_mv_xattr_bucket_cross_cluster(
4513                                 (unsigned long long)last_cluster_blkno,
4514                                 (unsigned long long)new_blkno);
4515
4516         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4517                                      last_cluster_blkno, new_blkno,
4518                                      to_move, first_hash);
4519         if (ret) {
4520                 mlog_errno(ret);
4521                 goto out;
4522         }
4523
4524         /* This is the first bucket that got moved */
4525         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4526
4527         /*
4528          * If the target bucket was part of the moved buckets, we need to
4529          * update first and target.
4530          */
4531         if (bucket_blkno(target) >= src_blkno) {
4532                 /* Find the block for the new target bucket */
4533                 src_blkno = new_blkno +
4534                         (bucket_blkno(target) - src_blkno);
4535
4536                 ocfs2_xattr_bucket_relse(first);
4537                 ocfs2_xattr_bucket_relse(target);
4538
4539                 /*
4540                  * These shouldn't fail - the buffers are in the
4541                  * journal from ocfs2_cp_xattr_bucket().
4542                  */
4543                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4544                 if (ret) {
4545                         mlog_errno(ret);
4546                         goto out;
4547                 }
4548                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4549                 if (ret)
4550                         mlog_errno(ret);
4551
4552         }
4553
4554 out:
4555         return ret;
4556 }
4557
4558 /*
4559  * Find the suitable pos when we divide a bucket into 2.
4560  * We have to make sure the xattrs with the same hash value exist
4561  * in the same bucket.
4562  *
4563  * If this ocfs2_xattr_header covers more than one hash value, find a
4564  * place where the hash value changes.  Try to find the most even split.
4565  * The most common case is that all entries have different hash values,
4566  * and the first check we make will find a place to split.
4567  */
4568 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4569 {
4570         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4571         int count = le16_to_cpu(xh->xh_count);
4572         int delta, middle = count / 2;
4573
4574         /*
4575          * We start at the middle.  Each step gets farther away in both
4576          * directions.  We therefore hit the change in hash value
4577          * nearest to the middle.  Note that this loop does not execute for
4578          * count < 2.
4579          */
4580         for (delta = 0; delta < middle; delta++) {
4581                 /* Let's check delta earlier than middle */
4582                 if (cmp_xe(&entries[middle - delta - 1],
4583                            &entries[middle - delta]))
4584                         return middle - delta;
4585
4586                 /* For even counts, don't walk off the end */
4587                 if ((middle + delta + 1) == count)
4588                         continue;
4589
4590                 /* Now try delta past middle */
4591                 if (cmp_xe(&entries[middle + delta],
4592                            &entries[middle + delta + 1]))
4593                         return middle + delta + 1;
4594         }
4595
4596         /* Every entry had the same hash */
4597         return count;
4598 }
4599
4600 /*
4601  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4602  * first_hash will record the 1st hash of the new bucket.
4603  *
4604  * Normally half of the xattrs will be moved.  But we have to make
4605  * sure that the xattrs with the same hash value are stored in the
4606  * same bucket. If all the xattrs in this bucket have the same hash
4607  * value, the new bucket will be initialized as an empty one and the
4608  * first_hash will be initialized as (hash_value+1).
4609  */
4610 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4611                                     handle_t *handle,
4612                                     u64 blk,
4613                                     u64 new_blk,
4614                                     u32 *first_hash,
4615                                     int new_bucket_head)
4616 {
4617         int ret, i;
4618         int count, start, len, name_value_len = 0, name_offset = 0;
4619         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4620         struct ocfs2_xattr_header *xh;
4621         struct ocfs2_xattr_entry *xe;
4622         int blocksize = inode->i_sb->s_blocksize;
4623
4624         trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4625                                               (unsigned long long)new_blk);
4626
4627         s_bucket = ocfs2_xattr_bucket_new(inode);
4628         t_bucket = ocfs2_xattr_bucket_new(inode);
4629         if (!s_bucket || !t_bucket) {
4630                 ret = -ENOMEM;
4631                 mlog_errno(ret);
4632                 goto out;
4633         }
4634
4635         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4636         if (ret) {
4637                 mlog_errno(ret);
4638                 goto out;
4639         }
4640
4641         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4642                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4643         if (ret) {
4644                 mlog_errno(ret);
4645                 goto out;
4646         }
4647
4648         /*
4649          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4650          * there's no need to read it.
4651          */
4652         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4653         if (ret) {
4654                 mlog_errno(ret);
4655                 goto out;
4656         }
4657
4658         /*
4659          * Hey, if we're overwriting t_bucket, what difference does
4660          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4661          * same part of ocfs2_cp_xattr_bucket().
4662          */
4663         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4664                                                 new_bucket_head ?
4665                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4666                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4667         if (ret) {
4668                 mlog_errno(ret);
4669                 goto out;
4670         }
4671
4672         xh = bucket_xh(s_bucket);
4673         count = le16_to_cpu(xh->xh_count);
4674         start = ocfs2_xattr_find_divide_pos(xh);
4675
4676         if (start == count) {
4677                 xe = &xh->xh_entries[start-1];
4678
4679                 /*
4680                  * initialized a new empty bucket here.
4681                  * The hash value is set as one larger than
4682                  * that of the last entry in the previous bucket.
4683                  */
4684                 for (i = 0; i < t_bucket->bu_blocks; i++)
4685                         memset(bucket_block(t_bucket, i), 0, blocksize);
4686
4687                 xh = bucket_xh(t_bucket);
4688                 xh->xh_free_start = cpu_to_le16(blocksize);
4689                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4690                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4691
4692                 goto set_num_buckets;
4693         }
4694
4695         /* copy the whole bucket to the new first. */
4696         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4697
4698         /* update the new bucket. */
4699         xh = bucket_xh(t_bucket);
4700
4701         /*
4702          * Calculate the total name/value len and xh_free_start for
4703          * the old bucket first.
4704          */
4705         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4706         name_value_len = 0;
4707         for (i = 0; i < start; i++) {
4708                 xe = &xh->xh_entries[i];
4709                 name_value_len += namevalue_size_xe(xe);
4710                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4711                         name_offset = le16_to_cpu(xe->xe_name_offset);
4712         }
4713
4714         /*
4715          * Now begin the modification to the new bucket.
4716          *
4717          * In the new bucket, We just move the xattr entry to the beginning
4718          * and don't touch the name/value. So there will be some holes in the
4719          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4720          * called.
4721          */
4722         xe = &xh->xh_entries[start];
4723         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4724         trace_ocfs2_divide_xattr_bucket_move(len,
4725                         (int)((char *)xe - (char *)xh),
4726                         (int)((char *)xh->xh_entries - (char *)xh));
4727         memmove((char *)xh->xh_entries, (char *)xe, len);
4728         xe = &xh->xh_entries[count - start];
4729         len = sizeof(struct ocfs2_xattr_entry) * start;
4730         memset((char *)xe, 0, len);
4731
4732         le16_add_cpu(&xh->xh_count, -start);
4733         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4734
4735         /* Calculate xh_free_start for the new bucket. */
4736         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4737         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4738                 xe = &xh->xh_entries[i];
4739                 if (le16_to_cpu(xe->xe_name_offset) <
4740                     le16_to_cpu(xh->xh_free_start))
4741                         xh->xh_free_start = xe->xe_name_offset;
4742         }
4743
4744 set_num_buckets:
4745         /* set xh->xh_num_buckets for the new xh. */
4746         if (new_bucket_head)
4747                 xh->xh_num_buckets = cpu_to_le16(1);
4748         else
4749                 xh->xh_num_buckets = 0;
4750
4751         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4752
4753         /* store the first_hash of the new bucket. */
4754         if (first_hash)
4755                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4756
4757         /*
4758          * Now only update the 1st block of the old bucket.  If we
4759          * just added a new empty bucket, there is no need to modify
4760          * it.
4761          */
4762         if (start == count)
4763                 goto out;
4764
4765         xh = bucket_xh(s_bucket);
4766         memset(&xh->xh_entries[start], 0,
4767                sizeof(struct ocfs2_xattr_entry) * (count - start));
4768         xh->xh_count = cpu_to_le16(start);
4769         xh->xh_free_start = cpu_to_le16(name_offset);
4770         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4771
4772         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4773
4774 out:
4775         ocfs2_xattr_bucket_free(s_bucket);
4776         ocfs2_xattr_bucket_free(t_bucket);
4777
4778         return ret;
4779 }
4780
4781 /*
4782  * Copy xattr from one bucket to another bucket.
4783  *
4784  * The caller must make sure that the journal transaction
4785  * has enough space for journaling.
4786  */
4787 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4788                                  handle_t *handle,
4789                                  u64 s_blkno,
4790                                  u64 t_blkno,
4791                                  int t_is_new)
4792 {
4793         int ret;
4794         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4795
4796         BUG_ON(s_blkno == t_blkno);
4797
4798         trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4799                                     (unsigned long long)t_blkno,
4800                                     t_is_new);
4801
4802         s_bucket = ocfs2_xattr_bucket_new(inode);
4803         t_bucket = ocfs2_xattr_bucket_new(inode);
4804         if (!s_bucket || !t_bucket) {
4805                 ret = -ENOMEM;
4806                 mlog_errno(ret);
4807                 goto out;
4808         }
4809
4810         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4811         if (ret)
4812                 goto out;
4813
4814         /*
4815          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4816          * there's no need to read it.
4817          */
4818         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4819         if (ret)
4820                 goto out;
4821
4822         /*
4823          * Hey, if we're overwriting t_bucket, what difference does
4824          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4825          * cluster to fill, we came here from
4826          * ocfs2_mv_xattr_buckets(), and it is really new -
4827          * ACCESS_CREATE is required.  But we also might have moved data
4828          * out of t_bucket before extending back into it.
4829          * ocfs2_add_new_xattr_bucket() can do this - its call to
4830          * ocfs2_add_new_xattr_cluster() may have created a new extent
4831          * and copied out the end of the old extent.  Then it re-extends
4832          * the old extent back to create space for new xattrs.  That's
4833          * how we get here, and the bucket isn't really new.
4834          */
4835         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4836                                                 t_is_new ?
4837                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4838                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4839         if (ret)
4840                 goto out;
4841
4842         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4843         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4844
4845 out:
4846         ocfs2_xattr_bucket_free(t_bucket);
4847         ocfs2_xattr_bucket_free(s_bucket);
4848
4849         return ret;
4850 }
4851
4852 /*
4853  * src_blk points to the start of an existing extent.  last_blk points to
4854  * last cluster in that extent.  to_blk points to a newly allocated
4855  * extent.  We copy the buckets from the cluster at last_blk to the new
4856  * extent.  If start_bucket is non-zero, we skip that many buckets before
4857  * we start copying.  The new extent's xh_num_buckets gets set to the
4858  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4859  * by the same amount.
4860  */
4861 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4862                                   u64 src_blk, u64 last_blk, u64 to_blk,
4863                                   unsigned int start_bucket,
4864                                   u32 *first_hash)
4865 {
4866         int i, ret, credits;
4867         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4868         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4869         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4870         struct ocfs2_xattr_bucket *old_first, *new_first;
4871
4872         trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4873                                      (unsigned long long)to_blk);
4874
4875         BUG_ON(start_bucket >= num_buckets);
4876         if (start_bucket) {
4877                 num_buckets -= start_bucket;
4878                 last_blk += (start_bucket * blks_per_bucket);
4879         }
4880
4881         /* The first bucket of the original extent */
4882         old_first = ocfs2_xattr_bucket_new(inode);
4883         /* The first bucket of the new extent */
4884         new_first = ocfs2_xattr_bucket_new(inode);
4885         if (!old_first || !new_first) {
4886                 ret = -ENOMEM;
4887                 mlog_errno(ret);
4888                 goto out;
4889         }
4890
4891         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4892         if (ret) {
4893                 mlog_errno(ret);
4894                 goto out;
4895         }
4896
4897         /*
4898          * We need to update the first bucket of the old extent and all
4899          * the buckets going to the new extent.
4900          */
4901         credits = ((num_buckets + 1) * blks_per_bucket);
4902         ret = ocfs2_extend_trans(handle, credits);
4903         if (ret) {
4904                 mlog_errno(ret);
4905                 goto out;
4906         }
4907
4908         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4909                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4910         if (ret) {
4911                 mlog_errno(ret);
4912                 goto out;
4913         }
4914
4915         for (i = 0; i < num_buckets; i++) {
4916                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4917                                             last_blk + (i * blks_per_bucket),
4918                                             to_blk + (i * blks_per_bucket),
4919                                             1);
4920                 if (ret) {
4921                         mlog_errno(ret);
4922                         goto out;
4923                 }
4924         }
4925
4926         /*
4927          * Get the new bucket ready before we dirty anything
4928          * (This actually shouldn't fail, because we already dirtied
4929          * it once in ocfs2_cp_xattr_bucket()).
4930          */
4931         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4932         if (ret) {
4933                 mlog_errno(ret);
4934                 goto out;
4935         }
4936         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4937                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4938         if (ret) {
4939                 mlog_errno(ret);
4940                 goto out;
4941         }
4942
4943         /* Now update the headers */
4944         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4945         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4946
4947         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4948         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4949
4950         if (first_hash)
4951                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4952
4953 out:
4954         ocfs2_xattr_bucket_free(new_first);
4955         ocfs2_xattr_bucket_free(old_first);
4956         return ret;
4957 }
4958
4959 /*
4960  * Move some xattrs in this cluster to the new cluster.
4961  * This function should only be called when bucket size == cluster size.
4962  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4963  */
4964 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4965                                       handle_t *handle,
4966                                       u64 prev_blk,
4967                                       u64 new_blk,
4968                                       u32 *first_hash)
4969 {
4970         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4971         int ret, credits = 2 * blk_per_bucket;
4972
4973         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4974
4975         ret = ocfs2_extend_trans(handle, credits);
4976         if (ret) {
4977                 mlog_errno(ret);
4978                 return ret;
4979         }
4980
4981         /* Move half of the xattr in start_blk to the next bucket. */
4982         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4983                                           new_blk, first_hash, 1);
4984 }
4985
4986 /*
4987  * Move some xattrs from the old cluster to the new one since they are not
4988  * contiguous in ocfs2 xattr tree.
4989  *
4990  * new_blk starts a new separate cluster, and we will move some xattrs from
4991  * prev_blk to it. v_start will be set as the first name hash value in this
4992  * new cluster so that it can be used as e_cpos during tree insertion and
4993  * don't collide with our original b-tree operations. first_bh and header_bh
4994  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4995  * to extend the insert bucket.
4996  *
4997  * The problem is how much xattr should we move to the new one and when should
4998  * we update first_bh and header_bh?
4999  * 1. If cluster size > bucket size, that means the previous cluster has more
5000  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5001  *    update the first_bh and header_bh if the insert bucket has been moved
5002  *    to the new cluster.
5003  * 2. If cluster_size == bucket_size:
5004  *    a) If the previous extent rec has more than one cluster and the insert
5005  *       place isn't in the last cluster, copy the entire last cluster to the
5006  *       new one. This time, we don't need to upate the first_bh and header_bh
5007  *       since they will not be moved into the new cluster.
5008  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5009  *       the new one. And we set the extend flag to zero if the insert place is
5010  *       moved into the new allocated cluster since no extend is needed.
5011  */
5012 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5013                                             handle_t *handle,
5014                                             struct ocfs2_xattr_bucket *first,
5015                                             struct ocfs2_xattr_bucket *target,
5016                                             u64 new_blk,
5017                                             u32 prev_clusters,
5018                                             u32 *v_start,
5019                                             int *extend)
5020 {
5021         int ret;
5022
5023         trace_ocfs2_adjust_xattr_cross_cluster(
5024                         (unsigned long long)bucket_blkno(first),
5025                         (unsigned long long)new_blk, prev_clusters);
5026
5027         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5028                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5029                                                           handle,
5030                                                           first, target,
5031                                                           new_blk,
5032                                                           prev_clusters,
5033                                                           v_start);
5034                 if (ret)
5035                         mlog_errno(ret);
5036         } else {
5037                 /* The start of the last cluster in the first extent */
5038                 u64 last_blk = bucket_blkno(first) +
5039                         ((prev_clusters - 1) *
5040                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
5041
5042                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5043                         ret = ocfs2_mv_xattr_buckets(inode, handle,
5044                                                      bucket_blkno(first),
5045                                                      last_blk, new_blk, 0,
5046                                                      v_start);
5047                         if (ret)
5048                                 mlog_errno(ret);
5049                 } else {
5050                         ret = ocfs2_divide_xattr_cluster(inode, handle,
5051                                                          last_blk, new_blk,
5052                                                          v_start);
5053                         if (ret)
5054                                 mlog_errno(ret);
5055
5056                         if ((bucket_blkno(target) == last_blk) && extend)
5057                                 *extend = 0;
5058                 }
5059         }
5060
5061         return ret;
5062 }
5063
5064 /*
5065  * Add a new cluster for xattr storage.
5066  *
5067  * If the new cluster is contiguous with the previous one, it will be
5068  * appended to the same extent record, and num_clusters will be updated.
5069  * If not, we will insert a new extent for it and move some xattrs in
5070  * the last cluster into the new allocated one.
5071  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5072  * lose the benefits of hashing because we'll have to search large leaves.
5073  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5074  * if it's bigger).
5075  *
5076  * first_bh is the first block of the previous extent rec and header_bh
5077  * indicates the bucket we will insert the new xattrs. They will be updated
5078  * when the header_bh is moved into the new cluster.
5079  */
5080 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5081                                        struct buffer_head *root_bh,
5082                                        struct ocfs2_xattr_bucket *first,
5083                                        struct ocfs2_xattr_bucket *target,
5084                                        u32 *num_clusters,
5085                                        u32 prev_cpos,
5086                                        int *extend,
5087                                        struct ocfs2_xattr_set_ctxt *ctxt)
5088 {
5089         int ret;
5090         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5091         u32 prev_clusters = *num_clusters;
5092         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5093         u64 block;
5094         handle_t *handle = ctxt->handle;
5095         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5096         struct ocfs2_extent_tree et;
5097
5098         trace_ocfs2_add_new_xattr_cluster_begin(
5099                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5100                 (unsigned long long)bucket_blkno(first),
5101                 prev_cpos, prev_clusters);
5102
5103         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5104
5105         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5106                                       OCFS2_JOURNAL_ACCESS_WRITE);
5107         if (ret < 0) {
5108                 mlog_errno(ret);
5109                 goto leave;
5110         }
5111
5112         ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5113                                      clusters_to_add, &bit_off, &num_bits);
5114         if (ret < 0) {
5115                 if (ret != -ENOSPC)
5116                         mlog_errno(ret);
5117                 goto leave;
5118         }
5119
5120         BUG_ON(num_bits > clusters_to_add);
5121
5122         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5123         trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5124
5125         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5126             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5127              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5128                 /*
5129                  * If this cluster is contiguous with the old one and
5130                  * adding this new cluster, we don't surpass the limit of
5131                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5132                  * initialized and used like other buckets in the previous
5133                  * cluster.
5134                  * So add it as a contiguous one. The caller will handle
5135                  * its init process.
5136                  */
5137                 v_start = prev_cpos + prev_clusters;
5138                 *num_clusters = prev_clusters + num_bits;
5139         } else {
5140                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5141                                                        handle,
5142                                                        first,
5143                                                        target,
5144                                                        block,
5145                                                        prev_clusters,
5146                                                        &v_start,
5147                                                        extend);
5148                 if (ret) {
5149                         mlog_errno(ret);
5150                         goto leave;
5151                 }
5152         }
5153
5154         trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5155                                                  v_start, num_bits);
5156         ret = ocfs2_insert_extent(handle, &et, v_start, block,
5157                                   num_bits, 0, ctxt->meta_ac);
5158         if (ret < 0) {
5159                 mlog_errno(ret);
5160                 goto leave;
5161         }
5162
5163         ocfs2_journal_dirty(handle, root_bh);
5164
5165 leave:
5166         return ret;
5167 }
5168
5169 /*
5170  * We are given an extent.  'first' is the bucket at the very front of
5171  * the extent.  The extent has space for an additional bucket past
5172  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5173  * of the target bucket.  We wish to shift every bucket past the target
5174  * down one, filling in that additional space.  When we get back to the
5175  * target, we split the target between itself and the now-empty bucket
5176  * at target+1 (aka, target_blkno + blks_per_bucket).
5177  */
5178 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5179                                      handle_t *handle,
5180                                      struct ocfs2_xattr_bucket *first,
5181                                      u64 target_blk,
5182                                      u32 num_clusters)
5183 {
5184         int ret, credits;
5185         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5186         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5187         u64 end_blk;
5188         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5189
5190         trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5191                                         (unsigned long long)bucket_blkno(first),
5192                                         num_clusters, new_bucket);
5193
5194         /* The extent must have room for an additional bucket */
5195         BUG_ON(new_bucket >=
5196                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5197
5198         /* end_blk points to the last existing bucket */
5199         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5200
5201         /*
5202          * end_blk is the start of the last existing bucket.
5203          * Thus, (end_blk - target_blk) covers the target bucket and
5204          * every bucket after it up to, but not including, the last
5205          * existing bucket.  Then we add the last existing bucket, the
5206          * new bucket, and the first bucket (3 * blk_per_bucket).
5207          */
5208         credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5209         ret = ocfs2_extend_trans(handle, credits);
5210         if (ret) {
5211                 mlog_errno(ret);
5212                 goto out;
5213         }
5214
5215         ret = ocfs2_xattr_bucket_journal_access(handle, first,
5216                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5217         if (ret) {
5218                 mlog_errno(ret);
5219                 goto out;
5220         }
5221
5222         while (end_blk != target_blk) {
5223                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5224                                             end_blk + blk_per_bucket, 0);
5225                 if (ret)
5226                         goto out;
5227                 end_blk -= blk_per_bucket;
5228         }
5229
5230         /* Move half of the xattr in target_blkno to the next bucket. */
5231         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5232                                         target_blk + blk_per_bucket, NULL, 0);
5233
5234         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5235         ocfs2_xattr_bucket_journal_dirty(handle, first);
5236
5237 out:
5238         return ret;
5239 }
5240
5241 /*
5242  * Add new xattr bucket in an extent record and adjust the buckets
5243  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5244  * bucket we want to insert into.
5245  *
5246  * In the easy case, we will move all the buckets after target down by
5247  * one. Half of target's xattrs will be moved to the next bucket.
5248  *
5249  * If current cluster is full, we'll allocate a new one.  This may not
5250  * be contiguous.  The underlying calls will make sure that there is
5251  * space for the insert, shifting buckets around if necessary.
5252  * 'target' may be moved by those calls.
5253  */
5254 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5255                                       struct buffer_head *xb_bh,
5256                                       struct ocfs2_xattr_bucket *target,
5257                                       struct ocfs2_xattr_set_ctxt *ctxt)
5258 {
5259         struct ocfs2_xattr_block *xb =
5260                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5261         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5262         struct ocfs2_extent_list *el = &xb_root->xt_list;
5263         u32 name_hash =
5264                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5265         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5266         int ret, num_buckets, extend = 1;
5267         u64 p_blkno;
5268         u32 e_cpos, num_clusters;
5269         /* The bucket at the front of the extent */
5270         struct ocfs2_xattr_bucket *first;
5271
5272         trace_ocfs2_add_new_xattr_bucket(
5273                                 (unsigned long long)bucket_blkno(target));
5274
5275         /* The first bucket of the original extent */
5276         first = ocfs2_xattr_bucket_new(inode);
5277         if (!first) {
5278                 ret = -ENOMEM;
5279                 mlog_errno(ret);
5280                 goto out;
5281         }
5282
5283         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5284                                   &num_clusters, el);
5285         if (ret) {
5286                 mlog_errno(ret);
5287                 goto out;
5288         }
5289
5290         ret = ocfs2_read_xattr_bucket(first, p_blkno);
5291         if (ret) {
5292                 mlog_errno(ret);
5293                 goto out;
5294         }
5295
5296         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5297         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5298                 /*
5299                  * This can move first+target if the target bucket moves
5300                  * to the new extent.
5301                  */
5302                 ret = ocfs2_add_new_xattr_cluster(inode,
5303                                                   xb_bh,
5304                                                   first,
5305                                                   target,
5306                                                   &num_clusters,
5307                                                   e_cpos,
5308                                                   &extend,
5309                                                   ctxt);
5310                 if (ret) {
5311                         mlog_errno(ret);
5312                         goto out;
5313                 }
5314         }
5315
5316         if (extend) {
5317                 ret = ocfs2_extend_xattr_bucket(inode,
5318                                                 ctxt->handle,
5319                                                 first,
5320                                                 bucket_blkno(target),
5321                                                 num_clusters);
5322                 if (ret)
5323                         mlog_errno(ret);
5324         }
5325
5326 out:
5327         ocfs2_xattr_bucket_free(first);
5328
5329         return ret;
5330 }
5331
5332 /*
5333  * Truncate the specified xe_off entry in xattr bucket.
5334  * bucket is indicated by header_bh and len is the new length.
5335  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5336  *
5337  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5338  */
5339 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5340                                              struct ocfs2_xattr_bucket *bucket,
5341                                              int xe_off,
5342                                              int len,
5343                                              struct ocfs2_xattr_set_ctxt *ctxt)
5344 {
5345         int ret, offset;
5346         u64 value_blk;
5347         struct ocfs2_xattr_entry *xe;
5348         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5349         size_t blocksize = inode->i_sb->s_blocksize;
5350         struct ocfs2_xattr_value_buf vb = {
5351                 .vb_access = ocfs2_journal_access,
5352         };
5353
5354         xe = &xh->xh_entries[xe_off];
5355
5356         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5357
5358         offset = le16_to_cpu(xe->xe_name_offset) +
5359                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5360
5361         value_blk = offset / blocksize;
5362
5363         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5364         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5365
5366         vb.vb_bh = bucket->bu_bhs[value_blk];
5367         BUG_ON(!vb.vb_bh);
5368
5369         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5370                 (vb.vb_bh->b_data + offset % blocksize);
5371
5372         /*
5373          * From here on out we have to dirty the bucket.  The generic
5374          * value calls only modify one of the bucket's bhs, but we need
5375          * to send the bucket at once.  So if they error, they *could* have
5376          * modified something.  We have to assume they did, and dirty
5377          * the whole bucket.  This leaves us in a consistent state.
5378          */
5379         trace_ocfs2_xattr_bucket_value_truncate(
5380                         (unsigned long long)bucket_blkno(bucket), xe_off, len);
5381         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5382         if (ret) {
5383                 mlog_errno(ret);
5384                 goto out;
5385         }
5386
5387         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5388                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5389         if (ret) {
5390                 mlog_errno(ret);
5391                 goto out;
5392         }
5393
5394         xe->xe_value_size = cpu_to_le64(len);
5395
5396         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5397
5398 out:
5399         return ret;
5400 }
5401
5402 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5403                                   struct buffer_head *root_bh,
5404                                   u64 blkno,
5405                                   u32 cpos,
5406                                   u32 len,
5407                                   void *para)
5408 {
5409         int ret;
5410         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5411         struct inode *tl_inode = osb->osb_tl_inode;
5412         handle_t *handle;
5413         struct ocfs2_xattr_block *xb =
5414                         (struct ocfs2_xattr_block *)root_bh->b_data;
5415         struct ocfs2_alloc_context *meta_ac = NULL;
5416         struct ocfs2_cached_dealloc_ctxt dealloc;
5417         struct ocfs2_extent_tree et;
5418
5419         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5420                                           ocfs2_delete_xattr_in_bucket, para);
5421         if (ret) {
5422                 mlog_errno(ret);
5423                 return ret;
5424         }
5425
5426         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5427
5428         ocfs2_init_dealloc_ctxt(&dealloc);
5429
5430         trace_ocfs2_rm_xattr_cluster(
5431                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
5432                         (unsigned long long)blkno, cpos, len);
5433
5434         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5435                                                len);
5436
5437         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5438         if (ret) {
5439                 mlog_errno(ret);
5440                 return ret;
5441         }
5442
5443         mutex_lock(&tl_inode->i_mutex);
5444
5445         if (ocfs2_truncate_log_needs_flush(osb)) {
5446                 ret = __ocfs2_flush_truncate_log(osb);
5447                 if (ret < 0) {
5448                         mlog_errno(ret);
5449                         goto out;
5450                 }
5451         }
5452
5453         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5454         if (IS_ERR(handle)) {
5455                 ret = -ENOMEM;
5456                 mlog_errno(ret);
5457                 goto out;
5458         }
5459
5460         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5461                                       OCFS2_JOURNAL_ACCESS_WRITE);
5462         if (ret) {
5463                 mlog_errno(ret);
5464                 goto out_commit;
5465         }
5466
5467         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5468                                   &dealloc);
5469         if (ret) {
5470                 mlog_errno(ret);
5471                 goto out_commit;
5472         }
5473
5474         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5475         ocfs2_journal_dirty(handle, root_bh);
5476
5477         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5478         if (ret)
5479                 mlog_errno(ret);
5480         ocfs2_update_inode_fsync_trans(handle, inode, 0);
5481
5482 out_commit:
5483         ocfs2_commit_trans(osb, handle);
5484 out:
5485         ocfs2_schedule_truncate_log_flush(osb, 1);
5486
5487         mutex_unlock(&tl_inode->i_mutex);
5488
5489         if (meta_ac)
5490                 ocfs2_free_alloc_context(meta_ac);
5491
5492         ocfs2_run_deallocs(osb, &dealloc);
5493
5494         return ret;
5495 }
5496
5497 /*
5498  * check whether the xattr bucket is filled up with the same hash value.
5499  * If we want to insert the xattr with the same hash, return -ENOSPC.
5500  * If we want to insert a xattr with different hash value, go ahead
5501  * and ocfs2_divide_xattr_bucket will handle this.
5502  */
5503 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5504                                               struct ocfs2_xattr_bucket *bucket,
5505                                               const char *name)
5506 {
5507         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5508         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5509
5510         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5511                 return 0;
5512
5513         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5514             xh->xh_entries[0].xe_name_hash) {
5515                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5516                      "hash = %u\n",
5517                      (unsigned long long)bucket_blkno(bucket),
5518                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5519                 return -ENOSPC;
5520         }
5521
5522         return 0;
5523 }
5524
5525 /*
5526  * Try to set the entry in the current bucket.  If we fail, the caller
5527  * will handle getting us another bucket.
5528  */
5529 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5530                                         struct ocfs2_xattr_info *xi,
5531                                         struct ocfs2_xattr_search *xs,
5532                                         struct ocfs2_xattr_set_ctxt *ctxt)
5533 {
5534         int ret;
5535         struct ocfs2_xa_loc loc;
5536
5537         trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5538
5539         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5540                                        xs->not_found ? NULL : xs->here);
5541         ret = ocfs2_xa_set(&loc, xi, ctxt);
5542         if (!ret) {
5543                 xs->here = loc.xl_entry;
5544                 goto out;
5545         }
5546         if (ret != -ENOSPC) {
5547                 mlog_errno(ret);
5548                 goto out;
5549         }
5550
5551         /* Ok, we need space.  Let's try defragmenting the bucket. */
5552         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5553                                         xs->bucket);
5554         if (ret) {
5555                 mlog_errno(ret);
5556                 goto out;
5557         }
5558
5559         ret = ocfs2_xa_set(&loc, xi, ctxt);
5560         if (!ret) {
5561                 xs->here = loc.xl_entry;
5562                 goto out;
5563         }
5564         if (ret != -ENOSPC)
5565                 mlog_errno(ret);
5566
5567
5568 out:
5569         return ret;
5570 }
5571
5572 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5573                                              struct ocfs2_xattr_info *xi,
5574                                              struct ocfs2_xattr_search *xs,
5575                                              struct ocfs2_xattr_set_ctxt *ctxt)
5576 {
5577         int ret;
5578
5579         trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5580
5581         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5582         if (!ret)
5583                 goto out;
5584         if (ret != -ENOSPC) {
5585                 mlog_errno(ret);
5586                 goto out;
5587         }
5588
5589         /* Ack, need more space.  Let's try to get another bucket! */
5590
5591         /*
5592          * We do not allow for overlapping ranges between buckets. And
5593          * the maximum number of collisions we will allow for then is
5594          * one bucket's worth, so check it here whether we need to
5595          * add a new bucket for the insert.
5596          */
5597         ret = ocfs2_check_xattr_bucket_collision(inode,
5598                                                  xs->bucket,
5599                                                  xi->xi_name);
5600         if (ret) {
5601                 mlog_errno(ret);
5602                 goto out;
5603         }
5604
5605         ret = ocfs2_add_new_xattr_bucket(inode,
5606                                          xs->xattr_bh,
5607                                          xs->bucket,
5608                                          ctxt);
5609         if (ret) {
5610                 mlog_errno(ret);
5611                 goto out;
5612         }
5613
5614         /*
5615          * ocfs2_add_new_xattr_bucket() will have updated
5616          * xs->bucket if it moved, but it will not have updated
5617          * any of the other search fields.  Thus, we drop it and
5618          * re-search.  Everything should be cached, so it'll be
5619          * quick.
5620          */
5621         ocfs2_xattr_bucket_relse(xs->bucket);
5622         ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5623                                            xi->xi_name_index,
5624                                            xi->xi_name, xs);
5625         if (ret && ret != -ENODATA)
5626                 goto out;
5627         xs->not_found = ret;
5628
5629         /* Ok, we have a new bucket, let's try again */
5630         ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5631         if (ret && (ret != -ENOSPC))
5632                 mlog_errno(ret);
5633
5634 out:
5635         return ret;
5636 }
5637
5638 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5639                                         struct ocfs2_xattr_bucket *bucket,
5640                                         void *para)
5641 {
5642         int ret = 0, ref_credits;
5643         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5644         u16 i;
5645         struct ocfs2_xattr_entry *xe;
5646         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5647         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5648         int credits = ocfs2_remove_extent_credits(osb->sb) +
5649                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5650         struct ocfs2_xattr_value_root *xv;
5651         struct ocfs2_rm_xattr_bucket_para *args =
5652                         (struct ocfs2_rm_xattr_bucket_para *)para;
5653
5654         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5655
5656         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5657                 xe = &xh->xh_entries[i];
5658                 if (ocfs2_xattr_is_local(xe))
5659                         continue;
5660
5661                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5662                                                       i, &xv, NULL);
5663                 if (ret) {
5664                         mlog_errno(ret);
5665                         break;
5666                 }
5667
5668                 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5669                                                          args->ref_ci,
5670                                                          args->ref_root_bh,
5671                                                          &ctxt.meta_ac,
5672                                                          &ref_credits);
5673
5674                 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5675                 if (IS_ERR(ctxt.handle)) {
5676                         ret = PTR_ERR(ctxt.handle);
5677                         mlog_errno(ret);
5678                         break;
5679                 }
5680
5681                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5682                                                         i, 0, &ctxt);
5683
5684                 ocfs2_commit_trans(osb, ctxt.handle);
5685                 if (ctxt.meta_ac) {
5686                         ocfs2_free_alloc_context(ctxt.meta_ac);
5687                         ctxt.meta_ac = NULL;
5688                 }
5689                 if (ret) {
5690                         mlog_errno(ret);
5691                         break;
5692                 }
5693         }
5694
5695         if (ctxt.meta_ac)
5696                 ocfs2_free_alloc_context(ctxt.meta_ac);
5697         ocfs2_schedule_truncate_log_flush(osb, 1);
5698         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5699         return ret;
5700 }
5701
5702 /*
5703  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5704  * or change the extent record flag), we need to recalculate
5705  * the metaecc for the whole bucket. So it is done here.
5706  *
5707  * Note:
5708  * We have to give the extra credits for the caller.
5709  */
5710 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5711                                             handle_t *handle,
5712                                             void *para)
5713 {
5714         int ret;
5715         struct ocfs2_xattr_bucket *bucket =
5716                         (struct ocfs2_xattr_bucket *)para;
5717
5718         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5719                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5720         if (ret) {
5721                 mlog_errno(ret);
5722                 return ret;
5723         }
5724
5725         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5726
5727         return 0;
5728 }
5729
5730 /*
5731  * Special action we need if the xattr value is refcounted.
5732  *
5733  * 1. If the xattr is refcounted, lock the tree.
5734  * 2. CoW the xattr if we are setting the new value and the value
5735  *    will be stored outside.
5736  * 3. In other case, decrease_refcount will work for us, so just
5737  *    lock the refcount tree, calculate the meta and credits is OK.
5738  *
5739  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5740  * currently CoW is a completed transaction, while this function
5741  * will also lock the allocators and let us deadlock. So we will
5742  * CoW the whole xattr value.
5743  */
5744 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5745                                         struct ocfs2_dinode *di,
5746                                         struct ocfs2_xattr_info *xi,
5747                                         struct ocfs2_xattr_search *xis,
5748                                         struct ocfs2_xattr_search *xbs,
5749                                         struct ocfs2_refcount_tree **ref_tree,
5750                                         int *meta_add,
5751                                         int *credits)
5752 {
5753         int ret = 0;
5754         struct ocfs2_xattr_block *xb;
5755         struct ocfs2_xattr_entry *xe;
5756         char *base;
5757         u32 p_cluster, num_clusters;
5758         unsigned int ext_flags;
5759         int name_offset, name_len;
5760         struct ocfs2_xattr_value_buf vb;
5761         struct ocfs2_xattr_bucket *bucket = NULL;
5762         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5763         struct ocfs2_post_refcount refcount;
5764         struct ocfs2_post_refcount *p = NULL;
5765         struct buffer_head *ref_root_bh = NULL;
5766
5767         if (!xis->not_found) {
5768                 xe = xis->here;
5769                 name_offset = le16_to_cpu(xe->xe_name_offset);
5770                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5771                 base = xis->base;
5772                 vb.vb_bh = xis->inode_bh;
5773                 vb.vb_access = ocfs2_journal_access_di;
5774         } else {
5775                 int i, block_off = 0;
5776                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5777                 xe = xbs->here;
5778                 name_offset = le16_to_cpu(xe->xe_name_offset);
5779                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5780                 i = xbs->here - xbs->header->xh_entries;
5781
5782                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5783                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5784                                                         bucket_xh(xbs->bucket),
5785                                                         i, &block_off,
5786                                                         &name_offset);
5787                         if (ret) {
5788                                 mlog_errno(ret);
5789                                 goto out;
5790                         }
5791                         base = bucket_block(xbs->bucket, block_off);
5792                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5793                         vb.vb_access = ocfs2_journal_access;
5794
5795                         if (ocfs2_meta_ecc(osb)) {
5796                                 /*create parameters for ocfs2_post_refcount. */
5797                                 bucket = xbs->bucket;
5798                                 refcount.credits = bucket->bu_blocks;
5799                                 refcount.para = bucket;
5800                                 refcount.func =
5801                                         ocfs2_xattr_bucket_post_refcount;
5802                                 p = &refcount;
5803                         }
5804                 } else {
5805                         base = xbs->base;
5806                         vb.vb_bh = xbs->xattr_bh;
5807                         vb.vb_access = ocfs2_journal_access_xb;
5808                 }
5809         }
5810
5811         if (ocfs2_xattr_is_local(xe))
5812                 goto out;
5813
5814         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5815                                 (base + name_offset + name_len);
5816
5817         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5818                                        &num_clusters, &vb.vb_xv->xr_list,
5819                                        &ext_flags);
5820         if (ret) {
5821                 mlog_errno(ret);
5822                 goto out;
5823         }
5824
5825         /*
5826          * We just need to check the 1st extent record, since we always
5827          * CoW the whole xattr. So there shouldn't be a xattr with
5828          * some REFCOUNT extent recs after the 1st one.
5829          */
5830         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5831                 goto out;
5832
5833         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5834                                        1, ref_tree, &ref_root_bh);
5835         if (ret) {
5836                 mlog_errno(ret);
5837                 goto out;
5838         }
5839
5840         /*
5841          * If we are deleting the xattr or the new size will be stored inside,
5842          * cool, leave it there, the xattr truncate process will remove them
5843          * for us(it still needs the refcount tree lock and the meta, credits).
5844          * And the worse case is that every cluster truncate will split the
5845          * refcount tree, and make the original extent become 3. So we will need
5846          * 2 * cluster more extent recs at most.
5847          */
5848         if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5849
5850                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5851                                                          &(*ref_tree)->rf_ci,
5852                                                          ref_root_bh, vb.vb_xv,
5853                                                          meta_add, credits);
5854                 if (ret)
5855                         mlog_errno(ret);
5856                 goto out;
5857         }
5858
5859         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5860                                        *ref_tree, ref_root_bh, 0,
5861                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5862         if (ret)
5863                 mlog_errno(ret);
5864
5865 out:
5866         brelse(ref_root_bh);
5867         return ret;
5868 }
5869
5870 /*
5871  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5872  * The physical clusters will be added to refcount tree.
5873  */
5874 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5875                                 struct ocfs2_xattr_value_root *xv,
5876                                 struct ocfs2_extent_tree *value_et,
5877                                 struct ocfs2_caching_info *ref_ci,
5878                                 struct buffer_head *ref_root_bh,
5879                                 struct ocfs2_cached_dealloc_ctxt *dealloc,
5880                                 struct ocfs2_post_refcount *refcount)
5881 {
5882         int ret = 0;
5883         u32 clusters = le32_to_cpu(xv->xr_clusters);
5884         u32 cpos, p_cluster, num_clusters;
5885         struct ocfs2_extent_list *el = &xv->xr_list;
5886         unsigned int ext_flags;
5887
5888         cpos = 0;
5889         while (cpos < clusters) {
5890                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5891                                                &num_clusters, el, &ext_flags);
5892                 if (ret) {
5893                         mlog_errno(ret);
5894                         break;
5895                 }
5896
5897                 cpos += num_clusters;
5898                 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5899                         continue;
5900
5901                 BUG_ON(!p_cluster);
5902
5903                 ret = ocfs2_add_refcount_flag(inode, value_et,
5904                                               ref_ci, ref_root_bh,
5905                                               cpos - num_clusters,
5906                                               p_cluster, num_clusters,
5907                                               dealloc, refcount);
5908                 if (ret) {
5909                         mlog_errno(ret);
5910                         break;
5911                 }
5912         }
5913
5914         return ret;
5915 }
5916
5917 /*
5918  * Given a normal ocfs2_xattr_header, refcount all the entries which
5919  * have value stored outside.
5920  * Used for xattrs stored in inode and ocfs2_xattr_block.
5921  */
5922 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5923                                 struct ocfs2_xattr_value_buf *vb,
5924                                 struct ocfs2_xattr_header *header,
5925                                 struct ocfs2_caching_info *ref_ci,
5926                                 struct buffer_head *ref_root_bh,
5927                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5928 {
5929
5930         struct ocfs2_xattr_entry *xe;
5931         struct ocfs2_xattr_value_root *xv;
5932         struct ocfs2_extent_tree et;
5933         int i, ret = 0;
5934
5935         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5936                 xe = &header->xh_entries[i];
5937
5938                 if (ocfs2_xattr_is_local(xe))
5939                         continue;
5940
5941                 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5942                         le16_to_cpu(xe->xe_name_offset) +
5943                         OCFS2_XATTR_SIZE(xe->xe_name_len));
5944
5945                 vb->vb_xv = xv;
5946                 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5947
5948                 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5949                                                         ref_ci, ref_root_bh,
5950                                                         dealloc, NULL);
5951                 if (ret) {
5952                         mlog_errno(ret);
5953                         break;
5954                 }
5955         }
5956
5957         return ret;
5958 }
5959
5960 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5961                                 struct buffer_head *fe_bh,
5962                                 struct ocfs2_caching_info *ref_ci,
5963                                 struct buffer_head *ref_root_bh,
5964                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
5965 {
5966         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5967         struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5968                                 (fe_bh->b_data + inode->i_sb->s_blocksize -
5969                                 le16_to_cpu(di->i_xattr_inline_size));
5970         struct ocfs2_xattr_value_buf vb = {
5971                 .vb_bh = fe_bh,
5972                 .vb_access = ocfs2_journal_access_di,
5973         };
5974
5975         return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5976                                                   ref_ci, ref_root_bh, dealloc);
5977 }
5978
5979 struct ocfs2_xattr_tree_value_refcount_para {
5980         struct ocfs2_caching_info *ref_ci;
5981         struct buffer_head *ref_root_bh;
5982         struct ocfs2_cached_dealloc_ctxt *dealloc;
5983 };
5984
5985 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5986                                            struct ocfs2_xattr_bucket *bucket,
5987                                            int offset,
5988                                            struct ocfs2_xattr_value_root **xv,
5989                                            struct buffer_head **bh)
5990 {
5991         int ret, block_off, name_offset;
5992         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5993         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5994         void *base;
5995
5996         ret = ocfs2_xattr_bucket_get_name_value(sb,
5997                                                 bucket_xh(bucket),
5998                                                 offset,
5999                                                 &block_off,
6000                                                 &name_offset);
6001         if (ret) {
6002                 mlog_errno(ret);
6003                 goto out;
6004         }
6005
6006         base = bucket_block(bucket, block_off);
6007
6008         *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6009                          OCFS2_XATTR_SIZE(xe->xe_name_len));
6010
6011         if (bh)
6012                 *bh = bucket->bu_bhs[block_off];
6013 out:
6014         return ret;
6015 }
6016
6017 /*
6018  * For a given xattr bucket, refcount all the entries which
6019  * have value stored outside.
6020  */
6021 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6022                                              struct ocfs2_xattr_bucket *bucket,
6023                                              void *para)
6024 {
6025         int i, ret = 0;
6026         struct ocfs2_extent_tree et;
6027         struct ocfs2_xattr_tree_value_refcount_para *ref =
6028                         (struct ocfs2_xattr_tree_value_refcount_para *)para;
6029         struct ocfs2_xattr_header *xh =
6030                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6031         struct ocfs2_xattr_entry *xe;
6032         struct ocfs2_xattr_value_buf vb = {
6033                 .vb_access = ocfs2_journal_access,
6034         };
6035         struct ocfs2_post_refcount refcount = {
6036                 .credits = bucket->bu_blocks,
6037                 .para = bucket,
6038                 .func = ocfs2_xattr_bucket_post_refcount,
6039         };
6040         struct ocfs2_post_refcount *p = NULL;
6041
6042         /* We only need post_refcount if we support metaecc. */
6043         if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6044                 p = &refcount;
6045
6046         trace_ocfs2_xattr_bucket_value_refcount(
6047                                 (unsigned long long)bucket_blkno(bucket),
6048                                 le16_to_cpu(xh->xh_count));
6049         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6050                 xe = &xh->xh_entries[i];
6051
6052                 if (ocfs2_xattr_is_local(xe))
6053                         continue;
6054
6055                 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6056                                                       &vb.vb_xv, &vb.vb_bh);
6057                 if (ret) {
6058                         mlog_errno(ret);
6059                         break;
6060                 }
6061
6062                 ocfs2_init_xattr_value_extent_tree(&et,
6063                                                    INODE_CACHE(inode), &vb);
6064
6065                 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6066                                                         &et, ref->ref_ci,
6067                                                         ref->ref_root_bh,
6068                                                         ref->dealloc, p);
6069                 if (ret) {
6070                         mlog_errno(ret);
6071                         break;
6072                 }
6073         }
6074
6075         return ret;
6076
6077 }
6078
6079 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6080                                      struct buffer_head *root_bh,
6081                                      u64 blkno, u32 cpos, u32 len, void *para)
6082 {
6083         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6084                                            ocfs2_xattr_bucket_value_refcount,
6085                                            para);
6086 }
6087
6088 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6089                                 struct buffer_head *blk_bh,
6090                                 struct ocfs2_caching_info *ref_ci,
6091                                 struct buffer_head *ref_root_bh,
6092                                 struct ocfs2_cached_dealloc_ctxt *dealloc)
6093 {
6094         int ret = 0;
6095         struct ocfs2_xattr_block *xb =
6096                                 (struct ocfs2_xattr_block *)blk_bh->b_data;
6097
6098         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6099                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6100                 struct ocfs2_xattr_value_buf vb = {
6101                         .vb_bh = blk_bh,
6102                         .vb_access = ocfs2_journal_access_xb,
6103                 };
6104
6105                 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6106                                                          ref_ci, ref_root_bh,
6107                                                          dealloc);
6108         } else {
6109                 struct ocfs2_xattr_tree_value_refcount_para para = {
6110                         .ref_ci = ref_ci,
6111                         .ref_root_bh = ref_root_bh,
6112                         .dealloc = dealloc,
6113                 };
6114
6115                 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6116                                                 ocfs2_refcount_xattr_tree_rec,
6117                                                 &para);
6118         }
6119
6120         return ret;
6121 }
6122
6123 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6124                                      struct buffer_head *fe_bh,
6125                                      struct ocfs2_caching_info *ref_ci,
6126                                      struct buffer_head *ref_root_bh,
6127                                      struct ocfs2_cached_dealloc_ctxt *dealloc)
6128 {
6129         int ret = 0;
6130         struct ocfs2_inode_info *oi = OCFS2_I(inode);
6131         struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6132         struct buffer_head *blk_bh = NULL;
6133
6134         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6135                 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6136                                                          ref_ci, ref_root_bh,
6137                                                          dealloc);
6138                 if (ret) {
6139                         mlog_errno(ret);
6140                         goto out;
6141                 }
6142         }
6143
6144         if (!di->i_xattr_loc)
6145                 goto out;
6146
6147         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6148                                      &blk_bh);
6149         if (ret < 0) {
6150                 mlog_errno(ret);
6151                 goto out;
6152         }
6153
6154         ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6155                                                 ref_root_bh, dealloc);
6156         if (ret)
6157                 mlog_errno(ret);
6158
6159         brelse(blk_bh);
6160 out:
6161
6162         return ret;
6163 }
6164
6165 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6166 /*
6167  * Store the information we need in xattr reflink.
6168  * old_bh and new_bh are inode bh for the old and new inode.
6169  */
6170 struct ocfs2_xattr_reflink {
6171         struct inode *old_inode;
6172         struct inode *new_inode;
6173         struct buffer_head *old_bh;
6174         struct buffer_head *new_bh;
6175         struct ocfs2_caching_info *ref_ci;
6176         struct buffer_head *ref_root_bh;
6177         struct ocfs2_cached_dealloc_ctxt *dealloc;
6178         should_xattr_reflinked *xattr_reflinked;
6179 };
6180
6181 /*
6182  * Given a xattr header and xe offset,
6183  * return the proper xv and the corresponding bh.
6184  * xattr in inode, block and xattr tree have different implementaions.
6185  */
6186 typedef int (get_xattr_value_root)(struct super_block *sb,
6187                                    struct buffer_head *bh,
6188                                    struct ocfs2_xattr_header *xh,
6189                                    int offset,
6190                                    struct ocfs2_xattr_value_root **xv,
6191                                    struct buffer_head **ret_bh,
6192                                    void *para);
6193
6194 /*
6195  * Calculate all the xattr value root metadata stored in this xattr header and
6196  * credits we need if we create them from the scratch.
6197  * We use get_xattr_value_root so that all types of xattr container can use it.
6198  */
6199 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6200                                              struct buffer_head *bh,
6201                                              struct ocfs2_xattr_header *xh,
6202                                              int *metas, int *credits,
6203                                              int *num_recs,
6204                                              get_xattr_value_root *func,
6205                                              void *para)
6206 {
6207         int i, ret = 0;
6208         struct ocfs2_xattr_value_root *xv;
6209         struct ocfs2_xattr_entry *xe;
6210
6211         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6212                 xe = &xh->xh_entries[i];
6213                 if (ocfs2_xattr_is_local(xe))
6214                         continue;
6215
6216                 ret = func(sb, bh, xh, i, &xv, NULL, para);
6217                 if (ret) {
6218                         mlog_errno(ret);
6219                         break;
6220                 }
6221
6222                 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6223                           le16_to_cpu(xv->xr_list.l_next_free_rec);
6224
6225                 *credits += ocfs2_calc_extend_credits(sb,
6226                                                 &def_xv.xv.xr_list);
6227
6228                 /*
6229                  * If the value is a tree with depth > 1, We don't go deep
6230                  * to the extent block, so just calculate a maximum record num.
6231                  */
6232                 if (!xv->xr_list.l_tree_depth)
6233                         *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6234                 else
6235                         *num_recs += ocfs2_clusters_for_bytes(sb,
6236                                                               XATTR_SIZE_MAX);
6237         }
6238
6239         return ret;
6240 }
6241
6242 /* Used by xattr inode and block to return the right xv and buffer_head. */
6243 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6244                                       struct buffer_head *bh,
6245                                       struct ocfs2_xattr_header *xh,
6246                                       int offset,
6247                                       struct ocfs2_xattr_value_root **xv,
6248                                       struct buffer_head **ret_bh,
6249                                       void *para)
6250 {
6251         struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6252
6253         *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6254                 le16_to_cpu(xe->xe_name_offset) +
6255                 OCFS2_XATTR_SIZE(xe->xe_name_len));
6256
6257         if (ret_bh)
6258                 *ret_bh = bh;
6259
6260         return 0;
6261 }
6262
6263 /*
6264  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6265  * It is only used for inline xattr and xattr block.
6266  */
6267 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6268                                         struct ocfs2_xattr_header *xh,
6269                                         struct buffer_head *ref_root_bh,
6270                                         int *credits,
6271                                         struct ocfs2_alloc_context **meta_ac)
6272 {
6273         int ret, meta_add = 0, num_recs = 0;
6274         struct ocfs2_refcount_block *rb =
6275                         (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6276
6277         *credits = 0;
6278
6279         ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6280                                                 &meta_add, credits, &num_recs,
6281                                                 ocfs2_get_xattr_value_root,
6282                                                 NULL);
6283         if (ret) {
6284                 mlog_errno(ret);
6285                 goto out;
6286         }
6287
6288         /*
6289          * We need to add/modify num_recs in refcount tree, so just calculate
6290          * an approximate number we need for refcount tree change.
6291          * Sometimes we need to split the tree, and after split,  half recs
6292          * will be moved to the new block, and a new block can only provide
6293          * half number of recs. So we multiple new blocks by 2.
6294          */
6295         num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6296         meta_add += num_recs;
6297         *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6298         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6299                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6300                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6301         else
6302                 *credits += 1;
6303
6304         ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6305         if (ret)
6306                 mlog_errno(ret);
6307
6308 out:
6309         return ret;
6310 }
6311
6312 /*
6313  * Given a xattr header, reflink all the xattrs in this container.
6314  * It can be used for inode, block and bucket.
6315  *
6316  * NOTE:
6317  * Before we call this function, the caller has memcpy the xattr in
6318  * old_xh to the new_xh.
6319  *
6320  * If args.xattr_reflinked is set, call it to decide whether the xe should
6321  * be reflinked or not. If not, remove it from the new xattr header.
6322  */
6323 static int ocfs2_reflink_xattr_header(handle_t *handle,
6324                                       struct ocfs2_xattr_reflink *args,
6325                                       struct buffer_head *old_bh,
6326                                       struct ocfs2_xattr_header *xh,
6327                                       struct buffer_head *new_bh,
6328                                       struct ocfs2_xattr_header *new_xh,
6329                                       struct ocfs2_xattr_value_buf *vb,
6330                                       struct ocfs2_alloc_context *meta_ac,
6331                                       get_xattr_value_root *func,
6332                                       void *para)
6333 {
6334         int ret = 0, i, j;
6335         struct super_block *sb = args->old_inode->i_sb;
6336         struct buffer_head *value_bh;
6337         struct ocfs2_xattr_entry *xe, *last;
6338         struct ocfs2_xattr_value_root *xv, *new_xv;
6339         struct ocfs2_extent_tree data_et;
6340         u32 clusters, cpos, p_cluster, num_clusters;
6341         unsigned int ext_flags = 0;
6342
6343         trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6344                                          le16_to_cpu(xh->xh_count));
6345
6346         last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6347         for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6348                 xe = &xh->xh_entries[i];
6349
6350                 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6351                         xe = &new_xh->xh_entries[j];
6352
6353                         le16_add_cpu(&new_xh->xh_count, -1);
6354                         if (new_xh->xh_count) {
6355                                 memmove(xe, xe + 1,
6356                                         (void *)last - (void *)xe);
6357                                 memset(last, 0,
6358                                        sizeof(struct ocfs2_xattr_entry));
6359                         }
6360
6361                         /*
6362                          * We don't want j to increase in the next round since
6363                          * it is already moved ahead.
6364                          */
6365                         j--;
6366                         continue;
6367                 }
6368
6369                 if (ocfs2_xattr_is_local(xe))
6370                         continue;
6371
6372                 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6373                 if (ret) {
6374                         mlog_errno(ret);
6375                         break;
6376                 }
6377
6378                 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6379                 if (ret) {
6380                         mlog_errno(ret);
6381                         break;
6382                 }
6383
6384                 /*
6385                  * For the xattr which has l_tree_depth = 0, all the extent
6386                  * recs have already be copied to the new xh with the
6387                  * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6388                  * increase the refount count int the refcount tree.
6389                  *
6390                  * For the xattr which has l_tree_depth > 0, we need
6391                  * to initialize it to the empty default value root,
6392                  * and then insert the extents one by one.
6393                  */
6394                 if (xv->xr_list.l_tree_depth) {
6395                         memcpy(new_xv, &def_xv, sizeof(def_xv));
6396                         vb->vb_xv = new_xv;
6397                         vb->vb_bh = value_bh;
6398                         ocfs2_init_xattr_value_extent_tree(&data_et,
6399                                         INODE_CACHE(args->new_inode), vb);
6400                 }
6401
6402                 clusters = le32_to_cpu(xv->xr_clusters);
6403                 cpos = 0;
6404                 while (cpos < clusters) {
6405                         ret = ocfs2_xattr_get_clusters(args->old_inode,
6406                                                        cpos,
6407                                                        &p_cluster,
6408                                                        &num_clusters,
6409                                                        &xv->xr_list,
6410                                                        &ext_flags);
6411                         if (ret) {
6412                                 mlog_errno(ret);
6413                                 goto out;
6414                         }
6415
6416                         BUG_ON(!p_cluster);
6417
6418                         if (xv->xr_list.l_tree_depth) {
6419                                 ret = ocfs2_insert_extent(handle,
6420                                                 &data_et, cpos,
6421                                                 ocfs2_clusters_to_blocks(
6422                                                         args->old_inode->i_sb,
6423                                                         p_cluster),
6424                                                 num_clusters, ext_flags,
6425                                                 meta_ac);
6426                                 if (ret) {
6427                                         mlog_errno(ret);
6428                                         goto out;
6429                                 }
6430                         }
6431
6432                         ret = ocfs2_increase_refcount(handle, args->ref_ci,
6433                                                       args->ref_root_bh,
6434                                                       p_cluster, num_clusters,
6435                                                       meta_ac, args->dealloc);
6436                         if (ret) {
6437                                 mlog_errno(ret);
6438                                 goto out;
6439                         }
6440
6441                         cpos += num_clusters;
6442                 }
6443         }
6444
6445 out:
6446         return ret;
6447 }
6448
6449 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6450 {
6451         int ret = 0, credits = 0;
6452         handle_t *handle;
6453         struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6454         struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6455         int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6456         int header_off = osb->sb->s_blocksize - inline_size;
6457         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6458                                         (args->old_bh->b_data + header_off);
6459         struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6460                                         (args->new_bh->b_data + header_off);
6461         struct ocfs2_alloc_context *meta_ac = NULL;
6462         struct ocfs2_inode_info *new_oi;
6463         struct ocfs2_dinode *new_di;
6464         struct ocfs2_xattr_value_buf vb = {
6465                 .vb_bh = args->new_bh,
6466                 .vb_access = ocfs2_journal_access_di,
6467         };
6468
6469         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6470                                                   &credits, &meta_ac);
6471         if (ret) {
6472                 mlog_errno(ret);
6473                 goto out;
6474         }
6475
6476         handle = ocfs2_start_trans(osb, credits);
6477         if (IS_ERR(handle)) {
6478                 ret = PTR_ERR(handle);
6479                 mlog_errno(ret);
6480                 goto out;
6481         }
6482
6483         ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6484                                       args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6485         if (ret) {
6486                 mlog_errno(ret);
6487                 goto out_commit;
6488         }
6489
6490         memcpy(args->new_bh->b_data + header_off,
6491                args->old_bh->b_data + header_off, inline_size);
6492
6493         new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6494         new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6495
6496         ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6497                                          args->new_bh, new_xh, &vb, meta_ac,
6498                                          ocfs2_get_xattr_value_root, NULL);
6499         if (ret) {
6500                 mlog_errno(ret);
6501                 goto out_commit;
6502         }
6503
6504         new_oi = OCFS2_I(args->new_inode);
6505         /*
6506          * Adjust extent record count to reserve space for extended attribute.
6507          * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6508          */
6509         if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6510             !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6511                 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6512                 le16_add_cpu(&el->l_count, -(inline_size /
6513                                         sizeof(struct ocfs2_extent_rec)));
6514         }
6515         spin_lock(&new_oi->ip_lock);
6516         new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6517         new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6518         spin_unlock(&new_oi->ip_lock);
6519
6520         ocfs2_journal_dirty(handle, args->new_bh);
6521
6522 out_commit:
6523         ocfs2_commit_trans(osb, handle);
6524
6525 out:
6526         if (meta_ac)
6527                 ocfs2_free_alloc_context(meta_ac);
6528         return ret;
6529 }
6530
6531 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6532                                           struct buffer_head *fe_bh,
6533                                           struct buffer_head **ret_bh,
6534                                           int indexed)
6535 {
6536         int ret;
6537         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6538         struct ocfs2_xattr_set_ctxt ctxt;
6539
6540         memset(&ctxt, 0, sizeof(ctxt));
6541         ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6542         if (ret < 0) {
6543                 mlog_errno(ret);
6544                 return ret;
6545         }
6546
6547         ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6548         if (IS_ERR(ctxt.handle)) {
6549                 ret = PTR_ERR(ctxt.handle);
6550                 mlog_errno(ret);
6551                 goto out;
6552         }
6553
6554         trace_ocfs2_create_empty_xattr_block(
6555                                 (unsigned long long)fe_bh->b_blocknr, indexed);
6556         ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6557                                        ret_bh);
6558         if (ret)
6559                 mlog_errno(ret);
6560
6561         ocfs2_commit_trans(osb, ctxt.handle);
6562 out:
6563         ocfs2_free_alloc_context(ctxt.meta_ac);
6564         return ret;
6565 }
6566
6567 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6568                                      struct buffer_head *blk_bh,
6569                                      struct buffer_head *new_blk_bh)
6570 {
6571         int ret = 0, credits = 0;
6572         handle_t *handle;
6573         struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6574         struct ocfs2_dinode *new_di;
6575         struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6576         int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6577         struct ocfs2_xattr_block *xb =
6578                         (struct ocfs2_xattr_block *)blk_bh->b_data;
6579         struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6580         struct ocfs2_xattr_block *new_xb =
6581                         (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6582         struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6583         struct ocfs2_alloc_context *meta_ac;
6584         struct ocfs2_xattr_value_buf vb = {
6585                 .vb_bh = new_blk_bh,
6586                 .vb_access = ocfs2_journal_access_xb,
6587         };
6588
6589         ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6590                                                   &credits, &meta_ac);
6591         if (ret) {
6592                 mlog_errno(ret);
6593                 return ret;
6594         }
6595
6596         /* One more credits in case we need to add xattr flags in new inode. */
6597         handle = ocfs2_start_trans(osb, credits + 1);
6598         if (IS_ERR(handle)) {
6599                 ret = PTR_ERR(handle);
6600                 mlog_errno(ret);
6601                 goto out;
6602         }
6603
6604         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6605                 ret = ocfs2_journal_access_di(handle,
6606                                               INODE_CACHE(args->new_inode),
6607                                               args->new_bh,
6608                                               OCFS2_JOURNAL_ACCESS_WRITE);
6609                 if (ret) {
6610                         mlog_errno(ret);
6611                         goto out_commit;
6612                 }
6613         }
6614
6615         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6616                                       new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6617         if (ret) {
6618                 mlog_errno(ret);
6619                 goto out_commit;
6620         }
6621
6622         memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6623                osb->sb->s_blocksize - header_off);
6624
6625         ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6626                                          new_blk_bh, new_xh, &vb, meta_ac,
6627                                          ocfs2_get_xattr_value_root, NULL);
6628         if (ret) {
6629                 mlog_errno(ret);
6630                 goto out_commit;
6631         }
6632
6633         ocfs2_journal_dirty(handle, new_blk_bh);
6634
6635         if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6636                 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6637                 spin_lock(&new_oi->ip_lock);
6638                 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6639                 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6640                 spin_unlock(&new_oi->ip_lock);
6641
6642                 ocfs2_journal_dirty(handle, args->new_bh);
6643         }
6644
6645 out_commit:
6646         ocfs2_commit_trans(osb, handle);
6647
6648 out:
6649         ocfs2_free_alloc_context(meta_ac);
6650         return ret;
6651 }
6652
6653 struct ocfs2_reflink_xattr_tree_args {
6654         struct ocfs2_xattr_reflink *reflink;
6655         struct buffer_head *old_blk_bh;
6656         struct buffer_head *new_blk_bh;
6657         struct ocfs2_xattr_bucket *old_bucket;
6658         struct ocfs2_xattr_bucket *new_bucket;
6659 };
6660
6661 /*
6662  * NOTE:
6663  * We have to handle the case that both old bucket and new bucket
6664  * will call this function to get the right ret_bh.
6665  * So The caller must give us the right bh.
6666  */
6667 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6668                                         struct buffer_head *bh,
6669                                         struct ocfs2_xattr_header *xh,
6670                                         int offset,
6671                                         struct ocfs2_xattr_value_root **xv,
6672                                         struct buffer_head **ret_bh,
6673                                         void *para)
6674 {
6675         struct ocfs2_reflink_xattr_tree_args *args =
6676                         (struct ocfs2_reflink_xattr_tree_args *)para;
6677         struct ocfs2_xattr_bucket *bucket;
6678
6679         if (bh == args->old_bucket->bu_bhs[0])
6680                 bucket = args->old_bucket;
6681         else
6682                 bucket = args->new_bucket;
6683
6684         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6685                                                xv, ret_bh);
6686 }
6687
6688 struct ocfs2_value_tree_metas {
6689         int num_metas;
6690         int credits;
6691         int num_recs;
6692 };
6693
6694 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6695                                         struct buffer_head *bh,
6696                                         struct ocfs2_xattr_header *xh,
6697                                         int offset,
6698                                         struct ocfs2_xattr_value_root **xv,
6699                                         struct buffer_head **ret_bh,
6700                                         void *para)
6701 {
6702         struct ocfs2_xattr_bucket *bucket =
6703                                 (struct ocfs2_xattr_bucket *)para;
6704
6705         return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6706                                                xv, ret_bh);
6707 }
6708
6709 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6710                                       struct ocfs2_xattr_bucket *bucket,
6711                                       void *para)
6712 {
6713         struct ocfs2_value_tree_metas *metas =
6714                         (struct ocfs2_value_tree_metas *)para;
6715         struct ocfs2_xattr_header *xh =
6716                         (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6717
6718         /* Add the credits for this bucket first. */
6719         metas->credits += bucket->bu_blocks;
6720         return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6721                                         xh, &metas->num_metas,
6722                                         &metas->credits, &metas->num_recs,
6723                                         ocfs2_value_tree_metas_in_bucket,
6724                                         bucket);
6725 }
6726
6727 /*
6728  * Given a xattr extent rec starting from blkno and having len clusters,
6729  * iterate all the buckets calculate how much metadata we need for reflinking
6730  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6731  */
6732 static int ocfs2_lock_reflink_xattr_rec_allocators(
6733                                 struct ocfs2_reflink_xattr_tree_args *args,
6734                                 struct ocfs2_extent_tree *xt_et,
6735                                 u64 blkno, u32 len, int *credits,
6736                                 struct ocfs2_alloc_context **meta_ac,
6737                                 struct ocfs2_alloc_context **data_ac)
6738 {
6739         int ret, num_free_extents;
6740         struct ocfs2_value_tree_metas metas;
6741         struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6742         struct ocfs2_refcount_block *rb;
6743
6744         memset(&metas, 0, sizeof(metas));
6745
6746         ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6747                                           ocfs2_calc_value_tree_metas, &metas);
6748         if (ret) {
6749                 mlog_errno(ret);
6750                 goto out;
6751         }
6752
6753         *credits = metas.credits;
6754
6755         /*
6756          * Calculate we need for refcount tree change.
6757          *
6758          * We need to add/modify num_recs in refcount tree, so just calculate
6759          * an approximate number we need for refcount tree change.
6760          * Sometimes we need to split the tree, and after split,  half recs
6761          * will be moved to the new block, and a new block can only provide
6762          * half number of recs. So we multiple new blocks by 2.
6763          * In the end, we have to add credits for modifying the already
6764          * existed refcount block.
6765          */
6766         rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6767         metas.num_recs =
6768                 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6769                  ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6770         metas.num_metas += metas.num_recs;
6771         *credits += metas.num_recs +
6772                     metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6773         if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6774                 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6775                             le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6776         else
6777                 *credits += 1;
6778
6779         /* count in the xattr tree change. */
6780         num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6781         if (num_free_extents < 0) {
6782                 ret = num_free_extents;
6783                 mlog_errno(ret);
6784                 goto out;
6785         }
6786
6787         if (num_free_extents < len)
6788                 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6789
6790         *credits += ocfs2_calc_extend_credits(osb->sb,
6791                                               xt_et->et_root_el);
6792
6793         if (metas.num_metas) {
6794                 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6795                                                         meta_ac);
6796                 if (ret) {
6797                         mlog_errno(ret);
6798                         goto out;
6799                 }
6800         }
6801
6802         if (len) {
6803                 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6804                 if (ret)
6805                         mlog_errno(ret);
6806         }
6807 out:
6808         if (ret) {
6809                 if (*meta_ac) {
6810                         ocfs2_free_alloc_context(*meta_ac);
6811                         *meta_ac = NULL;
6812                 }
6813         }
6814
6815         return ret;
6816 }
6817
6818 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6819                                 u64 blkno, u64 new_blkno, u32 clusters,
6820                                 u32 *cpos, int num_buckets,
6821                                 struct ocfs2_alloc_context *meta_ac,
6822                                 struct ocfs2_alloc_context *data_ac,
6823                                 struct ocfs2_reflink_xattr_tree_args *args)
6824 {
6825         int i, j, ret = 0;
6826         struct super_block *sb = args->reflink->old_inode->i_sb;
6827         int bpb = args->old_bucket->bu_blocks;
6828         struct ocfs2_xattr_value_buf vb = {
6829                 .vb_access = ocfs2_journal_access,
6830         };
6831
6832         for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6833                 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6834                 if (ret) {
6835                         mlog_errno(ret);
6836                         break;
6837                 }
6838
6839                 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6840                 if (ret) {
6841                         mlog_errno(ret);
6842                         break;
6843                 }
6844
6845                 ret = ocfs2_xattr_bucket_journal_access(handle,
6846                                                 args->new_bucket,
6847                                                 OCFS2_JOURNAL_ACCESS_CREATE);
6848                 if (ret) {
6849                         mlog_errno(ret);
6850                         break;
6851                 }
6852
6853                 for (j = 0; j < bpb; j++)
6854                         memcpy(bucket_block(args->new_bucket, j),
6855                                bucket_block(args->old_bucket, j),
6856                                sb->s_blocksize);
6857
6858                 /*
6859                  * Record the start cpos so that we can use it to initialize
6860                  * our xattr tree we also set the xh_num_bucket for the new
6861                  * bucket.
6862                  */
6863                 if (i == 0) {
6864                         *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6865                                             xh_entries[0].xe_name_hash);
6866                         bucket_xh(args->new_bucket)->xh_num_buckets =
6867                                 cpu_to_le16(num_buckets);
6868                 }
6869
6870                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6871
6872                 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6873                                         args->old_bucket->bu_bhs[0],
6874                                         bucket_xh(args->old_bucket),
6875                                         args->new_bucket->bu_bhs[0],
6876                                         bucket_xh(args->new_bucket),
6877                                         &vb, meta_ac,
6878                                         ocfs2_get_reflink_xattr_value_root,
6879                                         args);
6880                 if (ret) {
6881                         mlog_errno(ret);
6882                         break;
6883                 }
6884
6885                 /*
6886                  * Re-access and dirty the bucket to calculate metaecc.
6887                  * Because we may extend the transaction in reflink_xattr_header
6888                  * which will let the already accessed block gone.
6889                  */
6890                 ret = ocfs2_xattr_bucket_journal_access(handle,
6891                                                 args->new_bucket,
6892                                                 OCFS2_JOURNAL_ACCESS_WRITE);
6893                 if (ret) {
6894                         mlog_errno(ret);
6895                         break;
6896                 }
6897
6898                 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6899
6900                 ocfs2_xattr_bucket_relse(args->old_bucket);
6901                 ocfs2_xattr_bucket_relse(args->new_bucket);
6902         }
6903
6904         ocfs2_xattr_bucket_relse(args->old_bucket);
6905         ocfs2_xattr_bucket_relse(args->new_bucket);
6906         return ret;
6907 }
6908
6909 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6910                                 struct inode *inode,
6911                                 struct ocfs2_reflink_xattr_tree_args *args,
6912                                 struct ocfs2_extent_tree *et,
6913                                 struct ocfs2_alloc_context *meta_ac,
6914                                 struct ocfs2_alloc_context *data_ac,
6915                                 u64 blkno, u32 cpos, u32 len)
6916 {
6917         int ret, first_inserted = 0;
6918         u32 p_cluster, num_clusters, reflink_cpos = 0;
6919         u64 new_blkno;
6920         unsigned int num_buckets, reflink_buckets;
6921         unsigned int bpc =
6922                 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6923
6924         ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6925         if (ret) {
6926                 mlog_errno(ret);
6927                 goto out;
6928         }
6929         num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6930         ocfs2_xattr_bucket_relse(args->old_bucket);
6931
6932         while (len && num_buckets) {
6933                 ret = ocfs2_claim_clusters(handle, data_ac,
6934                                            1, &p_cluster, &num_clusters);
6935                 if (ret) {
6936                         mlog_errno(ret);
6937                         goto out;
6938                 }
6939
6940                 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6941                 reflink_buckets = min(num_buckets, bpc * num_clusters);
6942
6943                 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6944                                                  new_blkno, num_clusters,
6945                                                  &reflink_cpos, reflink_buckets,
6946                                                  meta_ac, data_ac, args);
6947                 if (ret) {
6948                         mlog_errno(ret);
6949                         goto out;
6950                 }
6951
6952                 /*
6953                  * For the 1st allocated cluster, we make it use the same cpos
6954                  * so that the xattr tree looks the same as the original one
6955                  * in the most case.
6956                  */
6957                 if (!first_inserted) {
6958                         reflink_cpos = cpos;
6959                         first_inserted = 1;
6960                 }
6961                 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6962                                           num_clusters, 0, meta_ac);
6963                 if (ret)
6964                         mlog_errno(ret);
6965
6966                 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6967                                                   num_clusters, reflink_cpos);
6968
6969                 len -= num_clusters;
6970                 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6971                 num_buckets -= reflink_buckets;
6972         }
6973 out:
6974         return ret;
6975 }
6976
6977 /*
6978  * Create the same xattr extent record in the new inode's xattr tree.
6979  */
6980 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6981                                    struct buffer_head *root_bh,
6982                                    u64 blkno,
6983                                    u32 cpos,
6984                                    u32 len,
6985                                    void *para)
6986 {
6987         int ret, credits = 0;
6988         handle_t *handle;
6989         struct ocfs2_reflink_xattr_tree_args *args =
6990                         (struct ocfs2_reflink_xattr_tree_args *)para;
6991         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6992         struct ocfs2_alloc_context *meta_ac = NULL;
6993         struct ocfs2_alloc_context *data_ac = NULL;
6994         struct ocfs2_extent_tree et;
6995
6996         trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6997
6998         ocfs2_init_xattr_tree_extent_tree(&et,
6999                                           INODE_CACHE(args->reflink->new_inode),
7000                                           args->new_blk_bh);
7001
7002         ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7003                                                       len, &credits,
7004                                                       &meta_ac, &data_ac);
7005         if (ret) {
7006                 mlog_errno(ret);
7007                 goto out;
7008         }
7009
7010         handle = ocfs2_start_trans(osb, credits);
7011         if (IS_ERR(handle)) {
7012                 ret = PTR_ERR(handle);
7013                 mlog_errno(ret);
7014                 goto out;
7015         }
7016
7017         ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7018                                           meta_ac, data_ac,
7019                                           blkno, cpos, len);
7020         if (ret)
7021                 mlog_errno(ret);
7022
7023         ocfs2_commit_trans(osb, handle);
7024
7025 out:
7026         if (meta_ac)
7027                 ocfs2_free_alloc_context(meta_ac);
7028         if (data_ac)
7029                 ocfs2_free_alloc_context(data_ac);
7030         return ret;
7031 }
7032
7033 /*
7034  * Create reflinked xattr buckets.
7035  * We will add bucket one by one, and refcount all the xattrs in the bucket
7036  * if they are stored outside.
7037  */
7038 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7039                                     struct buffer_head *blk_bh,
7040                                     struct buffer_head *new_blk_bh)
7041 {
7042         int ret;
7043         struct ocfs2_reflink_xattr_tree_args para;
7044
7045         memset(&para, 0, sizeof(para));
7046         para.reflink = args;
7047         para.old_blk_bh = blk_bh;
7048         para.new_blk_bh = new_blk_bh;
7049
7050         para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7051         if (!para.old_bucket) {
7052                 mlog_errno(-ENOMEM);
7053                 return -ENOMEM;
7054         }
7055
7056         para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7057         if (!para.new_bucket) {
7058                 ret = -ENOMEM;
7059                 mlog_errno(ret);
7060                 goto out;
7061         }
7062
7063         ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7064                                               ocfs2_reflink_xattr_rec,
7065                                               &para);
7066         if (ret)
7067                 mlog_errno(ret);
7068
7069 out:
7070         ocfs2_xattr_bucket_free(para.old_bucket);
7071         ocfs2_xattr_bucket_free(para.new_bucket);
7072         return ret;
7073 }
7074
7075 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7076                                         struct buffer_head *blk_bh)
7077 {
7078         int ret, indexed = 0;
7079         struct buffer_head *new_blk_bh = NULL;
7080         struct ocfs2_xattr_block *xb =
7081                         (struct ocfs2_xattr_block *)blk_bh->b_data;
7082
7083
7084         if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7085                 indexed = 1;
7086
7087         ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7088                                              &new_blk_bh, indexed);
7089         if (ret) {
7090                 mlog_errno(ret);
7091                 goto out;
7092         }
7093
7094         if (!indexed)
7095                 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7096         else
7097                 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7098         if (ret)
7099                 mlog_errno(ret);
7100
7101 out:
7102         brelse(new_blk_bh);
7103         return ret;
7104 }
7105
7106 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7107 {
7108         int type = ocfs2_xattr_get_type(xe);
7109
7110         return type != OCFS2_XATTR_INDEX_SECURITY &&
7111                type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7112                type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7113 }
7114
7115 int ocfs2_reflink_xattrs(struct inode *old_inode,
7116                          struct buffer_head *old_bh,
7117                          struct inode *new_inode,
7118                          struct buffer_head *new_bh,
7119                          bool preserve_security)
7120 {
7121         int ret;
7122         struct ocfs2_xattr_reflink args;
7123         struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7124         struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7125         struct buffer_head *blk_bh = NULL;
7126         struct ocfs2_cached_dealloc_ctxt dealloc;
7127         struct ocfs2_refcount_tree *ref_tree;
7128         struct buffer_head *ref_root_bh = NULL;
7129
7130         ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7131                                        le64_to_cpu(di->i_refcount_loc),
7132                                        1, &ref_tree, &ref_root_bh);
7133         if (ret) {
7134                 mlog_errno(ret);
7135                 goto out;
7136         }
7137
7138         ocfs2_init_dealloc_ctxt(&dealloc);
7139
7140         args.old_inode = old_inode;
7141         args.new_inode = new_inode;
7142         args.old_bh = old_bh;
7143         args.new_bh = new_bh;
7144         args.ref_ci = &ref_tree->rf_ci;
7145         args.ref_root_bh = ref_root_bh;
7146         args.dealloc = &dealloc;
7147         if (preserve_security)
7148                 args.xattr_reflinked = NULL;
7149         else
7150                 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7151
7152         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7153                 ret = ocfs2_reflink_xattr_inline(&args);
7154                 if (ret) {
7155                         mlog_errno(ret);
7156                         goto out_unlock;
7157                 }
7158         }
7159
7160         if (!di->i_xattr_loc)
7161                 goto out_unlock;
7162
7163         ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7164                                      &blk_bh);
7165         if (ret < 0) {
7166                 mlog_errno(ret);
7167                 goto out_unlock;
7168         }
7169
7170         ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7171         if (ret)
7172                 mlog_errno(ret);
7173
7174         brelse(blk_bh);
7175
7176 out_unlock:
7177         ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7178                                    ref_tree, 1);
7179         brelse(ref_root_bh);
7180
7181         if (ocfs2_dealloc_has_cluster(&dealloc)) {
7182                 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7183                 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7184         }
7185
7186 out:
7187         return ret;
7188 }
7189
7190 /*
7191  * Initialize security and acl for a already created inode.
7192  * Used for reflink a non-preserve-security file.
7193  *
7194  * It uses common api like ocfs2_xattr_set, so the caller
7195  * must not hold any lock expect i_mutex.
7196  */
7197 int ocfs2_init_security_and_acl(struct inode *dir,
7198                                 struct inode *inode,
7199                                 const struct qstr *qstr)
7200 {
7201         int ret = 0;
7202         struct buffer_head *dir_bh = NULL;
7203
7204         ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7205         if (ret) {
7206                 mlog_errno(ret);
7207                 goto leave;
7208         }
7209
7210         ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7211         if (ret) {
7212                 mlog_errno(ret);
7213                 goto leave;
7214         }
7215         ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7216         if (ret)
7217                 mlog_errno(ret);
7218
7219         ocfs2_inode_unlock(dir, 0);
7220         brelse(dir_bh);
7221 leave:
7222         return ret;
7223 }
7224 /*
7225  * 'security' attributes support
7226  */
7227 static size_t ocfs2_xattr_security_list(const struct xattr_handler *handler,
7228                                         struct dentry *dentry, char *list,
7229                                         size_t list_size, const char *name,
7230                                         size_t name_len)
7231 {
7232         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7233         const size_t total_len = prefix_len + name_len + 1;
7234
7235         if (list && total_len <= list_size) {
7236                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7237                 memcpy(list + prefix_len, name, name_len);
7238                 list[prefix_len + name_len] = '\0';
7239         }
7240         return total_len;
7241 }
7242
7243 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7244                                     struct dentry *dentry, const char *name,
7245                                     void *buffer, size_t size)
7246 {
7247         if (strcmp(name, "") == 0)
7248                 return -EINVAL;
7249         return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_SECURITY,
7250                                name, buffer, size);
7251 }
7252
7253 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7254                                     struct dentry *dentry, const char *name,
7255                                     const void *value, size_t size, int flags)
7256 {
7257         if (strcmp(name, "") == 0)
7258                 return -EINVAL;
7259
7260         return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_SECURITY,
7261                                name, value, size, flags);
7262 }
7263
7264 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7265                      void *fs_info)
7266 {
7267         const struct xattr *xattr;
7268         int err = 0;
7269
7270         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7271                 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7272                                       xattr->name, xattr->value,
7273                                       xattr->value_len, XATTR_CREATE);
7274                 if (err)
7275                         break;
7276         }
7277         return err;
7278 }
7279
7280 int ocfs2_init_security_get(struct inode *inode,
7281                             struct inode *dir,
7282                             const struct qstr *qstr,
7283                             struct ocfs2_security_xattr_info *si)
7284 {
7285         /* check whether ocfs2 support feature xattr */
7286         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7287                 return -EOPNOTSUPP;
7288         if (si)
7289                 return security_old_inode_init_security(inode, dir, qstr,
7290                                                         &si->name, &si->value,
7291                                                         &si->value_len);
7292
7293         return security_inode_init_security(inode, dir, qstr,
7294                                             &ocfs2_initxattrs, NULL);
7295 }
7296
7297 int ocfs2_init_security_set(handle_t *handle,
7298                             struct inode *inode,
7299                             struct buffer_head *di_bh,
7300                             struct ocfs2_security_xattr_info *si,
7301                             struct ocfs2_alloc_context *xattr_ac,
7302                             struct ocfs2_alloc_context *data_ac)
7303 {
7304         return ocfs2_xattr_set_handle(handle, inode, di_bh,
7305                                      OCFS2_XATTR_INDEX_SECURITY,
7306                                      si->name, si->value, si->value_len, 0,
7307                                      xattr_ac, data_ac);
7308 }
7309
7310 const struct xattr_handler ocfs2_xattr_security_handler = {
7311         .prefix = XATTR_SECURITY_PREFIX,
7312         .list   = ocfs2_xattr_security_list,
7313         .get    = ocfs2_xattr_security_get,
7314         .set    = ocfs2_xattr_security_set,
7315 };
7316
7317 /*
7318  * 'trusted' attributes support
7319  */
7320 static size_t ocfs2_xattr_trusted_list(const struct xattr_handler *handler,
7321                                        struct dentry *dentry, char *list,
7322                                        size_t list_size, const char *name,
7323                                        size_t name_len)
7324 {
7325         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7326         const size_t total_len = prefix_len + name_len + 1;
7327
7328         if (!capable(CAP_SYS_ADMIN))
7329                 return 0;
7330
7331         if (list && total_len <= list_size) {
7332                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7333                 memcpy(list + prefix_len, name, name_len);
7334                 list[prefix_len + name_len] = '\0';
7335         }
7336         return total_len;
7337 }
7338
7339 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7340                                    struct dentry *dentry, const char *name,
7341                                    void *buffer, size_t size)
7342 {
7343         if (strcmp(name, "") == 0)
7344                 return -EINVAL;
7345         return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_TRUSTED,
7346                                name, buffer, size);
7347 }
7348
7349 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7350                                    struct dentry *dentry, const char *name,
7351                                    const void *value, size_t size, int flags)
7352 {
7353         if (strcmp(name, "") == 0)
7354                 return -EINVAL;
7355
7356         return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_TRUSTED,
7357                                name, value, size, flags);
7358 }
7359
7360 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7361         .prefix = XATTR_TRUSTED_PREFIX,
7362         .list   = ocfs2_xattr_trusted_list,
7363         .get    = ocfs2_xattr_trusted_get,
7364         .set    = ocfs2_xattr_trusted_set,
7365 };
7366
7367 /*
7368  * 'user' attributes support
7369  */
7370 static size_t ocfs2_xattr_user_list(const struct xattr_handler *handler,
7371                                     struct dentry *dentry, char *list,
7372                                     size_t list_size, const char *name,
7373                                     size_t name_len)
7374 {
7375         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7376         const size_t total_len = prefix_len + name_len + 1;
7377         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7378
7379         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7380                 return 0;
7381
7382         if (list && total_len <= list_size) {
7383                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7384                 memcpy(list + prefix_len, name, name_len);
7385                 list[prefix_len + name_len] = '\0';
7386         }
7387         return total_len;
7388 }
7389
7390 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7391                                 struct dentry *dentry, const char *name,
7392                                 void *buffer, size_t size)
7393 {
7394         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7395
7396         if (strcmp(name, "") == 0)
7397                 return -EINVAL;
7398         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7399                 return -EOPNOTSUPP;
7400         return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_USER, name,
7401                                buffer, size);
7402 }
7403
7404 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7405                                 struct dentry *dentry, const char *name,
7406                                 const void *value, size_t size, int flags)
7407 {
7408         struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7409
7410         if (strcmp(name, "") == 0)
7411                 return -EINVAL;
7412         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7413                 return -EOPNOTSUPP;
7414
7415         return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_USER,
7416                                name, value, size, flags);
7417 }
7418
7419 const struct xattr_handler ocfs2_xattr_user_handler = {
7420         .prefix = XATTR_USER_PREFIX,
7421         .list   = ocfs2_xattr_user_list,
7422         .get    = ocfs2_xattr_user_get,
7423         .set    = ocfs2_xattr_user_set,
7424 };